Content uploaded by Stephen Morse
Author content
All content in this area was uploaded by Stephen Morse on Jul 23, 2015
Content may be subject to copyright.
1
Reductions in insecticide use from adoption of Bt cotton in South
Africa: impacts on economic performance and toxic load to the
environment
Running title: Reductions in insecticide use from Bt cotton
Richard Bennett1, Stephen Morse2, Yousouf Ismael1 and Bhavani Shankar1
1 Department of Agricultural and Food Economics, The University of Reading,
PO Box 237, Reading RG6 6AR, UK Tel: +44 (0) 118 3786478 Fax: +44 (0)
118975 6467
2 Corresponding author. Department of Geography, School of Human and
Environmental Sciences, The University of Reading, PO Box 227, Reading RG6
6AB Email: s.morse@reading.ac.uk Tel +44 (0) 118 3788736 Fax +44 (0) 118
9755865
2
SUMMARY
The study reported here presents the findings relating to commercial growing of
genetically-modified Bt cotton in South Africa by a large sample of smallholder
farmers over three seasons (1998/99, 1999/2000, 2000/01) following adoption.
The analysis presented constructs and compares groupwise differences for key
variables in Bt versus non-Bt technology and uses regressions to further analyse
the production and profit impacts of Bt adoption. Analysis of the distribution of
benefits between farmers due to the technology is also presented. In parallel with
these socio-economic measures, the toxic loads being presented to the
environment following the introduction of Bt cotton are monitored in terms of
insecticide active ingredient (ai) and the Biocide Index. The latter adjusts ai to
allow for differing persistence and toxicity of insecticides.
Results show substantial and significant financial benefits to smallholder
cotton growers of adopting Bt cotton over three seasons in terms of increased
yields, lower insecticide spray costs and higher gross margins. This includes one
particularly wet, poor growing season. In addition, those with the smaller
holdings appeared to benefit proportionately more from the technology (in terms
of higher gross margins) than those with larger holdings. Analysis using the
Gini-coefficient suggests that the Bt technology has helped to reduce inequality
amongst smallholder cotton growers in Makhathini compared to what may have
been the position if they had grown conventional cotton. However, while Bt
growers applied lower amounts of insecticide and had lower Biocide Indices (per
ha) than growers of non-Bt cotton, some of this advantage was due to a reduction
in non-bollworm insecticide. Indeed, the Biocide Index for all farmers in the
population actually increased with the introduction of Bt cotton.
3
The results indicate the complexity of such studies on the socio-economic
and environmental impacts of GM varieties in the developing world.
INTRODUCTION
Criticisms of the so-called Green Revolution included concerns that the
technologies largely discriminated against small producers, increased risk and
dependency and contributed to greater inequality of incomes. Similar criticisms
are now being levelled at genetically modified (GM) crops and their uptake in
developing countries (Orton 2003). At the same time, the potential benefits of
GM crops for developing countries are also being cited. Included here are
benefits such as higher profits and a reduction in damage to human health and the
environment resulting from less application of expensive, persistent and toxic
insecticides. There is now a growing body of literature regarding the former
point in a number of developing countries (James 2003; Qaim & Zilberman
2003). However, many of these are founded on trial data rather than what
farmers actually practice on their own farms and few such studies relate to
growers in Sub Saharan Africa, given that very few countries in that region have
adopted GM crops. With regard to the health and environmental benefits
resulting from the use of less insecticide, there are a number of studies which
prove the causal link between the growing of insect-resistant GM varieties and
the use of less insecticide. However, none of these have attempted to measure the
effects in terms of the ‘toxic load’ to the environment, preferring instead to
calculate the quantities of product and the cost. Given that plant resistance may
4
only be to one group of insects, growers still have to adopt other crop protection
technologies, typically insecticide. Reducing the need for some insecticides may
not have much overall benefit if GM growers still have to apply more
environmentally-damaging products.
There are various methods designed to measure ‘toxic load’ to an
environment, but perhaps the simplest methodology is that of Jansen et al. (1995)
in their calculation of what they refer to as the ‘Biocide Index’:
‘toxic load’ for pesticide
Biocide Index = ------------------------------------
duration of land use system
‘Toxic load’ is a function of the amount of pesticide applied each year, its
concentration of active ingredient, the toxicity of the ai (i.e. its toxicity code as
categorized by the World Health Organization; WHO) and how long the ai
remains active in the environment. The latter is taken to be the square root of the
duration in days (Jansen et al. 1995). Summation takes place for all of the
pesticides applied to the system. The assumption is: the lower the value of the
Biocide Index the better.
While the Biocide Index is relatively straightforward to calculate, it can
be criticized for its simplifying assumptions. For example, the WHO rating refers
to mammalian toxicity, but insecticides can also be highly toxic, perhaps even
more so, to fish, birds and beneficial insects. Nevertheless, it provides a starting
point to look for differences in toxic load with the introduction of insect-resistant
traits.
5
The aim of the research reported here was firstly to estimate the economic
benefits, if any, for small-scale farmers adopting a type of GM cotton in the
Republic of South Africa, and secondly to estimate what advantages accrue in
terms of the toxic load on the environment.
The Republic of South Africa (RSA) was the first country in Sub-Saharan
Africa to adopt a GM crop variety in the form of the Genetic Modification
Organism Act of 1997 which allowed the Bt cotton variety (NuCOTN 37-B with
Bollgard) to be grown in RSA from 1998. This variety contains a set of genes
that control the production of a natural insecticide (toxins produced by Bacillus
thuringensis) that acts specifically on the cotton bollworm complex (Lepidoptera
such as Helicoverpa zea, H. armigera, Diparopsis castenea, Earias biplaga and
E. insulana) and some Coleoptera. Cotton bollworm can severely reduce yields
and increase costs and for this reason large amounts of toxic sprays are usually
applied to the cotton crop (Rother 1998 refers to a “pesticide culture” in RSA).
By controlling bollworm using the Bt cotton variety, the amount and number of
sprays are greatly reduced. Moreover, despite applying a large number of sprays,
bollworm can remain difficult to control in the cotton crop because of the
importance of the timing of spray applications and the need for regular sprays.
The Bt variety should require fewer sprays and may also result in higher yields
relative to non-Bt varieties.
Indeed, previous studies have shown that farmers can benefit from
adopting Bt cotton in terms of both cost savings from reduced spray inputs and
also in terms of higher revenues by achieving higher yields than conventional
varieties (see, for example, Ismael et al. 2002; Manwan & Subagyo 2002; Naik
2001; Pray et al. 2002; Qaim 2003; Traxler et al. 2001). However, these studies
6
largely report results from trial data or from small samples of producers over
short time periods (e.g. 1 or 2 years). Such data may suffer from being
unrepresentative ‘snapshots’ of the actual growing practices and experiences of
the farm population over time.
The study reported here presents the findings relating to commercial
growing of Bt cotton by a large sample of smallholder farmers over three seasons
(1998/99, 1999/00, 2000/01) following adoption.
The study area
The Makhathini Flats area of RSA in KwaZulu Natal Province was chosen for
study because of a relatively rapid adoption of Bt cotton by growers in that area
from 1998 onwards. Indeed, by 2002, of the 30 000 hectares of Bt cotton grown
in RSA, nearly 6000 were grown in Makhathini Flats with an estimated 92 % of
cotton growers having adopted the Bt variety (James 2002). In the Makhathini
area, agriculture is very important and typically small households cultivate plots
of around 1–3 ha allocated to them by tribal chiefs. As well as subsistence crops
(such as maize and beans), cotton is widely grown as a cash crop and usually
occupies most of the farm area. There are around 5000 smallholder farmers in the
area, of which some 1400 are registered to grow cotton in any one year. The
availability of cotton inputs (such as seed and insecticides) and especially the
credit to buy them is very important to smallholders. At the time of the present
study, the Vunisa Cotton Company supplied smallholder growers in Makhathini
with inputs and credit and bought the cotton they produced. The company also
supplied some extension advice to growers.
7
Research methods
Data collection
Records obtained from Vunisa Cotton included data on 1283 growers in the
1998/99 season (some 89 % of all registered cotton growers in the area), 441 (32
% of growers) in 1999/2000 and 499 (33 % of growers) in 2000/2001. Records
showed age and sex of farmers, farm size/cotton area planted, credit obtained and
previous credit status, yields/marketable output of cotton achieved, price paid,
seed, pesticide and hired labour use (expenditures) and other information. These
records were hand-written and stored at Vunisa premises. Records were selected
at random and due to organizational changes concerning Vunisa cotton, smaller
sample sizes of records had to be used in 2000 and 2001 than those obtained for
1999 (Vunisa Cotton left the Makhathini area in 2002). In addition, electronic
records for all registered cotton growers in Makhathini were available from
Vunisa for the three seasons, which showed for each grower the area of cotton
planted, the type of seed used, the yield obtained and the age, sex and credit
rating of the grower. These were checked against the written records. The
validated data were used to calculate the levels (and costs) of inputs and output
for farmers adopting and not adopting Bt cotton over the three seasons.
Production function and gross margin regressions
Whilst summary statistics and gross margin (value of output less variable input
costs) calculations provide a broad overview, they are less effective in isolating
the effects of individual changes while controlling for the effects of other
variables. A multiple regression with cotton output as the dependent variable and
inputs as the independent variables (i.e. a production function) can provide a
8
much better characterization. Two important points regarding our production
function estimation strategy are noted below.
Firstly, although a production function is typically represented as a
regression of output on quantifiable inputs such as pesticide and seed, variables
such as soil quality and farmer managerial competence should also be in the set
of explanatory variables when farm level data are used. Data on such variables
are seldom available, however, and were not part of the Vunisa dataset.
However, where panel data are available that track the same farms over multiple
years, panel data econometric specifications can implicitly account for such
variables that remain largely fixed over the years. Hence the strategy used here
for production function estimation was to restrict attention to those farms that
had at least 2 years of data available and perform panel data production function
estimation on the restricted dataset. Although some information is lost in such
restriction, the advantage is that the effect of unobserved fixed characteristics
such as soil quality and farmer skill is implicitly accounted for.
The second point concerns the choice of functional form for production
function estimation. A variety of forms have been used in past empirical analysis,
including the quadratic, Cobb-Douglas, Translog and the CES. Choice among
these forms usually involves trading-off practical considerations such as ease of
estimation and interpretation against the extent of ‘flexibility’ available within
the forms to allow diverse productivity patterns. For instance, the popular Cobb-
Douglas production function is linear in logarithms and hence easy to estimate.
Additionally, estimated parameters are interpretable as output elasticities.
However, the Cobb-Douglas form, being log-linear, does not accommodate the
case when one or more inputs are used at zero levels. Nor is the Cobb-Douglas
9
able to allow for both the ‘second stage’ of production (diminishing but positive
marginal product) as well as the ‘third stage’ (negative marginal product).
Agricultural inputs are often used at zero levels and some inputs are known to
have a significant range where the marginal product is negative. This limits the
usefulness of the Cobb-Douglas for agricultural production analysis at the micro
level.
The quadratic production function, used in this research, overcomes both
these weaknesses of the Cobb-Douglas. Our dataset does contain several
observations where one or more inputs are used at zero levels, but this aspect
does not restrict the application of the quadratic form. The quadratic form also
allows for the third stage of production to be reached, i.e., marginal products to
be eventually negative.
Other alternatives include ‘flexible functional’ forms, the most popular of
which is the translog production function. The translog offers significant
flexibility, allowing the data to determine almost every key aspect of the
production relationship, instead of constraining them a-priori. For instance, the
translog would allow for unconstrained patterns of substitution between inputs,
which both the Cobb-Douglas as well as the quadratic do not. However, the log-
linear translog function does not accommodate zero inputs, which presents a
problem for our dataset. Additionally, checking for key aspects of the production
relationship, such as concavity, becomes substantially more difficult with the
translog function. While the quadratic form allows concavity (diminishing
marginal product) to be checked simply by observing the sign of an estimated
10
parameter, in the case of the translog this task becomes much more complex, and
the evaluation has to be done at every single data point. Thus the quadratic offers
a reasonable compromise between the simplest but inflexible Cobb-Douglas and
a completely flexible but less easily interpretable form such as the translog. For
these reasons, the quadratic form, successfully used in several production
function studies (e.g. Qaim and DeJanvry, 2002) was used in the estimation.
The sample for the regression was restricted to those farmers with more
than one year of data available (427 farmers, 26 of whom are represented for all
3 years, with a total sample size of 869). Using this sub-sample, a random effects
panel data regression was estimated, including season dummies to capture time-
specific effects (variables listed in Table 3). The random effects model was
chosen since the alternative fixed effects model results in several loss in degrees
of freedom when the cross section is large and the time dimension short.
The profit implications of the new technology are also important. Hence a
regression was employed to examine the effects of Bt adoption on gross margins
(variables listed in Table 4). Empirical production economics has a long tradition
of estimating profit functions, where variable profits (total revenue minus
variable costs) at the farm level are regressed on input and output prices, and
fixed factors specific to the farm. Gross margins, although not the same as
variable profits, are constructed along similar lines. This suggests that variables
used in explaining gross margins across farms should be similar to those used in
estimating profit functions. However, price information is not available in the
dataset and the sample is derived from a small region lacking price variability.
Attention can therefore be reasonably restricted to fixed factors and socio-
11
economic characteristics of the farm. As in the production function case, a
random effects panel data regression including time dummies is employed.
The Biocide Index
For the purposes of this work a modification of the Biocide Index was used in the
sense that integration first took place on a farm basis.
Biocide Index per farm = ai applied (kg) Environmental Protection
Agency (EPA) toxicity code square root duration (days)
This was then converted to a per hectare basis to allow comparison across
varieties and over season.
The WHO toxicity code parameters and the pesticides included in the
categories are shown in Table 1. The same weighting of the codes was employed
as that used by Jansen et al. (1995): code I has a weighting of 7 and code II has a
weighting of 5. Duration was measured as half-life in soil (days) 2. Where a
range of values existed, the lowest value was used on the assumption that the
insecticides would degrade faster in a hot and humid environment such as that of
Makhathini. Toxicity and half-life data were taken from the database of the
Extension Toxicology Network (EXTOXNET).
For example if a farmer applied a total of 5 litres of monocrotophos (400
g/l active ingredient) and 1 litre of cypermethrin (200 g/l active ingredient) to a
05 hectare plot of cotton in a season then the Biocide Index would be:
Monocrotophos = 5 04 4 7 = 56
Cypermethrin = 1 02 4 5 = 4
12
The half-life for monocrotophos in soil is taken to be 7 days and the half-
life for cypermethrin is taken to be 8 days. Doubling these and taking the square
root yields a multiplier of 4. Monocrotophos has a WHO code of I (multiplier =
7) while cypermethrin has a WHO code of II (multiplier = 5). Adding these two
together and dividing by the number of hectares provides the Biocide Index on a
per hectare basis = 60/05 = 120.
RESULTS
Summary statistics
Table 2 shows some summary statistics relating to the large Vunisa dataset. The
average age of cotton growers in the sample is around 45 with no difference in
age between Bt adopters and non-adopters. In terms of gender, there are higher
proportions of women registered growers than men across all three seasons,
especially in season 3. The proportion of male growers is higher in the case of
adopters than for non-adopters. However, these differences are not statistically
significantly different. Area planted shows an average of 19–6 ha, with the
larger holdings being targeted for Bt adoption in the first year (this was a policy
of Vunisa) but with a more balanced pattern of adoption in season 2. In season 3,
there again appears to be a bias toward larger holdings for adopters compared to
non-adopters. In fact, season 2 was an unusual growing year, which had
repercussions for season 3. Season 2 was a particularly wet year with relatively
poor yields and relatively high levels of bollworm infestation.
Average yields of Bt adopters are significantly higher in all three seasons
than for non-adopters. Even in the poor growing conditions of season 2, the
13
yields of Bt adopters were far higher than non-adopters. As a consequence,
revenues of adopters were also much higher. In contrast, seed costs are higher for
adopters across the three seasons (due to the relatively high cost of the Bt cotton
seed) but the cost of bollworm sprays are significantly lower for adopters than
non-adopters. It should be noted that the Bt gene does not provide complete
resistance to bollworm throughout the plant’s life. Larger instars of bollworm
larvae are relatively unaffected. Hence some application of bollworm insecticide
is necessary for growers of Bt. Surprisingly, Bt adopters also spent less on non-
bollworm sprays. Bt-based resistance only protects against bollworm and not
other insect pests such as jassids and aphids. Given the reduced spray by Bt
adopters (both bollworm and non-bollworm insecticide), spray labour costs are
significantly lower. Harvest labour costs are higher for Bt adopters due to the
higher yields obtained.
Taking account of all of these costs and subtracting from the revenues
received, it can be seen that Bt adopters achieved substantially and significantly
higher gross margins (revenue less variable costs) than non-adopters across all
three seasons. In season 2, non-adopters actually had a negative average gross
margin. For many growers this caused problems for them paying credit (used to
buy the cotton seed, sprays and hired labour for growing cotton) back to Vunisa
such that they were less able to borrow more money to grow cotton in season 3
(hence the reason for the lower average areas planted in season 3).
Production functions
The results of the production function regression are shown in Table 3. The
regressors are almost all strongly significant, and the squared terms for the three
14
inputs are negative, indicating concavity of the production function. Most
interaction terms, especially those of the Bt variety with inputs apart from non-
bollworm insecticide, were persistently individually insignificant; an F test
indicated joint insignificance and these terms were subsequently dropped.
The Bt dummy variable is large, positive and strongly significant,
indicating that the new technology has a very significant effect on yields even
after controlling for all other variables. The interpretation of the Bt coefficient is
that, on average, a smallholder would gain about 934 kg of cotton across the full
cotton acreage by switching to the Bt technology. This amounts to approximately
259 kg/ha, or 65 % yield gain for the average farmer, all else remaining fixed.
Interestingly, the Bt-Bollworm pesticide interaction variable coefficient is
positive, large and highly significant. This indicates that Bt technology and
bollworm insecticide application may actually be complementary to each other in
some ways. Such an effect is not inconceivable given that both are biocidal and
hence there may well be an element of synergy between them (e.g. insecticide
application may weaken an insect’s ability to overcome plant resistance and
induce a greater mortality than either method of control by itself). Interestingly,
some have argued that such synergy would be unlikely with Bt-based resistance
(van Emden 2003).
The regression also confirms that time/seasonal effects are very
important. The 1999/2000 season was extraordinarily wet and correspondingly
marked by larger than normal pest infestation. Producers responded to this by
significantly increasing their pesticide application compared to 1998/99 (Table
1). Also, those who adopted Bt cotton in 1999/2000 would have had the
additional crop protection provided by the new technology. The coefficient on
15
the year 2 dummy in Table 3 indicates that an average farm still lost about 928
kg of cotton, or 226 kg/ha compared to the baseline year 1, after controlling for
the various changes. More broadly, this estimate underlines the riskiness of rain-
fed smallholder cotton and points out the value, indeed the necessity, of multi-
season data when evaluating Bt cotton technology.
The profit implications for small producers are revealed by a panel data
regression of gross margins per hectare on individual-specific and seasonal
terms, cotton variety (Bt vs. non-Bt technology), cotton area, and an interaction
term between variety and area. (Table 4).
While the R2 is low (11%) the coefficient attached to the Bt dummy in
Table 4 is large and significant, and indicates that with all else held equal,
switching to the Bt variety increases gross margins by approximately 562
SAR/ha, on average. Again, the deleterious effects of the bad weather season
(1999/2000) are obvious in the value of the coefficient for that season’s dummy
being a large negative number (–478.8). The cotton area variable is positive and
significant (205), indicating that larger farmers do have some slight economic
advantage. Most interestingly, however, the coefficient for the Bt-area interaction
dummy is negative and strongly significant, implying that Bt technology
adoption plays a role in neutralizing the size advantage.
Distribution of benefits from the technology
The effect of the technology on inequality of income can be assessed using the
Gini-coefficient (Sen 1997). The Gini-coefficient measures the equality (or
inequality) of distribution of income or a resource amongst a population. A value
of zero indicates that income is evenly spread amongst all of the members of the
16
population, while a value of one corresponds to extreme inequality (income is
concentrated in a few members). The Gini-coefficients of the gross margins of
the cotton farmers that make up the sample distribution are shown in Table 5.
Values are shown for the whole sample and for the sub-samples represented by
farmers who adopt and don’t adopt the Bt cotton.
It can be seen from Table 5 that the Gini-coefficient increases from 055
to 060 over the three seasons for all cotton producers. This indicates an increase
in income inequality. However, although inequality is higher for adopters in
1998/99 (i.e. because the larger farmers were targeted in that year for adoption of
the Bt variety), by 1999/2000 inequality has reduced and is somewhat lower than
for cotton growers generally and in 2000/01 it falls yet further. In contrast,
perhaps due to the poor harvest of 1999/2000 (especially for non-Bt growers) and
the consequences for many cotton growers in 2000/01 (i.e. problems repaying
credit to buy cotton inputs), the income inequality of non-adopters increases
significantly from 054 to 066.
Although the Gini-coefficient is a relatively crude measure of income
inequality, it does appear, in this instance at least, that the Bt technology has
helped to reduce inequality amongst smallholder cotton growers in Makhathini
compared to what may have been the position if they had grown conventional
cotton.
Toxic loads into the environment
The quantities (kg ai/ha) of the insecticides used by the farmers over the three
seasons is shown in Table 6. In all three seasons non-bollworm insecticide
accounted for the majority of the total active ingredient applied during the
17
season. Interestingly, non-Bt growers gradually increased their application of
insecticide over all three growing seasons from 0105 kg/ha in 1998/99 to 0158
kg/ha in 1999/2000 and 02 kg/ha in 2000/01. The corresponding figures for Bt
growers were more stable at 0049, 0095 and 0075 kg/ha respectively.
The quantities of ai applied varied between the two cotton varieties, with
Bt growers using less bollworm insecticide (mostly cypermethrin, but also
deltamethrin, fenvalerate and endosulfan) than growers of non-Bt cotton. Even
so, Bt growers still used some insecticide for controlling bollworm. Bt growers
also used less non-bollworm insecticide (mostly monocrotophos but also
dimethoate) than growers of conventional cotton. This may in part be due to a
misunderstanding amongst Bt growers that the resistance is effective against a
wide range of pests.
The results of converting the quantities of ai into the Biocide Index are
also shown in Table 6. Given that overall there is less insecticide used on Bt
cotton, it is no surprise to see that the Biocide Index for this variety was
significantly lower than for the non-Bt variety over all three seasons, implying
that the insecticide regime for Bt cotton was far less damaging to the
environment. While this is encouraging, it should be noted that much of this
benefit was due to the reduced use of non-bollworm insecticide by Bt growers.
Given that the non-bollworm insecticides are particularly damaging (both
monocrotophos and dimethoate are ranked in category ‘I’ by the WHO) the
benefits would be less if Bt growers applied the same levels as conventional
growers. Bt growers may start doing this once they see that the variety provides
no resistance to foliar pests such as aphids.
18
It is also interesting to note that adoption of Bt cotton did not reduce the
total insecticide or Biocide Index per hectare over the three seasons. Between
1998/99 and 1999/2000 the Biocide Index for all cotton producers calculated on
a per hectare basis increased by nearly 50 %. This change was largely due to
increased use of non-bollworm insecticide amongst non-Bt adopters. If Bt
adopters had also increased their use of non-bollworm insecticide in line with
non-adopters then the increase in Biocide Index would have been even greater.
DISCUSSION
It should first be noted that the number of records used in these analyses
represent 89, 32 and 33 % in the three seasons respectively of the total
population of smallholder cotton growers in the area. Such sample sizes help to
reduce possible problems of sample bias. Moreover, when comparing the
characteristics (such as gender, age, cotton area planted) of these large samples
with the population, they were generally found to be representative. In addition,
the data are based on actual records (electronic and written) of input expenditures
and output rather than relying on farmer recall (which can be unreliable).
The results of analysis of the relatively large dataset reported here show
substantial and significant financial benefits to smallholder cotton growers of
adopting the Bt variety over three seasons. Bt growers had significant benefits
compared to growers of non-Bt cotton in terms of gross margin over the three
seasons. All other things being equal, Bt growers had a increase in gross margin
of SAR 562/ha over growers of non-Bt. This is a significant sum. Assuming a
typical daily wage of SAR 10/day in South Africa this equates to almost 2
19
months paid work. Of added interest is the finding that those with the smaller
holdings appeared to benefit proportionately more from the technology (in terms
of higher gross margins) than those with larger holdings. Given often-quoted
concerns about the bias of new technologies in favour of larger farmers,
particularly in developing countries, in this case at least, the genetically modified
technology would appear to have equity benefits as well as general benefits to
profitability. Indeed the lower Gini coefficient for Bt adopters suggests that gross
margin benefits were more evenly distributed within that population than in the
non-Bt sample.
However, while insecticide use (and Biocide Index) was lower for Bt
plots compared to non-Bt, this was due in part to an apparent misunderstanding
on the part of Bt adopters. There is no agronomic reason why Bt adopters should
reduce their use of non-bollworm insecticide, yet that is what they are doing. It is
the non-bollworm insecticides which have the largest impact on insecticide use –
more quantities of them are used and they have a higher toxicity. Therefore it
should not be assumed that the introduction of Bt cotton will inevitably reduce
toxic load to the environment arising from insecticide.
Clearly, the findings presented here relate to just one crop in one part of
the developing world, although substantial benefits to farmers of adopting Bt
cotton have also been reported by others in other countries (Manwan & Subagyo
2002; Naik 2001; Pray et al. 2002; Qaim 2003; Traxler et al. 2001). The impacts
of adoption of GM crops need to be assessed on a case-by-case basis, since they
will inevitably depend on the nature of the crop and the particular circumstances
of where and how it is grown.
20
CONCLUSION
The uptake of Bt cotton by smallholder farmers in the Makhathini Flats area of
South Africa has resulted in substantial and significant economic benefits to
growers over three seasons, with the smaller growers benefiting as much, if not
more, from the new technology as the larger growers. However, the toxic load to
the environment arising from insecticides increased over the same three seasons.
Care needs to be taken in extrapolating assumptions of environmental benefit
from an apparently logical stance that the introduction of Bt-based resistance
must reduce insecticide use. Much depends upon the type of insecticides being
used in the regime as a whole, and how farmers perceive their pest problems.
REFERENCES
EXTOXNET. EXTENSION TOXICOLOGY NETWORK. http://www.extoxnet.orst.edu
(date of access: 11/03/2003)
ISMAEL, Y., BENNETT, R. & MORSE, S. (2002). Benefits from Bt cotton use by
smallholder farmers in South Africa. AgBioForum 5, 1–5.
JAMES, C. (2002). Global review of commercialised transgenic crops featuring Bt
cotton. ISAA Brief No. 26. International Service for the Acquisition of Agri-
Biotech Applications, Ithaca, USA.
21
JAMES, C. (2003). Global review of commercialised transgenic crops: 2003.
ISAA Brief No. 30. International Service for the Acquisition of Agri-Biotech
Applications, Ithaca, USA.
JANSEN, D. M., STOORVOGEL, J. J. & SCHIPPER, R. A. (1995). Using
sustainability indicators in agricultural land use analysis: an example from
Costa Rica. Netherlands Journal of Agricultural Science 43, 61–82.
MANWAN, I. & SUBAGYO, T. (2002). Transgenic cotton in Indonesia: Challenges
and opportunities. Paper presented at the regional workshop for the South
East Asian Biotechnology Information Centers, Philippines 30–31 July 2002.
NAIK, G. (2001). An analysis of socio-economic impact of Bt technology on
Indian cotton farmers. Centre for Management in Agriculture, Indian Institute
of Management, Ahmedabad, India.
ORTON. L. (2003). GM crops – going against the grain. Actionaid (available at
www.actionaid.org). Date of access: 25/05/2003
PRAY, C., ROZELLE, S., HUANG, J. & WANG, Q. (2002). Plant biotechnology in
China. Science 295, 674–676.
QAIM, M. (2003). Bt cotton in India. Field trial results and economic projections.
World Development 31(12), 2115-2127.
QAIM, M. & DE JANVRY, A. (2002). Bt cotton in Argentina: analysing adoption
and farmers’ willingness to pay. Paper Presented at the American
Agricultural Economics Association (AAEA). Long Beach, California, USA.
28–31 July 2002.
QAIM, M. & ZILBERMAN, D. (2003). Yield effects of genetically modified crops
in developing countries. Science 299, 900–902.
22
ROTHER, H. A. (1998). Influences of pesticide risk perception on the health of
rural South African women and children. African Newsletter on Occupational
Health and Safety 10(2), 42-46.
SEN, A. (1997). On economic inequality. Expanded edition with annex.
Clarendon, Oxford.
TRAXLER, G., GODOY-AVILLA, S., FALCK-ZEPEDA, J. & ESPINOZA-ARELLANO, J.
J. (2001). Transgenic cotton in Mexico: economic and environmental
impacts. Paper Presented at the 5th International Conference, Biotechnology,
Science and Modern Agriculture: A new industry at the dawn of the century,
Ravello, Italy, 15–18 June 2001.
VAN EMDEN, H. F. (2003). GM crops: A potential for pest mis-management. Acta
Agriculturae Scandinavica Section B – Soil and plant science 53, 26–33.
23
Table 1. Some characteristics of the insecticides used for cotton production in Makhathini
Type of
insecticide
Concentration
(g ai/litre)
soil half-life in
days1
LD50 (oral, mg/kg)
WHO
category
Biocide Index
weighting
Minimum
Maximum
Non-bollworm insecticides
Monocrotophos
OP
400
7
17
20
I
7
Dimethoate
OP
400
4–16
60
400
I
7
Bollworm insecticides
Cypermethrin
PY
200
8–6
82
3000
II
5
Deltamethrin
PY
165
7–14
21
5000
II
5
Fenvalerate
PY
85
30
451
3200
II
5
Endosulfan
OC
350
35–150
2
220
I
7
OP = organophosphorius PY = pyrethroid OC = organochlorine
1 For the calculation of the Biocide Index the lowest value in the range was employed
24
Table 2. Age, gender, area planted, yields, revenue, input costs and gross margins (revenue – input costs) of Bt and non-Bt cotton
producers 1998/1999–2000/01 (95 % confidence limits are given in parentheses)
Season 1 (1998/1999)
Season 2 (1999/2000)
Season 3 (2000/2001)
Non-adopters
Adopters
P <
Non-adopters
Adopters
P <
Non-adopters
Adopters
P <
Age
Gender (% male)
Area planted (ha)
45 (07)
43
39 (02)
45 (26)
49
60 (13)
001
46 (132)
42
31 (02)
46 (25)
45
27 (05)
ns at P
<0.05
43 (16)
37
19 (02)
41 (15)
43
30 (04)
0001
Yield (kg/ha)
452 (269)
738 (1188)
0001
264 (234)
489 (688)
0001
501 (825)
783 (946)
0001
Revenue (SAR/ha)
984 (585)
1605
(2586)
0001
574 (508)
1064 (1495)
0001
1090 (1792)
1704 (2057)
0001
Input costs (SAR/ha)
Seed
138 (94)
278 (459)
0001
190 (196)
413 (654)
0001
176 (183)
260 (215)
0001
Pesticide (all)
153 (113)
72 (206)
0001
222 (151)
104 (189)
0001
305 (302)
113 (160)
0001
(a) bollworm
73 (56)
38 (124)
0001
91 (89)
5 (45)
0001
138 (296)
26 (68)
0001
(b) non-bollworm
81 (7)
34 (124)
0001
131 (92)
100 (187)
005
167 (241)
87 (134)
0001
Spray labour (all)
77 (58)
38 (107)
0001
108 (76)
49 (89)
0001
135 (129)
45 (69)
0001
(a) bollworm
38 (31)
20 (71)
0001
46 (49)
3 (30)
0001
50 (70)
11 (29)
0001
(b) non-bollworm
39 (34)
17 (62)
0001
62 (44)
46 (87)
001
85 (92)
34 (56)
0001
Harvest labour
113 (67)
184 (298)
0001
66 (58)
122 (172)
0001
125 (206)
196 (236)
0001
Total
481 (265)
572 (832)
005
586 (363)
688 (871)
005
741 (628)
613 (483)
001
Gross margin (SAR/ha)
502 (529)
1033 (208)
0001
11 (537)
376 (1444)
0001
348 (1515)
1090 (1698)
0001
Sample size (N)
1196
87
329
112
254
245
SAR = South African Rand (1US$=11 SAR in October 2002). Bollworm pesticide refers to pesticide that is sprayed specifically to
control bollworm, while non-bollworm pesticide is used for other pests (e.g. aphids).
25
Table 3. Production function on panel data: whole farm cotton output as a function
of inputs
Variable
Coefficients
P <
Intercept
19464
ns P < 01
Season 2 dummy
–92804
001
Season 3 dummy
–61026
001
Cotton Area
22645
001
Bollworm insecticide
8639
001
Non-bollworm insecticide
8266
001
Bt dummy
93399
001
(Cotton Area)2
–175
ns P < 01
(Bollworm Insecticide)2
–1540
001
(Non-bollworm Insecticide)2
–085
ns P < 01
Area * Bollworm insecticide
4167
001
Area * Non-bollworm insecticide
–825
001
Bollworm insecticide * Non-bollworm insecticide
028
ns P < 01
Bt dummy * Bollworm Insecticide
25202
001
R2
048
26
Table 4. Random effects regression model on panel data showing factors influencing
gross margin per hectare (SAR/ha)
Coefficients
P <
Intercept
3779
001
Season 2 Dummy
–4788
001
Season 3 Dummy
–1503
01
Cotton Area
205
005
Bt Dummy
5621
001
Bt Dummy*Cotton Area
–401
001
Adjusted R2
011
27
Table 5. Gini-coefficients of income inequality based on the distribution of gross
margins from cotton growing
Season
All cotton growers
Bt adopters
Non-adopters
1998/99
055
057
054
1999/2000
061
056
061
2000/01
060
049
066
Sample size
1283
441
499
Gini coefficient is a measure of distribution. A value of 0 indicates that distribution is
even amongst the population while a value of 1 indicates maximum inequality in
distribution.
28
Table 6. Application of insecticide active ingredient (ai) applied to non-Bt and Bt cotton over three seasons along with the conversion of ai to
Biocide Index (BI)
1998/1999
1999/2000
2000/2001
Insecticide
Non-Bt
Bt
P <
Non-Bt
Bt
P <
Non-Bt
Bt
P <
Monocrotophos ai
0684 (006)
0314 (0122)
0001
1153 (0083)
0942 (0175)
ns at P
<0.05
17 (0183)
0678 (0113)
0001
Dimethoate ai
0131 (0026)
0065 (0048)
ns at P
<0.05
0126 (0035)
0008 (0012)
0001
0 (0)
0 (0)
n/a
Total non-bollworm ai
0815 (0068)
0379 (0124)
0001
1278 (0091)
095 (0175)
005
17 (0183)
0678 (0113)
0001
Cypermethrin ai
0186 (0015)
0093 (0032)
0001
02 (0019)
0004 (0004)
0001
028 (0035)
0067 (0018)
0001
Deltamethrin ai
0006 (0002)
0002 (0002)
ns at P
<0.05
0 (00002)
0 (0)
n/a
0019 (0014)
0 (00006)
001
Fenvalerate ai
0001 (00005)
0 (00004)
n/a
0 (0)
0 (0)
n/a
0 (0)
0 (0)
n/a
Endosulfan ai
0037 (0014)
0019 (00195)
ns at P
<0.05
0101 (0032)
0 (0)
0001
0 (0)
0 (0)
n/a
Total bollworm ai
023 (0021)
0114 (0037)
0001
0301 (0038)
0004 (0004)
0001
0299 (0041)
0067 (0018)
0001
Total insecticide ai (variety)
1045 (0082)
0493 (0146)
0001
1579 (0109)
0954 (0175)
0001
1999 (0198)
0745 (0118)
0001
Total insecticide ai (season)
101 (008)
142 (01)
138 (013)
% non-bollworm ai
781
776
81
100
85
907
% bollworm ai
219
224
19
0
15
93
BI (non-bollworm)
2191 (1831)
1015 (3406)
0001
3491 (245)
2654 (4896)
005
4759 (5122)
1898 (3174)
0001
BI (bollworm)
595 (0856)
298 (1257)
001
965 (1832)
009 (0075)
0001
598 (0819)
134 (0355)
0001
BI (total insecticide; variety)
2786 (2298)
1313 (4008)
0001
4456 (334)
2663 (4895)
0001
5357 (5405)
2032 (3266)
0001
BI (total insecticide; season)
2686 (217)
4001 (288)
3725 (351)
29
Sample size
1196
87
329
112
254
245
Figures are the average and 95 % confidence interval (parentheses).Active ingredient is in kg/ha, while Biocide Index is adjusted per hectare.
; n/a = not applicable.
