Cooperation in the Commons: Community-Based Rangeland Management in Namibia

Full text

NBER WORKING PAPER SERIES
COOPERATION IN THE COMMONS:
COMMUNITY-BASED RANGELAND MANAGEMENT IN NAMIBIA
D. Layne Coppock
Lucas Crowley
Susan L. Durham
Dylan Groves
Julian C. Jamison
Dean Karlan
Brien E. Norton
R. Douglas Ramsey
Working Paper 29469
http://www.nber.org/papers/w29469
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
November 2021
The authors thank Nate Barker, Caton Brewster, Anais Dahmani, Pierre Durand, Alexander
Fertig, Sam Hambira, Matthew Haufiku, Stephen Kulungu, Sayan Kundu, Peter Lugthart, Max
Mauerman, Jared Otuke, Linda Papagallo, Amol Singh Raswan, Elvis Siyamba, Venoo Tjiseua,
Delia Welsh, and Sandy Yuan for research assistance and project management; Leon Burger,
Holly Dentz, and Cornelis van der Waal for their support implementing the cattle, qualitative, and
rangeland data collection exercises, respectively; Helmke von Bach, Donald Green, John Huber,
Indongo Indongo, Edmore Masaire, Colin Nott, Heinrich Pielok, and James Walsh for comments;
and Johannes Beck, Algerlynn Gill, and Jack Molyneaux for feedback and support throughout the
research process. This evaluation was made possible by funding from the Millennium Challenge
Corporation. The opinions expressed herein are those of the authors and do not necessarily reflect
the views of MCC or the U.S. government. The views expressed herein are those of the authors
and do not necessarily reflect the views of the National Bureau of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.
© 2021 by D. Layne Coppock, Lucas Crowley, Susan L. Durham, Dylan Groves, Julian C.
Jamison, Dean Karlan, Brien E. Norton, and R. Douglas Ramsey. All rights reserved. Short
sections of text, not to exceed two paragraphs, may be quoted without explicit permission
provided that full credit, including © notice, is given to the source.

Cooperation in the Commons: Community-based Rangeland Management in Namibia
D. Layne Coppock, Lucas Crowley, Susan L. Durham, Dylan Groves, Julian C. Jamison, Dean
Karlan, Brien E. Norton, and R. Douglas Ramsey
NBER Working Paper No. 29469
November 2021
JEL No. O12,O13,P11,Q15,Q24
ABSTRACT
Classic theories suggest that common pool resources are subject to overexploitation. Community-
based resource management approaches may ameliorate “tragedy of the commons” effects. Using
a randomized evaluation in Namibia’s communal rangelands, we find that a comprehensive four-
year program to support community-based rangeland and cattle management led to persistent and
large improvements for eight of thirteen indices of social and behavioral outcomes. Effects on
rangeland health, cattle productivity and household economics, however, were either negative or
nil. Positive impacts on community resource management may have been offset by communities’
inability to control grazing by non-participating herds and inhibited by an unresponsive rangeland
sub-system. This juxtaposition, in which measurable improvements in community resource
management did not translate into better outcomes for households or rangeland health,
demonstrates the fragility of the causal pathway from program implementation to intended
socioeconomic and environmental outcomes. It also points to challenges for improving climate
change–adaptation strategies.
D. Layne Coppock
Utah State University
Old Main Hill
Logan, UT 84322
layne.coppock@usu.edu
Lucas Crowley
228 Park Ave S, Suite 53674
Washington D.C. 20005
lukecrowley@gmail.com
Susan L. Durham
Utah State University
Old Main Hill
Logan, UT 84322
susan.durham@usu.edu
Dylan Groves
Columbia University
533 W. 218th St.
New York, NY 10034
dwg2115@columbia.edu
Julian C. Jamison
Department of Economics
University of Exeter Business School
Exeter EX4 4PU
United Kingdom
j.jamison@exeter.ac.uk
Dean Karlan
Kellogg Global Hub
Northwestern University
2211 Campus Drive
Evanston, IL 60208
and CEPR
and also NBER
dean.karlan@gmail.com
Brien E. Norton
Utah State University
Old Main Hill
Logan, UT 84322
brien_norton@comcast.net
R. Douglas Ramsey
Utah State University
Old Main Hill
Logan, UT 84322
doug.ramsey@usu.edu

1
Title: Cooperation in the commons: Community-based rangeland 1
management in Namibia 2
D. Layne Coppock1, Lucas Crowley2, Susan L. Durham3, Dylan Groves4, Julian C. Jamison5, 3
Dean Karlan6*, Brien E. Norton7, R. Douglas Ramsey7 4
1 Department of Environment and Society, Utah State University, Logan, UT 84322-5215, USA. 5
2 Innovations for Poverty Action, Washington D.C., 20005, USA. 6
3 Ecology Center, Utah State University, Logan, UT 84322-5205, USA. 7
4 Department of Political Science, Columbia University, New York City, NY, 10027, USA. 8
5 Department of Economics, University of Exeter, Exeter EX44LZ, U.K. 9
6 Kellogg School of Management, Northwestern University, Evanston, IL 60208, USA. 10
7 Department of Wildland Resources, Utah State University, Logan, UT 84322-5230, USA. 11
*To whom correspondence should be addressed. E-mail: karlan@northwestern.edu
12
13
14
Abstract 15
Classic theories suggest that common pool resources are subject to overexploitation. 16
Community-based resource management approaches may ameliorate “tragedy of the 17
commons” effects. Using a randomized evaluation in Namibia’s communal rangelands, we 18
find that a comprehensive four-year program to support community-based rangeland and 19
cattle management led to persistent and large improvements for eight of thirteen indices of 20
social and behavioral outcomes. Effects on rangeland health, cattle productivity and 21
household economics, however, were either negative or nil. Positive impacts on community 22
resource management may have been offset by communities’ inability to control grazing by 23
non-participating herds and inhibited by an unresponsive rangeland sub-system. This 24
juxtaposition, in which measurable improvements in community resource management did 25
not translate into better outcomes for households or rangeland health, demonstrates the 26
fragility of the causal pathway from program implementation to intended socioeconomic 27
and environmental outcomes. It also points to challenges for improving climate change–28
adaptation strategies. 29
30

2
Main text 31
In his seminal 1968 essay, “The Tragedy of the Commons,” Garrett Hardin argued that 32
poorly managed common resources are subject to overexploitation1. Hardin explained the 33
tragedy of the commons using the metaphor of “a pasture open to all” in which each herd owner 34
receives individual benefits from accumulating livestock while sharing the cost of overgrazing 35
with other community members. This “natural” promotion of self-interest harms the common 36
resource and ultimately brings ruin to all herders. Today, rangeland degradation is not only a 37
textbook metaphor for the tragedy of the commons theory, but highly relevant globally: Drylands 38
occupy 41% of the Earth’s land area, support two billion people, and are experiencing rapid 39
environmental degradation exacerbated by climate change, and in many cases attributable to 40
overuse from livestock and crop agriculture2. Strategies for coping with impending climate 41
change are critical for local and global policy. 42
Hardin concluded that the tragedy of the commons can be prevented only by coercive 43
government regulation or resource privatization. However, Elinor Ostrom and other critics of 44
Hardin’s thesis have documented numerous communities that successfully developed local 45
management systems to avoid overexploitation of commonly held resources, including 46
rangelands3–11. These findings have generated considerable enthusiasm for programs undertaken 47
by governmental and non-governmental organizations that provide external support for holistic, 48
community-based management of natural resources2,12,13. 49
But observing that some communities have developed successful systems of collective 50
management does not mean that collective management instigated by outside organizations will 51
succeed, and assessing the efficacy of such external interventions poses classic evaluation 52
challenges. It is difficult to identify the impact of interventions because of external factors such 53
as weather and macroeconomic conditions, and because of unobserved community or individual 54
traits that drive both program participation and successful community management. 55
Measurement is difficult because impacts are expected across many domains of a social-56
ecological system and at different points in time14. Related evidence from recent randomized 57
evaluations suggests that community-driven programs can successfully deliver infrastructure and 58
economic returns, but have less success sustainably affecting community governance and the 59
creation of social capital15. 60
We evaluated an integrated program in Namibia’s Northern Communal Areas (NCAs) 61
that promoted improved rangeland and livestock management among cattle-owning households. 62
To overcome attribution and measurement challenges, we conducted a large-scale, randomized 63
evaluation and included multi-disciplinary measurement of behavioral, economic, livestock, and 64
rangeland outcomes up to seven years after the program was initiated. The main questions posed 65
were: (1) Can external support cause improvements in community resource management that 66
persist two years after the support ends? (2) What is the effect of external support for community 67
resource management on rangeland health, cattle productivity, and household well-being? 68
69 Study context and design 70
Namibia’s NCAs have a population of about 1.2 million people, predominantly 71
pastoralists and agro-pastoralists, who herd cattle and small ruminants using traditional methods 72
and grow crops (i.e., millet, maize) under non-irrigated conditions16 Rangeland vegetation and 73
soils have been degraded by pressure from growing populations and reduced herd mobility (see 74
Supplementary Information section 2 for details). Low-input management results in 75
uncoordinated livestock grazing and overuse of local resources. Resource management in the 76

3
NCAs is further complicated by climate change17. For example, climate change may increase the 77
prevalence of drought and bush encroachment, which are already destabilizing rangeland 78
ecosystems in the NCAs2,18. 79
The economic and ecological challenges facing the NCAs are partially traceable to three 80
features of colonial-era land administration. First, in 1897 German colonial authorities 81
established a veterinary cordon fence (VCF) separating the NCAs from southern Namibia to 82
prevent the spread of livestock disease. Restrictions on movement and sale of livestock from 83
northern to southern Namibia remain in place today, severely limiting the development of the 84
formal livestock sector in the NCAs. Second, between 1897 and 1962, German and South 85
African colonial authorities expropriated land from hundreds of thousands of black Namibians 86
and relocated them to marginal communal lands known as “native reserves” on both sides of the 87
VCF 19,20. The native reserve policy restricted private land and capital accumulation by black 88
Namibians and eroded customary land governance institutions in communal areas19,21. Finally, in 89
1962 the South African government, which took over the administration of Namibia from 90
Germany following WWI, funded widespread borehole development in the NCAs to address 91
growing political unrest. This dramatic expansion of water infrastructure, which was carried out 92
with minimal concern for ecological consequences or investment in local resource governance, 93
severed the link between grazing movements and the availability of natural water sources and 94
catalyzed the growth of human and livestock populations, laying the groundwork for many of the 95
ecological challenges that northern Namibia faces today16,22. 96
The Community Based Rangeland and Livestock Management program (CBRLM) was 97
part of a four-year partnership between the Millennium Challenge Account-Namibia and the 98
Government of Namibia to reduce rangeland degradation and promote economic development. 99
From 2010 to 2014 the implementing partner, Gesellschaft für Organisation, Planung und 100
Ausbildung (GOPA), worked with communities to jointly develop locally tailored rangeland 101
grazing management, livestock management, and livestock marketing plans. GOPA then offered 102
multi-faceted support to communities that established committees to coordinate and monitor 103
these resource management plans. GOPA’s support included water-infrastructure development, 104
trainings on animal husbandry, livestock marketing, and rangeland management, livestock loans, 105
matching grants, and technical assistance from trained field facilitators. 106
The rangeland management approach underlying CBRLM centered on combined herding 107
and planned grazing. The program encouraged participating community members to combine 108
household cattle herds into larger herds and rotate them among pre-planned sites within the 109
grazing area. Planned rotation allows for vegetation rest and recovery and the establishment of 110
dry-season fodder reserves, while combined herding improves grazing coordination and reduces 111
the costs of herding. CBRLM field facilitators also encouraged enhanced livestock sales and 112
flexible stocking rates to optimize grazing pressure. According to CBRLM’s theory of change, 113
improved management practices and enhanced cattle sales would improve communities’ 114
economic well-being while reducing the risk of rangeland degradation (see Methods). 115
116

4
Fig. 1. Distribution of Rangeland Intervention Areas (RIAs) and Grazing Areas (GAs) for CBRLM in 118
northern Namibia. 119
120
In order to select study areas, GOPA mapped 38 Rangeland Intervention Areas (RIAs), 121
intervention zones with locally recognized boundaries and sufficiently low density of people, 122
livestock, and bush cover to enable the implementation of new group-grazing plans Each RIA 123
comprised 5-15 Grazing Areas (GAs), communal rangeland parcels shared by 5-35 households. 124
We randomly assigned 19 RIAs to treatment and 19 RIAs to control, and measured program 125
outcomes in 123 selected GAs (52 treatment and 71 control, see Methods). Figure 1 displays the 126
GAs in treatment and control RIAs; darker shades identify the GAs sampled for measurement. 127
Inference was computed using clustered standard errors and randomization inference, due to the 128
38-unit clustered design. 129
To measure resource management behaviors, we conducted 1,241 and 1,348 surveys of 130
cattle herd managers at program end and two years later, respectively. We confirmed key 131
practices with direct observation audits conducted after each survey. To assess impacts on 132
rangeland condition two years after program end, we collected vegetation and soil data via 133
randomly-sampled 1-ha sites during the wet (Apr-May) and dry (Sep-Oct) seasons. To assess 134
impacts on cattle health and productivity two years after program end, we weighed, aged, and 135
assessed body condition scores of 20,000 cattle in 730 herds during the dry season. Finally, to 136
assess impacts on household economic outcomes three years after program end, we conducted 137
1,345 household surveys. We used ordinary least squares regression with standard errors 138
clustered at the RIA level to estimate treatment effects. 139
140
Treatment effects on social and behavioral outcomes 141
Figure 2 illustrates impacts of CBRLM on standardized indices of social and behavioral 142
outcomes (see Methods for details of the composition and construction of indices). At program 143
end, we find large, statistically significant effects on eight of thirteen social indices: grazing 144
planning (+1.31sd, p < 0.001), grazing-plan adherence (+0.35sd, p < 0.001), herding practices 145
(+0.37sd, p = 0.003), herder management (+0.15sd, p = 0.07), cattle husbandry (+0.36sd, p = 146
0.002), community governance (+0.75sd, p <0.001), collective action (+1.53sd, p < 0.001), and 147
expertise (+0.30sd, p = 0.005). We do not observe statistically significant improvements in herd 148
restructuring (+0.00sd, p = 0.95), cattle marketing (-0.06sd, p = 0.37), community disputes 149

5
(+0.07sd, p = 0.34), trust (-0.02sd, p = 0.73), or perceptions of self and community efficacy 150
(+0.04sd, p = 0.67) (also see Extended Data Table 1). 151
To illustrate program influences on collective action we highlight two key outcomes: At 152
program end, planned grazing with peers increased by 28 percentage points (control mean = 153
22%, p < 0.001) while combining cattle with those of peers increased by 34 percentage points 154
(control mean = 38%, p < 0.001) (Extended Data Table 4). Patterns were validated via direct 155
observation audits (Extended Data Table 10). 156
Two years after program end, improvements in all four indices of rangeland grazing 157
management persisted: grazing planning (1.02sd, p < 0.001), grazing-plan adherence (0.32sd, p < 158
0.001), herding practices (0.30sd, p = 0.001), and herder management (0.43sd, p = 0.004)), as did 159
positive effects on community governance (0.55sd, p < 0.001), collective action (0.89sd, p < 160
0.001), and expertise (0.35sd, p < 0.001). Improvements in cattle husbandry were smaller and no 161
longer statistically significant (0.13sd, p = 0.19). Community disputes increased due to 162
disagreements both within and between grazing communities over access to program-generated 163
resources such as water developments and forage reserves (-0.29sd, p = 0.002) (Extended Data 164
Tables 1 and 4). 165
166
167
169
Fig. 2. Effects of CBRLM on 13 indices of social and behavioral outcomes at 0 – 1 years after program end (2014) 170
and 2 – 3 years after program end (2016). For each index the mid-point is the standardized treatment effect size, 171
with a corresponding 95% confidence interval. Supporting statistical results are shown in Extended Data Table 1. 172
173
174

6
Treatment effects on rangeland health, cattle productivity, and household economics 175
Figure 3 illustrates results concerning our second research question, namely whether 176
changes in resource management translated to improved rangeland health, cattle productivity, 177
and household economics. No statistically significant effects were observed for herd productivity 178
two years after program end or for household outcomes three years after program end. Of 10 179
rangeland outcomes measured two years after program end, four showed statistically significant 180
but negative effects. We observed these adverse effects on key rangeland outcomes during the 181
wet season, including 4 percentage points lower protected soil surface (control mean = 81% 182
protected, p = 0.05), 3 percentage points lower plant litter cover (control mean = 55%, p = 0.04), 183
8 percentage points lower herbaceous canopy cover (control mean = 45%, p = 0.07), and a 184
121kg/ha decrease in fresh plant biomass (control mean = 459kg/ha, p = 0.10). These are 185
indicators of declining ecosystem health. We also observed a 5 percentage-point reduction in 186
herbaceous canopy cover (control mean = 22%, p = 0.002) and a 6kg/ha reduction in fresh plant 187
biomass during the dry season (control mean = 233kg/ha p = 0.004), illustrating that the CBRLM 188
failed to enhance fodder reserves for risk management purposes (see Extended Data Table 6). 189
190
192

7
Fig. 3. Effect of CBRLM on 20 cattle, economic, and rangeland outcomes at 2 - 3 years or 3 years after program 193
end (2016, 2017). For each outcome, the mid-point is the standardized treatment effect size with a corresponding 194
95% confidence interval. Supporting statistical results are shown in Extended Data Table 2. 195
196
Discussion 197
We find that an external intervention to support community-based resource management 198
generated substantial and persistent improvements in rangeland grazing management, 199
community governance, and collective action. However, effects on rangeland, livestock, and 200
household attributes were mostly nil, and in some cases negative. 201
The null to negative effects on rangeland condition are most likely the result of CBRLM 202
increasing, rather than reducing, grazing intensity. For example, relative to control sites, sites in 203
treatment areas were 12 percentage points more likely to be heavily grazed in the wet season 204
(control mean = 13%, p = 0.003) and 10 percentage points more likely to be heavily grazed in 205
the dry season (control mean = 46%, p = 0.02) of 2016 (see Extended Data Table 9). While we 206
find no evidence that CBRLM increased the number of cattle herds or the number of cattle per 207
herd in treatment areas, we did observe that non-CBRLM-participating herd owners from inside 208
and outside treated areas exploited the treated GAs. Relative to herd owners in control areas, 209
herd owners in treatment GAs were seven percentage points more likely to report observing 210
“uninvited herds” in their GA in the previous year (control mean = 16%, p = 0.005). We 211
speculate that the incentives for outsiders to “poach” forage in treated areas were strong in the 212
dry season because of CBRLM investments in water infrastructure and encouragement of 213
CBRLM herd owners to set aside un-grazed forage reserves. Thus, one consideration for future 214
implementation and research is completeness of coverage: had implementation been able to 215
cover all areas, then this would have reduced the risk of such incursions. These effects were 216
compounded by the program’s failure to stimulate opportunistic livestock off-take through 217
livestock marketing. 218
Null effects on rangeland outcomes may also have resulted from an unresponsive 219
rangeland sub-system. In this sense, our findings mirror the outcomes from other integrated, 220
grazing management programs for commercial ranching in developed nations. Namely, 221
ecologically based processes exhibit significant temporal inertia relative to management and 222
social outcomes23–25. Temporal lags between primary and secondary productivity can be 223
exacerbated by the precipitation variability that characterizes northern Namibia26. Even if the 224
CBRLM grazing management schemes had been perfectly implemented with reduced stocking 225
rates, adequate protection from grass poachers and favorable rainfall regimes, rangeland 226
responsiveness to the treatment may have been limited by the nonequilibrium characteristics of 227
forage—dominated by annual grasses—and pervasive soil degradation (see Methods). 228
Nonetheless, further tracking of outcomes may be fruitful, and it is possible that positive 229
economic or ecological outcomes will manifest over longer periods of time. While we do not 230
observe early indicators of positive ecological or economic change, we also do not have a strong 231
prediction based on outside literature as to whether impacts will improve, worsen or remain the 232
same. We also recognize that improvements in social outcomes such as governance or collective 233
action may offer intrinsic benefits to communities. 234
Hardin proposed that effective management of the commons under population pressure 235
requires either coercive regulation or resource privatization1 (neither of which is politically 236

8
realistic in many contexts in low-income countries). Inspired by Ostrom’s theories of community 237
resource management, CBRLM took a third path by investing in local institutions to arrest 238
environmental degradation. 239
Our findings should temper overly optimistic views of what external interventions to 240
promote community-based resource management can achieve in dryland situations to cope with 241
climate change. Although it is important to note, as with any evaluation, our findings are 242
particular to the specific program studied. Should our results temper enthusiasm for the theory of 243
change, or are the results that did not match the aspirations more a consequence of specific 244
programmatic decisions or imperfect implementation? The program studied took a holistic 245
approach to CBRLM, whereas the broad concept of community-based resource management 246
clearly could encompass a different set of components. For instance, water infrastructure 247
development as implemented may have increased participation rates and provided direct benefits 248
to the communities but at the cost of increased incursions by outside herds. On implementation, 249
the process data do reveal high levels of participation and strong, positive feedback indicators, 250
suggesting strong implementation fidelity (although a question remains whether the theory of 251
change requires an even higher participation rate than achieved). 252
When designing future programs to support improved community-based responses to 253
climate change and ecological degradation, policymakers should integrate complementary 254
strengths, resources, and wisdom from local (e.g., traditional), regional and national authorities 255
to address commons management challenges27,28. One focal area should be how to better design 256
and enforce property rights for land, water, and grazing resources. The design of these rights 257
should reflect the varied levels (e.g., household versus community) at which different resources 258
are managed and utilized and incorporate historical perspectives about how social, economic, 259
and ecological sub-systems have evolved and interacted over time10,11,16,29–31. Innovative 260
livestock marketing programs could be considered to better address structural constraints and 261
incorporate cultural perspectives of producers. Finally, policymakers could explore well-tested 262
alternative livelihood programs to achieve development goals in light of the long-time horizon 263
and uncertain effects of programs to support new community-management systems32–34. 264
In addition to its theoretical and practical implications, this research demonstrates the 265
value of providing experimental evidence on impacts of community-based development 266
programs in a policy-relevant setting. Many experimental studies of resource management are 267
conducted using tightly managed plots under direct researcher control, limiting their relevance 268
for answering real-world policy questions25. On the other hand, field studies of community-based 269
resource management programs typically rely on non-experimental evidence that may be biased 270
due to self-selected participation or unobserved social, ecological, or economic factors. Given 271
the importance of resource management, particularly with increasing issues from climate change, 272
further research is needed to identify the contexts, approaches, and program components that 273
yield strong and inclusive impacts12. 274
275
276
277
278
279

9
Methods 280
281
Intervention design 282
283
Theory of change 284
At the heart of the of CBRLM’s theory of change is the assumption that improvements 285
in the ecological sub-system provide a sustainable resource base for increased livestock 286
production and marketing35. The ecological sub-system, however, depends on a functioning 287
economic sub-system because herd owners must be able to destock quickly in response to 288
adverse ecological circumstances. The theory holds that the most important constraint on the 289
economic sub-system is unproductive herds and low-quality cattle because farmers are unwilling 290
to sell their cattle when they command low market prices. Therefore, improvements in rangeland 291
grazing management need to be complemented by improvements in information and access to 292
livestock markets, herd structures, and animal husbandry practices. 293
Crucially, changes to the ecological, economic, and livestock sub-systems rely on 294
effective community governance and collective-action capacity in CBRLM communities. This is 295
because rangeland grazing management practices can be easily undermined by non-participating 296
herd owners inside or outside the GA. The theory therefore calls for investments at multiple 297
levels of the social-ecological system to ensure that improvements in certain program areas are 298
not undermined by failures in others35. The CBRLM implementers believed that previous 299
rangeland development programs were undermined by a failure to account for the linkages 300
among sub-systems, which motivated them to design a more holistic intervention35. 301
302
Intervention components 303
CBRLM was a multi-faceted package of administrative, educational, financial, and 304
technical support. Implementation of the package was designed as an experimental treatment to 305
assist in project assessment. To select study areas for evaluation, GOPA identified 38 RIAs with 306
sufficiently low density of people, livestock, and bush cover to enable the implementation of new 307
group-grazing plans, one of the core treatment components. The evaluation team randomly 308
assigned 19 RIAs to treatment and 19 RIAs to control (see Randomization for details). GOPA 309
implemented CBRLM in up to seven GAs within each treatment RIA. 310 Mobilization. GOPA conducted pre-mobilization meetings with TAs and other 311
stakeholders in the second half of 2010 to identify GA communities most likely to participate in 312
CBRLM35. Early mobilization efforts focused on soliciting community buy-in for the 313
cornerstone principles of CBRLM, including community-planned grazing, combined herding of 314
cattle, and efficient livestock management. There is also substantial evidence from qualitative 315
surveys that some community members were motivated to participate in the CBRLM by 316
prospects for water infrastructure development by GOPA35. 317
While almost 100 GAs were initially mobilized for the project, by 2014 GOPA was 318
targeting resources and support towards 58 GAs based on community receptivity and the 319
discretion of CBRLM management. In each GA, GOPA worked principally with households 320
owning 10 or more cattle, although other community members benefitted from participation in a 321
“Small Stock Pass-on Scheme” and a variety of training activities, which are described below. 322 Rangeland grazing management. The core aim of CBRLM was to shift how 323
communities approached livestock grazing, forage conservation, and risk management by 324
encouraging two key practices: planned grazing and combined herding. Planned grazing entails 325

10
rotating a community’s cattle to a new pasture on a regular basis in accordance with a written 326
plan. The goal was to preserve grass for the dry season and allow grazed pastures more time to 327
recover. Combined herding entails grouping many owners’ cattle into one large herd and herding 328
them in a tight bunch. This practice is meant to concentrate animal impact on rangeland, 329
minimize cattle losses, and increase the likelihood that cows are exposed to bulls, thus increasing 330
the pregnancy and calving rates of the entire herd. The scientific and practical rationale behind 331
these practices is reviewed in Supplementary Information section 2. 332
GOPA staff developed grazing plans with each participating community and taught them 333
planned grazing and combined herding via field-based training sessions. These followed a 334
“training of trainers” approach in which GOPA recruited field facilitators from each community, 335
taught them the principles of CBRLM, and tasked them with training their fellow participating 336
pastoralists. 337 Livestock management. GOPA taught participants some best practices in animal 338
husbandry, including structuring herds to maximize productivity (by increasing the proportion of 339
bulls and reducing the proportion of oxen and cattle over the age of 10 years), providing 340
vaccinations and supplements, and deworming35. Additionally, to support the introduction of 341
more bulls into herds, the project implemented a “bull scheme” in which participating 342
communities were given the opportunity to collectively buy certified breeding bulls at a 343
subsidized price. Communities were meant to repay the cost of the bulls either with cash or in-344
kind trades of goats. Goats collected in this repayment process fed into the small stock pass-on 345
scheme under which participating community members nominated households to receive goats 346
from GOPA. GOPA requested that communities nominate households that owned few or no 347
livestock and were led by youth and/or women. When GOPA received goats as payment for 348
loaned bulls, they would pass them on to nominated households. The recipients were then 349
expected to pass on the offspring of the goats they received to other disadvantaged households. 350 Cattle marketing. CBRLM also sought to increase participants’ marketing of cattle to 351
generate revenue from livestock raising and encourage offtake of unproductive animals35. 352
Community facilitators and project experts provided participating herd owners with information 353
about market opportunities and ideal herd composition, and encouraged flexible offtake in 354
response to forage shortages. In 2013, GOPA invested in the development of regional livestock 355
cooperatives that held local auctions and helped farmers transport their animals to markets. 356
Finally, GOPA invested in identifying international export opportunities for CBRLM farmers to 357
Zimbabwe and Angola, although these were generally not successful32. 358 Community development. The project sought to institutionalize community-level 359
governance to organize and enforce collective activities like planned grazing, water point 360
maintenance, and financing of livestock inputs. The central management unit of each GA was a 361
new Grazing Area Committee consisting of five to 10 elected community members. The project 362
encouraged participating communities to collectively cover operational expenses in their GA 363
through a GA fund managed by the committee. Among these expenses were the payments to 364
herders, costs of diesel for water pumps and maintenance of water infrastructure, financing 365
collective livestock vaccination campaigns, and any other collective expenses that would support 366
operation of the GA. CBRLM supported every GA fund with a 1:1 matched subsidy. The 367
matched subsidy was limited by a ceiling amount determined by the estimated number of cattle 368
in a GA. GOPA also instructed committees to maintain “GA record books” to track grazing 369
plans, record meeting minutes, and keep logs of community members’ participation and financial 370
contributions. 371

11
Water infrastructure. GOPA upgraded water infrastructure at a total of 84 sites 372
throughout the NCAs to facilitate planned grazing and combined herding. Water infrastructure 373
improvement included minor upgrades like water tanks and drinking troughs, and larger 374
investments such as the installation of diesel and solar pump systems, the drilling and installation 375
of boreholes, and the construction of pipelines, deep wells, and a large earthen dam32. 376
377
Intervention timeline 378
The timeline for major components of the research process and CBRLM roll-out is 379
illustrated in Supplementary Figure 1. The research team conducted the random assignments and 380
the implementation team began community mobilization in early 2010. Formal enrollment in 381
CBRLM began in early 2011. The program implementer conducted mobilization in two waves: 382
they mobilized 11 of 19 RIAs in 2010 and the remaining 8 RIAs in 2011. The evaluation team 383
conducted qualitative data collection to inform the design of social and cattle surveys prior to 384
project end 2014; social surveys in 2014 and 2016; rangeland surveys in the wet and dry seasons 385
of 2016; a cattle survey in 2016; and a household economic survey in 2017. 386
Cumulative GA-level implementation is illustrated in Supplementary Figure 2. The 387
project implementer first formally reported enrollment and field visits in April 2011. The 388
implementer achieved nearly full targeted enrollment (50 GAs) by November 11, although some 389
grazing areas were added or subtracted thereafter. Mobilization exceeded enrollment because 390
some grazing area communities chose not to participate in the program and some enrolled in the 391
program and then dropped out. The program averaged between 25 and 50 field visits per month 392
over the project period. A field visit consisted of a week-long community meeting about grazing-393
plan development and implementation, animal husbandry and budget training, and marketing 394
opportunities. 395
396 Randomization 397
The unit of randomization is the RIA, an intervention zone with a locally recognized 398
boundary. Each RIA falls under the jurisdiction of a single local governing body, known as a 399
Traditional Authority (TA). As noted above, RIAs contain five to 15 GAs where a community of 400
producers share water and forage resources. Grazing areas do not have legally defined 401
boundaries. A herd owner’s ability to move among GAs is variable. 402
GOPA mapped 41 RIAs prior to randomization. Three contiguous RIAs in the north-403
central region, composed of two treatment RIAs and one control RIA, were omitted from the 404
study post-randomization because reexamination of baseline density of bushland vegetation 405
deemed them unviable for CBRLM implementation. These are the three RIAs without sampled 406
GAs in Fig 1. The other 38 RIAs were randomly assigned to either receive the CBRLM 407
treatment (19 RIAs) or serve as controls (19 RIAs). 408
The randomization was stratified by TA to ensure that at least one RIA was assigned to 409
the treatment in each TA. The research team then re-randomized the sample units until seven 410
variables were balanced (a p-value of 0.33 or higher for an omnibus f-test of all seven variables) 411
between treatment and control: (1) Presence of forest; (2) number of households; (3) number of 412
cattle; (4) cattle density per unit area; (5) quality of water sources; (6) presence of community-413
based organizations (CBOs); and (7) overlap with complementary interventions (see 414
Supplementary Table 1). For future researchers, we recommend re-randomizing a set number of 415
times and choosing the re-randomization with the highest balance36. These variables and 416
indicator variables for TA are included as covariates in all analyses. 417

12
418
Sample selection 419
In the original sampling strategy, the project implementer was asked to predict the GAs 420
where they would implement the project if the RIA were assigned to treatment. However, there 421
was limited overlap between the GAs that the implementer predicted and the GAs where 422
CBRLM was ultimately implemented. Therefore, the evaluation team devised a revised sampling 423
strategy in 2013, which proceeded in four steps: 424
(1) Map GAs in sampled RIAs. The evaluation team traveled to all 38 RIAs and worked 425
with TAs and Namibian Agricultural Extension (AE) officers to map all the GAs in 426
each RIA. The team mapped 171 GAs in control RIAs and 213 GAs in treatment 427
RIAs. 428
(2) Collect pre-program data on GAs. The evaluation team collected information on pre-429
program characteristics of each GA from interviews with TAs and AE staff, the 430
Namibian national census37, and the Namibian Atlas38. The latter has a geo-431
referenced database on climate, ecology, and livestock for the nation. 432
(3) Predict CBRLM enrollment for treatment GAs. The researchers used these data in a 433
logistic regression to predict the probability that each GA would enroll in CBRLM 434
and would adopt the CBRLM interventions based on pre-program characteristics. For 435
example, the model found that GAs with more existing water infrastructure, strong 436
social cohesion, and adequate cell phone service were more likely to be enrolled in 437
the program. The variables used to predict CBRLM adoption were: (1) Presence of 438
water installations (yes/no); (2) carrying capacity of the land (above/below the 439
regional median); (3) community’s readiness to change (high/very high); (4) 440
community’s social cohesion (high/very high); (5) spillover effects from neighbors; 441
(6) quality of herders and herder turnover; (7) presence of members of the Himba 442
ethnic group; (8) the TA’s readiness to change; (9) cell phone coverage; and (10) 443
primary housing material (mud, clay, or brick). 444
(4) Generate sample of GAs in treatment and control RIAs. The evaluation team applied 445
the statistical model (above) to all GAs in the sample and set a cut-off point to 446
separate GAs that were likely to adopt the CBRLM program versus those that were 447
unlikely to do so. In treatment RIAs, the model predicted 52 GAs, of which 37 were 448
formally enrolled in CBRLM and 15 were not. In control RIAs, 71 GAs met or 449
exceeded the cutoff; they offer the best counter-factual estimate of which GAs would 450
have enrolled in the program had their RIA received treatment. 451 452
453
Data collection 454
The names, survey questions, and variable constructions for all outcomes included in the 455
analysis are available at the AEA RCT Registry (ID number: AEARCTR-0002723). See 456
Supplementary Information section 1 for a list of definitions of variables depicted in Figure 2 457
and Figure 3. 458 459
Social surveys 460
Social surveys were intended to assess the effect of CBRLM on community behaviors, 461
community dynamics, knowledge, and attitudes. All data were collected using electronic tablets 462
with the SurveyCTO software39. 463

13
The primary unit of analysis for household respondents is the manager of the cattle kraal 464
(holding pen). Researchers conducted surveys with kraal managers, rather than heads of 465
households, for three reasons. First, many kraals contain cattle owned by multiple households, 466
and decisions about grazing practices, cattle treatment, and participation in grazing groups are 467
generally made at the kraal level. Second, many cattle-owning households do not directly 468
oversee the day-to-day activities of their cattle (many live outside the GA), and so would be 469
unable to answer questions about key outcomes, such as livestock management behaviors and 470
community dynamics40. Finally, enrollment in CBRLM occurred at the kraal, rather than 471
household, level. 472
In 2014, the research team worked with local headmen and other community members to 473
generate a complete census of kraals in every sampled Grazing Area (GA) that contained 10 or 474
more cattle at the start of the program (an eligibility requirement for enrollment in CBRLM). The 475
research team randomly sampled up to 11 community members for participation in the 2014 476
kraal manager survey. Surveys were conducted in the manager’s local language and lasted 477
approximately 45 minutes. Alongside the 2014 survey, teams of two surveyors visited all grazing 478
areas where at least one respondent reported participating in a community grazing group or 479
community combined herd to corroborate reported behaviors through direct observation. 480
To assess the persistence of CBRLM’s effects on behaviors, community dynamics, 481
knowledge, and attitudes, the research team conducted a follow-up survey of kraal managers in 482
2016, two years after program end. The survey team randomly sampled two additional kraals in 483
each grazing area to account for the possibility of attrition. The 2016 survey lasted 484
approximately one hour on average, and included an expanded list of questions about 485
governance, social conflict, and collective action as well as new survey modules on cattle 486
marketing, cattle movement, and livestock management. In 2017, the research team randomly 487
sampled three kraals in each grazing area to conduct direct observation audits of key rangeland 488
grazing-management behaviors. 489
To assess the effects of CBRLM on economic outcomes, the research team conducted a 490
household-level survey in 2017, three years after program end. The survey instrument asked 491
detailed questions on topics that could not be answered by kraal managers, such as household 492
consumption, income, food security, and savings. To select households for this survey, during 493
the 2016 survey the research team asked kraal managers to list all households that owned cattle 494
in the manager’s kraal, then randomly selected one household from each kraal. Alongside the 495
2017 survey, the research team conducted an in-depth survey with the local headman of all 123 496
GAs in the sample. The headman survey focused on historical background about the grazing 497
area, as well as the headman’s perceptions of rangeland and livestock issues. 498
499
Cattle data 500
The cattle component was intended to assess effects of CBRLM on cattle numbers, body 501
condition, and productivity. The variables of key interest involved the average liveweight and 502
body condition, calving rates, and average market value of cattle, as well as overall herd 503
structures. 504
The data collection protocols closely followed standards from livestock assessments 505
elsewhere in Sub-Saharan Africa41. The research team randomly selected up to six kraals in each 506
GA to participate in the cattle survey. The survey team mobilized selected herds during multiple 507
community visits to ensure all herds were accounted for. Herd owners were compensated for the 508

14
costs of rounding up animals and weighed cattle received anti-parasite treatment (“dipping”)42. A 509
total of 19,875 cattle from 669 herds were weighed. 510
The data-collection process for each herd proceeded in six steps. First, surveyors worked 511
with herd managers to round up all cattle that regularly stayed in the selected cattle kraal. Once 512
cattle had been brought to the designated location for data collection, they were passed through a 513
mobile crush pen and scale. As each animal passed through the crush pen, a survey team member 514
recorded the animal type (i.e., bull, ox, cow, calf) and used a SurveyCTO randomizer to calculate 515
whether the animal was randomly selected for assessment. The random number generator was set 516
to randomly select approximately 30 cattle from each herd for weighing. If the animal was 517
selected, the survey team kept the animal on the scale and recorded its weight and body 518
condition. A semi-subjective 1-5 scale, commonly used by livestock buyers in the NCAs (see 519
Supplementary Fig. 3), was adjusted to a 0-4 scale used to determine formal market pricing. The 520
team then placed the animal in a neck clamp and estimated the animal’s age by dentition (but 521
extremely young calves were aged visually). Each animal was marked as it moved through the 522
crush pen to ensure that it was assessed only once. In addition to assessing randomly selected 523
animals, the survey team weighed and aged all bulls in the herd. The cattle survey yielded 524
average cattle weight, age, and body condition for 19,875 animals across all treatment and 525
control GAs, as well as estimates of calving rates, ratios of bulls to cows, and ratios of 526
productive to unproductive animals. 527
528
Rangeland data 529
The rangeland ecology research was intended to assess treatment effects on vegetation 530
and soil surface conditions. Full research details, including field technician training protocols, 531
are available elsewhere43. The data collection approach followed methods commonly used in 532
Africa44,45. Extended definitions of variables depicted in Fig. 3 and Extended Data Table 2 are 533
available in the Supplementary Information section 1. 534
The rationale for how the ecological variables presented in Fig. 3 translate into 535
assessments of rangeland condition or health is based on forage and soil characteristics from a 536
livestock production perspective26. The highest quality forages for cattle on rangelands are 537
perennial grasses, since annual grasses are more ephemeral in terms of nutritive value and 538
productivity. Herbaceous forbs often have the poorest forage quality for large grazers because of 539
their low fiber content and risks of containing toxic chemicals. When rangelands are degraded by 540
over-grazing, perennial grasses are reduced and replaced by annual grasses and forbs. This trend 541
reflects animal diet selectivity that favors consumption of the perennial plants. Reversing such 542
trends via management interventions can be difficult. The main option is to reduce grazing 543
pressure and hope that perennial grasses can outcompete annuals and become reestablished over 544
time. Another option is to implement a grazing rotation that allows perennial grasses to recover 545
after a grazing period. 546
Increases in annual grasses are documented to occur as one outcome of chronic 547
overgrazing in Namibia46,47. In 2016, annual grasses were 5-times more abundant than perennial 548
grasses in our study area. When over-grazing occurs, most plant material is harvested and less is 549
available for the pool of organic matter (OM) for the topsoil. Less OM (e.g., plant litter) on the 550
soil surface means that more soil is also exposed to wind and rain, accelerating erosion. The GAs 551
in our research occur on various soil types and landscapes, some of which are more susceptible 552
to erosion than others. Silty soils on slopes are vulnerable to erosion, for example, while sandy 553
soils on level sites are less vulnerable26. 554

15
On-the-ground sampling was conducted in all 123 selected GAs along an 800-km zone 555
running West to East. Elevations ranged from 750 to 1,700 masl (West) and 1,050 to 1,120 masl 556
(East). Within each sampled GA, up to 12 1-ha (square) sampling sites were initially chosen 557
using coordinates generated randomly from latitude and longitude coordinates in a satellite 558
image of the GA48. About 17% of sites were later removed from the sample based on their close 559
proximity to landscape disturbances or inaccessibility by field technicians. Overall, 972 sites 560
were analyzed in the wet season and 885 in the dry season of 2016, two years after the 561
implementation phase of CBRLM had ended. 562
The geographic center-point for a sampling site was generated using a spatially 563
constrained random distribution algorithm applied to the satellite image, and the field team 564
navigated to the center-point coordinates using GPS technology. The team took photographs and 565
recorded descriptive information including elevation, slope, aspect, other landscape features, 566
vegetation type, dominant plant species, soil type, soil erosion, and degree of grazing or 567
browsing pressure, and proximity to high impact areas such as trails, water points, and villages. 568
At the center point, the survey team then established two perpendicular transects, each 569
100 m in length and crossing at the middle. The resulting four, 50-m transect lines ran according 570
to each cardinal direction (N, S, E, W) as determined with a compass. Technicians then placed 1-571
m notched sampling sticks at randomized locations along each transect line and recorded what 572
plants or other materials (i.e., stone, wood, leaf litter, animal dung, etc.) were located under or 573
above the notches of the sampling sticks. These data points were tabulated to calculate percent 574
cover for various categories of vegetation; there were n=200 data points per site based on 40 575
stick placements and 5 notches per stick. This method enabled precise calculation of cover 576
values for herbaceous (i.e., grass, forb) and diminutive woody plants (i.e., small shrubs, 577
seedlings, saplings, etc.). Tree cover was estimated from point data collected via a small 578
adjustment in the approach43. Herbaceous species were identified in wet seasons but not in dry 579
seasons due to senescence during the latter. 580
Quadrat sampling supplemented the notched stick approach. Random placements of a 1-581
m2 quadrat frame within the sampling site allowed for 20 estimates of a soil surface condition 582
score ranging from 1 (poor) to 2 (moderate) or 3 (good)43. Poor was indicated by smooth soil 583
surfaces, absence of litter, having poor infiltration and signs of erosion such as rills, pedestals, or 584
terracettes; Good was indicated by rough soil surfaces, abundant litter, seedlings evident, and 585
lack of evidence of erosion. Herbaceous biomass was estimated in the quadrats and weighed to 586
estimate herbaceous biomass. 587
588
589
Statistics 590
591
Index creation 592
Index construction for socioeconomic variables was composed of several steps49. For 593
each response variable we first signed all component variables such that a higher sign is a 594
positive outcome, i.e., in line with CBRLM’s intended impacts. Then we standardized each 595
component by subtracting its control group mean and dividing by its control group standard 596
deviation. We computed the mean of the standardized components of the index and standardized 597
the sum once again by the control group sum’s mean and standard deviation. When the value of 598
one component in an index was missing, we computed the index average from the remaining 599
components. See Extended Data Tables 3-6 for index components. 600

16
601
Calculation of Average Treatment Effects 602
The estimate of interest is the Average Treatment Effect (ATE), or the average change in 603
an outcome generated by assignment to CBRLM. We estimated the ATE using standard 604
Ordinary Least Squares regression and control for variables used in stratification. Regressions 605
for rangeland outcome variables include a unique set of controls, including rainfall over the 606
project period, rainfall in the year of data collection, grazing area cattle density, grazing area 607
ecological zones, and a remote-sensing estimate of pre-project biomass. The core model takes 608
the form: 609

�=+1+ 610
where T represents treatment assignment and X represents pre-treatment covariates used to test 611
for balance during re-randomizations. The results capture the intention-to-treat (ITT) effect 612
rather than the effect of treatment-on-treated (TOT). ITT is more appropriate than TOT in this 613
context for two principal reasons. First, it is more relevant for policymakers – the effect of 614
policies should account for imperfect compliance. Second, “uptake” is not well-defined, and 615
certainly not a binary concept, for CBRLM since many communities and community members 616
complied partially, complied with some but not all components, and complied for some but not 617
all of the time. 618
619
Standard errors and p-values 620
We report two-tailed p-values for all analyses. For each outcome, we show the two-tailed 621
p-value from a standard Ordinary Least Squares (OLS) regression with standard errors clustered 622
at the level of the RIA, the unit of randomization50. We also calculate two-tailed p-values using 623
Randomization Inference (RI). To calculate RI p-values, we re-run the randomization procedure 624
(described above) 10,000 times and generate an Average Treatment Effect (ATE) under each 625
hypothetical randomization. The p-value is the percent of re-randomizations that generate a 626
treatment effect that is either equal to, or larger in absolute value than, the true ATE. 627 628
Multiple hypotheses correction 629
We calculate q-values to account for families of outcome indices with multiple 630
hypotheses51. The q-value represents the minimum false discovery rate at which the null 631
hypothesis would be rejected for a given test. We pre-specified five families of indices: 632 1. Behavioral outcomes (all in 2014): Grazing planning, Grazing-plan adherence, 633
Herding practices, and Herder management 634
2. Behavioral outcomes (all in 2016): Grazing planning, Grazing-plan adherence, 635
Herding practices, and Herder management 636
3. Primary material outcomes: Cattle herd value (2016), Herd productivity (2016), 637
Household income (2017), Household expenditures (2017), Household livestock 638
wealth (2017) 639
4. Secondary material outcomes: Time use (2017), Resilience (2017), Female 640
empowerment (2017), Diet (2017), and Herd structure (2016) 641
5. Mechanisms: Collective Action (2014, 2016), Community Governance (2014, 2016), 642
Community disputes (2014, 2016), Trust (2014), Self and community efficacy (2014, 643
2017), and Knowledge (2016) 644
645

17
Heterogeneous treatment effects analysis 646
We are interested in whether the effect of CBRLM was impacted by lower rainfall in 647
some grazing areas during the project period. We evaluated heterogeneous treatment effects by 648
rainfall in grazing areas using a variety of measures of rainfall, including aggregate rainfall 649
during the project period and deviation in aggregate rainfall from the ten-year mean during the 650
project period. 651
For simplicity, Extended Data Table 7 presents the results of analysis of the interaction 652
between treatment and a binary indicator of low rainfall. To construct this indicator, for each GA 653
we first compute the absolute difference between mean rainfall during the project and mean 654
rainfall during the 10 years prior (2000 – 2010). We divide the absolute difference by mean 655
rainfall during the 10 years prior to produce a relative (%) difference. We then determine the 656
median relative difference over all GAs. For each GA, we assign the value 1 to the low rainfall 657
indicator if the relative difference for the GA is less than the median relative difference over all 658
GAs; we assign 0 otherwise. The results are consistent when we use alternative rainfall 659
measures. 660 661
Spillovers analysis 662
Because CBRLM grazing areas were more likely to experience external incursions by 663
cattle herds from outside the community, we test for spillovers. Specifically, we are interested in 664
whether control grazing areas near treatment areas were affected by having a treatment grazing 665
area nearby. We conducted the spillovers analysis only on control group grazing areas. For each 666
control group grazing area, we measured the distance to the border of the nearest treatment 667
grazing area. We created a binary measure taking the value 1 if the distance between the control 668
group grazing area and nearest treatment group grazing area is below the median distance, and 0 669
otherwise. We find no evidence of spillover effects. The results are presented in Extended Data 670
Table 8. 671 672
Ethical considerations: Approval for this study was obtained from the Institutional Review 673
Boards at Yale University (1103008148), Innovations for Poverty Action (253.11March-001), 674
and Northwestern University (STU00205556-CR0001). The program was conceived, designed, 675
and implemented by the Millennium Challenge Account compact between the Millennium 676
Challenge Corporation and the Government of Namibia. The research team did not participate in 677
program design or implementation. Communities and individual farmers were informed that they 678
were free to withdraw from participation in evaluation activities at any time. The random 679
assignment of the program was appropriate given the uncertainty around the program’s effect, 680
and the Government of Namibia committed to implementing the program in control areas if the 681
evaluation showed positive results. 682
The research team took a number of steps to ensure the autonomy and well-being of 683
study participants. First, we designed the survey and data collection protocols after significant 684
qualitative field work to ensure that questions about sensitive issues (e.g., cattle wealth, cattle 685
losses, attitudes towards the Traditional Authority) were phrased appropriately and did not 686
engender adverse emotional or social consequences. Second, all survey activities were reviewed 687
and approved by the MCA compact, Regional Governors, and Traditional Authorities. Third, 688
surveys were conducted with informed consent and in private to ensure that information 689
remained private and respondents were as comfortable as possible during the survey. Finally, the 690

18
research team disseminated findings on market prices and rangeland condition to communities 691
and regional Agriculture Extension Officers. 692
We received no negative reports about the community reception of the survey from surveyors 693
during the evaluation. Two cows were injured during the cattle weighing exercise, and the owner 694
was financially compensated in line with a compensation agreement made with all farmers prior 695
to the cattle weighing exercise. 696 697
Data availability: Hypotheses and analytical methods for this research were pre-registered prior 698
to analysis through the American Economic Association’s RCT registry and are available online 699
(https://www.socialscienceregistry.org/trials/2723). Data used for this research are accessible at 700
the Millennium Challenge Corporation website 701
(https://data.mcc.gov/evaluations/index.php/catalog/138/study-description) and will be posted on 702
the Innovations for Poverty Action dataverse. In the publicly available data, some numerical 703
outliers have been censored in order to preserve the anonymity of the survey respondents. Access 704
to uncensored data is available upon request from the corresponding author, subject to approval 705
by the Millennium Challenge Corporation. 706
707
Code availability: Data analysis was conducted in R and Stata. All code needed to replicate the 708
figures and tables in this paper and the Supplementary Information is available, with 709
accompanying datasets, through the Millennium Challenge Corporation at 710
(https://data.mcc.gov/evaluations/index.php/catalog/138/study-description) and will be posted on 711
the Innovations for Poverty Action dataverse. 712
713
714
References 715 716
1. Hardin, G. The tragedy of the commons. Science 162, 1243–1248 (1968). 717
2. Climate Change and Land: an IPCC special report on climate change, desertification, land 718
degradation, sustainable land management, food security, and greenhouse gas fluxes in 719
terrestrial ecosystems. https://www.ipcc.ch/srccl/ (2019). 720
3. Ostrom, E. Governing the Commons: The Evolution of Institutions for Collective Action. 721
(Cambridge University Press, 2015). 722
4. Sneath, D. State Policy and Pasture Degradation in Inner Asia. Science 281, 1147–1148 723
(1998). 724
5. Acheson, J. M. The Lobster Gangs of Maine. (UPNE, 1988). 725

19
6. Wade, R. Village Republics: Economic Conditions for Collective Action in South India. 726
(Cambridge University Press, 1988). 727
7. Netting, R. M. Balancing on an Alp: Ecological Change and Continuity in a Swiss Mountain 728
Community. (CUP Archive, 1981). 729
8. McCay, B. J. The Question of the Commons: The Culture and Ecology of Communal 730
Resources. (University of Arizona Press, 1987). 731
9. Ostrom, E., Burger, J., Field, C. B., Norgaard, R. B. & Policansky, D. Revisiting the 732
Commons: Local Lessons, Global Challenges. 284, 6 (1999). 733
10. McCabe, J. T. Turkana pastoralism: A case against the Tragedy of the Commons. Hum Ecol 734
18, 81–103 (1990). 735
11. Moritz, M., Scholte, P., Hamilton, I. M. & Kari, S. Open Access, Open Systems: Pastoral 736
Management of Common-Pool Resources in the Chad Basin. Hum Ecol 41, 351–365 (2013). 737
12. Slough, T. et al. Adoption of community monitoring improves common pool resource 738
management across contexts. PNAS 118, (2021). 739
13. Ferraro, P. J. & Agrawal, A. Synthesizing evidence in sustainability science through 740
harmonized experiments: Community monitoring in common pool resources. PNAS 118, 741
(2021). 742
14. Schlüter, M., Hinkel, J., Bots, P. & Arlinghaus, R. Application of the SES Framework for 743
Model-based Analysis of the Dynamics of Social-Ecological Systems. Ecology and Society 744
19, (2014). 745
15. Casey, K. Radical Decentralization: Does Community-Driven Development Work? Annu. 746
Rev. Econ. 10, 139–163 (2018). 747

20
16. Bollig, M. Shaping the African Savannah: From Capitalist Frontier to Arid Eden in 748
Namibia. (Cambridge University Press, 2020). doi:10.1017/9781108764025. 749
17. Inman, E. N., Hobbs, R. J. & Tsvuura, Z. No safety net in the face of climate change: The 750
case of pastoralists in Kunene Region, Namibia. PLOS ONE 15, e0238982 (2020). 751
18. Weinzierl, T., Wehberg, J., Böhner, J. & Conrad, O. Spatial Assessment of Land 752
Degradation Risk for the Okavango River Catchment, Southern Africa. Land Degradation & 753
Development 27, 281–294 (2016). 754
19. Werner, W. A Brief History of Land Disposession in Namibia. Journal of Southern African 755
Studies 19, 135–146 (1993). 756
20. Werner, W. No one will become rich: Economy and society in the Herero reserves in 757
Namibia, 1915-1946. (P. Schlettwein Publishing, 1998). 758
21. Silvester, J. Beasts, boundaries & buildings: the survival & creation of pastoral economies in 759
Southern Namibia 1915–35. in Namibia under South African Rule: Mobility & Containment 760
1915–46 95–116 (Ohio University Press, 1998). 761
22. Werner, W. Tenure reform in Namibia’s communal areas. Journal of Namibian Studies : 762
History Politics Culture 18, 67–87 (2015). 763
23. Gosnell, H., Grimm, K. & Goldstein, B. E. A half century of Holistic Management: what 764
does the evidence reveal? Agric Hum Values 37, 849–867 (2020). 765
24. Huntsinger, L. Private correspondence with Lynn Huntsinger, University of California at 766
Berkeley. (2020). 767
25. Briske, D. D. et al. Origin, Persistence, and Resolution of the Rotational Grazing Debate: 768
Integrating Human Dimensions Into Rangeland Research. Rangeland Ecology & 769
Management 64, 325–334 (2011). 770

21
26. Holechek, J., Pieper, R. D. & Herbel, C. H. Range Management: Principles and Practices. 771
(Prentice Hall, 1989). 772
27. Gemedo-Dalle, Isselstein, J. & Maass, B. L. Indigenous ecological knowledge of Borana 773
pastoralists in southern Ethiopia and current challenges. International Journal of Sustainable 774
Development & World Ecology 13, 113–130 (2006). 775
28. Manzano, P. et al. Toward a holistic understanding of pastoralism. One Earth 4, 651–665 776
(2021). 777
29. Henrichsen, D. Claiming Space and Power in Pre-colonial Central Namibia: The Relevance 778
of Herero Praise Songs. (Basler Afrika Bibliographien, 1999). 779
30. Robinson, L. W. A Complex-Systems Approach to Pastoral Commons. Hum Ecol 37, 441–780
451 (2009). 781
31. Undargaa, S. Pastoralism and Common Pool Resources: rangeland co-management, 782
property rights and access in ... Mongolia. (ROUTLEDGE, 2018). 783
32. Coppock, D. L. et al. Community Based Rangeland and Livestock Management Evaluation 784
Report. https://data.mcc.gov/evaluations/index.php/catalog/138/study-description (2020). 785
33. Banerjee, A. et al. A multifaceted program causes lasting progress for the very poor: 786
Evidence from six countries. Science 348, 1260799 (2015). 787
34. Gugissa, D. A., Ingenbleek, P. T. M. & van Trijp, H. C. M. Market knowledge as a driver of 788
sustainable use of common-pool resources: A lab-in-the-field study among pastoralists in 789
Ethiopia. Ecological Economics 185, 107039 (2021). 790
35. Gesellschaft für Organisation, Planung und Ausbildung. Community Based Rangeland and 791
Livestock Management Inception Report. 792

22
https://www.yumpu.com/en/document/view/7305588/cbrlm-inception-report-millennium-793
challenge-account-namibia (2013). 794
36. Bruhn, M. & McKenzie, D. In pursuit of balance: Randomization in practice in development 795
field experiments. American Economic Journal: Applied Economics 1, 200–232 (2009). 796
37. Namibia Statistics Agency. Namibia 2011 Population and Housing Census [PUMS dataset]. 797
(Namibia Statistics Agency [producer and distributor], 2013). 798
38. Mendelsohn, J. M. Atlas of Namibia a portrait of the land and its people. (2002). 799
39. Hartung, C. et al. Open data kit: tools to build information services for developing regions. 800
in Proceedings of the 4th ACM/IEEE International Conference on Information and 801
Communication Technologies and Development 1–12 (Association for Computing 802
Machinery, 2010). doi:10.1145/2369220.2369236. 803
40. Groves, D. & Tjiseua, V. The mismeasurement of cattle ownership in Namibia’s Northern 804
Communal Areas. Nomadic Peoples 24, 255–271 (2020). 805
41. Machila, N., Fèvre, E. M., Maudlin, I. & Eisler, M. C. Farmer estimation of live bodyweight 806
of cattle: Implications for veterinary drug dosing in East Africa. Preventive Veterinary 807
Medicine 87, 394–403 (2008). 808
42. Moyo, B. & Masika, P. J. Tick control methods used by resource-limited farmers and the 809
effect of ticks on cattle in rural areas of the Eastern Cape Province, South Africa. Trop Anim 810
Health Prod 41, 517–523 (2009). 811
43. Norton, B. Rangeland Data Collection Protocol: For Evaluation of Ecological Parameters for 812
the CBRLM Program in Northern Namibia Conducted by IPA. (2020). 813
44. Riginos, C., Herrick, J. & van der Waal, C. Monitoring Rangeland health, A Manual for 814
Namibian Rangelands. http://www.namibiarangelands.com/downloads/ (2014). 815

23
45. Riginos, C. & Herrick, J. Monitoring Rangeland Health: A Guide for pastoralists and Other 816
Land Managers in Eastern Africa, Version II. (2010). 817
46. Klintenberg, P. & Verlinden, A. Water points and their influence on grazing resources in 818
central northern Namibia. Land Degradation & Development 19, 1–20 (2008). 819
47. Sander, H., Bollig, M. & Schulte, A. Himba paradise lost: Stability, degradation and 820
pastoralist management of the Omuhonga basin (Northwestern Namibia). Die Erde 129, 821
301–315 (1998). 822
48. Yu, C. L., Li, J., Karl, M. G. & Krueger, T. J. Obtaining a balanced area sample for the 823
Bureau of Land Management rangeland survey. JABES 25, 250–275 (2020). 824
49. Kling, J. R., Liebman, J. B. & Katz, L. F. Experimental analysis of neighborhood effects. 825
Econometrica 75, 83–119 (2007). 826
50. Abadie, A., Athey, S., Imbens, G. W. & Wooldridge, J. When should you adjust standard 827
errors for clustering? http://www.nber.org/papers/w24003 (2017) doi:10.3386/w24003. 828
51. Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: A practical and powerful 829
approach to multiple testing. Journal of the Royal Statistical Society. Series B 830
(Methodological) 57, 289–300 (1995). 831
Acknowledgements: The authors thank Nate Barker, Caton Brewster, Anais Dahmani, Pierre 832
Durand, Alexander Fertig, Sam Hambira, Matthew Haufiku, Stephen Kulungu, Sayan Kundu, 833
Peter Lugthart, Max Mauerman, Jared Otuke, Linda Papagallo, Amol Singh Raswan, Elvis 834
Siyamba, Venoo Tjiseua, Delia Welsh, and Sandy Yuan for research assistance and project 835
management; Leon Burger, Holly Dentz, and Cornelis van der Waal for their support 836
implementing the cattle, qualitative, and rangeland data collection exercises, respectively; 837
Helmke von Bach, Donald Green, John Huber, Indongo Indongo, Edmore Masaire, Colin Nott, 838
Heinrich Pielok, and James Walsh for comments; and Johannes Beck, Algerlynn Gill, and Jack 839
Molyneaux for feedback and support throughout the research process. This evaluation was made 840
possible by funding from the Millennium Challenge Corporation. The opinions expressed herein 841
are those of the authors and do not necessarily reflect the views of MCC or the U.S. government. 842
843

24
Author contributions: D.L.C: Analysis, Writing, Supervision; L.C.: Conceptualization, 844
Methodology, Supervision; S.L.D.: Analysis, Methodology, Writing; D.G.: Conceptualization, 845
Analysis, Methodology, Writing, Supervision; D.K.: Conceptualization, Analysis, Methodology, 846
Writing, Supervision; J.C.J.: Conceptualization, Methodology, Writing, Supervision; B.E.N.: 847
Methodology, Writing; R.D.R.: Analysis, Methodology, Writing. 848
849
Competing interests: None of the authors declares any competing interests. 850
851
Additional information: Supplementary Information is available for this paper. 852
Correspondence and requests for materials should be addressed to Dean Karlan 853
(karlan@northwestern.edu). 854
855
856
List of extended data tables: 857
Extended Data Table 1: Treatment effect on social indices 858
Extended Data Table 2: Treatment effect on rangeland health, cattle productivity, and 859
household economics 860
Extended Data Table 3: Treatment effect on social indices and their components (Panel A) 861
Extended Data Table 4: Treatment effect on social indices and their components (Panels B & 862
C) 863
Extended Data Table 5: Treatment effect on indices of rangeland health, cattle productivity and 864
household economics, and their components (Panel A) 865
Extended Data Table 6: Treatment effect on indices of rangeland health, cattle productivity and 866
household economics, and their components (Panel B) 867
Extended Data Table 7: Treatment effect heterogeneity by rainfall for rangeland health, cattle 868
productivity and household economics 869
Extended Data Table 8: Geographic spillover effects, for rangeland health 870
Extended Data Table 9: Mechanisms 871
Extended Data Table 10: Audits 872

Extended Data Table 1: Treatment effect on social indices
Panel A: Behaviors
Dependent variable coef. SE p-val. RI p-val. q-val. N coef. SE p-val. RI p-val. q-val. N
Grazing planning 1.31 0.24 <0.001 0.002 0.001 1,199 1.02 0.21 <0.001 0.002 0.001 1,218
Grazing plan adherence 0.35 0.09 <0.001 0.034 0.001 1,199 0.32 0.06 <0.001 0.002 0.001 1,240
Herding practices 0.37 0.12 0.003 0.013 0.001 1,199 0.30 0.08 0.001 0.023 0.002 1,243
Herder management 0.15 0.08 0.069 0.133 0.072 1,199 0.43 0.14 0.004 0.058 0.005 1,243
Cattle husbandry * 0.36 0.11 0.002 0.029 .1,199 0.13 0.09 0.190 0.354 .1,249
Herd restructuring * 0.00 0.07 0.952 0.977 .1,199 -0.02 0.03 0.604 0.777 .1,243
Cattle marketing * -0.06 0.06 0.374 0.655 .1,199 0.07 0.05 0.184 0.474 .1,245
Panel B: Community dynamics,
knowledge, and attitudes
Dependent variable coef. SE p-val. RI p-val. q-val. N coef. SE p-val. RI p-val. q-val. N
Community governance 0.75 0.14 <0.001 0.007 0.001 1,199 0.55 0.12 <0.001 0.004 0.001 1,245
Collective action 1.53 0.26 <0.001 0.002 0.001 1,199 0.89 0.23 <0.001 0.002 0.002 1,245
Community disputes 0.07 0.07 0.339 0.458 0.466 1,140 -0.29 0.09 0.002 0.108 0.004 1,243
Trust -0.02 0.07 0.729 0.786 0.803 1,198 . . . . . .
Expertise 0.30 0.10 0.005 0.044 0.009 1,199 0.35 0.09 <0.001 0.011 0.002 1,248
Self & community efficacy 0.04 0.09 0.668 0.754 0.803 1,196 0.00 0.08 0.970 0.980 0.971 1,009
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of an index of social or behavioral outcomes on treatment status. It is an intent-to-treat (ITT) estimate
relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are calculated using randomization inference. Each regression
includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block stratification and the RIA-level variables used in re-randomization to ensure
balance, which are: vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with
prior intervention areas, has a quality water source, and has a community based organization. Indices are the standardized (mean = 0 and sd = 1), unweighted average of standardized
components. See Materials and Methods for details of index construction. Variables for the "trust" index were not collected in the survey 2 - 3 years after program end. All p-values are two-
tailed. * indicates variables for which multiple hypothesis correction was not specified in the pre-analysis plan.
0 - 1 years after program end 2 - 3 years after program end
0 - 1 years after program end 2 - 3 years after program end

Dependent variable coef. SE p-val. RI p-val. q-val. N
Herd value 0.00 0.11 0.988 0.994 0.982 653
Herd productivity 0.02 0.09 0.826 0.904 0.982 1,285
Weekly household income 0.08 0.07 0.230 0.418 0.975 1,210
Weekly household expenditure 0.02 0.05 0.663 0.608 0.975 1,210
Household livestock wealth -0.06 0.05 0.207 0.502 0.975 1,210
Dependent variable coef. SE p-val. RI p-val. q-val. N
Herd structure -0.02 0.07 0.746 0.841 0.984 653
Time use 0.04 0.10 0.703 0.818 0.984 1,210
Resilience -0.02 0.07 0.786 0.885 0.984 1,210
Female empowerment -0.01 0.08 0.880 0.909 0.984 1,210
Meat and dairy consumption 0.00 0.04 0.990 0.993 0.997 1,210
Dependent variable coef. SE p-val. RI p-val. q-val. N
Erosion:
Wet season site erosion (1 = no erosion, 0 = erosion) -0.08 0.10 0.389 0.661 . 972
Ground cover:
Wet season unexposed soil surface (%, logit-transformed) -0.21 0.10 0.051 0.160 . 972
Wet season plant litter cover (%, logit-transformed) -0.18 0.08 0.035 0.201 . 972
Dry season plant litter cover (%, logit-transformed) -0.09 0.12 0.444 0.715 . 885
Herbaceous cover:
Wet season herbaceous canopy cover (%, logit-transformed) -0.26 0.14 0.072 0.270 . 972
Dry season herbaceous canopy cover (%, logit-transformed) -0.23 0.07 0.002 0.079 . 885
Wet season fresh plant biomass at site (kg/ha, log-transformed) -0.26 0.16 0.104 0.294 . 966
Dry season fresh plant biomass at site (kg/ha, log-transformed) -0.21 0.07 0.004 0.112 . 792
Relative canopy cover of perennial and annual grasses:
Wet season perennial to annual canopy ratio (log-transformed) -0.05 0.08 0.486 0.750 . 972
Relative canopy cover of grasses and forbs:
Wet season grass to forb canopy ratio (log-transformed) -0.23 0.10 0.025 0.260 . 972
Weeds:
Wet season % of shrubs that are not stinkbush (%, logit-transformed) 0.02 0.08 0.770 0.922 . 870
Wet season grass to Aristida canopy cover ratio (log-transformed) * -0.14 0.13 0.259 0.467 . 752
Woody vegetation:
Wet season shrub canopy cover (%, logit-transformed) -0.01 0.14 0.956 0.972 . 972
Dry season shrub canopy cover (%, logit-transformed) -0.09 0.15 0.569 0.734 . 885
Extended Data Table 2: Treatment effect on rangeland health, cattle productivity, and
household economics
Panel C: Rangeland outcomes (standardized)
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a physical program outcome on treatment status. It is an intent-to-treat (ITT)
estimate relative to the control group. Data in Panels A and B were collected from surveys of heads of household and cattle managers, and data in Panel C were
collected from randomly selected transects as described in the Methods. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values
are calculated using randomization inference. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was
used for block stratification and the RIA-level variables used in re-randomization to ensure balance, which are: quality of water source, an indicator for whether the
RIA has a community based organization, vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible households,andan
indicator for whether the RIA overlaps with prior intervention areas. Indices are the standardized (mean = 0 and sd = 1), unweighted average of standardized
components. Monetary variables have been scaled to weekly Namibian dollar (NAD) amounts. At the time of data collection (2017) the exchange rate was 13.3
NAD to 1 USD. Rangeland outcomes have been transformed as noted in parentheses to better meet assumptions of normality and homogeneity of variance. See
Materials and Methods and the Supplementary Materials for details of index and variable construction. Multiple hypothesis correction was not specified for
rangeland outcomes in the pre-analysis plan. All p-values are two-tailed. * Aristida is a genus of grasses that are undesirable forage plants in this context.
Panel A: Primary outcomes (indices) 2 - 3 years after program end
Panel B: Secondary outcomes (indices) 2 - 3 years after program end
2 - 3 years after program end

Extended Data Table 3: Treatment effect on social indices and their components (Panel A)
Panel A: Behavioral outcomes
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Grazing planning 1.31 0.24 <0.001 0.002 0.00 1,199 1.02 0.21 <0.001 0.002 0.00 1,218
Manager has grazing plan 0.08 0.04 0.032 0.215 0.67 1,199 0.13 0.03 <0.001 0.002 0.62 1,217
Manager can show written grazing plan 0.27 0.05 <0.001 0.001 0.01 1,182 0.20 0.05 <0.001 0.002 0.03 1,218
Manager has grazing plan for next season 0.18 0.03 <0.001 0.006 0.45 1,199 . . . . . .
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Grazing plan adherence 0.35 0.09 <0.001 0.034 0.00 1,199 0.32 0.06 <0.001 0.002 0.00 1,240
Manager followed grazing plan * 0.17 0.03 <0.001 0.017 0.40 1,199 0.09 0.03 0.002 0.024 0.25 1,218
Number of months followed plan (past year) 0.88 0.39 0.030 0.178 5.00 1,186 1.63 0.32 <0.001 0.005 4.03 1,181
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Herding practices 0.37 0.12 0.003 0.013 0.00 1,199 0.30 0.08 0.001 0.023 0.00 1,243
Someone herds manager's cattle 0.06 0.04 0.113 0.192 0.78 1,199 0.02 0.03 0.455 0.780 0.82 1,225
Herder stays with cattle throughout day * 0.11 0.03 <0.001 0.020 0.40 1,199 0.09 0.03 0.002 0.024 0.25 1,218
Cattle herded from water point in bunch 0.16 0.06 0.007 0.041 0.21 1,199 . . . . . .
Cattle herded in bunch when grazing 0.13 0.04 0.004 0.023 0.14 1,199 0.11 0.04 0.019 0.045 0.16 1,243
No cattle missing from manager's herd 0.00 0.03 0.916 0.960 0.56 1,199 . . . . . .
(-1)*Ratio of cattle lost/stolen to cattle owned -0.01 0.03 0.848 0.877 -0.14 1,187 -0.01 0.01 0.373 0.538 -0.06 1,234
Grazing plan intended to protect grass . . . . . . 0.13 0.05 0.010 0.045 0.19 819
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Herder management 0.15 0.08 0.069 0.133 0.00 1,199 0.43 0.14 0.004 0.058 0.00 1,243
Manager communicates weekly with herders 0.05 0.04 0.203 0.442 0.67 1,198 . . . . . .
Manager pays herders in cash 0.09 0.04 0.019 0.106 0.28 1,198 0.04 0.05 0.405 0.725 0.55 1,243
Total cash & in-kind payment to herders (NAD) 64.97 35.64 0.076 0.132 252.95 1,196 60.45 69.11 0.387 0.585 463.78 1,204
Total spent on gear provided to herders (NAD) . . . . . . -4.93 102.86 0.962 0.975 462.14 994
Total gear provided to herders (# of items) -0.04 0.09 0.651 0.781 1.00 1,195 . . . . . .
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Cattle husdandry 0.36 0.11 0.002 0.029 0.00 1,199 0.13 0.09 0.190 0.354 0.00 1,249
Cattle visit water point at least once per day 0.17 0.05 <0.001 0.020 0.18 1,199 . . . . . .
Any non-mandatory cattle vaccination 0.07 0.05 0.158 0.366 0.54 1,199 0.04 0.05 0.416 0.603 0.59 1,242
Cumulative number of cattle vaccinations 0.17 0.09 0.071 0.257 0.83 1,199 . . . . . .
Total spent on cattle vaccines (NAD) . . . . . . 163.86 71.88 0.028 0.146 603.19 1,220
Cattle have been dewormed 0.08 0.04 0.032 0.124 0.17 1,199 0.02 0.04 0.608 0.652 0.30 1,243
Number of cattle dietary supplements provided 0.11 0.09 0.236 0.464 0.93 1,199 0.18 0.12 0.165 0.345 1.39 1,242
Cattle checked for ticks at least monthly 0.04 0.03 0.172 0.512 0.35 1,199 -0.02 0.04 0.636 0.770 0.38 1,243
Total investment in animal treatment (NAD) . . . . . . -50.68 95.97 0.601 0.809 462.07 1,222
Fraction of cattle eartagged . . . . . . 0.04 0.03 0.172 0.276 0.84 653
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Herd restructuring 0.00 0.07 0.952 0.977 0.00 1,199 -0.02 0.03 0.604 0.777 0.00 1,243
Sold cattle to improve herd structure 0.00 0.03 0.952 0.977 0.30 1,199 0.00 0.01 0.604 0.777 0.05 1,243
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Cattle marketing -0.06 0.06 0.374 0.655 0.00 1,199 0.07 0.05 0.184 0.474 0.00 1,245
Any live cattle sold (past year) 0.00 0.03 0.978 0.990 0.58 1,199 0.04 0.02 0.067 0.226 0.36 1,243
Total number of live cattle sold (past year) -0.47 0.41 0.263 0.614 3.66 1,190 0.18 0.26 0.506 0.698 1.67 1,245
Total value of live cattle sold (NAD, past year) -2,321 1,809 0.208 0.567 11,471 1,157 1,246 1,055 0.245 0.561 7,108 1,226
0 - 1 years after program end 2 - 3 years after program end
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a behavioral program outcome, as measured in a survey of grazing area managers, on treatment
status. It is an intent-to-treat (ITT) estimate relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are calculated using
randomization inference. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block stratification and the RIA-level
variables used in re-randomization to ensure balance: vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible households, and binary indicators
for whether the RIA overlaps with prior intervention areas, has a quality water source, and has a community based organization. Each index is the standardized (mean = 0 and sd = 1),
unweighted average of the standardized components listed below it; see Materials and Methods for a complete description of index creation. Empty cells indicate that a variable was not
collected in that survey round. Monetary variables are in Namibian dollar (NAD) amounts. 0 -1 years after program end (2014), the exchange rate was 10.8 NAD to 1 USD, and 2 - 3 years
after program end was 14.7 NAD to 1 USD. Component variables without description of units are binary, with positive responses coded as 1. All p-values are two-tailed. * indicates that
the survey question used to construct the variable asked about behaviors during the past rainy season in the survey conducted 0-1 years after program end, and behaviors during the past
year in the survey conducted 2-3 years after program end.

Extended Data Table 4: Treatment effect on social indices and their components (Panel B)
Panel B: Community dynamics,
knowledge, and attitudes
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Community governance 0.75 0.14 <0.001 0.007 0.00 1,199 0.55 0.12 <0.001 0.004 0.00 1,245
GA community groups, past 5 yrs (# of groups) . . . . . . 0.36 0.06 <0.001 0.010 1.54 1,243
GA community groups currently (# of groups) . . . . . . 0.32 0.08 <0.001 0.049 1.47 1,243
Manager's cumulative membership (# of groups) 0.46 0.09 <0.001 0.026 0.70 1,199 0.30 0.08 <0.001 0.060 0.78 1,244
Group performance (# of satisfying groups) . . . . . . 0.86 0.21 <0.001 0.041 3.69 1,243
Farmers enforce water point payments . . . . . . 0.03 0.05 0.578 0.742 0.65 1,243
Farmers pay for water according to usage . . . . . . 0.02 0.06 0.759 0.821 0.19 1,239
Grazing plan formally enforced . . . . . . 0.05 0.02 0.010 0.083 0.04 1,243
Someone personally enforces grazing plan * 0.30 0.05 <0.001 0.004 0.13 1,198 0.26 0.05 <0.001 0.003 0.13 1,217
Non-community grazing not allowed . . . . . . 0.07 0.02 0.005 0.070 0.16 1,230
Conflict resolution is group-based . . . . . . 0.09 0.02 <0.001 0.041 0.60 1,243
Satisfied with group conflict resolution (1 - 3 scale) . . . . . . -0.07 0.04 0.147 0.235 2.67 1,225
Approves of traditional authority -0.01 0.03 0.681 0.845 0.25 1,175 . . . . . .
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Collective action 1.53 0.26 <0.001 0.002 0.00 1,199 0.89 0.23 <0.001 0.002 0.00 1,245
Manager pays herders communally 0.08 0.01 <0.001 0.023 0.02 1,199 0.11 0.03 <0.001 0.036 0.28 1,240
Pays for vaccines communally 0.15 0.04 <0.001 0.013 0.03 1,199 . . . . . .
Pays for cattle care communally . . . . . . 0.05 0.07 0.457 0.646 0.32 1,243
Attended water committee >4x yearly * 0.05 0.03 0.098 0.162 0.11 1,199 0.04 0.02 0.094 0.156 0.12 1,239
Contributed money to water committee 0.11 0.03 <0.001 0.025 0.19 1,199 0.04 0.04 0.320 0.503 0.25 1,243
Water committee contribution amt (NAD) . . . . . . 43.72 67.97 0.524 0.609 138.89 1,230
Attended development committee >4x yearly 0.01 0.01 0.343 0.609 0.06 1,199 0.02 0.01 0.185 0.498 0.05 1,238
Contributed money to development committee 0.04 0.01 <0.001 0.070 0.05 1,196 . . . . . .
Development committee contribution amt (NAD) . . . . . . -0.14 1.57 0.930 0.967 5.25 1,233
Practiced rainy season combined herding * 0.34 0.04 <0.001 0.004 0.38 1,188 0.19 0.07 0.008 0.033 0.36 1,217
Intentionally combined cattle with specific herd * 0.34 0.06 <0.001 0.004 0.20 1,199 . . . . . .
Ratio of GA herds to herds in combined herd * 0.23 0.05 <0.001 0.003 0.05 1,089 0.12 0.04 0.001 0.011 0.04 1,216
Ratio of manager cattle to cattle in combined herd * 0.21 0.06 <0.001 0.007 0.03 1,039 0.12 0.03 <0.001 0.009 0.03 1,186
Grazing plan is decided on by group * 0.28 0.05 <0.001 0.004 0.22 1,189 0.24 0.05 <0.001 0.006 0.26 1,218
Shared grazing plan exists for rainy season * 0.19 0.04 <0.001 0.012 0.32 1,199 . . . . . .
Ratio of farmers in group grazing plan to GA herds * 0.18 0.04 <0.001 0.020 0.13 1,171 0.16 0.05 0.002 0.018 0.15 1,218
Attended grazing committee >4x yearly 0.16 0.03 <0.001 0.009 0.03 1,199 0.10 0.02 <0.001 0.002 0.02 1,243
Contributed money to grazing committee 0.16 0.04 <0.001 0.007 0.02 1,197 0.05 0.01 <0.001 0.013 0.02 1,243
Grazing committee contribution amt (NAD) . . . . . . 11.12 4.85 0.028 0.157 4.90 1,239
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Community disputes 0.07 0.07 0.339 0.458 0.00 1,140 -0.29 0.09 0.002 0.108 0.00 1,243
Community conflicts decreased (past 3 yrs) * 0.03 0.03 0.339 0.458 0.30 1,140 . . . . . .
Conflicts w/ farmers inside GA (-1*[# conflicts]) . . . . . . -0.12 0.03 <0.001 0.082 -1.15 1,243
Conflicts w/ farmers outside GA (-1*[# conflicts]) . . . . . . -0.08 0.03 0.012 0.182 -1.08 1,243
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Trust -0.02 0.07 0.729 0.786 0.00 1,198 .. . . . .
Manager believes people can be trusted -0.05 0.04 0.249 0.414 0.49 1,188 . . . . . .
No decrease in # of people manager trusts 0.03 0.03 0.351 0.603 0.64 1,177 . . . . . .
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Expertise 0.30 0.10 0.005 0.044 0.00 1,199 0.35 0.09 <0.001 0.011 0.00 1,248
Cattle expert available for disease questions 0.18 0.05 <0.001 0.025 0.43 1,199 0.17 0.06 0.003 0.020 0.31 1,234
Cattle expert available for general questions 0.14 0.06 0.017 0.034 0.19 1,199 . . . . . .
Correctly ages cow based on dental condition . . . . . . 0.08 0.02 <0.001 0.036 0.13 1,243
Manager identifies ideal bull to cow ratio -0.03 0.03 0.331 0.405 0.20 1,198 0.02 0.02 0.386 0.596 0.85 1,243
Cattle weight guess (-1*[% error]) . . . . . . 0.27 0.10 0.010 0.142 -0.54 416
Cattle market price guess (-1*[% error]) . . . . . . -0.02 0.02 0.418 0.587 -0.33 409
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N coef. SE p-val. RI p-val. Ctrl mean N
Index: Self & community efficacy 0.04 0.09 0.668 0.754 0.00 1,196 0.00 0.08 0.970 0.980 0.00 1,009
Own actions affect cattle health & value 0.00 0.03 0.903 0.928 0.78 1,196 0.01 0.03 0.776 0.863 0.58 1,009
Own actions affect rangeland quality 0.03 0.05 0.471 0.642 0.61 1,195 -0.02 0.03 0.576 0.637 0.49 1,009
Community engagement affects cattle health . . . . . . -0.02 0.04 0.683 0.820 0.64 1,009
Community actions affect rangeland . . . . . . 0.03 0.04 0.455 0.682 0.64 1,009
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a behavioral program outcome, as measured in a survey of grazing area
managers, on treatment status. It is an intent-to-treat (ITT) estimate relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of
randomization. RI p-values are calculated using randomization inference. Each regression includes as controls a categorical variable for traditional authority (an
administrative unit) that was used for block stratification and the RIA-level variables used in re-randomization to ensure balance, which are: vegetation type, number of
livestock, livestock density, the log of the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a
quality water source, and has a community based organization. Each index is the standardized (mean = 0 and sd = 1), unweighted average of the standardized
components listed below it; see Materials and Methods for a complete description of index creation. Empty cells indicate that a variable or index was not collected in
that survey round. Monetary variables are in Namibian dollar (NAD) amounts. 0 -1 years after program end (2014), the exchange rate was 10.8 NAD to 1 USD, and 2 -
3 years after program end was 14.7 NAD to 1 USD. Component variables without description of units are binary, with positive responses coded as 1. All p-values are
two-tailed. * indicates that the survey question used to construct the variable asked about behaviors during the past rainy season in the survey conducted 0-1 years
after program end, and behaviors during the past year in the survey conducted 2-3 years after program end.
2 - 3 years after program end0 - 1 years after program end

Panel A: Primary outcomes
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Herd value 0.00 0.11 0.988 0.994 0.00 653
Total number of cattle per kraal 0.23 3.61 0.950 0.971 34.15 653
Total meat production per kraal (kg) -33 1,083 0.976 0.984 9,010 653
Total herd market value (NAD) -8,953 116,241 0.939 0.960 1,007,571 653
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Herd productivity 0.02 0.09 0.826 0.904 0.00 1,285
Calving rate among productive calves 0.00 0.03 0.940 0.961 0.74 641
Change in herd size (# of cattle, rainy season) 0.47 1.27 0.715 0.780 -8.23 1,243
Weekly milk products produced (kg, rainy season ) 4.71 6.55 0.477 0.578 26.06 1,153
Sub-index: cattle weight -0.06 0.09 0.480 0.622 0.00 653
Sub-index: cattle condition -0.31 0.21 0.145 0.463 0.00 653
Sub-index: Cattle weight -0.06 0.09 0.480 0.622 0.00 653
Average cow weight (kg) 0.13 4.96 0.978 0.987 299.60 641
Average ox weight (kg) 4.66 7.25 0.524 0.623 380.38 587
Average male calf weight (kg) 1.95 2.36 0.415 0.724 118.65 564
Average female calf weight (kg) -2.17 2.58 0.407 0.580 116.84 578
Average heifer weight (kg) -6.68 4.47 0.144 0.323 245.58 576
Average steer weight (kg) -11.15 6.04 0.073 0.271 241.01 363
Average bull weight (kg) 16.11 12.59 0.209 0.343 386.04 361
Sub-index: Cattle body condition -0.31 0.21 0.145 0.463 0.00 653
Average cow body condition (0 - 5 scale) -0.12 0.08 0.139 0.450 0.44 641
Average ox body condition (0 - 5 scale) -0.15 0.11 0.195 0.520 0.98 587
Average male calf body condition (0 - 5 scale) -0.04 0.05 0.437 0.711 0.27 564
Average female calf body condition (0 - 5 scale) -0.10 0.06 0.072 0.354 0.26 577
Average heifer body condition (0 - 5 scale) -0.19 0.11 0.090 0.385 0.65 576
Average steer body condition (0 - 5 scale) -0.28 0.11 0.013 0.232 0.69 364
Average bull body condition (0 - 5 scale) -0.09 0.15 0.539 0.705 1.03 362
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Additive index: Weekly per capita household income (NAD) 39.81 32.59 0.230 0.418 201.09 1,210
Total crop revenue (NAD, scaled from 12 months) 2.76 2.43 0.263 0.393 4.32 1,210
Total formal employment profits (NAD, scaled from 12 months) 43.53 67.14 0.521 0.738 340.82 1,210
Total value of all food produced at home (NAD, weekly) -2.80 33.72 0.934 0.970 201.48 1,210
Total value of non-sold byproducts (NAD, weekly) -0.04 0.05 0.349 0.349 0.19 1,210
Value of own cattle used for plowing (NAD, scaled from 12 months) -2.35 3.27 0.477 0.641 33.15 1,195
Total cattle sale revenue (NAD, scaled from 12 months) 6.24 27.83 0.824 0.881 79.24 1,210
Total cattle byproduct sale revenue (NAD, scaled from 12 months) 0.48 0.51 0.354 0.679 1.94 1,210
Amount of remittances received (NAD, scaled from 12 months) 4.73 2.29 0.046 0.237 15.20 1,172
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Additive index: Weekly per capita household expenditure (NAD) 28.66 65.17 0.663 0.608 402.70 1,210
Total amount borrowed (NAD, scaled from 12 months) -46.94 24.29 0.061 0.373 77.25 1,210
Total nonfood expenditure (NAD, scaled from 12 months) -40.91 74.52 0.586 0.743 306.23 1,210
Total nonfood expenditure (NAD, scaled from 30 days) 125.20 61.57 0.049 0.144 426.57 1,210
Total crop expenditure (NAD, scaled from 12 months) 0.54 0.40 0.181 0.495 3.32 1,183
Expenditure hiring animals for plowing (NAD, scaled from 12 months) 0.09 0.22 0.691 0.826 1.20 1,210
Amount sent in remittances (NAD, scaled from 12 months) 5.06 3.67 0.176 0.432 21.89 1,210
Total expenditure on water (NAD, scaled from 12 months) 0.08 0.91 0.927 0.967 6.60 1,176
Total value of food purchased (NAD) 4.67 90.06 0.959 0.970 314.33 1,210
Amount spent purchasing cattle (NAD, scaled from 12 months) 0.54 6.89 0.938 0.972 29.93 1,210
Amount spent transporting sold cattle (NAD, scaled from 12 months) 0.07 0.13 0.620 0.654 0.13 1,210
Total cattle upkeep expenditure (NAD, scaled from 12 months) 9.90 20.99 0.640 0.817 176.18 1,210
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Household livestock wealth -0.06 0.05 0.207 0.502 0.00 1,210
Total cattle wealth (livestock units) -4.40 3.13 0.168 0.391 30.62 1,176
Total non-cattle wealth (livestock units) -0.07 0.49 0.885 0.935 6.35 1,210
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a behavioral program outcome on treatment status. It is an intent-to-treat
(ITT) estimate relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are calculated using
randomization inference. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block
stratification and the RIA-level variables used in re-randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log
of the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water source, and
has a community based organization. Herd value, herd productivity, and household livestock wealth indices are the standardized (mean = 0 and sd = 1),
unweighted average of the standardized components listed below each index. Income and expenditure indices are the sum of components, adjusted for
household size. See Materials and Methods for a complete description of index creation. Monetary variables are in Namibian dollar (NAD) amounts. 0 -1 years
after program end (2014), the exchange rate was 10.8 NAD to 1 USD, and 2 - 3 years after program end was 14.7 NAD to 1 USD. Cattle body condition scores
are on a 0 - 5 scale used by Meat Corporation of Namibia, with 0 being low fat content and 5 being high. Component variables without description of units are
binary, with positive responses coded as 1. All p-values are two-tailed.
2 - 3 years after program end
Extended Data Table 5: Treatment effect on indices of rangeland health, cattle productivity and
household economics, and their components (Panel A)

Panel B: Secondary outcomes
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Herd structure -0.02 0.07 0.746 0.841 0.00 653
Ratio of bulls to cows is higher than 1:40 -0.10 0.03 0.001 0.104 0.61 646
(-1)*Ratio of oxen to total cattle 0.01 0.01 0.649 0.742 -0.15 653
(-1)*Ratio of unproductive cattle to total cattle 0.02 0.01 0.206 0.586 -0.13 653
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Time use 0.04 0.10 0.703 0.818 0.00 1,210
Days spent herding (typical week scaled to annual, adult) -8.40 10.49 0.429 0.558 81.70 1,210
Days spent working on crops (past year, adult) 2.91 2.37 0.228 0.460 0.88 1,210
Days formally employed (past year, adult) 3.62 4.57 0.433 0.586 34.74 1,210
(-1)*Days spent herding (typical week scaled to annual, child) -2.76 4.50 0.543 0.680 -15.43 970
(-1)*Days spent working on crops (past year, child) -0.27 0.30 0.381 0.594 -0.17 970
(-1)*Days formally employed (past year, child) -0.24 0.33 0.461 0.773 -0.22 970
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Resilience -0.02 0.07 0.786 0.885 0.00 1,210
FAO food security index (-3 - 0; -3 = severely insecure) -0.12 0.09 0.205 0.572 -1.62 1,207
Did not lack money for school fees (past year) 0.02 0.02 0.343 0.622 0.89 1,210
Savings available to cover emergency expense (NAD) -31.05 211.14 0.884 0.929 1,486 1,210
Savings and credit available to cover emergency expense (NAD) -341.20 216.17 0.123 0.407 2,829 1,210
Household saves money 0.04 0.05 0.390 0.636 0.70 1,165
Total household savings (NAD) -1,189 2,279 0.605 0.731 6,720 1,034
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Female empowerment -0.01 0.08 0.880 0.909 0.00 1,210
Any female HH member owns cattle -0.03 0.04 0.382 0.597 0.48 1,210
Fraction of HH cattle owned by women -0.01 0.03 0.681 0.798 0.25 1,111
Any new female goat owner in HH (past 3 years) 0.02 0.02 0.457 0.616 0.13 1,210
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Index: Meat and dairy consumption 0.00 0.04 0.990 0.993 0.00 1,210
Per capita meat consumption (kg, past week) -1.12 2.00 0.579 0.684 6.77 1,210
Per capita dairy consumption (kg, past week) 0.09 0.31 0.763 0.868 1.15 1,197
Panel C: Rangeland outcomes
Dependent variable coef. SE p-val. RI p-val. Ctrl mean Treat mean N
Erosion:
Wet season site erosion (1 = no erosion, 0 = erosion) -0.04 0.05 0.389 0.661 0.517 0.434 972
Ground cover:
Wet season protected soil surface (%, logit-transformed) -0.34 0.17 0.051 0.160 0.807 0.762 972
Wet season plant litter cover (%, logit-transformed) -0.22 0.10 0.035 0.201 0.547 0.514 972
Dry season plant litter cover (%, logit-transformed) -0.18 0.23 0.444 0.715 0.620 0.573 885
Herbaceous cover:
Wet season herbaceous canopy cover (%, logit-transformed) -0.53 0.29 0.072 0.270 0.446 0.369 972
Dry season herbaceous canopy cover (%, logit-transformed) -0.52 0.16 0.002 0.079 0.216 0.171 885
Wet season fresh plant biomass at site (kg/ha, log-transformed) -0.45 0.27 0.104 0.294 459 338 966
Dry season fresh plant biomass at site (kg/ha, log-transformed) -0.48 0.16 0.004 0.112 233 227 792
Relative canopy cover of perennial and annual grasses:
Wet season perennial to annual canopy ratio (log-transformed) -0.18 0.26 0.486 0.750 22.800 16.816 972
Relative canopy cover of grasses and forbs:
Wet season grass to forb canopy ratio (log-transformed) -0.33 0.14 0.025 0.260 43.329 33.563 972
Weeds:
Wet season % of shrubs that are not stinkbush (%, logit-transformed) 0.02 0.07 0.770 0.922 0.991 0.964 870
Wet season grass to Aristida canopy cover ratio (log-transformed) * -0.18 0.16 0.259 0.467 12.962 12.935 752
Woody vegetation:
Wet season shrub canopy cover (%, logit-transformed) -0.01 0.19 0.956 0.972 0.084 0.074 972
Dry season shrub canopy cover (%, logit-transformed) -0.13 0.23 0.569 0.734 0.108 0.089 885
2 - 3 years after program end
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a program outcome on treatment status. It is an intent-to-treat (ITT) estimate relative to
the control group. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are calculated using randomization inference. Each regression
includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block stratification and the RIA-level variables used in re-
randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible households, and binary indicators
for whether the RIA overlaps with prior intervention areas, has a quality water source, and has a community based organization. Each index is the standardized (mean = 0 and
sd = 1), unweighted average of the standardized components listed below it; see Materials and Methods for a complete description of index creation. Monetary variables are in
Namibian dollar (NAD) amounts. 0 -1 years after program end (2014), the exchange rate was 10.8 NAD to 1 USD, and 2 - 3 years after program end was 14.7 NAD to 1USD.
Component variables without description of units are binary, with positive responses coded as 1. Rangeland outcomes have been transformed (but not standardized as in
Extended Data Table 2) as noted in parentheses to better meet assumptions of normality and homogeneity of variance; treatment and control means are sample means
computed from data on untransformed scales. All p-values are two-tailed. * Aristida is a genus of grasses that are undesirable forage plants in this context.
2 - 3 years after program end
Extended Data Table 6: Treatment effect on indices of rangeland health, cattle productivity and
household economics, and their components (Panel B & C)

Panel A: Physical outcomes (2 - 3 years)
Dependent variable coef.1 SE p-val. coef.2 SE p-val. coef.3 SE p-val. RI p-val Ctrl mean N
Herd value 0.12 0.11 0.271 -0.18 0.18 0.318 -0.17 0.16 0.314 0.521 0.00 653
Herd productivity -0.12 0.09 0.212 -0.31 0.15 0.044 0.20 0.16 0.224 0.477 0.00 653
Weekly household income 58.22 38.66 0.141 40.78 52.69 0.444 -37.12 63.03 0.560 0.755 201.1 1,210
Weekly household expenditure -33.96 74.49 0.651 -23.77 113.8 0.836 118.5 127.5 0.359 0.549 402.7 1,210
Household livestock wealth -0.03 0.06 0.624 -0.03 0.16 0.841 -0.05 0.09 0.565 0.749 0.00 1,210
Herd structure -0.12 0.09 0.212 -0.31 0.15 0.044 0.20 0.16 0.224 0.477 0.00 653
Time use 0.27 0.16 0.089 0.62 0.29 0.037 -0.48 0.26 0.068 0.168 0.00 1,210
Resilience -0.17 0.09 0.076 0.00 0.13 0.969 0.28 0.12 0.028 0.177 0.00 1,210
Female empowerment 0.06 0.13 0.666 0.08 0.14 0.591 -0.14 0.14 0.347 0.521 0.00 1,210
Panel B: Rangeland outcomes (2 - years)
Dependent variable coef.1 SE p-val. coef.2 SE p-val. coef.3 SE p-val. RI p-val Ctrl mean N
Erosion:
Wet season site erosion (1 = no erosion, 0 = erosion) 0.01 0.08 0.887 0.01 0.10 0.877 -0.14 0.09 0.129 0.319 0.52 972
Ground cover:
Wet season protected soil surface (%, logit-trans.) -0.53 0.22 0.019 -0.28 0.17 0.103 0.43 0.25 0.099 0.295 0.81 972
Wet season plant litter cover (%, logit-trans.) -0.24 0.13 0.075 0.32 0.11 0.008 0.11 0.17 0.543 0.632 0.55 972
Dry season plant litter cover (%, logit-trans.) 0.00 0.42 0.994 0.02 0.31 0.950 -0.31 0.49 0.531 0.687 0.62 885
Herbaceous cover:
Wet season herbaceous canopy cover (%, logit-trans.) -1.22 0.36 0.002 -0.79 0.26 0.004 1.26 0.47 0.011 0.141 0.45 972
Dry season herbaceous canopy cover (%, logit-trans.) -0.84 0.21 <0.001 -0.84 0.22 <0.001 0.58 0.20 0.007 0.126 0.22 885
Wet season fresh plant biomass at site (kg/ha, log-trans.) -0.67 0.28 0.024 -0.47 0.29 0.113 0.41 0.32 0.209 0.455 459.37 966
Dry season fresh plant biomass at site (kg/ha, log-trans.) -0.78 0.20 <0.001 -0.67 0.11 <0.001 0.68 0.26 0.014 0.124 232.59 792
Relative canopy cover of perennial and annual grasses:
Wet season perennial to annual canopy ratio (log-trans.) 0.44 0.46 0.347 0.17 0.50 0.730 -0.87 0.64 0.184 0.294 22.80 972
Relative canopy cover of grasses and forbs:
Wet season grass to forb canopy ratio (log-trans.) -0.43 0.23 0.068 -0.09 0.32 0.783 0.21 0.33 0.530 0.640 43.33 972
Weeds:
Wet season % of shrubs that are not stinkbush (%, logit-
trans.) 0.05 0.09 0.567 0.28 0.15 0.065 -0.03 0.15 0.853 0.852 0.99 870
Wet season grass to Aristida canopy cover ratio (log-
trans.) * -0.26 0.19 0.186 -0.49 0.18 0.011 0.08 0.19 0.698 0.873 12.96 752
Woody vegetation:
Wet season shrub canopy cover (%, logit-trans.) 0.01 0.26 0.967 -0.38 0.18 0.039 -0.10 0.32 0.747 0.811 0.08 972
Dry season shrub canopy cover (%, logit-trans.) -0.09 0.33 0.794 -0.48 0.33 0.162 -0.03 0.40 0.934 0.942 0.11 885
Extended Data Table 7: Treatment effect heterogeneity by rainfall for rangeland health, cattle productivity and
household economics
Notes: Each row displays results from a separate regression in which the dependent variable is a rangeland outcome and the independent variables are treatment status and
an indicator variable for low rainfall. Coef. 1 indicates the coefficient on treatment, which is an intent-to-treat (ITT) estimate relative to control. Coef. 2 indicates the coefficient
on an indicator variable for low rainfall, which is equal to 1 if a grazing area was below the median of all grazing areas in terms of percent difference in the grazing area's rainfall
during the project period relative to the mean of the grazing area's rainfall over the 10 years prior to the program. Coef. 3 shows the interaction of the low-rainfall indicator with
treatment. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are calculated using randomization inference. Each regression includes as
controls a categorical variable for traditional authority (an administrative unit) that was used for block stratification and the RIA-level variables used in re-randomization to
ensure balance: vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps
with prior intervention areas, has a quality water source, and has a community based organization. See Materials and Methods for additional details of this analysis. All p-values
are two
-
tailed.
Treatment x low rainfall
indicator
Treatment Low rainfall
indicator Treatment x low rainfall
indicator
Treatment Low rainfall
indicator

Extended Data Table 8: Geographic spillover effects, rangeland health
Dependent variable coef. SE p-val. Distant mean Near mean N
Erosion:
Wet season site erosion (1 = no erosion, 0 = erosion) -0.03 0.06 0.627 0.47 0.56 553
Ground cover:
Wet season protected soil surface (%, logit-transformed) -0.52 0.32 0.126 0.79 0.82 553
Wet season plant litter cover (%, logit-transformed) -0.31 0.21 0.164 0.54 0.55 553
Dry season plant litter cover (%, logit-transformed) -0.24 0.42 0.582 0.60 0.63 499
Herbaceous cover:
Wet season herbaceous canopy cover (%, logit-transformed) -0.29 0.34 0.409 0.41 0.48 553
Dry season herbaceous canopy cover (%, logit-transformed) -0.32 0.43 0.475 0.17 0.25 499
Wet season fresh plant biomass (kg/ha, log-transformed) 0.12 0.22 0.589 459 463.82 550
Dry season fresh plant biomass (kg/ha, log-transformed) -0.52 0.24 0.042 265 207.94 445
Relative canopy cover of perennial and annual grasses:
Wet season perennial to annual canopy ratio (log-transformed) -0.33 0.80 0.683 27.28 19.07 553
Relative canopy cover of grasses and forbs:
Wet season grass to forb canopy ratio (log-transformed) -0.53 0.23 0.038 42.97 44.19 553
Weeds:
Wet season % of shrubs that are not stinkbush (%, logit-transformed) 0.07 0.14 0.627 0.98 1.00 498
Wet season grass to Aristida canopy cover ratio (log-transformed) * -0.19 0.20 0.364 11.06 15.00 443
Woody vegetation:
Wet season shrub canopy cover (%, logit-transformed) 0.14 0.15 0.367 0.09 0.08 553
Dry season shrub canopy cover (%, logit-transformed) -0.08 0.27 0.783 0.13 0.09 499
Notes: Each row displays results from a separate regression in which the sample is all rangeland data collection sites in control GAs and the dependent variable
is a rangeland outcome. The independent variable is an indicator of whether the distance between the GA in which the site is located and the nearest treatment
GA is less than median distance to the nearest treatment GA among all control GAs; the coef. column shows the estimated effect of a site's GA being closer to
a treatment GA than the median. The distant mean column shows the endline mean for distant control GAs. Standard errors are clustered at the RIA level, i.e.,
the unit of randomization. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block
stratification and the RIA-level variables used in re-randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log
of the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water source, and
has a community based organization. Rangeland outcomes have been transformed as noted in parentheses to better meet assumptions of normality and
homogeneity of variance; distant and near means are sample means of the untransformed variables. See Materials and Methods for additional details of this
analysis. All p-values are two-tailed. * Aristida is a genus of grasses that are undesirable forage plants in this context.
Rangeland outcomes (2 - 3 years after program end) Effect of control GA being located < median
distance from a treatment GA

Extended Data Table 9: Mechanisms
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Evidence of heavy grazing on herbaceous plants (wet season) 0.12 0.04 0.003 0.032 0.13 972
Evidence of heavy grazing on herbaceous plants (dry season) 0.10 0.04 0.016 0.106 0.46 972
Evidence of any grazing on herbaceous plants (wet season) 0.04 0.03 0.151 0.336 0.92 972
Evidence of any grazing on herbaceous plants (dry season) 0.00 0.03 0.953 0.980 0.87 972
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Cattle numbers
Number of herds currently in GA -1.49 1.80 0.413 0.580 21.94 1,210
Number of cattle currently in GA -178 130 0.178 0.433 1,011 1,245
Reduced farmer movement
Manager moved cattle outside GA in past year -0.04 0.03 0.290 0.549 0.20 1,242
Fraction of herd that manager moved outside GA in past year -0.04 0.04 0.295 0.567 0.19 1,238
Number of months in which manager moved cattle outside GA (past 12 months) -0.19 0.17 0.273 0.535 0.92 1,243
Number of years in which manager moved cattle outside GA (past 6 years) -0.08 0.16 0.636 0.782 0.76 1,243
Outside encroachment
Outside farmers brought cattle to GA in past year 0.05 0.03 0.105 0.408 0.37 1,207
Outside farmers brought cattle to GA in past year without permission 0.07 0.02 0.005 0.070 0.16 1,230
Freq. at which herders saw outside herders in GA in past wet season (1 - 6 scale) 0.15 0.30 0.617 0.785 2.69 280
Freq. at which herders saw outside herders in GA in past dry season (1 - 6 scale) 0.40 0.27 0.151 0.241 2.77 277
Herders saw outside herder in GA more than once a week in past wet season 0.07 0.07 0.326 0.550 0.28 280
Herders saw outside herder in GA more than once a week in past dry season 0.13 0.07 0.056 0.196 0.31 277
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a program outcome on treatment status. It is an intent-to-treat (ITT)
estimate relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are calculated using
randomization inference. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block
stratification and the RIA-level variables used in re-randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log of
the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water source, and has a
community based organization. The 1 - 6 scale used to measure frequency at which herders saw outside herders in the GA is as follows: 0 = "never", 1 = "less
than once a month", 2 = "once a month", 3 = "multiple times per month", 4 = "once a week", 5 = "multiple times per week", 6 = "daily". Variables without
description of units are binary. All p-values are two-tailed.
Panel A: Direct evidence of grazing intensity
Panel B: Potential causes of increased grazing intensity
Treatment effect 2 years after program end
Treatment effect 2 - 3 years after program end

Extended Data Table 10: Audits
Panel A: 0 - 1 years after program end
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Combined herding observed in GA 0.28 0.08 <0.001 0.010 0.10 123
Number of herds in combined herd 2.47 0.74 0.002 0.020 0.35 123
Number of cattle in combined herd 52.85 17.10 0.004 0.020 14.15 122
Combined herd herded in bunched shape 0.20 0.09 0.033 0.010 0.04 123
Combined herd is accompanied by herders 0.37 0.09 <0.001 0.000 0.06 123
Number of herd owners listed in grazing group meeting minutes 2.60 0.70 <0.001 0.010 0.96 123
Number of herd owners listed in grazing group contribution list 1.92 0.54 0.001 0.020 0.39 123
Number of herd owners in water group meeting minutes -1.03 1.54 0.509 0.770 3.41 123
Number of herd owners in water group contribution list 1.31 0.81 0.112 0.140 2.93 123
Number of herd owners in development group meeting minutes 0.86 0.73 0.247 0.550 2.10 123
Number of herd owners in development group contributions list 0.97 0.46 0.040 0.180 0.55 123
Panel B: 3 years after program end
Dependent variable coef. SE p-val. RI p-val. Ctrl mean N
Herders observed combined herding 0.12 0.06 0.047 0.100 0.16 358
Herders observed returning from grazing with cattle 0.09 0.05 0.072 0.220 0.40 357
Herders observed actively herding cattle while grazing 0.05 0.04 0.252 0.320 0.26 358
# Herders observed actively herding cattle during grazing 0.18 0.10 0.075 0.120 0.29 358
Herders report following grazing plan 0.12 0.05 0.013 0.120 0.49 345
Herders report following written grazing plan 0.12 0.04 0.009 0.130 0.06 355
Herders report following group grazing plan 0.12 0.05 0.015 0.100 0.20 355
Combined cash and in-kind payments each herder receives 123.10 87.79 0.169 0.430 631.93 261
Herd owner listed in grazing group meeting minutes 0.10 0.05 0.029 0.100 0.04 1,359
Herd owner listed in grazing group contributions list 0.09 0.05 0.090 0.220 0.06 1,359
Herd owner listed in water group meeting minutes 0.07 0.06 0.250 0.470 0.17 1,359
Herd owner listed in water group contributions list 0.09 0.06 0.150 0.340 0.26 1,359
Herd owner listen in development group meeting minutes -0.01 0.02 0.472 0.750 0.06 1,359
Herd owner listed in development group contributions list -0.03 0.02 0.187 0.430 0.07 1,359
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a program outcome on treatment status. It is
an intent-to-treat (ITT) estimate relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of
randomization. RI p-values are calculated using randomization inference. Each regression includes as controls a categorical
variable for traditional authority (an administrative unit) that was used for block stratification and the RIA-level variables used in re-
randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log of the number of
CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water
source, and has a community based organization. Variables without description of units are binary, with positive responses coded as
1. See Materials and Methods for additional details. All
p
-values are two-tailed.
Treatment effect
Treatment effect

1
Supplementary Information for:
Cooperation in the commons: Community-based rangeland
management in Namibia
D. Layne Coppock1, Lucas Crowley2, Susan L. Durham3, Dylan Groves4, Julian C. Jamison5,
Dean Karlan6*, Brien E. Norton7, R. Douglas Ramsey7
1 Department of Environment and Society, Utah State University, Logan, UT 84322-5215, USA. 2 Innovations for
Poverty Action, Washington D.C., 20005, USA. 3 Ecology Center, Utah State University, Logan, UT 84322-5205,
USA. 4 Department of Political Science, Columbia University, New York City, NY, 10027, USA. 5 Department of
Economics, University of Exeter, Exeter EX44LZ, U.K. 6 Kellogg School of Management, Northwestern University,
Evanston, IL 60208, USA. 7 Department of Wildland Resources, Utah State University, Logan, UT 84322-5230,
USA.*To whom correspondence should be addressed. E-mail: karlan@northwestern.edu
Contents:
1. Supplementary Methods
a. Study timelines
b. Data collection
i. Primary outcome variable definitions
ii. Cattle scoring key
2. Supplementary Text
a. Context
i. Historical context
ii. Program context
iii. Ecological context
b. Scientific rationale for planned grazing
c. Comparing CBRLM and holistic management approaches
i. Community governance
ii. Commercialization of livestock production
3. Supplementary References
4. Supplementary Tables 1 – 5
a. Supplementary Table 1: Randomization balance
b. Supplementary Table 2: Program participation and attrition
c. Robustness checks
i. Supplementary Table 3: Treatment effect on social indices, with inverse
probability weighting
ii. Supplementary Table 4: Treatment effect on physical outcomes, with
inverse probability weighting
d. Supplementary Table 5: Treatment effect heterogeneity by rainfall for social
indices

2
1. Supplementary Methods
a. Study timelines
Supplementary Fig. 1. Timelines for community-based rangeland management (CBRLM)
mobilization, implementation, and the research components.
Supplementary Fig. 2. Timelines for grazing area (GA) mobilization, enrollment, and visits by
implementation staff.

3
b. Data collection
i. Primary outcome variable definitions
Definitions of social variables depicted in Fig. 2 (see Main Text) and Extended Data Table 1 are
as follows:
(a) Grazing planning is an index of three variables measuring whether the manager has a
grazing plan and whether the grazing plan is written.
(b) Grazing plan adherence is an index of two variables measuring whether and for how long
the herd manager followed a pre-defined grazing plan while herding cattle.
(c) Herding practices is an index of seven variables measuring whether the herd manager
follows herding practices recommended by the program, such staying with the cattle
throughout the day and herding cattle in a bunch.
(d) Herder management is an index of five variables measuring the extent to which the herd
owner provides oversight and material support to herders.
(e) Cattle husbandry is an index of nine variables measuring whether the herd manager
follows cattle husbandry practices recommended by the program, such as vaccinating and
deworming cattle.
(f) Herd restructuring is a measure of whether herd owners have made any decisions to buy
or sell cattle in order to change the structure of their herd, as opposed to reasons such as
immediate financial need or sick cattle.
(g) Livestock marketing is an index of three variables measuring whether the herd owners
sold any cattle, the number of cattle sold, and the total value of cattle sold.
(h) Community governance is an index of 12 variables measuring whether respondent
perceives their community to be governed by institutional rules.
(i) Collective action is a measure of 19 variables measuring whether respondents engaged in
collective management behaviors, such as group grazing planning, combined herding,
group payment for vaccines.
(j) Community disputes is an index of three variables measuring the number of unresolved
community disputes with other farmers inside and outside of the grazing area.
(k) Trust is an index of two variables measuring whether the respondent trusts other
individuals in the community.
(l) Expertise is an index of six variables measuring herd manager expertise and access to
expertise about cattle husbandry and marketing.
(m) Self and community efficacy is an index of four variables measuring herd manager’s
beliefs that their actions or the actions of their community can influence cattle and
rangeland outcomes.
Definitions of social variables depicted in Fig. 3 (see Main Text) and Extended Data Table 2 are
as follows:
(a) Income is the total income earned by the household per week.
(b) Expenditure is the total consumption and expenditure by the household per week.
(c) Household livestock wealth is an index of cattle and non-cattle livestock units owned by
the household.

4
(d) Time use is an index of six variables representing time spent on economically productive
activities by adults in the household (positive) and children in the household (negative).
(e) Resilience is an index of six variables measuring the household’s resilience to economic
hardship, including food security and savings.
(f) Female empowerment is an index of three variables measuring economic empowerment
of women in the household.
(g) Meat/dairy consumption is an index of two variables measuring household consumption
of meat and dairy products.
Definitions of cattle variables depicted in Fig. 3 (see Main Text) and Extended Data Table 2 are
as follows:
(h) Cattle herd value is an index of three variables measuring the value of the cattle herd in
total number, total weight, and total market value.
(i) Herd productivity is an index of seven variables measuring the health and productivity of
the cattle herd, including calving rate, herd expansion, milk production, and average
weight and body condition.
(j) Herd structure is an index of three variables measuring whether the herd has a higher
ratio of bulls to cows, total cattle to oxen, and total cattle to old and unproductive cattle.
Definitions of variables depicted in Fig. 3 (see Main Text) and Extended Data Table 2 are as
follows, and methods are reviewed below:
(a) No site erosion is the estimated degree of soil surface disturbance;
(b) Protected soil surface is the percentage of ground area shielded by plant material or rock;
(c) Plant litter cover is the percentage of ground area shielded by dead plant material;
(d) Herbaceous canopy cover is the percentage of ground area shaded by grass and forb
foliage;
(e) Perennial to annual ratio is the ratio of respective canopy coverages for perennial and
annual grasses;
(f) Grass to forb ratio is the ratio of total grass canopy cover to total forb canopy cover;
(g) No stinkbush (Pechuel-Loeschea leubnitziae) is an indicator of noxious weedy species, as
measured by percent canopy coverage;
(h) Grass to Aristida ratio is the ratio of respective canopy coverages for total grasses
excluding Aristida and all Aristida species—Aristida in this context are undesirable
forage plants; and
(i) Shrub canopy cover is the percentage of ground surface shaded by shrub foliage.

5
ii. Cattle scoring key
Supplementary Fig. 3. Field guides for (A) assessing cattle body condition scores (1-5) and (B)
cattle age1.
2. Supplemental Text
a. Context
i. Historical context
The Community-Based Rangeland and Livestock Management (CBRLM) program was
implemented in Namibia’s Northern Communal Areas (NCAs)2. Pastoral livestock production is
the predominant agricultural system in the NCAs3, although higher rainfall allows for mixed crop
and livestock farming in the NCA’s central and eastern regions. Today, the NCAs contain
approximately 20% of Namibia’s 840,000 square km of land but 45% of Namibia’s 2.6 million
cattle, and 55% of Namibia’s 2.1 million citizens4.
Most arable land in the NCAs is communally owned, meaning that it is the formal
property of the state and fencing it is illegal, except for a limited allowance for homesteads and
cultivated fields3. In recent years, population pressure, illegal fencing, and proliferation of
boreholes have accelerated degradation of an already fragile ecosystem5. However, the resource
governance challenges facing communal lands in Namibia should be understood in the context of
Namibia’s history, from pre-colonial to colonial to post-independence.
The origins of the Northern Communal Areas are traceable to two distinct systems of
economic production and political authority in pre-colonial Namibia6. Southern, central, and

6
northwestern Namibia was predominantly inhabited by transhumant pastoral communities with
limited political centralization5,7. In contrast, populations in north-central and north-eastern
Namibia, most of what we call the Northern Communal Areas today, combined settled
agriculture and animal husbandry and were ruled by centralized tribal kingdoms7.
Under German colonialism, the differences between northern and southern systems of
economic production were formalized into distinct systems of legal land ownership and political
authority8. German colonists arrived in Namibia in 1883 and by 1902 had seized most of
Namibia’s southern and central territory7. However, German colonizers did not move into
northern Namibia, both because colonial authorities did not think northern Namibia contained
valuable natural resources and because Ovambo tribal authorities in northern Namibia were more
politically powerful than the Southern, decentralized pastoralist groups6. In 1897, Germany
formally demarcated the border between southern Namibia and the Northern Communal Areas
by establishing a Veterinary Cordon Fence (VCF) to contain a Rinderpest epidemic8. A decade
later, Germany prohibited white land settlement in the NCAs, and Germany’s political influence
in the NCAs remained restricted to indirect arrangements with traditional authorities9.
When South Africa began administering Namibia after World War I, it maintained and
expanded Germany’s policy of land expropriation in the south and indirect rule in the north.
South Africa also relocated large portions of the indigenous population living south of the VCF
onto marginal communal lands called “native reserves” that were governed by indigenous
authorities10,11. Native reserves were located both north and south of the VCF. The South African
government did not support, and in some cases actively hindered, agriculture and animal
husbandry development on native reserves in order to ensure a reservoir of cheap indigenous
migratory laborers for white-owned farms, mines, and businesses10,11. The native reserve policy
has left a lasting legacy on livestock farmers in the central and eastern NCAs, who still suffer
from low quality grazing land, underdeveloped livestock markets, and limited training in animal
husbandry.
While communal land policy in the NCAs is traceable to colonial policy in the first half
of the twentieth century, the present ecological crisis facing the NCAs is also influenced by
South African policy changes initiated in the 1960s in response to the emergence of indigenous
resistance to South African apartheid5,6. In 1962, the Odendaal Commission recommended that
the South African government consolidate scattered native reserves into ethnic homelands and
increase investment in economic development in communal areas7. In the NCA’s, the Odendaal
Commission led to the dramatic expansion of investments in borehole drilling and road
networks, especially in north-western Namibia, and the first steps towards privatization of
communal lands in Oshikoto and Kavango West5,9. These changes upended existing systems of
grazing and water management by delinking grazing patterns from the availability of natural
water sources and enabling growth of human and livestock populations5. However, the South
African government failed to establish a framework of customary rights for regulating access to
land, water, and grazing resources to manage to these dynamics5,9.
Debates around the development of livestock markets, land tenure reform, and natural
resource conservation continued after Namibian independence in 1990. For example, laws
restricting the movement and sale of livestock from the NCAs south of the VCF remain deeply
contested because they are perceived to limit the economic potential of livestock production in
communal areas12. The Communal Land Reform Act of 2002 acknowledged the authority of
traditional leaders to manage customary land rights and established Communal Land Boards to
register new and customary land allocations. However, progress in registration in the ensuing

7
decades has been slow and current customary land rights focus narrowly on grazing access to the
exclusion of other communal resources, such as water and fire wood9. In the absence of group
land rights, many communities in the NCAs have developed resource government strategies
using conservancies, community forests, and Water Point Associations, but the powers of these
organizations are severely circumscribed9. The ecological and economic challenges facing the
livestock sector in the NCAs at the time CBRLM was introduced should be understood in this
historical context. Far from a static socio-economic and ecological system, the NCAs have been
deeply influenced by pre-colonial, colonial, and post-independence land administration.
Supplementary Fig. 4. CBRLM project regions and human population density (20).
ii. Program context
The CBRLM was funded under the auspices of the Millennium Challenge Account-
Namibia, and was implemented by a consulting firm called Gesellschaft für Organisation,
Planung und Ausbildung (GOPA)2. The CBRLM spanned seven administrative regions
including: Kunene, Omusati, Oshana, Oshikoto, Ohangwena, Kavango West and Kavango East.
Together these cover an area of about 170,000 km2 (Supplementary Fig. 4). The area can be
approximated by a rectangle that is 800 km long (East to West) and 200 km wide (South to
North). The NCAs have a predominantly warm and dry climate with a pronounced seasonal
distribution of precipitation. Ecological details are reviewed later in this section.
Within the seven administrative regions listed above, the CBRLM operated across 11
areas governed by Traditional Authorities (TAs)2. TAs allocate communal land, regulate
communal land use, and formulate and enforce customary law4. Within each TA the GOPA
implementation team mapped Rangeland Intervention Areas (RIAs) where CBRLM could be
implemented. Wherever possible, RIAs conformed to the boundaries of pre-existing Communal
Area Conservancies or Community Forests2. Where no Communal Area Conservancy or
Community Forest existed, the implementation team worked with TAs to map appropriately
sized intervention areas in their jurisdiction13. Each RIA contains five to 15 Grazing Areas
(GAs). The GAs are communal rangeland parcels used by five to 50 cattle kraals; herd owners in
each GA share forage and water resources. The cattle kraals are overnight holding pens for cattle
herds owned by one to five households (usually extended family members). Households that

8
share a kraal usually designate or hire a herd manager who is responsible for day-to-day
management of cattle but does not generally make decisions with regards to buying, selling, or
health treatments without the consent of the cattle owners. The size, makeup, and economic
status of herding households varies greatly across Northern Namibia14. Most GAs have a local
headman who is a member of the TA and is responsible for admission of new herd owners to the
GA as well as the management of community disputes. In practice, the extent of the power of the
local headman varies substantially among GAs.
CBRLM was intended to improve cattle raising by facilitating herd restructuring, animal
husbandry, and cattle marketing. GOPA hoped that the intervention would improve the
productivity and economic viability of cattle rearing in the NCAs15. Previous research points to
low bull-to-cow ratios, low calving rates, and inadequate weaning practices as causes for poor
productivity16,17. Others have argued that greater integration between small-scale communal
pastoralism and livestock markets could also alleviate such problems. However, there are
significant practical barriers to raising cattle for profit in northern Namibia; many cattle
producers are absentee owners and marketing transaction costs can be a hindrance18.
Such challenges are reflected in the broader literature on African pastoral development.
Some critics of cattle commercialization projects argue that raising cattle for the formal market
on communal land is not economically viable, and that development interventions should
enhance herd productivity for its own sake15. There is also debate over factors that keep
communal pastoralists from selling cattle in the formal economy. One argument is that for
pastoralists the primary economic value of cattle comes not from income-generating potential but
rather from their use as insurance19. In this view, cattle are a reliable store of wealth and animals
are primarily sold during crisis. Others argue that reluctance to sell cattle comes from their value
as social capital20.
Water development is another key issue. The question of how to protect and sustainably
maintain water resources is urgent in Namibia. Like many developing countries, the Namibian
government has adopted a community-management approach to the maintenance of boreholes in
rural areas21,22. However, water users often fail to pay their fees, especially in areas where
governance is weak. Moreover, during times of drought water users often ignore externally
imposed regulations in favor of traditional customs of reciprocity21,22.
As will be described, the CBRLM project was conceived not only as a check on
environmental degradation, but as a means of community self-help. New GA committees were
created and incentivized to pool financial resources to fund cattle production inputs like
vaccines, feed supplements, and herder salaries; CBRLM also invested in the development of
local marketing cooperatives. As such, the CBRLM is an example of a partnership to create
processes referred to as community-driven development. This is an increasingly popular concept
in international development (see Main Text), but the literature on its efficacy is mixed23.
Related evidence from recent randomized evaluations suggests that community-driven
development can successfully deliver infrastructure and economic returns, but has less success
sustainably affecting community governance and creation of social capital24,25 and may even
crowd-out pre-existing local institutions dependent on the beliefs of constituents with respect to
local politics26.
The Namibian government has previously pursued several projects meant to reduce
rangeland degradation and improve livestock production in the NCAs. A project called the
Northern Regions Livestock Development Project (NOLIDEP) took place from 1995-2003 and
had a general focus on commercialization of livestock production, with specific attention to

9
community capacity building and provision of strategic inputs such as rural veterinary clinics
and water points27. Another effort, referred to as the Sustainable Animal and Rangeland
Development Program (SARDEP), has existed in Namibia for over two decades with a focus on
creating more sustainable linkages between rural producers and service institutions, as well as
supporting dialogue to create national policies regarding sustainable use of natural resources28.
iii. Ecological context
In terms of ecological systems, the NCAs are diverse3. The topography is generally flat
with only the extreme western region exhibiting topographic variation towards the Great
Escarpment in Kunene (Supplementary Fig. 4). Precipitation has a distinct East to West gradient,
with the West being drier than the East3. Across the entire study region, annual precipitation
averages just under 400mm with high variability3. The main wet season occurs from January to
March with precipitation steadily dropping after April. A distinct dry period occurs from May to
September. During June through August the study region typically receives only scant
precipitation3.
Soils are diverse and are dominated by sandy, silty, or clay substrates3. Vegetation
community types include grasslands, shrublands, bushlands, and savannas3. Localized heavy
livestock grazing over many years is associated with the sedentarization of human settlement and
borehole development29,30. Woody encroachment and conversion of herbaceous perennial
communities to annual plants are common ecological responses to overuse of these
rangelands29,30.
As with most drylands of the world, low and highly variable precipitation is the norm in
northern Namibia. Drought, defined here as one or more years of below-average precipitation
that negatively affect socioeconomic attributes, is common. Resource use systems such as
pastoralism have evolved to cope with drought.
Our study region in northern Namibia has experienced a significant decline in rainfall in
the past eight years (e.g., 2013 to 2020) when compared to the previous 31 years beginning in
1981. This is illustrated by superimposing the monthly rainfall distributions in Supplementary
Fig. 5. Precipitation data are based on the Climate Hazards Group InfraRed Precipitation with
Station (CHIRPS) data set consisting of daily modeled precipitation from January 1981 to the
present with a ground resolution of 5.5 km. The overall decline in precipitation is on the order of
36%, with notable decreases from the main wet season months of January, February, and March.
Annual variation has been substantial over the past 39 years—and possibly increasing—as
illustrated in Supplementary Fig. 6. These data suggest that CBRLM was implemented and
evaluated during a particularly dynamic period. The project implementation phase of 2010 to
2014 may have been wetter than average, while the evaluation phase of 2014 to 2017 may have
been drier than average. The implications of such dynamism for pastoral development outcomes
from CBLRM are explored in subsequent sections.

10
Supplementary Fig. 5. Average monthly precipitation in northern Namibia for two periods,
1981 to 2012 and 2013 to 2020. Data are organized according to water year that commences
October 131.
Supplementary Fig. 6. Annual precipitation patterns from 1981 to 2020 for northern Namibia31.

11
b. Scientific rationale for planned grazing
Rotational grazing (often lumped into the category of ‘planned grazing’ or ‘prescribed
grazing’) has become a popular resource management strategy for averting environmental
degradation and increasing sustainable levels of forage and livestock production. The essential
practice of rotational grazing consists of combining herds that would otherwise graze
independently into one or more large herds. Herders then move these large herds around the
landscape, spending a short period in one location before moving to a new location. Allan
Savory32,33 espoused the idea that this form of intensively managed grazing replicates co-
evolved, sustainable relationships between grasses and large grazing animals, and that such
interactions can be used to restore damaged rangelands.
The core idea is that grasses have evolved to withstand frequent herbivory and will be
most productive when defoliated at a judicious frequency. Therefore, grasses in any given area
should be subjected to intensive, short bursts of heavy defoliation and then allowed time to
recover before subsequent waves of grazing. In a planned grazing rotation, livestock may occupy
a grazing location for just a few days to a week or so—in some forms of rotational grazing the
grazing period is just one day—and are herded together at high stocking density. This high-
density grazing creates a ‘herd effect’ imposing concentrated disturbance to the soil that Savory
believes is an important factor contributing to rangeland rehabilitation32. Savory has been one of
the more high-profile advocates of rotational grazing since 1978 when he first presented his
ideas, most of which were based on a book by French agronomist André Voisin34 at the First
International Rangeland Congress. Voisin pointed out that the concept of rotational grazing has
been around since at least the late 18th century, but it has become common practice for ranchers
and pastoralists on a global scale during only the last few decades.
Practitioners of rotational grazing see the benefits occurring more in terms of extended
rest periods that allow the vegetation to recover from defoliation, rather than the impact of
animal hooves disturbing the soil and breaking up dead plant material on the surface. They also
observe changes in plant composition of the pasture or rangeland in which palatable species tend
to increase at the expense of less palatable and weedy plants. In his review of relevant literature,
Norton35 noted that the rest periods protecting plants from grazing allow greater total forage
production, and that increased above-ground photosynthetic biomass builds a larger root system
penetrating deeper into the soil profile. However, Briske et al.36 reported that while there has
long been widespread concurrence among range scientists, federal land managers, and
commercial ranchers regarding the efficacy of rotational grazing on US rangelands relative to
continuous grazing, this distinction has not been supported by hard scientific evidence from
grazing trials on research stations.
Many research trials comparing rotational grazing to continuous grazing have failed to
find a consistent and significant benefit to either forage yield or livestock production36,37. Trials
were conducted on research stations where the experimental paddocks were small and research
herds likewise small, sometimes only 3-4 head of cattle. Another feature of research grazing
trials is that the number of paddocks in the rotation was often as low as 3 (in deferred rotations)
and rarely more than 14. Following the guidelines in Voisin34, grazing periods should be limited
to around seven days followed by a rest period of 30 days, which defines a rotation around just
five paddocks. As the grazing period is reduced and the rest period increased, the number of
paddocks required by the rotation rises. A grazing period of two days in a rotation of 60
paddocks means that livestock spend on average only six days grazing per year in each paddock,

12
and the paddock is rested for almost 360 days per year, which can lead the biological mechanics
of rotational grazing to cause a doubling of sustainable stocking rate and greater tolerance of low
rainfall seasons or years38. Voisin34 worked on dairy farms in temperate France; in a rangeland
setting, on the other hand, for most of the year rest periods need to be much longer than 30 days
to allow adequate time for recovery under the irregular and sparse rainfall patterns of a semi-arid
environment. Similarly, herds need to consist of dozens of animals or more to achieve a grazing
herd’s natural cohesive social behavior.
A key factor of livestock management is missing from grazing research in small
experimental units, namely, the spatial dimension: scientists could assume that both available
forage and forage utilization by grazing animals were spatially homogeneous, which is untrue in
a landscape context. When the distribution of grazing livestock across a spatially heterogenous
landscape is entered into the discussion, rotational grazing is clearly superior to continuous
grazing39,40. Even Briske et al. have admitted that research station results could not apply to a
commercial-scale operation41. Livestock in a small paddock can explore the entire area of
available pasture on a daily basis, and forage utilization is spatially more even than across a
landscape where patch-grazing is usually the norm if animal movement is unconstrained. One
would expect the simple factor of small paddocks to enhance livestock production, and it does.
Norton35 reported examples where he compared the experimental stocking rates to the stocking
rates for livestock on commercial ranches near the station: rates approaching twice the
recommended commercial rate could be sustained on the research station for many years without
adverse ecological consequences to either the continuous or the rotation treatment. Alternatively,
in a much larger paddock the livestock concentrate their grazing activity in preferred patches and
much of the pasture is neglected. The stocking rate calculated for the entire paddock is much
lower than the de facto stocking rate imposed on the preferred patches or zones where most of
the grazing is taking place. A critical aspect of rotational grazing is to prevent patch grazing that
opens up pastures to patch degradation (i.e., localized overgrazing), weed invasion, and erosion.
In a nutshell, the theory of rotational grazing has three elements: 1) Controlled defoliation
frequency achieved by short grazing periods followed by long rest periods; 2) high-density
grazing forcing even utilization by using combined herds for short grazing periods, with stocking
rate calculated for the entire rotation area; and 3) even spatial distribution of grazing animals in a
rotational sequence around landscape units. The outcomes comprise: 1) Greater forage
production; 2) higher livestock productivity from bigger animals or higher fecundity or both; and
3) increased ecological health of the pasture in terms of biodiversity and drought tolerance. A
good illustration of the benefits of rotational grazing that incorporates a number of dimensions of
the livestock management/pasture interaction was published by Odadi et al. from work in Kenya.
Odadi et al.42 describe an ecological assessment of rotational grazing conducted within a
communal pastoral area in northern Kenya divided into unfenced ‘paddocks’. The assessment
followed five years of planned rotational grazing and employed an experimental approach with
three pairs of sites. One of each site pair was subjected to rotational “planned grazing,” while the
other consisted of unplanned grazing (i.e., control). The planned system included bunched
grazing of livestock, multiple unfenced paddocks, and a 50% recommended level of forage
utilization prior to moving animals among paddocks. Overall, they concluded that the planned
grazing system had positive effects on all plant and animal indicators42.
In a later paper, Odadi et. al.43 focused specifically on the effectiveness of bunched herds
in a low-level rotation. Odadi and his colleagues compared herds that grazed in loose bunches
managed by one herder with herds that grazed in tight bunches enforced by three herders. All

13
other aspects of the grazing system were similar for both types of herding. The results were
noteworthy. Cattle herded in tight bunches traveled shorter distances, had higher nutrient intake
per unit of distance traveled, grazed less selectively consuming less of the preferred species with
intake spread over a wider array of species, but had higher weight gain. The higher cattle live
weights generated more income greater than the extra cost of herding. The benefits of herding in
tight bunches were financial as well as ecological.
Herding for rotational grazing as practiced in the CBRLM GAs was much looser and
often abandoned once the herd had been shepherded to a designated grazing site. In general,
rotational grazing in the GAs was implemented less rigorously by communities than by the
trained staff implementing the program described in Odadi et al. The CBRLM program should
be understood as an evaluation of a program designed to mobilize best practices in rotational
grazing through external support, rather than an evaluation of rotational grazing when applied
rigorously.
c. Comparing CBRLM and holistic management approaches
The approach to grazing management in the CBRLM proposal for the NCAs15 was
inspired by the Holistic Management (HM) model of Allan Savory32,33. In his 1988 textbook33,
Savory emphasizes the need to first identify community or household goals and then make
detailed plans to achieve those goals, which should include financial and life-style goals as well
as resource productivity, socio-economic sustainability, and household welfare. Furthermore,
Savory stresses that resource managers should be flexible, monitoring performance and revising
plans and activities on a regular basis. This flexibility and process of revising plans and actions is
an essential component of the HM strategy.
In the case of CBRLM, although it adopted rotational grazing and socio-economic
integration and household prosperity in project design15, the overarching goals were largely
determined by external development agents instead of being generated by recipient communities,
and full-scale revision of plans and activities was not possible within the short time-frame of
project implementation, even if it had been accommodated in CBRLM design. Therefore,
CBRLM did not employ the full HM template, although it followed some aspects of HM. It also
transpired that communities were unable to strictly enforce grazing management protocols of
combined herds and planned grazing; independent herder actions and trespassing by external
herds that poached conserved forage compromised the recommended rotational grazing practice
(see Main Text). Insofar as CBRLM is a test of well-executed rotational grazing at community
and landscape scales, testing the efficacy of planned grazing management was frustrated and
anticipated outcomes thwarted. CBRLM also failed as an example of HM because key features
of HM were omitted, but even if CBRLM had faithfully followed HM, there was insufficient
project time for adequate execution and evaluation of the full HM approach. This evaluation
should be considered a test of external inducements to engage in rotational grazing at community
and landscape scales, rather than a direct test of rigorously implemented HM.
In general, CBRLM, however, can be lauded for pursuing a development effort that
connected many elements of a complex social-ecological system (SES) in a core theory of
change (TOC) (see previous section). The study of outcomes—very unusual in development
programs—was thus a means to assess lessons learned. Our research has indicated that while
persistent changes in many social features of this pastoral community occurred with respect to
commons management planning, changes in the household economy, cattle production system,

14
and rangeland condition were not observed. This is not surprising, however, given the relatively
short time frame for assessment and bio-physical time lags in a setting strongly affected by
variable rainfall and other perturbations (see Main Text).
How a complex SES responds to externally generated intervention has received little
detailed study, particularly in the context of dryland settings. Rangeland management scholars
note anecdotally that while practitioners (e.g., ranchers) adopting HM paradigms in the western
US often perceive positive outcomes with regards to social or psychological aspects of their
increased investments in resource planning, hard evidence of associated improvements in the
natural environment as a result of treatment is often lacking44. A similar perspective is voiced by
Gosnell et al. in their recent meta-analysis of global studies on HM45. Although they note the
dearth of truly integrated SES research, their review points to a distinct dichotomy between the
ecological and social domains of HM; namely, that while many controversies prevail over the
pros and cons of ecological impacts arising from HM, far more consensus exists concerning the
positive benefits in the social sphere including attention to goal setting, human capacity building,
enhanced social networking, and creating social resilience. Our research findings generally
conform to this perspective.
i. Community governance
One of the key assumptions of CBRLM was that community governance needed to be
fortified to help combat environmental degradation as related to poor grazing management.
There is a growing agreement among researchers and development practitioners that a
weakening of traditional community governance is a major problem in the world’s dry lands.
Traditional governance in pastoral areas includes efforts to mitigate social disputes, allocate
natural resources, and organize labor to address community challenges20. When these attributes
are lost social cohesion can suffer and resource degradation occurs. Population growth, shifts in
cultural norms, increases in resource-based conflicts, emergence of local, ultra-wealthy elites
(who do as they please), expansion of absentee herd-ownership, and an undermining of local
traditional authorities by regional or national governments are some of the major internal and
external factors involved46–49. The problem is recognized by development agencies, who have
increasingly focused on restoring aspects of traditional governance in local situations to improve
natural resource management. Such processes include efforts to strengthen governance via
participatory combinations of traditional and contemporary leadership that reflects differing
knowledge bases and access to resources48.
ii. Commercialization of livestock production
Another key assumption of CBRLM was that the communities would be responsive to
efforts to boost cattle productivity via changes in animal husbandry, with an eye towards more
marketed offtake and increased producer incomes. While this presumed process makes perfect
sense to an outside expert trained in livestock development, there are false assumptions
concerning cultural values and differing economic goals for traditional producers that undermine
such plans in places like the NCAs of Namibia.
The struggles of pushing for more commercialized animal offtake from pastoral areas
have been well known for decades, but largely ignored by project donors who follow top-down
models of project design and implementation from a western perspective48. Cattle marketing is

15
often pursued by governments seeking exports to boost foreign exchange coffers48. New projects
based on false assumptions thus keep coming down the pipeline. The fundamental, inimical
nature of subsistence pastoralism versus commercial livestock production is best depicted by
Behnke50. Major differences occur in terms of inputs, outputs, goals, and even human
demographics. While indeed pastoral systems are changing46,51, it continues to be a truism that
traditional herdowners (e.g., men) typically aspire to accumulate large stock such as cattle. More
cattle may allow for a higher likelihood of surviving droughts or other crises, and there is little
doubt that large herds can convey high social status to herd owners in many cultures48. The flip
side is that large herds can dominate local resource consumption, thus exacerbating household
wealth stratification46. Large herds can also suffer enormous death losses during droughts20,46.
In contrast to cattle, however, small stock such as goats or sheep are more routinely sold
by pastoralists to meet modest cash needs. Small stock are also more readily produced in more
ecologically degraded environments20. Commercialization will thus tend to be more successful
for small stock when compared to that for cattle, and this can have a gender dimension as women
are then more likely to market these animals and use the proceeds to improve the livelihoods of
themselves and their children52. Such processes are more aligned with the rural-development
ambitions of project donors and development experts. Traditional pastoral systems are low-input,
high-risk enterprises. For cattle, they are not “cow-calf” operations as seen in modern
commercialized ranching. In pastoral systems, immature animals are typically retained and
matures are sold at advanced ages when they have attained a maximum body size. And when
mature cattle are sold the objective is often to use the proceeds to buy more immatures to meet
herd-building goals20.
Veterinary interventions for cattle are often embraced by producers because they
facilitate herd-building goals, not commercialization or cash-generation goals. Alternative
investments to large stock such as cattle are needed to diversify assets in support of household
resilience and improvement in rangeland management, and this can include bank accounts, urban
investments such as real estate or small businesses, and support for children who leave the
traditional system and become formally educated. Such options become more attractive when
“boom and bust” cycles for cattle productivity tilt the portfolio choices against more re-
investment in livestock versus the relative stability of investments in non-pastoral options less
connected to stocking rates or the weather46. A robust mix of different investments is the key for
managing risk.
Besides socioeconomic barriers, the cattle producers of northern Namibia also face
significant operational or structural barriers for marketing. These may include weak trading
networks and low farm-gate prices15. The Veterinary Cordon Fence, imposed by colonial
authorities and still enforced to manage the risks of epidemic diseases, limits access of producers
in the NCAs to more lucrative markets in the southern parts of Namibia12.
3. Supplementary References
1. Ensminger, M. E. The Stockman’s Handbook. (The Interstate, 1959).
2. Anon. Community Based Rangeland and Livestock Management Final Report. (2014).

16
3. Mendelsohn, J. M. Atlas of Namibia a portrait of the land and its people. (David Philip,
2002).
4. Mendelsohn, J. Farming Systems in Namibia. (Research & Information Services of Namibia,
2006).
5. Bollig, M. Shaping the African Savannah: From Capitalist Frontier to Arid Eden in
Namibia. (Cambridge University Press, 2020). doi:10.1017/9781108764025.
6. Werner, W. A Brief History of Land Disposession in Namibia. J. South. Afr. Stud. 19, 135–
146 (1993).
7. Wallace, M. & Kinahan, J. A history of Namibia: from the beginning to 1990. (2011).
8. Lechler, M. & McNamee, L. Indirect Colonial Rule Undermines Support for Democracy:
Evidence From a Natural Experiment in Namibia. Comp. Polit. Stud. 51, 1858–1898 (2018).
9. Werner, W. Tenure reform in Namibia’s communal areas. J. Namib. Stud. Hist. Polit. Cult.
18, 67–87 (2015).
10. Werner, W. No one will become rich: Economy and society in the Herero reserves in
Namibia, 1915-1946. (P. Schlettwein Publishing, 1998).
11. Silvester, J. Beasts, boundaries & buildings: the survival & creation of pastoral economies in
Southern Namibia 1915–35. in Namibia under South African Rule: Mobility & Containment
1915–46 95–116 (Ohio University Press, 1998).
12. Miescher, G. Namibia’s Red Line: The History of a Veterinary and Settlement Border.
(Palgrave Macmillan, New York, 2012).
13. Coppock, D. L. et al. Community Based Rangeland and Livestock Management Evaluation
Report. https://data.mcc.gov/evaluations/index.php/catalog/138/study-description (2020).

17
14. Namibia Statistics Agency. Namibia 2011 Population and Housing Census [PUMS dataset].
(Namibia Statistics Agency [producer and distributor], 2013).
15. Gesellschaft für Organisation, Planung und Ausbildung. Community Based Rangeland and
Livestock Management Inception Report.
https://www.yumpu.com/en/document/view/7305588/cbrlm-inception-report-millennium-
challenge-account-namibia (2013).
16. Hangara, G. N., Teweldemedhin, M. Y. & Groenewald, I. B. Major constraints for cattle
productivity and managerial efficiency in communal areas of Omaheke Region, Namibia.
Int. J. Agric. Sustain. 9, 495–507 (2011).
17. Frayne, B. Rural productivity and urban survival in Namibia: Eating away from home. J.
Contemp. Afr. Stud. 23, 51–76 (2005).
18. Shiimi, T., Taljaard, P. R. & Jordaan, H. Transaction costs and cattle farmers’ choice of
marketing channel in North-Central Namibia. Agrekon 51, 42–58 (2012).
19. Kazianga, H. & Udry, C. Consumption smoothing? Livestock, insurance and drought in rural
Burkina Faso. J. Dev. Econ. 79, 413–446 (2006).
20. Coppock, D. The Borana Plateau of Southern Ethiopia: Synthesis of Pastoral Research,
Development and Change. Int. Livest. Cent. Afr. Addis Ababa 374 (1994).
21. Schnegg, M. & Linke, T. Living Institutions: Sharing and Sanctioning Water among
Pastoralists in Namibia. World Dev. 68, 205–214 (2015).
22. Schnegg, M. & Bollig, M. Institutions put to the test: Community-based water management
in Namibia during a drought. J. Arid Environ. 124, 62–71 (2016).
23. Casey, K. Radical Decentralization: Does Community-Driven Development Work? Annu.
Rev. Econ. 10, 139–163 (2018).

18
24. Casey, K., Glennerster, R. & Miguel, E. Reshaping Institutions: Evidence on Aid Impacts
Using a Preanalysis Plan*. Q. J. Econ. 127, 1755–1812 (2012).
25. Humphreys, M., Sánchez de la Sierra, R. & Van der Windt, P. Exporting democratic
practices: Evidence from a village governance intervention in Eastern Congo. J. Dev. Econ.
140, 279–301 (2019).
26. Baldwin, K., Karlan, D., Udry, C. R. & Appiah, E. How Political Insiders Lose Out When
International Aid Underperforms: Evidence from a Participatory Development Experiment
in Ghana. http://www.nber.org/papers/w26930 (2020) doi:10.3386/w26930.
27. Northern Regions Livestock Development Project Interim Evaluation Report.
https://www.ifad.org/es/web/ioe/evaluation/asset/39835302 (2002).
28. Programme (Namibia), S. A. and R. D. Coping in a Fragile Environment: The SARDEP
Experience : the Sustainable Animal and Range Development Programme (SARDEP) in the
Communal Areas of Namibia. (SARDEP, Ministry of Agriculture, Water and Rural
Development, 2001).
29. Sander, H., Bollig, M. & Schulte, A. Himba paradise lost: Stability, degradation and
pastoralist management of the Omuhonga basin (Northwestern Namibia). Erde 129, 301–315
(1998).
30. Klintenberg, P. & Verlinden, A. Water points and their influence on grazing resources in
central northern Namibia. Land Degrad. Dev. 19, 1–20 (2008).
31. Funk, C. et al. The climate hazards infrared precipitation with stations—a new
environmental record for monitoring extremes. Sci. Data 2, 150066 (2015).
32. Savory, A. A Holistic Approach to Ranch Management Using Short Duration Grazing. in
Proceedings of First International Rangeland Congress 555–557 (1978).

19
33. Savory, A. Holistic Resource Management. (Island Press, 1988).
34. Voisin, A. Grass Productivity. (Island Press, 1988).
35. Norton, B. E. The application of grazing management to increase sustainable livestock
production : the McClymont lecture.
http://www.asap.asn.au/livestocklibrary/1998/MCCMONT.PDF (1998).
36. Briske, D. D. et al. Rotational Grazing on Rangelands: Reconciliation of Perception and
Experimental Evidence. Rangel. Ecol. Manag. 61, 3–17 (2008).
37. Hawkins, H.-J. A global assessment of Holistic Planned GrazingTM compared with season-
long, continuous grazing: meta-analysis findings. Afr. J. Range Forage Sci. 34, 65–75
(2017).
38. Norton, B. Personal Observation.
39. Norton, B. E., Barnes, M. & Teague, R. Grazing Management Can Improve Livestock
Distribution. Rangelands 35, 45–51 (2013).
40. Teague, R. & Barnes, M. Grazing management that regenerates ecosystem function and
grazingland livelihoods. Afr. J. Range Forage Sci. 34, 77–86 (2017).
41. Briske, D. D. et al. Origin, Persistence, and Resolution of the Rotational Grazing Debate:
Integrating Human Dimensions Into Rangeland Research. Rangel. Ecol. Manag. 64, 325–
334 (2011).
42. Odadi, W. O., Fargione, J. & Rubenstein, D. I. Vegetation, Wildlife, and Livestock
Responses to Planned Grazing Management in an African Pastoral Landscape. Land Degrad.
Dev. 28, 2030–2038 (2017).
43. Odadi, W. O., Riginos, C. & Rubenstein, D. I. Tightly Bunched Herding Improves Cattle
Performance in African Savanna Rangeland. Rangel. Ecol. Manag. 71, 481–491 (2018).

20
44. Huntsinger, L. Private correspondence with Lynn Huntsinger, University of California at
Berkeley. (2020).
45. Gosnell, H., Grimm, K. & Goldstein, B. E. A half century of Holistic Management: what
does the evidence reveal? Agric. Hum. Values 37, 849–867 (2020).
46. Coppock, D. L., Bailey, D., Ibrahim, M. & Tezera, S. Diversified Investments of Wealthy
Ethiopian Pastoralists Include Livestock and Urban Assets That Better Manage Risk☆.
Rangel. Ecol. Manag. 71, 138–148 (2018).
47. Fernandez-Gimenez, M. E. Reconsidering the Role of Absentee Herd Owners: A View from
Mongolia. Hum. Ecol. 27, 1–27 (1999).
48. Davies, J. et al. Improving governance of pastoral lands. Gov. Tenure Tech. Guide FAO Eng
No 6 (2016).
49. Coppock, D. L. et al. Rangeland Systems in Developing Nations: Conceptual Advances and
Societal Implications. in Rangeland Systems: Processes, Management and Challenges (ed.
Briske, D. D.) 569–641 (Springer International Publishing, 2017). doi:10.1007/978-3-319-
46709-2_17.
50. Behnke, R. H. Production rationales: The commercialization of subsistence pastoralism.
Nomadic Peoples 3–33 (1983).
51. Galvin, K. A. Transitions: Pastoralists Living with Change. Annu. Rev. Anthropol. 38, 185–
198 (2009).
52. Coppock, D. L., Desta, S., Tezera, S. & Gebru, G. Capacity building helps pastoral women
transform impoverished communities in Ethiopia. Science 334, 1394–1398 (2011).
4. Supplementary Tables 1 – 5 (following pages)

Supplementary Table 1: Randomization balance
Panel A: Data collected at RIA level
RIA characteristic Ctrl mean Treat mean p-val. RI p-val. % missing N
4.47 4.61 0.445 0.307 0.00 38
RIA has good water source * 0.79 0.74 0.674 0.658 0.00 38
RIA has community-based organizations * 0.74 0.79 0.568 0.545 0.00 38
Forest present in RIA 0.42 0.42 1.000 0.870 0.00 38
Grassland present in RIA 0.11 0.11 1.000 0.980 0.00 38
Livestock density (kg/ha) * 16.79 16.88 0.939 0.953 0.00 38
Number of livestock * 17,380 16,497 0.903 0.824 0.00 38
0.37 0.42 0.530 0.456 0.00 38
p-value, joint F-test: 0.998 p-value, joint F-test, RI: >0.999
Panel B: Data collected at GA level
GA characteristic Ctrl mean Treat mean p-val. RI p-val. % missing N
Community is willing to change 0.76 0.88 0.186 0.193 2.63 38
Traditional authority is ready for change 0.54 0.67 0.995 0.995 13.16 38
Community has social cohesion 0.63 0.67 0.756 0.721 0.00 38
Community is worried about spillover/grass poaching 0.49 0.65 0.166 0.094 2.63 38
Community perceives herder turnover as high 0.25 0.40 0.389 0.342 7.89 38
GA has cell phone reception 0.20 0.13 0.331 0.315 5.26 38
Community believes herders perform well 0.42 0.21 0.159 0.090 0.00 38
Cattle carrying capacity at or above regional norm 0.84 0.88 0.356 0.430 0.00 38
Proportion of HHs near water point made of mud/clay/brick 0.06 0.03 0.206 0.116 5.26 38
Full water point installed 0.72 0.66 0.754 0.771 7.89 38
Himba people live in community 0.25 0.36 0.454 0.381 5.26 38
Vegetation biomass production (1-9; 9 = extremely high production) 6.88 6.89 0.854 0.840 0.00 38
Non-cattle livestock density (mean #/square km) 1.12 1.27 0.874 0.834 0.00 38
Cattle density (mean #/square km) 7.63 8.01 0.925 0.904 0.00 38
Annual rainfall deficit (evaporation minus rainfall, in mm) 9.18 9.32 0.323 0.264 0.00 38
GA area (square km) 7,540.76 6,321.75 0.185 0.184 0.00 38
Ethnolinguistic fractionalization (inverted Herfindahl index) 0.00 0.01 0.380 0.247 0.00 38
Number of kraals per grazing area 25.25 22.84 0.452 0.326 0.00 38
Proportion plant cover of any kind 0.84 0.85 0.750 0.636 0.00 38
Rainfall (mm) in year ending in August 2016 353.30 355.33 0.753 0.698 0.00 38
p-value, joint F-test: 0.662 p-value, joint F-test, RI: >0.999
Panel C: Data collected from herd managers
Herd owner characteristic Ctrl mean Treat mean p-val. RI p-val. % missing N
Herd owner age (years) 54.46 54.32 0.178 0.125 1.92 1,176
Herd owner completed primary education 0.39 0.44 0.804 0.773 0.00 1,199
p-value, joint F-test: 0.396 p-value, joint F-test, RI: 0.557
Panel D: Data collected from heads of household
Household characteristic Ctrl mean Treat mean p-val. RI p-val. % missing N
Household head is male 0.80 0.79 0.783 0.784 11.04 1,209
Household head age (years) 55.94 57.47 0.927 0.917 11.63 1,201
Household head education level (0 - 9 scale; 0=none) 2.13 2.42 0.555 0.549 11.04 1,209
Household speaks Rukwangli 0.17 0.19 0.120 0.125 11.04 1,209
Household speaks Herero 0.30 0.27 0.920 0.910 11.04 1,209
p-value, joint F-test: 0.551 p-value, joint F-test, RI: 0.837
RIA-level statistics (pre-program)
Notes: Treatment and control means are sample means for each subgroup. Each p-value is two-tailed and comes from an OLS regression of treatment
on the associated balance variable, and indicates the probability of observing a test statistic as extreme or more extreme than the observed test
statistic given a true null hypothesis of no treatment effect. In each joint F-test, treatment status is regressed on all the variables in the associated panel
of the table. RI p-values are calculated using randomization inference. Standard errors are not clustered in Panels A and B because RIAs are the unit
of observation and the unit of randomization, but in Panels C and D are clustered at the RIA level. Each regression in Panels A and B controls for a
categorical variable for traditional authority (an administrative unit) that was used for block stratification. Panels C and D include as controls this
categorical variable for traditional authority and the RIA-level variables used in re-randomization to ensure balance: vegetation type, number of
livestock, livestock density, the log of the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior
intervention areas, has a quality water source, and has a community based organization. RIA-level regressions in Panels A and B do not include this
full set of randomization controls to avoid having more predictors than observations. In Panel B, missing values are coded as 0. In Panels C and D,
missing values are coded as zeros and regressions include a binary variable equal to 1 for observations in which the balance variable was missing and
zero otherwise. Variables without description of units are binary. * indicates that a variable was used for re-randomization to ensure balance.
Log of the number of CBRLM-eligible households *
RIA overlaps geographically with prior interventions *
Individual-level statistics (3 years after program end)
Individual-level statistics (0 - 1 years after program end)
RIA-level statistics (pre-program)

Supplementary Table 2: Program participation and attrition
Panel A: GA-level participation
Dependent variable Ctrl mean Treat mean p-val RI p-val. N
GA formally enrolled in CBRLM 0.00 0.71 <0.001 <0.001 123
Panel B: GA manager-level participation
Dependent variable Ctrl mean Treat mean p-val. RI p-val. N
Manager has heard of CBRLM program 0.63 0.91 <0.001 0.002 1,234
Manager was offered chance to participate in CBRLM 0.13 0.67 <0.001 <0.001 1,208
Manager participated in CBRLM 0.05 0.56 <0.001 <0.001 1,222
Panel C: Attrition
Dependent variable Ctrl mean Treat mean p-val. RI p-val. N
Attrited 0 - 1 years after end (behavioral survey 1) 0.03 0.04 0.336 0.407 1,241
Attrited 2 - 3 years after end (behavioral survey 2) 0.08 0.07 0.476 0.608 1,348
Attrited 2 - 3 years after end (cattle survey) 0.12 0.09 0.193 0.294 730
Attrited 3 years after end (household survey) 0.10 0.10 0.465 0.627 1,345
Notes: Each p-value is two-tailed and comes from an OLS regression of a variable measuring participation in the CBRLM program on
treatment status, and indicates the probability of observing a test statistic as extreme or more extreme than the observed test statistic
given a true null hypothesis of no treatment effect. RI p-values are calculated using randomization inference. Standard errors are
clustered at the RIA level, i.e., the unit of randomization. Each regression includes as controls a categorical variable for traditional
authority (an administrative unit) that was used for block stratification and the RIA-level variables used in re-randomization to ensure
balance, which are: vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible households, and
binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water source, and has a community based
organization. Variables without description of units are binary.

Panel A: Behaviors
Dependent variable coef. SE p-val. RI p-val. q-val. N coef. SE p-val. RI p-val. q-val. N
Grazin
g
p
lannin
g
1.36 0.23 <0.001 0.002 0.001 1,199 1.04 0.20 <0.001 0.002 0.001 1,218
Grazin
g
p
lan adherence 0.38 0.08 <0.001 0.027 0.001 1,199 0.32 0.06 <0.001 0.002 0.001 1,240
Herdin
g
p
ractices 0.40 0.12 0.001 0.014 0.002 1,199 0.31 0.08 <0.001 0.023 0.001 1,243
Herder mana
g
ement 0.17 0.08 0.044 0.101 0.045 1,199 0.43 0.14 0.003 0.058 0.004 1,243
Cattle husbandr
y
* 0.38 0.11 <0.001 0.029 .1,199 0.12 0.09 0.186 0.341 .1,249
Herd restructurin
g
* -0.01 0.07 0.927 0.960 .1,199 -0.02 0.04 0.506 0.746 .1,243
Cattle marketin
g
* -0.05 0.06 0.378 0.649 .1,199 0.07 0.05 0.210 0.484 .1,245
Panel B: Community
dynamics, knowledge, and
attitudes
Dependent variable coef. SE p-val. RI p-val. q-val. N coef. SE p-val. RI p-val. q-val. N
Communit
y
g
overnance 0.78 0.14 <0.001 0.008 0.001 1,199 0.55 0.11 <0.001 0.006 0.001 1,245
Collective action 1.59 0.24 <0.001 0.002 0.001 1,199 0.89 0.22 <0.001 0.002 0.001 1,245
Communit
y
dis
p
utes 0.07 0.07 0.303 0.444 0.418 1,140 -0.28 0.08 <0.001 0.088 0.002 1,243
Trust -0.03 0.06 0.641 0.715 0.784 1,198 ... . ..
Knowled
g
e 0.30 0.10 0.007 0.054 0.012 1,199 0.37 0.09 <0.001 0.009 0.001 1,248
Self & communit
y
efficac
y
0.03 0.10 0.783 0.831 0.858 1,196 -0.01 0.07 0.857 0.916 0.858 1,009
Supplementary Table 3: Treatment effect on social and behavioral indices, with inverse probability
weighting
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a behavioral program outcome on treatment status. It is an intent-to-treat (ITT)
estimate relative to the control group. Standard errors are clustered at the RIA level, i.e., the unit of randomization. Regressions are corrected for differences in
probability of treatment assignment within stratification blocks using inverse probability weighting, and RI p-values are calculated using randomization inference; see
Methods for explanations of these methods. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for
block stratification and the RIA-level variables used in re-randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log of
the number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water source, and has a
community based organization. Indices are the standardized (mean = 0 and sd = 1), unweighted average of standardized components. Variables for the "trust" index
were not collected in the survey 2 - 3 years after program end. All p-values are two-tailed. * indicates variables for which multiple hypothesis correction was not specified
in the pre-analysis plan.
0 - 1 years after program end 2 - 3 years after program end
0 - 1 years after program end 2 - 3 years after program end

Dependent variable coef. SE p-val. RI p-val. q-val. N
Herd value 0.01 0.11 0.955 0.969 0.955 653
Herd productivity 0.03 0.08 0.748 0.874 0.935 1,285
Weekly household income 0.10 0.07 0.163 0.353 0.408 1,210
Weekly household expenditure 0.03 0.05 0.567 0.506 0.935 1,210
Household livestock wealth -0.07 0.05 0.121 0.423 0.408 1,210
Dependent variable coef. SE p-val. RI p-val. q-val. N
Herd structure -0.01 0.07 0.875 0.932 0.945 653
Time use 0.04 0.10 0.699 0.832 0.945 1,210
Resilience -0.03 0.07 0.642 0.806 0.945 1,210
Female empowerment -0.02 0.08 0.804 0.849 0.945 1,210
Meat and dairy consumption 0.00 0.04 0.945 0.965 0.945 1,210
Dependent variable coef. SE p-val. RI p-val. q-val. N
Erosion:
Wet season site erosion (1 = no erosion, 0 = erosion) -0.09 0.10 0.360 0.646 . 972
Ground cover:
Wet season protected soil surface (%, logit-transformed) -0.21 0.11 0.061 0.184 . 972
Wet season plant litter cover (%, logit-transformed) -0.18 0.08 0.029 0.191 . 972
Dry season plant litter cover (%, logit-transformed) -0.08 0.11 0.466 0.729 . 885
Herbaceous cover:
Wet season herbaceous canopy cover (%, logit-transformed) -0.23 0.13 0.092 0.303 . 972
Dry season herbaceous canopy cover (%, logit-transformed) -0.23 0.07 0.002 0.076 . 885
Wet season fresh plant biomass (kg/ha, log-transformed) -0.23 0.15 0.142 0.326 . 966
Dry season fresh plant biomass (kg/ha, log-transformed) -0.21 0.07 0.004 0.116 . 792
Relative canopy cover of perennial and annual grasses:
Wet season perennial to annual canopy ratio (log-transformed) -0.06 0.07 0.389 0.710 . 972
Relative canopy cover of grasses and forbs:
Wet season grass to forb canopy ratio (log-transformed) -0.21 0.10 0.037 0.289 . 972
Weeds:
Wet season % of shrubs that are not stinkbush (%, logit-transformed) 0.00 0.08 0.980 0.993 . 870
Wet season grass to Aristida canopy cover ratio (log-transformed) * -0.12 0.13 0.358 0.554 . 752
Woody vegetation:
Wet season shrub canopy cover (%, logit-transformed) 0.02 0.15 0.866 0.917 . 972
Dry season shrub canopy cover (%, logit-transformed) -0.06 0.15 0.704 0.822 . 885
Supplementary Table 4: Treatment effect on rangeland health, cattle productivity and household
economics, with inverse probability weighting
2 - 3 years after program endPanel B: Secondary outcomes (indices)
Notes: Each coef. is the coefficient on the treatment variable in an OLS regression of a program outcome on treatment status. It is an intent-to-treat (ITT) estimate relative to
the control group. Data in Panels A and B were collected using surveys of heads of household and cattle managers, and data in Panel C were collected as described in the
Methods. Standard errors are clustered at the RIA level, i.e., the unit of randomization. Regressions include corrections for differences in probability of treatment assignment
within stratification blocks using inverse probability weighting, and RI p-values were calculated using randomization inference; see Materials and Methods for explanations of
these methods. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for block stratification and the RIA-
level variables used in re-randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log of the number of CBRLM-eligible
households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality watersource, and has a community based organization. Indices are
the standardized (mean = 0 and sd = 1), unweighted average of standardized components. Monetary variables have been scaled to weekly Namibian dollar (NAD) amounts.
At the time of data collection (2017) the exchange rate was 13.3 NAD to 1 USD. Rangeland outcomes have been transformed (but not standardized as in Extended Data
Table 2) as noted in parentheses to better meet assumptions of normality and homogeneity of variance, but treatment and control means are sample means computedfrom
data on untransformed scales. Multiple hypothesis correction was not specified for rangeland outcomes in the pre-analysis plan. All p-values are two-tailed. * Aristida is a
genus of grasses that are undesirable forage plants in this context.
2 years after program endPanel C: Rangeland outcomes (standardized)
Panel A: Primary outcomes (indices) 2 - 3 years after program end

Panel A: Social and Behavioral
Outcomes (0 - 1 years)
Dependent variable coef.1 SE p-val. coef.2 SE p-val. coef.3 SE p-val. RI p-val. N
Grazing planning 1.70 0.32 <0.001 0.07 0.32 0.826 -0.75 0.42 0.086 0.409 1,199
Grazing plan adherence 0.42 0.08 <0.001 0.18 0.13 0.174 -0.14 0.15 0.331 0.560 1,199
Herding practices 0.36 0.12 0.004 0.12 0.23 0.596 0.02 0.18 0.928 0.954 1,199
Herder management 0.17 0.09 0.067 -0.01 0.19 0.944 -0.04 0.13 0.772 0.869 1,199
Cattle husbandry 0.51 0.12 <0.001 0.14 0.16 0.396 -0.27 0.17 0.113 0.470 1,199
Herd restructuring 0.07 0.11 0.503 0.03 0.12 0.795 -0.11 0.13 0.401 0.579 1,199
Cattle marketing -0.01 0.08 0.920 0.15 0.14 0.301 -0.09 0.11 0.439 0.551 1,199
Community governance 0.92 0.19 <0.001 -0.02 0.24 0.943 -0.32 0.25 0.207 0.536 1,199
Collective action 1.65 0.27 <0.001 0.41 0.31 0.190 -0.25 0.45 0.585 0.771 1,199
Community disputes 0.13 0.07 0.065 0.01 0.12 0.912 -0.10 0.12 0.406 0.656 1,140
Trust 0.04 0.07 0.595 -0.01 0.14 0.927 -0.11 0.11 0.337 0.548 1,198
Knowledge 0.51 0.13 <0.001 0.42 0.18 0.029 -0.39 0.17 0.026 0.226 1,199
Self & community efficacy 0.04 0.12 0.725 0.02 0.19 0.930 -0.01 0.15 0.960 0.981 1,196
Panel B: Social and Behavioral
Outcomes (2 - 3 years)
Dependent variable coef.1 SE p-val. coef.2 SE p-val. coef.3 SE p-val. RI p-val. N
Grazing planning 1.53 0.26 <0.001 0.80 0.27 0.006 -1.02 0.30 0.002 0.181 1,218
Grazing plan adherence 0.53 0.09 <0.001 0.21 0.15 0.173 -0.40 0.10 <0.001 0.156 1,240
Herding practices 0.46 0.12 <0.001 0.32 0.13 0.017 -0.32 0.16 0.057 0.214 1,243
Herder management 0.47 0.14 0.002 0.33 0.15 0.035 -0.10 0.20 0.641 0.834 1,243
Cattle husbandry 0.06 0.10 0.536 0.04 0.11 0.745 0.11 0.15 0.461 0.695 1,249
Herd restructuring -0.01 0.06 0.822 0.21 0.08 0.014 -0.02 0.08 0.847 0.915 1,243
Cattle marketing 0.01 0.08 0.861 -0.17 0.10 0.096 0.12 0.12 0.343 0.606 1,245
Community governance 0.63 0.14 <0.001 0.16 0.18 0.385 -0.17 0.20 0.407 0.683 1,245
Collective action 1.07 0.20 <0.001 0.37 0.29 0.198 -0.37 0.40 0.353 0.602 1,245
Community disputes -0.39 0.11 0.001 0.18 0.24 0.462 0.19 0.13 0.149 0.437 1,243
Knowledge 0.43 0.10 <0.001 -0.09 0.14 0.548 -0.16 0.15 0.297 0.538 1,248
Self & community efficacy 0.09 0.11 0.430 0.23 0.21 0.272 -0.20 0.18 0.298 0.473 1,009
Panel C: Physical outcomes
(2 - 3 years)
Dependent variable coef.1 SE p-val. coef.2 SE p-val. coef.3 SE p-val. RI p-val. N
Herd value 0.12 0.11 0.271 -0.18 0.18 0.318 -0.17 0.16 0.314 0.521 653
Herd productivity -0.15 0.13 0.274 -0.22 0.21 0.308 0.35 0.21 0.097 0.291 1,285
Weekly household income 58.22 38.66 0.141 40.78 52.69 0.444 -37.12 63.03 0.560 0.755 1,210
Weekly household expenditure -33.96 74.49 0.651 -23.77 113.83 0.836 118.46 127.50 0.359 0.549 1,210
Household livestock wealth -0.03 0.06 0.624 -0.03 0.16 0.841 -0.05 0.09 0.565 0.749 1,210
Herd structure -0.12 0.09 0.212 -0.31 0.15 0.044 0.20 0.16 0.224 0.477 653
Time use 0.27 0.16 0.089 0.62 0.29 0.037 -0.48 0.26 0.068 0.168 1,210
Resilience -0.17 0.09 0.076 0.00 0.13 0.969 0.28 0.12 0.028 0.177 1,210
Female empowerment 0.06 0.13 0.666 0.08 0.14 0.591 -0.14 0.14 0.347 0.521 1,210
Food consumption 0.03 0.07 0.662 -0.17 0.12 0.144 -0.05 0.07 0.505 0.659 1,210
Supplementary Table 5: Treatment effect heterogeneity by rainfall, social outcomes and cattle productivity
Notes: Each row displays results from a separate regression in which the dependent variable is an index of behavioral, household, or cattle outcomes, and the
independent variables are treatment status and an indicator variable for low rainfall. Coef. 1 indicates the coefficient on treatment, which is an intent-to-treat (ITT) estimate
relative to control. Coef. 2 indicates the coefficient on an indicator variable for low rainfall, which is equal to 1 if a grazing area was below the median of all grazing areas in
terms of percent difference in the grazing area's rainfall during the project period relative to the mean of the grazing area's rainfall over the 10 years prior to the program.
Coef. 3 shows the interaction of the low-rainfall indicator with treatment. Standard errors are clustered at the RIA level, i.e., the unit of randomization. RI p-values are
calculated using randomization inference. Each regression includes as controls a categorical variable for traditional authority (an administrative unit) that was used for
block stratification and the RIA-level variables used in re-randomization to ensure balance, which are: vegetation type, number of livestock, livestock density, the log of the
number of CBRLM-eligible households, and binary indicators for whether the RIA overlaps with prior intervention areas, has a quality water source, and has a community
based organization. Indices are the standardized (mean = 0 and sd = 1), unweighted average of standardized components. Monetary variables are in Namibian dollar
(NAD) amounts. 0 -1 years after program end (2014) the exchange rate was 10.8 NAD to 1 USD, and 2 - 3 years after program end was 14.7 NAD to 1 USD. See
Materials and Methods and the Supplementary Materials for additional details. All p-values are two-tailed.
Treatment Low rainfall indicator
Treatment x low rainfall indicator
Treatment x low rainfall indicator
Treatment x low rainfall indicator
Treatment Low rainfall indicator
Treatment Low rainfall indicator