Content uploaded by Greg Mason
Author content
All content in this area was uploaded by Greg Mason on Jan 24, 2019
Content may be subject to copyright.
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later
Wayne Simpson, Greg Mason, Ryan Godwin
Canadian Public Policy, Volume 43, Number 1, March/mars 2017, pp. 85-104
(Article)
Published by University of Toronto Press
For additional information about this article
Access provided by University of Manitoba (21 Apr 2017 20:23 GMT)
https://muse.jhu.edu/article/651481
COMMENTARY
The Manitoba Basic Annual
Income Experiment: Lessons Learned
40 Years Later
WAYNE SIMPSON, GREG MASON, AND RYAN GODWIN
Department of Economics, University of Manitoba, Winnipeg, Manitoba
Le lancement re
´cent, en Ontario, du Projet pilote portant sur le revenu de base, et le revenu de base mis
en place en Finlande pour les cho
ˆmeurs sont deux exemples qui refle
`tent l’inte
´re
ˆt croissant pour diverses
formes de revenu annuel garanti (RAG). Il s’agit d’une ide
´e qui circule depuis des de
´cennies, et qui est
maintenant ravive
´e face aux questions que posent les ine
´galite
´s croissantes et les bouleversements
du marche
´du travail. Or, le Mincom (Manitoba Basic Annual Income Experiment, ou Projet pilote de
revenu annuel de base du Manitoba), qui a e
´te
´re
´alise
´il y a une quarantaine d’anne
´es et a constitue
´
une expe
´rience sociale audacieuse, avait justement e
´te
´conc¸u pour e
´valuer les re
´percussions, en matie
`re
d’offre de travail, d’un impo
ˆt sur le revenu ne
´gatif, qui est une forme de RAG. Dans cet article, les auteurs
examinent cette expe
´rience pour clarifier ce que le Mincome nous a permis (ou non) d’apprendre sur les
re
´actions des individus et des me
´nages qui ont profite
´de ce revenu garanti. Les auteurs montrent ainsi
comment cette expe
´rience peut nous aider a
`re
´pondre a
`des questions lie
´es aux projets actuels de revenu
garanti ou de base, et expliquent comment les chercheurs peuvent aujourd’hui avoir acce
`s aux donne
´es
que l’expe
´rience a permis de recueillir.
Mots cle
´s: expe
´rience sociale, Mincome, impo
ˆt sur le revenu ne
´gatif, revenu annuel garanti, revenu de
base, re
´percussions en matie
`re d’offre de travail
The recent announcements of the Ontario Basic Income Pilot and Finland’s cash grants to jobless persons
reflect the growing interest in some form of guaranteed annual income (GAI). This idea has circulated for
decades and has now been revived, no doubt prompted by concerns of increased inequality and employ-
ment disruptions. The Manitoba Basic Annual Income Experiment (Mincome), conducted some 40 years
ago, was an ambitious social experiment designed to assess a range of behavioural responses to a negative
income tax, a specific form of GAI. This article reviews that experiment, clarifying what exactly Mincome
did and did not learn about how individuals and households reacted to the income guarantees. This
article reviews the potential for Mincome to answer questions about modern-day income experiments
and describes how researchers may access these valuable data.
Keywords: social experiment, Mincome, negative income tax, guaranteed annual income, basic income,
labour supply response
Over the years, but particularly in the 1960s and 1970s, considerable effort has been spent in the examination of this
system [of personal income and transfers] and its component programs, with a view to identifying potential improve-
ments in the areas of fairness, effectiveness, efficiency, and simplicity. These interests led to one of the most significant
social experiments in Canadian history, namely the Manitoba Basic Annual Income Experiment (Mincome).
—Judith Maxwell, chairman of the Economic Council of Canada, in the Foreword to Hum and Simpson (1991, ix).
Reproduced with permission from the Minister of Public Works and Government Services Canada, 3 February 2017.
Introduction
Although made in the previous century, Maxwell’s com-
ments about redistributive taxation policy continue to
resonate today. Emerging concerns about poverty in the
United States and Canada in the 1960s led to a wave of
social experimentation in the 1970s to evaluate proposals
for a negative income tax program to alleviate or elimi-
nate the problem, including the Manitoba Basic Annual
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
Income Experiment (Mincome) between 1975 and 1978.
These income maintenance experiments provided im-
portant evidence of the feasibility, impact, and effective-
ness of programs of this nature, providing important
lessons about income maintenance policy and attendant
issues. Yet these significant policy events did not im-
mediately translate into effective policies to address
poverty, which remains an important unresolved social
issue.
Although emerging international and Canadian re-
search (Atkinson and Morelli 2014; Fortin et al. 2012)
has concentrated on rising income inequality and distri-
butional issues more generally, the related issues around
poverty and income support remain an important com-
ponent of the policy debate. This debate now includes
the 2009 recommendation from Quebec’s Comite
´con-
sultatif de lutte contre la pauvrete
´et l’exclusion sociale
for a ‘‘revenu minimum garanti’’ pegged at 80 percent
of Statistics Canada’s Market Basket Measure (Clavet,
Duclos, and Lacroix 2013) and the publication of a pro-
posal for the Ontario Basic Income Pilot (Segal, 2016).
The Government of Canada has communicated mixed
signals that some form of guaranteed minimum income
could be included in discussions with the provinces to
tackle poverty and improve Canada’s social safety net
(Curry, 2016). Mincome provides enduring lessons about
the feasibility of such a program, the scope for testing,
and what can be learned from such social experiments.
Our article is structured as follows. The next section,
‘‘Manitoba Basic Annual Income Experiment,’’ briefly
summarizes the Mincome experiment chronologically
from conceptualization to completion, considering both
its social context and the technical details of its execu-
tion while providing links to more extensive discussion
elsewhere. The third section, ‘‘Demise and Resurrection
of Mincome,’’ then reviews what happened after the
experiment was abruptly terminated in 1978, including
the temporary mothballing of the evidence, the sub-
sequent development and release of the digitized data,
and what can be learned about the conduct of large
social experiments from that sequence of events. The
fourth section, ‘‘Mincome Research,’’ summarizes the
findings from the Mincome research, including both
the early research on work disincentives and marital
stability and the subsequent research on the impact of
Mincome on broader social outcomes. The ‘‘Using the
Mincome Database’’ section discusses the research ques-
tions that Mincome can still support and some issues
in statistical modelling that researchers need to under-
stand, and it describes how to access these data. The
final section, ‘‘Old and New Lessons from the Mincome
Experiment,’’ presents some key lessons that current
social experiments such as the Ontario Basic Income
Pilot should consider. Appendix A provides details on
the available documentation and data.
Manitoba Basic Annual Income Experiment
Prompted by the analysis of poverty from the Economic
Council of Canada (1968), the Special Senate Committee
on Poverty (1971) recommended implementation of a
federally financed and administered Guaranteed Annual
Income (GAI) program to address poverty. The Manitoba
government led by Ed Schreyer was receptive to the idea
of an experimental project to test such a program and
submitted a proposal for funding to the Department
of National Health and Welfare in March 1973. The
minority federal Liberal government led by Pierre Tru-
deau, which had called for a review of Canada’s social
security system in the 1973 Throne Speech, was also
receptive to the idea of a trial guaranteed income plan.
The two governments entered discussions and reached
formal agreement on the budget for a Basic Annual
Income Experiment Project in June 1974.
1
Mincome had two primary goals, the first explicit
and the second of which evolved as the experiment pro-
gressed. The first goal was to ‘‘evaluate the economic
and social consequences of an alternative social welfare
system based on the concept of a negative income tax’’
and specifically to ‘‘examine the labour supply responses
of households and individuals to a guaranteed annual
income’’ (Hum, Laub, and Powell 1979, 1). Over time a
second goal, not explicitly stated in the design documents,
evolved, which was to understand the administrative
and logistical challenges involved in implementing such
a system across the population.
Although Manitoba appears to have initially conceived
the idea as a modest demonstration or pilot project,
Canada and Manitoba eventually agreed that Mincome
would be a randomized controlled trial.
2
The precise
design was left to researchers engaged by a separate legal
entity created to manage the logistics of administering
the payments, collect a wide range of socio-economic
data from participants and non-participants, and analyze
the results. Although sponsorship was jointly federal
and provincial, it appears that the federal government
focused primarily on the scientific aspects of the expe-
riment, and the province supervised the operational
aspects (payments and data collection).
The early design ideas behind the Mincome experi-
ment appear in Hikel, Laub, and Powell (1974) and
Hum, Laub, and Powell (1979). Very generally, qualify-
ing low-income households in Winnipeg and rural areas
were allocated at random to receive either a treatment of
one of several guaranteed income plans or no treatment,
to form a control or comparison group. The treatments
consisted of an income guarantee of maximum support
for families with no other source of income and a tax
rate that determined how much of the guarantee would
be removed per dollar of family income and net worth.
The net benefit also depended on family structure and
size.
86 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
The basic design reflected the approach used by the
other negative income tax experiments underway in the
United States at the time.
3
It presented the scientific
argument for random allocation combined with a control
group to allow uncontaminated (econometrically consis-
tent) estimation of behavioural response to the guaranteed
income treatments. Labour economists considered the ex-
perimental design crucial to providing credible estimates
of the impact of the treatments on labour supply response
and the important question of work disincentives that
dominated the debate over the GAI in the 1970s.
Mincome created a panel data experiment, with waves
of interviews occurring over a four-year period. Figure 1
presents the panel data structure of this experiment.
The initial screening interviews narrowed eligibility to
the baseline survey that collected information intended
to populate the experimental design. The enrolment inter-
view was the formal start to the experiment, and the
subsequent follow-up interviews, termed periodic surveys,
monitored labour market interactions and other socio-
economic variables for the households (heads as well as
any member aged older than 15 years). Periodic inter-
views occurred approximately every three months, and
all participants also completed monthly income report-
ing forms (IRFs). Mincome engaged in a very elaborate
and comprehensive data collection process that necessi-
tated extensive logistical and quality controls.
The basic design matrix for the main Winnipeg sam-
ple appears in Table 1. Note that Plan X was deemed too
expensive and never implemented, and because Plan 6
resulted in low uptake at the first payment because of
the high tax back and low basic income, participants
were rolled into Plan 7 after Periodic Survey 1.
4,5
The treatment groups in Mincome began with an
array of three guarantee levels and three tax rates. The
guarantee levels were $3,800, $4,600, and $5,400 for a
family of two adults and two children aged younger
than 15 years, adjusted according to family size for
smaller and larger families.
6
Plan 9 included the control
group that received no payments except an honourarium
for completing the periodic surveys.
The Mincome agreement also included a unique pro-
vision for a pilot project in Dauphin, where everyone
would be eligible for a single specific treatment (Plan 3).
This saturation site did not involve random allocation
of treatments or a control group but was advertised as
open to all who qualified. The intent was to provide
additional evidence of ‘‘administrative and community
issues resulting from a less artificial environment’’ than
the ‘‘dispersed samples’’ (Hum and Simpson 1991, 45)
of a small fraction of Winnipeg and rural Manitoba com-
munities.
7
Thus, the Dauphin saturation site was initially
seen as a way to test uptake if a GAI were universally im-
plemented. Other social scientists viewed the Dauphin
site as an opportunity to study community and family
dynamics when everyone was eligible and awareness of
Figure 1: Panel Data Structure of the Manitoba Basic Annual
Income Experiment
Source: Mason (2016)
Table 1: Mincome Guarantee Levels and Tax Rates by Plan Type (Just Before Periodic Survey 1): Main Winnipeg Sample
Tax Rate on Total Income
Guarantee at enrolment, $ 35% 50% 75%
3,800 Plan 1 (n¼55) Plan 3 (n¼61) Plan 6 (n¼49)
4,600 Plan 2 (n¼67) Plan 4 (n¼70) Plan 7 (n¼29)
5,400 Plan X Plan 5 (n¼56) Plan 8 (n¼45)
Plan 9 (Control; n¼94)
Note: Mincome ¼Manitoba Basic Annual Income Experiment.
Source: Hum, Laub, and Powell (1979), Laub, Metcalf, and Sabourin (1979), and authors’ calculations from Mincome payments data
(Appendix B).
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 87
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
who was receiving the support was high. In contrast,
participants in the Winnipeg and rural dispersed sites
were essentially anonymous.
The guarantees were adjusted for inflation, which
was substantial during the period of the experiment.
8
The tax rates at which the income guarantee would de-
cline with family income were 35 percent, 50 percent,
and 75 percent, resulting in nine prospective plans. In
the design process, the most generous plan (Plan X;
$5,400, 35 percent) was dropped because cost considera-
tions limited its policy relevance (Hikel et al. 1974, 27).
As mentioned, the least generous Plan 6 ($3,800, 75
percent) had insufficient take-up and was folded into
a more attractive Plan 7 ($4,600, 75 percent) after the
first year of the experiment (Hum, Laub, Metcalf, and
Sabourin 1979, 46), leaving seven treatment plans plus
a control group to be assigned across the urban dis-
persed (Winnipeg) site.
9
The smaller rural dispersed
sample was restricted to a control group and a single
treatment, Plan 3 ($3,800, 50 percent), aligned with the
Dauphin saturation site treatment. The rural dispersed
control group could therefore provide a useful com-
parison group for the Dauphin treatment sample under
certain conditions.
10
We discuss this further in the fifth
section of the article.
The first stage of the experiment involved the identi-
fication (screening) of appropriate low-income families
and their allocation to a treatment or a control group.
This was a relatively simple exercise in Dauphin, where
every family resident in that town as of 1 July 1974 was
eligible for a single plan, subject to age and normal in-
come ceilings of age 64 years and $9,000 for a family of
four, and 586 families were eventually enrolled accord-
ing to payments data (Crest et al., 1979, 7).
11
For the dis-
persed Winnipeg and rural samples, a screening survey
followed by a baseline survey was conducted to collect
data on earnings, work history, and income for families
and unattached individuals. Information on earnings
and work history from the baseline survey were used
to estimate normal or permanent income to identify
admissible low-income families and to determine initial
payments (Hum, Laub, and Powell 1979, 47–48). Admis-
sible families excluded those with an average annual
income, adjusted for family size, that exceeded $13,000
for a family of four (equivalent to $64,600 in 2016) plus
families with a head older than age 57 years, a language
barrier, or disabilities; families in the armed forces or a
religious order; or families in rooming house arrange-
ments (Sabourin 1979, 5).
Because allocation of families to treatment plans or
the control group had implications for the variation in
outcomes across plan features and for program costs
that depend on family income as well as treatment, the
experiment used an assignment model developed by
Conlisk and Watts (1979) for the New Jersey negative
income tax experiment. The model, which introduces
some data complexity, as we discuss in the fifth section,
determined the optimal allocation of admissible families
to the experimental treatment and control groups to
achieve accurate estimation of a response (such as labour
supply) given the design features (such as the plan
features and the family’s normal income), a budget con-
straint, and other sampling and policy constraints. A
total of 1,255 families were allocated to the experiment
at the urban and rural dispersed sites, 704 to the treat-
ment plans and 370 to the control group in Winnipeg,
103 to the treatment plan and 78 to the control group
in the rural sample (Hum, Laub, Metcalf, and Sabourin
1979, 63–64).
An important feature of all the income maintenance
experiments was the reliance on self-reported data. Sur-
veys, administered in person, were the prime data col-
lection methods, because the experiments predated the
availability of administrative files. This is in sharp con-
trast to the current potential to access administrative
information such as tax and health records to support
policy experimentation today.
Families enrolled in the experiment
12
were interviewed
three times a year for three years, for a total of nine
periodic surveys in addition to the baseline survey, as
depicted in Figure 1. The periodic surveys collected a
wide variety of information. Although labour supply
was uppermost in the minds of policy-makers and a
top priority for Mincome, the surveys also collected
extensive information on related labour market topics—
earnings, job search, human capital investment, un-
earned income, disability and unemployment benefits,
and leisure time use—as well as assets and liabilities,
family composition, geographic mobility, housing arrange-
ments, and socio-psychological factors, including attitudes
toward work, perceived standard of living and quality
of life, and housing satisfaction.
13
During the first year of the experiment, as the extent
of attrition and non-participation became apparent, it
was determined that a supplementary sample would be
necessary to provide sufficient observations in certain
treatment cells, primarily the lower income cells for
double-headed families in Winnipeg. The supplementary
sample enrolled an additional 196 families across the
treatment cells and 97 families in the control group
(Hum, Laub, Metcalf, and Sabourin 1979, 65), raising
total enrolment in the Winnipeg dispersed sample to
900 treatment and 467 control families, or 1,367 families
in total. The Supplementary Sample lagged the original
sample by one year because the baseline survey for the
Supplementary Sample was completed at the end of
1975 and the first periodic survey for the Supplementary
Sample roughly coincided with the fourth periodic sur-
vey of the original sample shown in Figure 1.
14
88 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
Table 2 presents summary baseline statistics for the
experimental participants (those in both treatment groups
and the control group) for the three sites (Dauphin,
Winnipeg, and rural sites). Remember these are 1974
dollars.
The payments mechanism was a critical feature of
the experiment administered through a distinct payment
agency, Mincome Manitoba, with offices in Winnipeg
for the two dispersed samples and in Dauphin for the
saturation sample (Crest et al. 1979, 10–15). Beyond being
an experiment, Mincome was an incorporated entity.
Those who agreed to participate went onto the Mincome
‘‘payroll,’’ receiving a monthly income corresponding to
the guarantee plus any additional income from pensions,
work, and so forth. Payments from Mincome Manitoba
were provided monthly on the basis of completion of
an IRF.
15
Those who were receiving social assistance
and other income support benefits before the experiment
were assured that their monthly income would not drop
and that their eligibility for these benefits would remain
if they subsequently decided to withdraw from the
experiment.
On the basis of family composition (number of heads
and number and age of dependents), income (Y) and
wealth (W) recorded on the IRF, and payments Pto
those in treatment cells identified by an experimentally
assigned guarantee Gand tax rate twere calculated
according to the formula
16
P¼GtY rW:ð1Þ
Although intended to be a simple benefits calculation
based on reported income, consistent with the classic
negative income tax design, adjustments to Equation (1)
were necessary. First, any federal or provincial income
support payments received—such as Unemployment
Insurance, Canada Pension Plan or Old Age Security
benefits, and War Veterans Allowances—were deducted
from the monthly payment. Second, there was a fixed
payment adjustment at rate rfor family net worth across
all treatments, such that net worth below $3,000 was not
taxed, but net worth between $3,000 and $10,000 was
taxed at 4 percent, net worth between $10,000 and
$30,000 was taxed at 8 percent, and net worth in excess
of $30,000 was taxed at 16 percent.
17
Third, there was
an annual reconciliation for overpayments and under-
payments, including regular federal and provincial taxes
owing, based on the calendar year (Hum, Crest, and
Komus 1979, 26–28).
18
Mincome Manitoba prepared the
tax returns of all treated families in the experiment.
A family in a particular treatment cell, assigned a
specific guarantee on the basis of family composition
and a specific tax rate, would receive payments as long
as its eligible income fell below the break-even level B
given by the ratio of the guarantee to the tax rate, G/t.
Thus, a family with fluctuating income might have a
Mincome payment one month but not the next when
income exceeded B. This potential for a fluctuating level
of support distinguishes negative income tax forms of
the GAI from other approaches that make fixed pay-
ments.
19
Other changes in payments could occur because
of changes in family composition, which would alter the
reporting unit and the level of treatment. For example, a
double-headed family that split into a lone parent family
and an unattached individual would retain its treatment
cell, but an adjustment to the guarantee level would be
made to reflect the change in family composition of the
two new families. Family unions, such as the case of
an unattached individual or lone parent joining with
another lone parent in a specific treatment plan, would
be able to form a new unit with the specified treatment
plan after a suitable waiting period (Hum, Crest, and
Komus 1979, 12). Moreover, some families did not com-
plete the experiment, and their payments were stopped.
20
That, in the proverbial nutshell, was the Mincome
experiment from conceptualization to completion. Those
seeking further details should consult the Mincome ex-
perimental papers listed in Appendix A.
Demise and Resurrection of Mincome
The original proposed budget was $17 million, with the
federal government responsible for three-quarters of it.
Canada and Manitoba were jointly responsible for all
issues related to research, and Manitoba assumed the
remainder of the cost and sole responsibility for opera-
tional matters (data collection and administration of
payments). Because Canada had no experience with
social experimentation and only a distant view of the
unfolding US trials, it would hardly be surprising if the
budgeting exercise rested on shaky ground. What is
more surprising today, when provisions for cost over-
runs are routine even for commonly occurring public
projects, is that the $17 million became a hard-and-fast
number as the experiment proceeded and the full cost
Table 2: Selected Participant Attributes at the Baseline
Survey
Attribute Mean
Average Age Male Head, y 36.5
Average Age Female Head, y 34.5
Households Owning Their Home, n801
Average Home Value, $ 8,000
Average Rent Paid (Monthly), $ 92
Number Single Headed Households 396
Average Family Income, $ 7,520
Median Family Income, $ 6,270
Source: Authors’ calculations based on Mincome baseline summary
file (see Appendix B).
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 89
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
of completing all its features in an inflationary envi-
ronment became apparent. As a consequence, Mincome
was scaled down to preserve its highest priority functions,
as we discuss later.
In 1978, the federal government expressed concerns
about the cost of the project. Other factors may have
weighed in its decision to withdraw from the experi-
ment because the research began to incorporate a range
of social and psychological substudies that diverged from
its original purpose. Manitoba had also become less keen
to maintain the experiment, and with the loss of its major
financial partner, it agreed to terminate data collection
and analysis. The decision to terminate resulted in a lay-
off of most of the staff; at its peak, Mincome employed
about 200 full- and part-time employees, most engaged
in payments administration, interviewing, and question-
naire processing. This termination was very stressful for
all Mincome staff, who were quite passionate about the
experiment and the GAI as a policy.
In the final year, the remaining staff, primarily re-
search and computer programmers, completed the tech-
nical documentation and file digitization to support
future analysis. It is important to note that during the
operational phase from 1974 to 1978, most Mincome
staff became engaged in the logistics of income main-
tenance, tax reconciliation, and managing a complex
survey research process to gather detailed information
on household behaviour and attitudes. Very little analysis
was completed on the central labour supply hypotheses
that formed the original purpose of the study. By 1979,
Mincome staff had completed technical documentation
and had coded the selected survey information electron-
ically.
21
The technical documents, which are essential
reading for anyone wanting to use the longitudinal data,
are listed in Appendix A. The hard-copy records, consist-
ing of completed surveys and voluminous payment and
taxation records as well as computer tapes, were trans-
ferred to the federal government. The state of transferred
hard-copy records remains chaotic to this day.
The fact that no analysis emerged after a multimillion-
dollar expenditure was clearly an embarrassment to the
government. In 1980, discussions started between the
federal government and the University of Manitoba to
develop a process to retrieve the Mincome data and
make it accessible to the research community. These
discussions culminated in the creation of ISER at the
University of Manitoba in 1981. The federal government
provided five years of funding to ISER, in part to restore
the Mincome data and promote its use to the research
community.
22
Work started immediately on the task of
converting the tape format (PDP-11) used by Mincome
to a more conventional IBM 360 format. It immediately
became apparent that although the technical documen-
tation (see Appendix A) had been completed to a very
high standard, the computer files were undeveloped
and not readable on mainframe equipment. Code books
were rudimentary and aligned poorly with the elec-
tronic fields. A team from the Department of Computer
Science at the University of Manitoba provided the key
support that eventually cracked the data codes and re-
constructed the Mincome survey data into an electronic
format usable by researchers.
By 1983, all the relevant data had been reconfigured
with detailed code books to support analysis. ISER under-
took a communications process through 1984 to outline
and promote the research possibilities to the research com-
munity. Between 1985 and 1991, a variety of research
studies were completed, but this was followed by a quiet
period, with occasional studies throughout the years.
By the mid-1980s, Mincome data development ceased.
First, the ISER evolved into a for-profit research com-
pany, University of Manitoba Research, wholly owned
by the university. This new entity had no responsibility
for the Mincome digital files, technical documentation,
and working papers, which were transferred to the Uni-
versity of Manitoba Library Archives. Second, interest
in a GAI waned. With the cessation of the US income
maintenance experiments and the completion of key
studies from Mincome, the policy community concluded
that implementation of this form of income support was
unlikely in the foreseeable future. Third, the 1980s and
1990s witnessed relatively high growth in real incomes
in North America (Brzozowski et al. 2010), and the issue
of income inequality receded as a policy of interest.
23
By 1990, the Mincome project had effectively ceased.
At some point, the libraries at the University of British
Columbia and University of Toronto secured the Mincome
data and code books. Although the presence of the data
and documentation at these sites may have served to
remind the research community of the Mincome experi-
ment and the potential value in its data, little research
has resulted. In Spring 2016, the University of Manitoba
Libraries ‘‘repatriated’’ the Mincome data and supported
their installation in a Dataverse system.
24
All the relevant
documentation and digitized data are now available
in one location, accessible to the research community
worldwide.
25
Mincome has come through two cycles of rebirth
since the experiment was terminated in 1979. First, the
data were rehabilitated by ISER, supporting the realiza-
tion of some key research on labour supply before
slipping into obscurity in the early 1990s. In this version,
the data remained on large data tapes and required
mainframe computing to support the processing, but
they were not available to the research community after
1990. Second, in the past five years, the data have been
translated into Excel format, which substantially improves
access. The installation of Mincome in the Dataverse
90 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
system at the University of Manitoba means that re-
searchers anywhere have access to the data and all
relevant documentation, and they can process the infor-
mation conveniently. The fifth section of this article
summarizes the access point for Mincome in its current
incarnation.
Mincome Research
Lack of resources for research in the Mincome budget
did not shelve analysis of the experimental data. Research
flowed steadily, if more slowly and with more limited
exposure, after the development of the Mincome data-
base at the University of Manitoba’s ISER. In this
section, we outline the published research to date in
three parts that correspond roughly to the chronological
order of the research undertaken. First, we review re-
search using the Mincome baseline surveys, which con-
stitute the earliest and most straightforward data from
the experiment. We then examine the research using the
panel data from the Winnipeg dispersed sample, which
concentrated on addressing the labour supply response
issues that were an important priority in the design of
Mincome. Finally, we consider the more recent research
related to the Dauphin saturation site and the rural dis-
persed sample.
As a pre-experimental cross-sectional survey, the base-
line survey provided the earliest and most easily under-
stood and analyzed data from Mincome. Although the
data do not pertain to the experiment per se, they are
part of the Mincome database and provided a rich set
of details on labour supply and work histories, as well
as a variety of other economic and sociodemographic
topics, at a time when such data were in short supply.
26
Mason (1984) outlines the income and wealth data avail-
able from the Mincome baseline, the distribution of
net worth across family types (double headed, single
headed, and single individuals), and the relationship
between income and wealth for these family types. He
finds an inverted U–shaped relationship with age that
is consistent with the life cycle model of household
wealth accumulation for double-headed families in both
the urban and the rural samples but not for single-headed
(lone parent) households. The results are stronger when
home ownership is included in the calculation of net
worth, because housing is the dominant form of wealth
accumulation for the lower income sample of families in
Mincome.
Ehnes and Simpson (1983) use the labour supply and
work history data in the Mincome baseline survey to
analyze the labour supply behaviour of men in double-
headed families. They estimate a standard family labour
supply model of hours worked that depends on the
wage of the male and female heads, unearned income,
age, education, number of dependent children, and race.
Prescott, Swidinsky, and Wilton (1986) also use the rich
Mincome baseline data to study the labour supply be-
haviour of women in lower income households. They
estimate a simultaneous model of wage offers and labour
supply that allows for non-random selection of workers
and non-workers. Wage offers depend on education,
age, and location (Winnipeg, Dauphin, or rural), whereas
average weekly hours worked depend on own wage,
husband’s wage (if household is double headed), other
income, wealth, children (three age categories), and loca-
tion. In addition, because the Mincome baseline captures
child care costs, Prescott et al. subtract these costs from
the gross female wage and find that this net wage
variable improves model performance. Consistent with
theoretical expectations, the female wage is positively
correlated with hours worked, and other income, presence
of preschool children, and husband’s wage (for double-
headed households) are negatively correlated.
At the heart of the justification and design of Min-
come and the US experiments was the goal of estimating
the labour supply response, or work-disincentive effects,
associated with a negative income tax program involv-
ing an income guarantee and a benefit tax rate. It was
therefore crucial to analyze the panel data for the dis-
persed samples, which provide rich information on labour
supply response in the context of a series of treatments
with differing guarantees and linear tax rates as well
as a control group. Using standard academic resources
and grant support, including the Social Sciences and
Humanities Research Council and the Economic Council
of Canada, Hum and Simpson (1991, 1993) provide an
analysis of labour supply response in the Winnipeg dis-
persed sample. We consider this research in two stages,
first considering the analysis of the direct impact of
the experimental treatments and then discussing the
analysis of labour supply elasticity estimates provided
by the Winnipeg data.
The virtue of random allocation of participants to
treatment and control cells is that it is possible in prin-
ciple to obtain direct estimates of the impact of the
experimental treatment in comparison with those receiv-
ing existing public income support in the control group.
The basic model explains the change in labour supply
(hours worked), Dh, between some experimental period
and the pre-experimental baseline as a function of the
set of experimental treatments T, appropriate control
variables Dxthat are not fixed through time, and random
error Dxin an analysis of variance framework (Hum and
Simpson 1993, S276–77):
h¼TbTþxbþð2Þ
The experimental treatments ensure variation in
guarantee levels and tax rates over time for the same
individual as well as across individuals, depending on
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 91
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
their experimental assignment, to produce more efficient
estimates of the experimental effects b
T
. The control vari-
ables might include time effects and variables to capture
potential bias arising from the Conlisk-Watts assignment
model (Keeley and Robins, 1978) or participation and
attrition (Robins and West, 1982).
27
Using a variety of
specifications, Hum and Simpson (1993) find the overall
effects of the Mincome experiment to be modest in the
Winnipeg sample—1 percent for men, 3 percent for
wives, and 5 percent for unmarried women—and statis-
tically insignificant when time effects are controlled.
These effects are smaller than the weighted average of
effects from the US experiments—6 percent for husbands,
19 percent for wives, and 13 percent for single mothers
(Burtless 1986; Robins 1985)—although only the effects
for the Seattle and Denver experiments are statistically
significant, and many of the estimates from the New
Jersey and Gary experiments cited in these reviews are
comparably small.
28
Comparison of the experimental re-
sults along these lines is complicated because treatments
differ between the US and Canadian experiments. These
summary comparisons can only convey some sense of
the impact of a typical or average experimental plan
on hours worked rather than the effect of any specific
guaranteed income treatment.
Moreover, these direct or non-structural analyses of
experimental impact cannot be compared with the non-
experimental evidence on work disincentives, which is
based on analysis of structural labour supply models.
Keeley (1981) argues that general social policy analysis,
and assessment of the impact of any proposed income
maintenance policy on labour supply, will rely on struc-
tural labour supply models that can be estimated with
either experimental or non-experimental data. In partic-
ular, the experimental evidence can be used to estimate
the impact on labour supply of variation in unearned
income and after-tax wages arising from the treatment
guarantee and tax rate. That is, for a change in after-tax
wages of Dwand a change in unearned income of Dy,
the standard labour supply model implies that one can
represent the experimental treatment variables in Equa-
tion (2) by variables that reflect the change in wages Dw
and the change in income hDwþDy, where the latter
term consists of the change in unearned income com-
bined with the change in earned income that arises
from the wage change measured at pre-experimental
hours worked h(Hum and Simpson 1993, S277):
29
Dh¼DwsþðhDwþDyÞyþDxbþD: ð3Þ
Although this formulation can be used for any panel
data that capture hours worked, wages, and other family
income, the experimental evidence has particular advan-
tages in generating substantial exogenous changes in
after-tax wages and other income as well as imposing
linear tax regimes for those treated that simplify and
potentially improve estimates of the crucial behavioural
parameters, the own compensated wage effect h
s
and
the income effect h
y
.
Hum and Simpson (1993) estimate variants of Equa-
tion (3) that include changes in the number of preschool
children and time effects.
30
As might be expected from
the non-structural estimates, the estimated labour supply
elasticities are small and often insignificant. These results
are not dissimilar to those obtained for the New Jersey
and Gary experiments, although they show that US
elasticity estimates were generally larger and more often
consistent with theoretical expectations. The experimental
evidence, along with prominent critiques by Mroz (1987),
Jakubson (1988), and MaCurdy, Green, and Paarsch
(1990), have led to both a sharpening and a lowering of
labour supply estimates that has endured to this day.
31
Although concerns about the work disincentives of
a guaranteed income program were paramount, there
were also fears that the program might provide adverse
incentives to break up families. These fears were fuelled
by Groeneveld, Tuma, and Hannan (1980), who found
that those receiving the guaranteed income treatments
in the Seattle and Denver experiments experienced a
significantly higher marital dissolution rate relative to the
control group. Hum and Choudhry (1992) and Choudhry
and Hum (1995) argue that there are counteracting effects
of a guaranteed income plan and that the issue is ulti-
mately an empirical one. They apply path and survival
analysis to the Mincome data and find contrary evidence
that the treatment group experienced slightly greater
marital stability than the controls, results that align better
with the reanalysis of marital dissolution in the Seattle
and Denver experiments by Cain and Wissoker (1990).
Bennett (1986) analyzes how well participants under-
stood the Mincome program, based on a small sample
of respondents to specific questions about knowledge of
the impact of the guarantee, tax rates, and break-even
income applicable to their circumstances. He finds that
most participants who completed the experiment were
generally aware of the impact of the Mincome parameters
and how they affected their economic situation.
The saturation site in Dauphin remains a unique
feature of Mincome. It was introduced after the main
project was underway and was intended to assess admin-
istrative issues and community effects in a fashion not
possible from the dispersed experimental sample design
in Winnipeg and rural areas.
32
Although researchers
ignored these data for the analysis of labour market im-
pacts, the Dauphin experience has figured more promi-
nently in recent research and media reports, largely
through the innovative work of Evelyn Forget (2011,
2013). Her work centres on the impact of a GAI on
health behaviour and outcomes. Because Mincome col-
lected no health-related data, even self-reported health
status, she relies on a quasi-experimental approach based
92 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
on administrative health records to estimate the impact
of the Mincome program in Dauphin. The underlying
assumption is that the Dauphin saturation site and all
its inhabitants can be used as a treatment area and con-
trasted with counterfactuals drawn from the remainder
of the population. Using administrative health records
spanning the Mincome payments period, Forget com-
pared health behaviour and outcome variables of indi-
viduals residing in Dauphin and selected comparison
areas. Propensity score matching refines the data points
within the treatment group and the comparison group
to improve the orthogonality between treatment and
controls. Forget then estimates difference-in-differences
regression models of the effect of residing in Dauphin,
as opposed to residing in the comparison areas, for hos-
pitalization rates and physician claims overall and for
the subcategories of accidents and injuries and mental
health diagnoses. This quasi-experimental approach esti-
mates that Mincome accounted for an 8.5 percent reduc-
tion in hospitalization rates, concentrated in accidents
and injuries and mental health diagnoses, and a reduc-
tion in physician claims for mental health disorders.
Forget (2011) also followed up on evidence from Hum
and Simpson (1991) of larger work disincentives for
tertiary workers by looking at high school completion
rates to see whether young men were leaving the work-
force primarily to invest in human capital. She also con-
siders possible impacts on fertility, birth weight, and
family dissolution but finds no evidence of a Mincome
effect. Forget’s research has served both to highlight the
Mincome experiment and to raise the possibility that a
GAI could result in important health outcomes.
33
Calnitsky (2016) and Calnitsky and Latner (2016) have
recently used the data gathered from Dauphin and the
rural dispersed sample to investigate community attitudes
and community effects on labour supply. Calnitsky (2016)
concludes that Mincome was viewed differently, and
decidedly more favourably, than welfare because it did
not single out or stigmatize participation, a result that
he suggests should strengthen social acceptance of a
guaranteed income program. Calnitsky and Latner (2016)
use the Mincome baseline and payments data to analyze
the impact of the Dauphin treatment on labour force
participation compared with the dispersed rural sample.
Their basic difference-in-differences analysis estimates
the total Dauphin treatment effect as an 11.3 percent
decline in participation in Dauphin relative to a rural
control group. The Dauphin social interaction effect is
estimated as the difference between the Dauphin treat-
ment effect and the rural treatment effect, which is 3.2
percent, or about 30 percent of the total Dauphin treat-
ment effect.
Using the Mincome Database
University of Manitoba Libraries now maintains a Data-
verse system for all Mincome-related materials includ-
ing the technical manuals prepared by Mincome staff in
1979, working papers prepared by ISER between 1982
and 1986, the code books for the six Mincome data files
that are available for analysis, and the Mincome data in
Excel format. Appendix B presents an overview of the
Mincome code books and data files held at the University
of Manitoba, along with specific links to the websites.
The most important document to consult before proceed-
ing with any analysis is the Mincome User Manual. This
document offers important information on the experi-
ment and summarizes the 13 technical manuals.
By far the easiest data set to use is the baseline set.
34
These data are from a cross-sectional survey that clearly
separates the three sites, with interviewing occurring
over a relatively contiguous time frame of six months.
The remaining data sets are all longitudinal files, featur-
ing 11 recurring sets of variables corresponding to the
11 survey periods. The data are all in Excel files and,
aside from the baseline data, present all of the periodic
surveys. A unique family number serves as the key to
concatenating the data sets.
Those seeking to use the data will need to consider
four issues:
1. Variable construction is complex and pertains to
institutional arrangements in 1974. For example,
both the provincial and the municipal govern-
ments delivered social assistance at that time. This
arrangement has since been amalgamated into a
single provincial delivery system. Users need to
be able to step back and interpret variables in the
context of the times.
2. Second, the longitudinal files feature recurring sets
of variables corresponding to each survey. It is
important to note that the time between surveys
may vary for any particular respondent, and the
surveys overlapped. In other words, two respond-
ents could be interviewed simultaneously for two
successive surveys. Mincome staff endeavoured
to maintain separation and fixed time sequencing
of interviewing, but the vagaries of field opera-
tions interfered with these plans to some extent.
3. Although we have only mentioned this in passing,
researchers will need to integrate the supplemen-
tary and main samples. Modecai Kurz, who pro-
vided an assessment of the Mincome sample (Kurz,
1979), considers the inclusion of the supplementary
sample as essential to supporting efficient estima-
tion based on the Mincome data.
4. Perhaps the most challenging aspect of the analysis
of the Mincome panel data is potential sample
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 93
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
selection bias. In many ways, the data issues in
Mincome are similar to those of the US experiments
of the time and other non-experimental panel data
that have been collected. Yet, although the rates
of non-completions, refusals, and attrition between
Mincome and the US experiments are similar
(Kurz, 1979), less attention has been paid to
sample selection bias arising from attrition in the
Mincome research than in some of the US ex-
periments, and new analytical techniques have
emerged. We now review this specific challenge
in a little more detail.
In general, sample selection bias in longitudinal experi-
ments or surveys can arise before or during the experi-
ment. Before data are collected, researchers must carefully
select a sample of participants, and some of these partic-
ipants may refuse, leading to possible bias from non-
coverage at the sample selection stage. Sample selection
bias due to noncoverage may be the fault of the experi-
mental designers, or it may arise from the behaviour of
participants if they refuse to be included in the study.
Sample selection bias at the outset, due to non-
coverage, can arise if those selected and initially enrolled
in the experiment are not a fair representation of the
population. Avoidance of this issue requires a carefully
designed experiment. The Mincome experiment has been
lauded for its careful and meticulous experimental design
(Hum, Laub, and Powell 1979; Kurz 1979), and selection
bias at the experimental design level was thought to
be unlikely. Mincome’s baseline interview was relatively
more complicated and provocative than that of other in-
come maintenance experiments, however, and its refusal
rate of 16.1 percent was high compared with the 6.9
percent refusal rate for the New Jersey experiment (Kurz,
1979). Future income experiments should understand
the trade-off between an additional interview question
and a potential sample selection bias.
During an experiment, sample selection bias may also
arise because of attrition; participants exit experiments
for a variety of reasons, including moves, deaths, and re-
fusals. In longitudinal surveys in general, it is the norm
rather than the exception that participants will drop out
at various stages. As such, the statistical methods for
dealing with the loss of observations in panel data are
now well developed, unlike at the time of the Mincome
experiment. We therefore consider the sources of sample
selection bias and the likelihood of its presence in the
Mincome experiment as well as potential tests and meth-
odological options to address the problem.
The exit of participants, at any stage of the experi-
ment, can be considered a problem of missing data. A
critical consideration in dealing with missing data is
whether the reason for missingness is random or non-
random. The data may be missing completely at random
(Little and Rubin 1987), in which case the problem is
‘‘ignorable’’ (Griliches 1986). In this situation, a simple
approach is complete case analysis, which throws away
incomplete observations. Although this was a common
way of proceeding at Mincome’s inception, it induces
a loss of information that translates to a decrease in
precision and efficiency. Contemporary methods seek to
recover the loss of information by using partial observa-
tions. In panel data, this situation has been dubbed the
‘‘unbalanced panel,’’ as in Baltagi (2013).
It is unlikely, however, that the missing data in Min-
come are missing at random, and many studies involv-
ing similar data would agree, as we discuss later. If the
data are not missing at random, then attention must be
given to the mechanism that causes the ‘‘missingness,’’
otherwise estimation may be biased (and inconsistent).
First, we consider the case of selective attrition, which is
more easily treated because there is more information
on the characteristics of the ‘‘attriters’’ than on those
who exit the experiment at the screening or baseline
survey.
As with the other income maintenance experiments,
there was considerable attrition in Mincome (Sabourin,
1979). The first year of Mincome saw 17.6 percent attri-
tion resulting from refusals and 3.3 percent resulting
from moves in Winnipeg. In the second year, 8.1 percent
attrition resulted from refusals and 4.1 percent resulted
from moves (Kurz, 1979). In total, 224 respondents of
591 left the experiment over the 11 surveys, with a similar
pattern across the other sites. Although these attrition
rates are higher than those of the other income main-
tenance experiments of the time, they are comparable to
the attrition rates in many other familiar surveys, such
as the Panel Study of Income Dynamics, the Current
Population Survey in the United States, and the British
Household Panel Survey.
35
Recall that if attrition is random, then the missing
observations are ignorable. One test for ignorability is a
test of a difference in characteristics between attriters
and non-attriters. Kurz (1979) attributes attrition rates
resulting from the payment package, the enrolment pro-
cess, mobility, and the organization of Mincome and
argues that there is no clear ‘‘bias-creating pattern of
attrition and refusals’’ (23), but the literature on attrition
bias was not well developed at that time. A preliminary
logistic regression of attrition on socio-economic charac-
teristics using the baseline survey suggests that attrition
is indeed non-random and is potentially non-ignorable.
For example, the probability of attrition appears to be
higher the more adults (excluding heads) there are in
the family and lower the more children there are aged
younger than six years. Not surprisingly, attrition prob-
ability also appears to be lower the more generous the
benefit plan is.
94 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
There has been considerable research on dealing with
sample selection for panel data, beginning with Hausman
and Wise (1979).
36
The Hausman and Wise paper was
motivated by the Gary Income Maintenance Experiment
and is pertinent for modelling the Mincome data, pro-
viding a basis for both testing and estimation procedures.
The literature stemming from Hausman and Wise now
provides several options for testing and modelling attri-
tion bias. For example, Verbeek and Nijman (1992) and
Wooldridge (1995) provide methods for the detection
and correction of attrition bias. Nicoletti (2006) argues
that testing for attrition bias is valid only when addi-
tional information is available on the reason for missing-
ness, which is the case for the Mincome data because of
the information contained in the baseline survey. We
refer anyone interested in a more extensive treatment of
this topic to Baltagi (2013).
The supplementary sample of 293 households presents
an additional opportunity to test and correct sample
selection bias resulting from attrition (Hirano et al., 2001).
In anticipation of attrition, contemporary longitudinal
studies use refreshment samples, and the data become
a rotating panel.Kurz (1979) formally tests for demo-
graphic differences between the supplementary and
original samples and concludes that the two samples
come from the same population. Hence, methods for
rotating panels (Baltagi, 2013) incorporating the supple-
mentary sample should be applicable to the Mincome
panel data.
We now turn to the other possible sources of sample
selection bias: non-completions and refusals of the screen-
ing interview and non-completions, moves, and refusals
of the baseline interview. Only 71.0 percent of screening
and 54.5 percent of baseline interviews were completed.
This is a potentially serious problem because the methods
for correcting sample selection bias rely on at least some
information about the missing participants. Furthermore,
the validity of tests for attrition bias will rely on the
information contained in the baseline interview. Kurz
(1979) argues that the reasons for incomplete or absent
screening and baseline interviews and the reasons for
attrition are similar, in which case information on the
attriters can be used to correct sample selection bias
resulting from not only attrition but also incomplete
and absent screening and baseline interviews. This pro-
vides another possible avenue for future research using
Mincome.
Although the primary concern is correcting for partic-
ipants who are absent from the sample (unit non-
response), a secondary concern is dealing with incomplete
interviews (item non-response). Discarding incomplete
interviews would at the least be inefficient but could re-
sult in biased and inconsistent estimation for the same
reasons associated with non-random attrition. For the
incomplete interviews to be usable, either the missing
values need to be imputed or the complete observations
need to be weighted. Fortunately, the literature on item
non-response has flourished since the time of Mincome,
and many methods are available for handling it. Two
popular broad approaches are maximum likelihood and
multiple imputation. We refer anyone interested in a
definitive overview of this issue and its treatment to
Allison (2001).
Finally, it is important to note that the presence of
sample selection bias depends critically on the estima-
tion goal. If the reasons for attrition are independent
from, or weakly related to, the outcome of interest, bias
will be non-existent or minimal. Consider the specific
goal of estimating the labour supply response to differ-
ent guarantee–tax combinations. The impetus for the
Hausman and Wise (1979) investigation of attrition bias
was the idea that participants who benefit more from
the experiment are less likely to exit, thereby potentially
overstating the treatment effect. They found evidence of
selective attrition but found that it had little effect on
labour supply response in the Gary experiment. Simi-
larly, Robins and West (1986) find little effect of attrition
bias in the Seattle and Denver experiments.
In other contexts, attrition bias is minimal as well.
Lillard and Panis (1998) find selectivity in attrition be-
haviour in the Panel Study of Income Dynamics, but
the resulting biases are minimal. A similar juxtaposition
between attrition selectivity and minimal attrition bias
has been found in a variety of panel data (Fitzgerald,
Gottschalk, and Moffitt 1998; van den Berg and Linde-
boom, 1998; Zabel, 1998), and there is very little research
to suggest otherwise. Thus, there is certainly a precedent
for ignoring sample selection bias and treating the Min-
come data as an unbalanced panel, but we encourage a
thorough treatment of the data using modern empirical
methods that would entertain the possibility of sample
selection bias.
Although there are well-developed statistical methods
for dealing with sample selection bias, these methods
can only go so far. Future income experiments should
recognize that the more obtrusive the data collection,
the more likely it is that participants will refuse the
experiment at various stages and that these refusals will
be non-random. The risk of each interview or survey
question should be carefully weighed, although this
risk may be mitigated through the use of administrative
data, as mentioned in the following section. A blanket of
questions may increase the breadth of research possibil-
ities, but it may also invalidate all of them as a result of
sample selection bias.
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 95
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
Old and New Lessons from the Mincome
Experiment
Mincome offers researchers important lessons about large-
scale social experiments. We separate these lessons into
conceptual, operational, and strategic categories.
Conceptual Lessons from Mincome
Many studies have examined the challenges of social ex-
perimentation in the wake of the income maintenance
experiments that included Mincome.
37
A core issue is
that participants know they are part of an experiment
and will confront their pre-experimental environment at
its conclusion. Mincome participants were aware that
program benefits would only last three years, which
limits what can be learned about the long-term labour
supply responses associated with a permanent GAI.
38
Thus, measured labour supply response may be either
understated or overstated. Individuals who elect to
withdraw from the labour market, typically those close
to retirement or who require major skills upgrading or
who have become disabled, may decide that the GAI
is an attractive option. Alternatively, as long as the
guarantee is not overly generous, recipients may elect to
upgrade skills and try to make the leap to well-paying
employment. Short-term income maintenance experi-
ments or pilots offer limited insight into these opposing
reactions.
39
In addition, the net outcome of individual
labour supply responses may have effects on the wage
structure of the labour market. For example, if low-
skilled workers can leave part-time and informal work,
wages for these occupations may rise, and this develop-
ment would likely have a ripple effect on higher skilled
occupations. The dispersed sampling design of the Min-
come and US experiments offers no insight into these
general equilibrium effects. Similarly, there is no infor-
mation in Mincome or the other income maintenance
experiments that would allow researchers to assess longer
term outcomes, especially important issues around in-
come and health outcomes.
As with the other income maintenance experiments
of the 1970s, Mincome relied solely on survey data to
screen, enrol, and monitor participants. This has the
important benefit of allowing researchers to define vari-
ables precisely to collect the information needed to test
the essential elements of a policy, but reliance on survey
data introduces a range of sampling and response issues
that need to be addressed, as we discussed in the previ-
ous section. Furthermore, surveys require strict attention
to informed consent and confidentiality, which can pro-
voke non-random refusals at the outset.
Fortunately, modern data collection can address some
of these problems. In particular, administrative data play
an increasingly important role. Future social experiments
and pilots to assess social policy could anchor the data
collection in an administrative data set such as health,
social assistance, and taxation records, augmented by
surveys to collect information not otherwise available.
By creating a consolidated administrative data set to
support the screening of participants into treatments
and controls, one can reduce response errors at the start.
Social experiments that use anonymized administrative
files need not engage in informed consent procedures.
Administrative data sets that can support the unobtru-
sive downstream monitoring of the participant should
reduce attrition, which was sufficient in Mincome to
require a supplementary sample to sustain the experi-
mental design.
40
In sum, the conceptual lessons lie in the inherent
short-term nature of a social experiment that seeks to
address long-term responses and the biases that occur
as a result of respondent selection and continued moni-
toring. Increased reliance on administrative data may
mitigate panel losses in future experiments, but it is
inevitable that the limited sample and length of a social
experiment will provide challenges in interpreting the
evidence.
Operational Lessons for Social Experimentation
using Panel Data
Mincome employed about 200 full- and part-time staff
to administer an experiment involving approximately
1,700 households. It was an incorporated entity that
had delegated authority from federal and provincial tax
departments to collect and remit income taxes. The staff
were involved in a range of tasks including monthly
payment delivery, tax remittance, interviews (quarterly
behavioural surveys and monthly earned income track-
ing), human resources (for the experiment itself ), and
research. The experiment cost almost $50,000 per partic-
ipant in 2016 dollars.
Mincome planners were caught somewhat off guard
by the ever-increasing minutiae of screening, enrolment,
creation of a parallel payments and taxation system, and
the complexity of managing periodic surveys. These un-
foreseen costs bumped up against an inflexible budget.
A central lesson from Mincome for future experiments
and pilots is that these are formidably complex and
costly projects. The technical documentation makes it
clear that Mincome researchers were unwilling to com-
promise their commitment to a very high technical and
scientific standard but that the research output suffered.
In resource-constrained contexts, researchers who are
unaware of the strict demands of panel data experi-
ments may be tempted to manage budgets by taking short
cuts and other compromises that reduce data validity and
reliability, sometimes in unknown and undocumented
96 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
ways. An enduring strength of Mincome is the trans-
parency of its methods and the detailed documentation
that supports research, even 40 years after data collection.
Such social experiments result in large data sets.
These data sets require focused management to ensure
the integrity and quality of the information, both for
current analysis and to preserve the records for future
study.
Strategic Lessons
Mincome concluded in 1978 without substantive research
and no remaining budget for research. The technical
documents prepared in 1979 contained hints of results,
but substantive research results only developed slowly
as limited resources and academic interest dictated. In
retrospect, two lessons emerge. First, the budget encom-
passed a high proportion of direct income replacement
payments. Therefore, the net cost of the experiment itself
was considerably lower than commonly cited. If the
annual reports had used net cost as the bottom line of
the experiment, it may have mitigated the politicians’
fiscal concern. Second, and more important, issuing
research results earlier during the experiment may also
have reduced the impression of a cash burn with nothing
to show for it. Only recently has the issue of knowledge
translation emerged as a priority for all forms of scientific
research. Clear communication of results is a specific skill
area, possessed only by select researchers. Future social
experiments must pay attention to knowledge transfer.
It is likely that the Mincome researchers realized the
political need to offer interim results, but several factors
may have impeded such release. First, the unfolding
complexity of the operations may have diverted energy
away from research and the release of results. For exam-
ple, the debate over whether to create a supplementary
sample was intense and persisted for several months,
which undoubtedly sapped research resources. Second,
the Mincome researchers were committed to doing things
better than the prior income maintenance studies in the
United States. A feeling existed within the organization
that Mincome should not only build on the previous
research but also design and execute the experiment to
a demonstrably higher standard. This focus on technical
excellence may have encouraged the researchers not to
release preliminary results that might require amend-
ment later. Third, researchers of the day were concerned
about possible behavioural modification if participants
in the experiment were made more aware of their being
observed, the so-called Hawthorne or observer effect
that might arise from the public release of research before
completion of the experiment. Finally, all the income
maintenance experiments showed that a certain number
of participants reduced their labour supply in the short
run. Leading with this result may have overshadowed
other outcomes and also overstated the longer term
labour supply adjustment. Preliminary results can be
misinterpreted, and revisions in light of better informa-
tion can place researchers in an untenable position, espe-
cially among journalists keen to issue a quick story. In
light of these risks, most researchers will understand-
ably choose a conservative approach to releasing results.
At the same time, politicians and the public justifiably
require transparency for large expenditures. Even if the
net cost per participant were 50 percent of the full cost
($25,000 in 2016 dollars), a new experiment or pilot
launched today and running for three years with 1,000
participants (the bare minimum) could still cost $25
million. If Mincome teaches us anything, therefore, it is
that a communications strategy of releasing preliminary
results is essential to maintaining political and public
support for large projects. Mitigating the risks of fund-
ing cuts requires that expected outcomes be communi-
cated early and placed in context. In the current era,
this would mean using social media and cultivating
journalists known to be able to present complex stories.
It also would mean breaking down the outcomes into
a sequence of initial changes that can be expected to
support longer term outcomes. Such theories of change
support the framing of initial results in the context of
longer term implications.
Concluding Thoughts
The 1960s and 1970s were heady days when solutions to
poverty were believed to be achievable, and social exper-
imentation around a negative income tax or guaranteed
income program was part of the way forward. Although
poverty has endured and the limitations of social experi-
mentation are more apparent, Mincome remains one of
Canada’s premiere social research studies. The technical
manuals attest to the scientific rigour of the research and
data, notwithstanding the challenges of assembling and
interpreting experimental data that we enumerate in the
preceding two sections. The fact that Ontario is proceed-
ing with a pilot to test a basic annual income in 2016
shows that Mincome remains pertinent to today’s policy
context.
Although Mincome reflects the public policy and
socio-economic context of four decades ago, the quality
of the information is high, and the supporting documen-
tation is strong. Basic labour supply assessment has
been completed, but remaining issues that could be
explored include effects on labour supply for specific
age groups and the differential response of rural and
urban residents. The effects on wealth remain another
important area, because measures of inequality focus
on income and do not analyze changes in wealth. Only
a limited amount of research has been done to examine
family stability over time or the changing attitudes of
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 97
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
participants. Mincome offers information on male and
female heads, which supports a thorough analysis of
joint decision making within the household. This is an
important advantage because modern privacy legislation
often limits researchers’ ability to construct households
from administrative data extracts.
At the same time, it must be acknowledged that
Mincome information dates from 40 years ago, and both
labour markets and the broader socio-economic demo-
graphics have changed. Researchers using these data
must be aware of the limits to extending interpretation
across time.
With the demise of major longitudinal data sets such
as the Survey of Labour and Income Dynamics, and with
replacement data slow in coming, Mincome offers the
research community a unique and high-quality data re-
source with which to address still pressing public policy
issues. These data and documentation are available
through the University of Manitoba Libraries Dataverse
(http://dataverse.lib.umanitoba.ca/dataverse/mincome),
and we invite the research community to explore this
unique data set.
Acknowledgements
We thank Derek Hum, the research director of the Mincome
project, for many useful comments on an earlier version of
this article.
Notes
1 For further details on the agreement and the social develop-
ments underlying it, see Hum and Simpson (1991, 43–44).
2 The early history of the design and development of Min-
come remains buried in the administrative documentation
retained by Library and Archives Canada. These docu-
ments have never been subject to systematic review.
3 These were the Seattle Income Maintenance Experiment,
Denver Income Maintenance Experiment, and the New
Jersey Income Maintenance Experiment.
4 Figures represent the assigned cell populations at the first
payment. Attrition in Plan 6 was severe in Period 1, which
prompted the combination of Plans 6 and 7 for Periodic
Survey 2 and subsequent interviews.
5 There is a second dimension of taxation on wealth, but this
was used to adjust payments and was not integrated into
the experimental design.
6 See Technical Report No.1 (Hum, Laub, and Powell 1979,
27). These guarantees correspond to about $18,900, $22,800,
and $26,800 at 2016 prices and constitute about 50 percent,
60 percent, and 70 percent of the Statistics Canada (1975,
19) Low Income Cutoff of $7,655 in 1974 for a family of
four in a community of 30,000–100,000. The guarantee was
adjusted according to family size from 38 percent of the
guarantee for a family of four for the smallest families
(single adults) to 247 percent of this value for the largest
families (Hum, Laub and Powell 1979, 31). The guarantees
were also increased by $600 for a family of four in July 1975
to improve program participation (Crest et al. 1979, 26).
7 For additional discussion of the considerations that led to
the Dauphin site, readers are referred to Technical Reports
No. 2 (Hum, Laub, Metcalf, and Sabourin 1979) and No. 8
(Sabourin et al. 1979). See Appendix A for a list of technical
documentation and working papers pertaining to Mincome.
8 Annual inflation rates in Canada ranged from 7.2 percent
to 10.7 percent between 1975 and 1978 (Canadian Inflation
2014).
9 The assignment across plans is summarized in Sabourin
(1979, 77).
10 See Technical Reports No. 2 (Hum, Laub, Metcalf, and
Sabourin 1979, 10–11, 43–44) and No. 6 (Sabourin 1979,
32–48) for additional discussion of the rural dispersed
sample assignment and Technical Reports No. 6 (Sabourin
1979, 49–61) and No. 8 (Sabourin et al. 1979) for additional
discussion of the saturation site sample in Dauphin.
11 Technical Report No. 8 (Sabourin et al. 1979) reports initial
enrolment of only 322 and explores the reasons for low
participation throughout the experiment at the Dauphin
saturation site. Kurz (1979, 27–37) also discusses participa-
tion issues in Dauphin and estimates that at least 1,400
families were eligible for Mincome payments. He notes that
the low participation rate occurred even though families did
not need to be selected by Mincome because Dauphin was a
saturation site, and any resident could self-enrol at any time.
12 Enrolment procedures are discussed in Technical Report
No. 4 (Crest et al. 1979, 30–38).
13 Summary tables of questionnaire content and periodicity
are provided in Rasmussen, Anderson, and Wright (1983,
18–19). A more detailed account of data development and
availability is provided in the next section.
14 Because the periodic survey in the original sample was not
stabilized until its third rendition, there are some differ-
ences in coverage between the first two periodic surveys
in the original sample and those in the supplementary
sample (Rasmussen, Anderson, and Wright 1983). Further
discussion of this issue and the supplementary sample are
contained in Technical Reports No. 2 (Hum, Laub, Metcalf,
and Sabourin 1979, 46–50) and No. 6 (Sabourin 1979, 62–
74). An evaluation of the supplementary sample design is
provided by Kurz (1979, 39–41).
15 The IRF and its accompanying schedules was simplified in
July 1975 to improve program participation (Crest et al.
1979, 25–26). To see whether income reporting affected
behaviour, the control group was roughly divided into
half who completed the IRF and half who did not (Hum,
Laub, Metcalf and Sabourin 1979, 10). Families in the
control group were provided with a small payment to
complete the interviews and monthly IRFs.
16 See Technical Report No. 1 (Hum, Laub and Powell 1979,
30–34) or No. 3 (Hum, Crest and Komus 1979, 24–26) for
details on the calculation of G, Y, and W.
17 Farmers in the rural dispersed sample were permitted an
additional exemption of $20,000 (Crest et al. 1979, 19). The
Mincome experiment was unique in its use of a wealth tax
as well as the adoption of a saturation site, the inclusion of
single individuals in its dispersed sample, and some ex-
perimentation with administrative parameters (Hikel et al.
1974, 49). The Mincome analysis of the appropriate treat-
ment of farm income and wealth would provide useful
98 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
background for the design of a modern guaranteed income
in which self-employment and non-wage income would be
more prominent.
18 Those eligible for payments in any month would not pay
regular taxes. For operational details of the annual recon-
ciliation process, see Technical Report No. 4 (Crest et al.
1979, 62–69). See also Technical Report No. 9 (Basilevsky,
Hum, and Sabourin 1979) for further details on income
reporting and deductions.
19 Hum (1981) provides an extensive discussion of this problem
and other design issues associated with negative income
tax experiments in general and Mincome in particular.
20 Evidence on the extent of new family unit formation and
inactive units is provided in the payments data in Crest
et al. (1979, 8).
21 In 1979, Mincome digitized only a portion of the collected
survey data, primarily relating to labour markets, changes
in income and net worth, and selected attitudinal and
household composition variables. The choice of what to
digitize reflects the Mincome researchers’ belief about
what information was essential to test work-incentive be-
haviours and changes to the economic foundation of the
household. Considerable information resides in the hard
copies of survey and IRF forms stored in the federal archives
that remain undigitized. The work of the Institute for Social
and Economic Research (ISER) focussed on transforming the
existing digital files (set up in PDP-11 formats) to run on the
mainframes of the day, creating codebooks, and promoting
the potential of the data to other researchers.
22 The University of Manitoba insisted that a portion of the
funding be devoted to a range of other academic pursuits,
beyond just restoring the Mincome data.
23 The Macdonald Royal Commission recommended a GAI in
its 1985 recommendations (Canada, 1985).
24 Dataverse is a Harvard University product that supports
the archiving and dissemination of digital objects. The Uni-
versity of Manitoba Libraries is a licensee.
25 Library and Archives Canada still maintains the adminis-
trative files and completed survey forms, essentially in the
same form as in which they were originally received from
Mincome in 1979.
26 New cross-sectional microdata on labour supply and work
histories from the 1981 long-form Census and from the Sur-
vey of Work History, the forerunner to the Labour Market
Activity Survey and the Survey of Labour and Income
Dynamics, were also emerging from Statistics Canada at
this time.
27 Issues arise because the assignment model and participation
both depend on family income and prospective payments,
which in turn depend on labour supply and introduce the
potential for bias. Many other socio-economic and demo-
graphic covariates might be included, but insofar as they
are either uncorrelated with the treatment variables by
experimental assignment or fixed in time, they would be
expected to have no effect on the estimates of the treatment
effects.
28 See Table 2 in Hum and Simpson (1993, S279) for a summary
of the non-structural labour supply response estimates from
the US experiments and Mincome. The weighted average US
effect is dominated by the final experiments in Seattle and
Denver, which had the largest samples, the most complex
designs, and the largest reported experimental effects (Hum
and Simpson 1991, 63).
29 Virtual income hDwþDyreflects the income guarantee
consistent with a linear tax schedule and pre-experimental
labour supply.
30 A complete discussion of the structural labour supply models
analyzed using the Winnipeg sample is provided in Hum
and Simpson (1991, 69–84).
31 See, for example, the review of recent empirical labour
supply research for the US Congressional Budget Office
(McClelland and Mok 2012). The estimated elasticities are
small and within a narrow range for both men and women
compared with the research reviewed in Hum and Simpson
(1991, 5–39).
32 No deep scientific justification supported the selection of
Dauphin as a saturation sites. Rather, based on undocu-
mented anecdotal accounts, it seems that the provincial
government of the day simply wanted a farm-oriented com-
munity to participate in Mincome.
33 The media have been captivated by the notion of Dauphin
as a saturation site. Many stories have circulated that gov-
ernment spent a wad of cash and then abruptly terminated
the project, leaving all the information to fester in dusty
boxes. Nothing in Professor Forget’s papers or presenta-
tions supports such misleading views.
34 The screening and enrolment surveys were not digitized by
Mincome before data archiving.
35 See Lee (2003) for a survey of attrition rates in panel surveys.
36 See Greene (2012) for an overview of the development of
this literature.
37 A standard reference is Campbell and Russo (1999).
38 The Seattle and Denver income maintenance experiments
included a component that was intended to run for 20
years to explore this issue, but that sample was terminated
much earlier with a settlement for the affected families. For
those families who expected the treatment benefits to last
20 years, Robins (1984) finds that labour supply response
is smaller than in 3- and 5-year samples, but the small size
of the 20-year sample hampers formal testing.
39 Metcalf (1974) analyzes the response bias from a limited-
duration experiment in the context of a model of labour
supply over the life cycle.
40 Any future social experiment would likely need to merge
administrative and survey data, which would require appro-
priate ethical procedures.
References
Allison, P. D. 2001. Missing data. Quantitative Applications in
the Social Sciences Vol. 136. Thousand Oaks, CA: Sage.
Atkinson, A.B., and S. Morelli. 2014. ‘‘Chartbook of Economic
Inequality.’’ Society for the Study of Economic Inequality
Working Paper 2014–324, Verona, Italy. At: http://
www.chartbookofeconomicinequality.com/ http://
dx.doi.org/10.2139/ssrn.2422269.
Baltagi, B. 2013. Econometric Analysis of Panel Data. New York:
Wiley.
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 99
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
Basilevsky, A., D. Hum, and D. Sabourin. 1979. Income
Reporting Behaviour in a Negative Income Tax Program: A
Comparison of Retrospective and Prospective Reporting Methods
in Mincome Manitoba. Technical Report No. 9 of the
Manitoba Basic Annual Income Experiment. Winnipeg:
University of Manitoba, Institute for Social and Economics
Research. At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/JE9UGG.
Basilevsky, A., and R. Sproule. 1979. The Accuracy of Income
Reporting in Mincome Manitoba. Technical Report No. 10 of
the Manitoba Basic Annual Income Experiment. Winnipeg:
University of Manitoba, Institute for Social and Economics
Research. At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/9XJUGP.
Bennett, S.E. 1986. ‘‘Do Public Program Clients Understand
Public Program Rules: The Manitoba Basic Annual Income
Experiment.’’ Canadian Public Administration/Adminstration
publique du Canada 29(3):462–68. http://dx.doi.org/
10.1111/j.1754-7121.1986.tb00820.x.
Billett, C., D. Komus, D. Hum, A. Basilevsky, and R. Sproule.
1979. Issues in the Administration of Mincome Manitoba: Three
Preliminary Assessments. Technical Report No. 11 of the
Manitoba Basic Annual Income Experiment. Winnipeg:
University of Manitoba, Institute for Social and Economics
Research. At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/YCCBMI.
Brzozowski, M., M. Gervais, P. Klein, and M. Suzuki. 2010.
‘‘Consumption, Income, and Wealth Inequality in Canada.’’
Review of Economic Dynamics 13(1):52–75. http://
dx.doi.org/10.1016/j.red.2009.10.006.
Burtless, G. 1986. ‘‘The Work Response to a Guaranteed
Income: A Survey of Experimental Evidence.’’ In Lessons
from the Income Maintenance Experiments, ed. Alicia
Munnell, 22–52. Boston: Federal Reserve Bank of Boston.
Cain, G., and D. Wissoker. 1990. ‘‘A Reanalysis of Marital
Stability in the Seattle-Denver Income-Maintenance
Experiment.’’ American Journal of Sociology 95(5):1235–69.
http://dx.doi.org/10.1086/229428.
Calnitsky, D. 2016. ‘‘‘More Normal than Welfare’: The
Mincome Experiment, Stigma, and Community
Experience.’’ Canadian Review of Sociology 53(1):26–71.
Medline:26890446 http://dx.doi.org/10.1111/cars.12091.
Calnitsky, D., and J. Latner. 2016. ‘‘Basic Income in a Small
Town: Understanding the Elusive Effects on Work.’’ Social
Problems.
Campbell, D.T, and M.J. Russo. 1999. Social Experimentation.
Thousand Oaks, CA: Sage.
Canada. 1985. Royal Commission on the Economic Union and
Development Prospects for Canada. Volume 1. At http://
publications.gc.ca/site/eng/472251/publication.html.
Canadian Inflation. 2014. ‘‘Historical Canadian Inflation Rates:
1915–2014.’’ At http://canadianinflation.com/historical-
canadian-inflation-rates/.
Choudhry, S., and D. Hum. 1995. ‘‘Graduated Work Incentives
and How They Affect Marital Stability: The Canadian
Evidence.’’ Applied Economics Letters 2(10):367–71. http://
dx.doi.org/10.1080/758518991.
Clavet, N.-J., J.-Y. Duclos, and G. Lacroix. 2013. ‘‘Fighting
Poverty: Assessing the Effect of Guaranteed Minimum
Income Proposals in Quebec.’’ Canadian Public Policy/
Analyse de politiques 39(4):491–516. http://dx.doi.org/
10.3138/CPP.39.4.491.
Conlisk, J., and H. Watts. 1979. ‘‘A Model for Optimizing
Experimental Designs for Estimating Response Surfaces.’’
Journal of Econometrics 11(1):27–42. http://dx.doi.org/
10.1016/0304-4076(79)90052-6.
Crest, D., C. Billett, D. Hum, D. Komus, and A. Quarry. 1979.
The Administration of the Payments System of Mincome
Manitoba. Technical Report No. 4. of the Manitoba Basic
Annual Income Experiment. Winnipeg: University of
Manitoba, Institute for Social and Economics Research.
At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/UYWPHK.
Curry, B. 2016. ‘‘Minister Eyes Guaranteed Minimum Income
to Tackle Poverty.’’ The Globe and Mail, 5 February 5.
At http://www.theglobeandmail.com/news/politics/
guaranteed-income-has-merit-as-a-national-policy-
minister-says/article28588670/.
Economic Council of Canada. 1968. Fifth Annual Review.
Ottawa: Queen’s Printer.
Ehnes, J., and W. Simpson. 1983. ‘‘Work Incentives and
Welfare: Evidence from the Mincome Baseline for Adult
Males.’’ Western Economic Review 2(3):17–25.
Fitzgerald, J., P. Gottschalk, and R. A. Moffitt. 1998. An
Analysis of Sample Attrition in Panel Data: The Michigan Panel
Study of Income Dynamics. NBER Working Paper 0220.
Cambridge, MA: National Bureau of Economic Research.
Forget, E. 2011. ‘‘The Town with No Poverty: The Health
Effects of a Canadian Guaranteed Annual Income Field
Experiment.’’ Canadian Public Policy/Analyse de politiques
37(3):283–305. http://dx.doi.org/10.3138/cpp.37.3.283.
Forget, E.L. 2013. ‘‘New Questions, New Data, Old
Interventions: The Health Effects of a Guaranteed Annual
Income.’’ Preventive Medicine 57(6):925–28.
Medline:23764242 http://dx.doi.org/10.1016/
j.ypmed.2013.05.029.
Fortin, N., D.A. Green, T. Lemieux, K. Milligan, and W.C.
Riddell. 2012. ‘‘Canadian Inequality: Recent Developments
and Policy Options.’’ Canadian Public Policy/Analyse de
politiques 38(2):121–45. http://dx.doi.org/10.3138/
cpp.38.2.121.
Greene, W.H. 2012. Econometric Analysis. 7th ed. Upper Saddle
River, NJ: Pearson Education.
Griliches, Z. 1986. ‘‘Economic Data Issues.’’ In Handbook of
Econometrics, Vol. 3, ed. Z. Griliches and M.D. Intriligator,
1465–1514. Amsterdam: Elsevier.
Groeneveld, L., N. Tuma, and M. Hannan. 1980. ‘‘The Effects
of Negative Income Tax Programs on Marital Dissolution.’’
Journal of Human Resources 15(4):654–74. http://dx.doi.org/
10.2307/145406.
Hausman, J.A., and D.A. Wise. 1979. ‘‘Attrition Bias in
Experimental and Panel Data: the Gary Income
Maintenance Experiment.’’ Econometrica 47(2):455–73.
http://dx.doi.org/10.2307/1914193.
Hikel, R.S., M.E. Laub, and B.J. Powell. (1974) ‘‘The
Development and Design of the Basic Annual Income
Experiment in Manitoba: A Preliminary Report.’’ Research
Project. Ottawa and Winnipeg: Health and Welfare Canada
and Department of Health and Social Development.
100 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
Hirano, K., G.W. Imbens, G. Ridder, and D.B. Rubin. 2001.
‘‘Combining Panel Data Sets with Attrition and
Refreshment Samples.’’ Econometrica 69(6):1645–59. http://
dx.doi.org/10.1111/1468-0262.00260.
Hum, D. 1981. ‘‘Canada’s Administrative Experience with
Negative Income Taxation.’’ Canadian Taxation (Spring):1–16.
Hum, D., and S. Choudhry. 1992. ‘‘Income, Work and Marital
Dissolution: Canadian Experimental Evidence.’’ Journal of
Comparative Family Studies 23(2):249–65.
Hum, D., D. Crest, and D. Komus. 1979. The Design of the
Payments System of Mincome Manitoba. Technical Report No.
3 of the Manitoba Basic Annual Income Experiment.
Winnipeg: University of Manitoba, Institute for Social and
Economics Research. At http://
dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/BWRHBQ.
Hum, D., M. Laub, C. Metcalf, and D. Sabourin. 1979. The
Sample Design and Assignment Model. Technical Report
No. 2 of the Manitoba Basic Annual Income Experiment.
Winnipeg: University of Manitoba, Institute for Social and
Economics Research. At http://dataverse.lib.
umanitoba.ca/dataset.xhtml?persistentId=doi:10.5203/
FK2/VPZILS.
Hum, D.P.J., M.E. Laub, and B.J. Powell. 1979. The Objectives
and Design of the Manitoba Basic Annual Income Experiment.
Technical Report No. 1 of the Manitoba Basic Annual
Income Experiment. Winnipeg: University of Manitoba,
Institute for Social and Economics Research. At http://
dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/XAGGJT.
Hum, D., and W. Simpson. 1991. Income Maintenance, Work
Effort and the Canadian Mincome Experiment. Ottawa:
Economic Council of Canada.
Hum, D., and W. Simpson. 1993. ‘‘Economic Response to a
Guaranteed Annual Income: Experience from Canada and
the United States.’’ Journal of Labor Economics 11(1 Pt 2):
S263–96. http://dx.doi.org/10.1086/298335.
Jakubson, G. 1988. ‘‘The Sensitivity of Labor Supply Parameter
Estimates to Unobserved Individual Effects: Fixed- and
Random-Effects Estimates in a Nonlinear Model using
Panel Data.’’ Journal of Labor Economics 6(3):302–29. http://
dx.doi.org/10.1086/298185.
Keeley, M. 1981. Labor Supply and Public Policy. New York:
Academic Press.
Keeley, M., and P. Robins. (1978) ‘‘The Design of Social
Experiments: A Critique of the Conlisk-Watts Assignment
Model,’’ Research Memorandum No. 57. Menlo Park, CA:
Stanford Research Institute Centre for the Study of Welfare
Policy.
Kurz, M. 1979. An Evaluation of the Experimental Sample of
Mincome Manitoba. Technical Report No. 5 of the Manitoba
Basic Annual Income Experiment. Winnipeg: University of
Manitoba, Institute for Social and Economics Research.
At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/Q1VF86.
Lee, U. 2003. Panel Attrition in Survey Data: A Literature Review.
Vol. 41. Rondebosch, South Africa: University of Cape
Town, Centre for Social Science Research.
Lillard, L.A., and C.W. Panis. 1998. ‘‘Panel Attrition from the
Panel Study of Income Dynamics: Household Income,
Marital Status, and Mortality.’’ Journal of Human Resources
33(2):437–57. http://dx.doi.org/10.2307/146436.
Little, R.J., and D.B. Rubin. 1987. Statistical Analysis with
Missing Data. New York: Wiley.
MaCurdy, T., D. Green, and H. Paarsch. 1990. ‘‘Assessing
Empirical Approaches for Analyzing Taxes and Labor
Supply.’’ Journal of Human Resources 25(3):415–90. http://
dx.doi.org/10.2307/145990.
Mason, G. 2016. Mincome User Manuals Dataverse. At http://
dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/XLOXQF.
Mason, G. 1984. ‘‘Wealth, Income and Economic Welfare: A
Preliminary Review of Mincome Baseline Data.’’ Research
Paper, Institute for Social and Economic Research,
University of Manitoba, May 28.
McClelland, R., and S. Mok. 2012. ‘‘A Review of Recent
Research on Labor Supply Elasticities.’’ Congressional
Budget Office Working Paper 2012–12, Congressional
Budget Office, Washington, DC.
Metcalf, C. 1974. ‘‘Predicting the Effects of Permanent
Programs from a Limited Duration Experiment.’’ Journal of
Human Resources 9(4):530–55. http://dx.doi.org/10.2307/
144784.
Mroz, T. 1987. ‘‘The Sensitivity of an Empirical Model of
Married Women’s Hours of Work to Economic and
Statistical Assumptions.’’ Econometrica 55(4):765–99. http://
dx.doi.org/10.2307/1911029.
Nicoletti, C. 2006. ‘‘Nonresponse in Dynamic Panel Data
Models.’’ Journal of Econometrics 132(2):461–89. http://
dx.doi.org/10.1016/j.jeconom.2005.02.008.
Prescott, D., R. Swidinsky, and D.A Wilton. 1986. ‘‘Labour
Supply Estimates for Low-Income Female Heads of
Household using Mincome Data.’’ Canadian Journal of
Economics/Revue canadienne d’e´conomique 19(1):134–41.
http://dx.doi.org/10.2307/135175.
Rasmussen, E., A.B. Anderson, and J.D. Wright. 1983. Mincome
Field Operations and Interviewing Techniques. Technical
Report No. 12 of the Manitoba Basic Annual Income
Experiment. Winnipeg: University of Manitoba, Institute
for Social and Economics Research. At http://
dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/YDGFHH.
Rasmussen, E., A.B. Anderson, J.D. Wright, and E. Sang. 1983.
Quality Control and Mincome Data. Technical Report No. 13
of the Manitoba Basic Annual Income Experiment.
Winnipeg: University of Manitoba, Institute for Social and
Economics Research. At http://
dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/EU7QKS.
Robins, P. 1984. ‘‘The Labor Supply Response of Twenty-Year
Families in the Denver Income Maintenance Experiment.’’
Review of Economics and Statistics 66(3):491–95. http://
dx.doi.org/10.2307/1925008.
Robins, P.K. 1985. ‘‘A Comparison of the Labor Supply
Findings from the Four Negative Income Tax
Experiments.’’ Journal of Human Resources 20(4):567–82.
http://dx.doi.org/10.2307/145685.
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 101
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
Robins, P.K., and R. West. 1982. ‘‘Program Participation and
Labor-Supply Response.’’ Journal of Human Resources
15:488–523.
Robins, P.K., and R.W. West. 1986. ‘‘Sample Attrition and
Labor Supply Response in Experimental Panel Data: A
Study of Alternative Correction Procedures.’’ Journal of
Business & Economic Statistics 4(3):329–38.
Sabourin, D. 1979. Sample Development Over Time, Participation
and Attrition. Technical Report No. 6 of the Manitoba Basic
Annual Income Experiment. Winnipeg: University of
Manitoba, Institute for Social and Economics Research.
At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/AXLATZ.
Sabourin, D., and D. Hum. 1979. An Analysis of Non-Response to
the Manitoba Basic Annual Income Experiment. Technical
Report No. 7 of the Manitoba Basic Annual Income
Experiment. Winnipeg: University of Manitoba, Institute
for Social and Economics Research. http://
dataverse.lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/KJCFS0.
Sabourin, D., D. Hum, W. Harrar, and A. Basilevsky. 1979.
Program Participation in the Saturation Site of the Manitoba
Basic Annual Income Experiment. Technical Report No. 8 of
the Manitoba Basic Annual Income Experiment. Winnipeg:
University of Manitoba, Institute for Social and Economics
Research. At http://dataverse.lib.umanitoba.ca/
dataset.xhtml?persistentId=doi:10.5203/FK2/IHA3UI.
Segal, H. 2016. ‘‘Finding a Better Way: A Basic Income Pilot for
Ontario.’’ At https://files.ontario.ca/discussionpaper_
nov3_english_final.pdf.
Special Senate Committee on Poverty. 1971. Report on Poverty.
Ottawa: Parliament of Canada.
Statistics Canada. 1975. Income Distributions by Size in Canada.
Cat. No. 13-207. Ottawa: Statistics Canada.
van den Berg, G.J., and M. Lindeboom. 1998. ‘‘Attrition in
Panel Survey Data and the Estimation of Multi-State Labor
Market Models.’’ Journal of Human Resources 33(2):458–78.
http://dx.doi.org/10.2307/146437.
Verbeek, M., and T. Nijman. 1992. ‘‘Testing for Selectivity Bias
in Panel Data Models.’’ International Economic Review
33(3):681–703. http://dx.doi.org/10.2307/2527133.
Wooldridge, J.M. 1995. ‘‘Selection Corrections for Panel Data
Models under Conditional Mean Independence
Assumptions.’’ Journal of Econometrics 68(1):115–32. http://
dx.doi.org/10.1016/0304-4076(94)01645-G.
Zabel, J.E. 1998. ‘‘An Analysis of Attrition in the Panel Study of
Income Dynamics and the Survey of Income and Program
Participation with an Application to a Model of Labor
Market Behavior.’’ Journal of Human Resources 33(2):479–
506. http://dx.doi.org/10.2307/146438.
Appendix A: Mincome Experimental Papers
These papers consist of the technical reports of the Mani-
toba Basic Annual Income Experiment and are available
online at the University of Manitoba Libraries Dataverse
at http://dataverse.lib.umanitoba.ca/dataverse/TechDocs.
Full references are provided in the reference list for this
article.
Technical Report No. 1, The Objectives and Design of the
Manitoba Basic Annual Income Experiment, by Derek
Hum, Michael E. Laub, and Brian J. Powell. 1979. 69 p.
http://dataverse.lib.umanitoba.ca/dataset.xhtml?
persistentId=doi:10.5203/FK2/XAGGJT.
Technical Report No. 2, The Sample Design and Assign-
ment Model, by Derek Hum, Michael Laub, Charles
Metcalf, and Don Sabourin. 1979. 65 p. http://
dataverse.lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/VPZILS.
Technical Report No. 3, The Design of the Payments
System of Mincome Manitoba, by Derek Hum, David
Crest, and David Komus. 1979. 35 p. http://dataverse.
lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/BWRHBQ.
Technical Report No. 4, The Administration of the Payments
System of Mincome Manitoba, by David Crest, Carol Billett,
Derek Hum, David Komus, and Andrew Quarry. 1979.
69 p. http://dataverse.lib.umanitoba.ca/dataset.xhtml?
persistentId=doi:10.5203/FK2/UYWPHK.
Technical Report No. 5, An Evaluation of the Experimental
Sample of Mincome Manitoba, by Mordecai Kurz. 1979.
68 p. http://dataverse.lib.umanitoba.ca/dataset.xhtml?
persistentId=doi:10.5203/FK2/Q1VF86.
Technical Report No. 6, Sample Development over Time,
Participation and Attrition, by Donald Sabourin. 1979.
81 p. http://dataverse.lib.umanitoba.ca/dataset.xhtml?
persistentId=doi:10.5203/FK2/AXLATZ.
Technical Report No. 7, An Analysis of Non-Response to
the Manitoba Basic Annual Income Experiment, by Donald
Sabourin and Derek Hum. 1979. 58 p. http://dataverse.
lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/KJCFS0.
Technical Report No. 8, Program Participation in the Satura-
tion Site of the Manitoba Basic Annual Income Experiment, by
Donald Sabourin, Derek Hum, William Harrar, and
Alexander Basilevsky. 1979. 60 p. http://dataverse.
lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/IHA3UI.
102 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
Technical Report No. 9, Income Reporting Behaviour in a
Negative Income Tax Program: A Comparison of Retrospec-
tive and Prospective Reporting Methods in Mincome Mani-
toba, by Alexander Basilevsky, Derek Hum, and Donald
Sabourin. 1979. 45 p. http://dataverse.lib.umanitoba.
ca/dataset.xhtml?persistentId=doi:10.5203/FK2/
JE9UGG.
Technical Report No. 10, The Accuracy of Income Report-
ing in Mincome Manitoba, by Alexander Basilevsky and
R. Sproule. 1979. 35 p. http://dataverse.lib.umanitoba.
ca/dataset.xhtml?persistentId=doi:10.5203/FK2/
9XJUGP.
Technical Report No. 11, Issues in the Administration of
Mincome Manitoba: Three Preliminary Assessments, by
Carol Billet, David Komus, Derek Hum, Alexander
Basilevsky, and Robert Sproule. 1979. 34 p. http://
dataverse.lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/YCCBMI.
Technical Report No. 12, Mincome Field Operations and
Interviewing Techniques, by Edgar Rasmussen, Andy B.
Anderson, and James D. Wright. 1983. 44 p. http://
dataverse.lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/YDGFHH.
Technical Report 13, Quality Control and Mincome Data,
by Edgar Rasmussen, Andy B. Anderson, James D.
Wright, and Eric Sang. 1983. 34 p. http://dataverse.
lib.umanitoba.ca/dataset.xhtml?persistentId=
doi:10.5203/FK2/EU7QKS.
Mincome User Manual Version 1, by Institute for Social
and Economic Research. 1983. http://dataverse.lib.
umanitoba.ca/dataset.xhtml?persistentId=doi:10.5203/
FK2/XLOXQF.
Appendix B: Mincome File Codebooks
and Data
The code books are detailed data dictionaries for the
Mincome files. They are available at the University of
Manitoba Libraries Dataverse at http://dataverse.lib.
umanitoba.ca/dataverse/Codebooks.
The data are in Excel format with two header rows—
(a) variable numbering corresponding to the numbering
in the code books and (b) a variable name. Users may
change the name but should retain the original file with
numbering to ensure data checking. The data are avail-
able at http://dataverse.lib.umanitoba.ca/dataverse/
Mincome_Data.
MINC1—Mincome Baseline Summary File
MINC1 contains 91 variables constructed from data in
the first major survey undertaken by the Mincome
guaranteed annual income experiment—the baseline
survey. The file contains three types of variables: those
pertaining to the household (size, composition, etc.),
those pertaining to the male head, and those pertaining
to the female head. For double-headed households,
information on both heads exists.
MINC2—Mincome Payments Summary File
MINC2 contains 495 variables constructed from the
payments database of the Mincome guaranteed annual
income experiment. The file consists of a header record
of 14 variables for each household, followed by 37
months of records covering December 1974–December
1977 inclusive, each month repeating the same 13
variables.
MINC3—Mincome Baseline-Payments Data
MINC3 concatenates the baseline (MINC1) and payments
(MINC2) data. This file contains 750 observations and 596
variables (91 from the payments data, 494 from the base-
line survey, and 11 additional header records). A family
(which could be a single- or double-headed household,
with or without children) was included in MINC3 only
if baseline information existed and if the family had
received 24 months of payments.
MINC4—Longitudinal Labour Data File
MINC4 contains 338 variables constructed from the 11
major surveys (conducted approximately every three
months) used to track labour market and other behav-
ioural data. The data consist of 26 household variables,
154 variables tracking 14 variables for male heads for
11 surveys, 154 variables tracking the 14 variables for
female heads for 11 surveys, and four records of header
information. Double-headed households have informa-
tion on both heads.
MINC5—Income and Net Worth
MINC5 contains 445 variables constructed from the data
in the 11 major surveys undertaken between December
1974 and December 1977. The surveys tracked changes
in financial assets, real property, and income for each of
the 11 surveys, conducted approximately every 3
months.
The Manitoba Basic Annual Income Experiment: Lessons Learned 40 Years Later 103
doi:10.3138/cpp.2016-082 8Canadian Public Policy / Analyse de politiques, March / mars 2017
MINC6—Family Composition and Attitudes
MINC6 has data only from the Winnipeg site and
includes 1,408 variables constructed from the 11 surveys.
There are 287 variables on family composition, 411
variables on male and female head attitudes, and 327
variables on double-headed households. There are 11
recurring cycles of about 35 attitudinal variables for
male and female heads. Mincome collected attitudinal
data on such issues as locus of control, housing satisfac-
tion, weekend activity, awareness and understanding of
the MINCOME experiment, attitudes toward work, and
decision making in the household.
Caution: These codebooks date from 1984 and con-
tain instructions on accessing the data using IBM Job
Control Language at the University of Manitoba at that
time. The current version of the data is in Excel format
and may be read into most statistical software currently
in use. The Excel data sets have two header rows,
numbering the field and listing the variable name as
identified in the applicable codebook. The Excel spread-
sheets are all under 5 megabytes in size and therefore
well within the limits of desktop and laptop computers.
104 Simpson, Mason, and Godwin
8Canadian Public Policy / Analyse de politiques, March / mars 2017 doi:10.3138/cpp.2016-082
