ArticlePDF Available

Genetically Modified Cotton and Farmers' Health in China


Abstract and Figures

This study provides the first evidence of a direct link between the adoption of a genetically modified (GM) crop and improvements in human health. Estimation of the impact of Bacillus thuringiensis (Bt) cotton adoption on pesticide use from data from a survey of cotton farmers in northern China, 1999-2001, showed that Bt cotton adoption reduced pesticide use. Assessment of a health-production function showed that predicted pesticide use had a positive impact on poisoning incidence. Taken together, these results indicate that the adoption of Bt cotton can substantially reduce the risk and the incidence of poisonings.
Content may be subject to copyright.
This study provides the first evidence of a direct link
between the adoption of a genetically modified (GM)
crop and improvements in human health. Estimation
of the impact of Bacillus thuringiensis (Bt) cotton adop-
tion on pesticide use from data from a survey of cotton
farmers in northern China, 1999–2001, showed that Bt
cotton adoption reduced pesticide use. Assessment of a
health-production function showed that predicted pes-
ticide use had a positive impact on poisoning inci-
dence. Taken together, these results indicate that the
adoption of Bt cotton can substantially reduce the risk
and the incidence of poisonings. Key words: Asia; China;
biotechnology; pesticide; farmers; health; poisoning;
genetically modified.
ritics of biotechnology have made health, food
safety, and environmental fears the centerpiece
of their attacks. There is no evidence that
anyone’s health has been harmed by modifications
resulting from biotechnology, but they imply that
hidden scourges such as mad cow disease lurk in
biotechnology-modified food and will some day harm us,
the implication being that it is safer to use current tech-
nology than to take the risks associated with biotechnol-
ogy. This implies that current technology is safe.
One component of the current technology for grow-
ing crops in developing countries is heavy use of chem-
ical pesticides, which kill and sicken many farmers and
farm laborers each year and cause debilitating sick-
nesses for years after exposure.
In China during the
period 1992–1996, the last five years for which aggre-
gate data are available, there were an average of 54,000
poisonings of farmers annually and approximately 490
deaths of farm workers or farmers by pesticide poison-
ings each year.
In addition, pesticides make their way
to consumers as residues on fruits, vegetables, and
grains and through contaminated water supplies. It is
clear in many countries that the use of pesticides,
which are the current alternative to biotechnology, car-
ries many immediate as well as longer-term risks to
human health.
The real choice for policymakers, farmers, and con-
sumers is not between a technology that may have risks
in the future and a completely safe technology. The
real choice is between genetically modified crops that
so far have proven safe but may be found to have some
health and environmental risks in the future and con-
ventional pest management technology that demon-
strably is implicated in thousands of cases of sickness
and hundreds of deaths each year.
To make this choice, however, policymakers need to
know how serious the risks from pesticides are and
whether the adoption of the specific types of biotech-
nology that are currently available will actually reduce
pesticide used by farmers and, in turn, the number of
poisonings. The main objective of this study was to
measure the relationship between pesticide poisonings
and the use of one specific product of biotechnology,
cotton that is has been genetically modified (GM) with
a gene from a soil bacterium, Bacillus thuringiensis (Bt),
that is deadly to bollworms and other similar pests. The
use of Bt cotton is dramatically reducing the use of
insecticides by farmers in industrialized
and develop-
ing countries.
However, no-one has conclusively docu-
mented the links between use of Bt technology, insecti-
cide use, and poisonings.
In previous papers using the data from a 1999 survey
of 286 farmers,
we have documented the lower mean
level of pesticide use and lower levels of reported poi-
soning by farmers who used Bt cotton compared with
those who used conventional techniques. We did not
explicitly model the relationship between Bt cotton
adoption and poisonings in such a way that we could
test the influence of Bt cotton use holding other impor-
tant influences on poisoning constant. In the crop
years 2000 and 2001 we increased our sample size to
about 400 farmers in Northeastern China. Based on
the data from three years and more provinces we are
able to confirm the hypotheses suggested in the earlier
studies that the adoption of at least some types of
genetically-modified crops can reduce the risk of pesti-
cide poisoning.
Genetically Modified Cotton and
Farmers’ Health in China
Received from the Department of Agricultural, Food and
Resource Economics, Rutgers University, New Brunswick, New Jersey
(FH, CEP, YL); and the Center for Chinese Agricultural Policy, Insti-
tute of Geographical Sciences and Natural Resource Research, Chi-
nese Academy of Sciences, Beijing, China (JH, CF, RH). Supported
by grants from the Rockefeller Foundation, the China Natural Sci-
ence Foundation, and Monsanto.
Address correspondence and reprint requests to: Carl E. Pray,
Department of Agricultural, Food and Resource Economics, Rutgers
University, 55 Dudley Road, New Brunswick, NJ 08901-8520, U.S.A.
VOL 10/NO 3, JUL/SEP 2004 Genetically Modified Cotton and Health 297
In recent years pesticide use has been the main method
of controlling insect pests in China. Both the active
ingredients and the formulated pesticides are mainly
produced by Chinese companies. Foreign suppliers
have been limited by Chinese regulations to approxi-
mately 20% of the market. The pesticides are distrib-
uted to farmers by government input supply organiza-
tions and the extension service. The government
extension service not only supplies the technology but
also does scouting for pests and provides advice to farm-
ers about when to spray and what to spray. At present
the Chinese pesticide market is probably the largest in
the world based on quantity used, and China competes
with the United States for the highest-value market.
More pesticide is applied per hectare to cotton than
to any other major field crop, although the amount
used is less than for most vegetable crops (Table 1).
Government officials are aware of the dangers of pesti-
cides and have put policies in place to reduce the worst
dangers. For example, they banned the use of chlori-
nated hydrocarbons such as DDT, endosulfan, and BHC
in 1983 to eliminate their impacts on the environment
and their longer-term health risks. However, the gov-
ernment did not ban the use of some very dangerous
organophosphate pesticides. Through the extension
system the government has tried to promote integrated
pest management practices with the goals of reducing
pesticide use and using pesticides more effectively. Nev-
ertheless, pesticide use continues to grow rapidly.
After the government banned the use of chlorinated
hydrocarbons in the early 1980s, organophosphates
were the main type of pesticide used to control boll-
worm. However, bollworms that were resistant to most
organophosphates evolved and farmers had to shift to a
new type of pesticide called pyrethroids. These pesti-
cides were effective for a while against bollworm and had
the added advantage of being relatively safe for the farm-
ers that applied them. However, by the mid 1990s boll-
worms had developed resistance to the pyrethroids, also.
In 1997 Bacillus thuringiensis (Bt) cotton, represent-
ing an entirely new technique for controlling pests, was
introduced to the farmers. This type of cotton had
been genetically engineered to contain the Bt gene
that produces a protein that kills many lepidoptera
insects. This type of pest includes many serious pests of
agricultural plants, including the bollworm (Helicov-
erpa armigera), which is the main cotton pest in north-
eastern China. Both the Chinese Academy of Agricul-
tural Sciences and a joint venture between Monsanto,
Delta, and Pineland and the Hebei Provincial Seed
Company developed varieties of Bt cotton for farmers.
Farmers found that Bt cotton gave much better protec-
tion against bollworm than chemical pesticides—it
increased yields while reducing the costs of insect con-
trol, thereby increasing the farmers’ net income.
As a
result, farmers have adopted it rapidly (see Figure 1).
In Hebei and Shandong, Bt cotton is grown on nearly
all farms in the cotton-growing area. In the country as
a whole at least 30% of the cotton area is planted with
Bt cotton varieties.
The reduction in pesticide use due to the adoption of
Bt cotton has been substantial. Huang et al.
mented the reduction in pesticide use with data from a
1999 survey of 286 farmers from Hebei and Shandong
provinces. They also found that a far larger percentage
of farm families who grew conventional cotton
reported sickness due to pesticide use compared with
those who used only Bt cotton. However, the total
TABLE 1. Pesticide Use on Key Crops in China, 1980–1998*
Per Hectare Pesticide Cost (US$ at 1995 Prices)
Rice Wheat Maize Cotton Tomato Cucumber
1980 11 3 1 31 NA NA
1985 14 3 1 35 NA NA
1990 16 5 2 46 45 56
1995 25 8 7 101 105 97
1998 25 9 7 88 136 129
Source: Huang et al.
Figure 1—Total hectares of cotton grown (left, light
bars) and Bt cotton grown (right, darker bars),
number of cases was not large enough to econometri-
cally test the relationship between Bt cotton use and
pesticide use.
In 2000 and 2001 second and third surveys of Bt
cotton’s impact were carried out. The number of farm-
ers was increased in 2000 by adding farms in the Henan
Province. In 2001 the survey continued to interview
farmers from Hebei, Shandong, and Henan and also
added farmers from Anhui and Jiangsu. Farmers were
chosen by first selecting counties in which Bt and con-
ventional cotton were grown and then randomly choos-
ing villages and farmers. The same questionnaire was
used in 2000 and 2001 that had been used in 1999
except that in 2000 and 2001 the farmers were asked
for more information about their experience with Bt
cotton adoption over the preceding five years and
about sickness from pesticide poisoning in the preced-
ing five years.
Average pesticide use of Bt cotton growers and the
numbers of poisonings of Bt and non-Bt cotton users in
2000 and 2001 were higher than in 1999. Farmers who
grew only Bt cotton applied about 18 kg of formulated
pesticide per ha, while farmers who grew only conven-
tional cotton sprayed about 46 kg per ha (Table 2). Nine
percent of the farmers who exclusively used Bt cotton
reported poisoning, while almost a third of the farmers
who exclusively used non-Bt cotton reported poisoning.
The linkages between Bt cotton adoption, reduction
of pesticide use, and reduced poisoning incidence are
further strengthened by the evidence presented in
Tables 3 and 4. Table 3 categorizes the pesticides used
by chemical type. The use of organophosphates showed
the greatest decline. A number of organophosphates
are rated highly for acute toxicity—category I in the
Chinese and international systems, which rate pesticides
from I to IV according to acute toxicity. Table 4 shows
the toxicity levels and the numbers of users reporting
poisonings for the insecticides that had caused the most
poisonings during the preceding five years. Five of the
top six pesticides, ranked by number of farmers report-
ing poisonings, were organophosphates. Furthermore,
the most popular pyrethroid pesticide, cypremethrin, is
a category II pesticide. It is not surprising, then, that a
decline the amount of organophosphates used would
result in a reduction in poisonings.
Our objective was to identify the impact of Bt cotton
adoption on the incidence of poisoning of farmers
holding other factors constant. It is hypothesized that
the adoption of Bt cotton varieties affects poisoning
incidence via two channels. First, adoption of Bt cotton
varieties affects poisoning incidence by reducing the
quantity of pesticide used. Pesticide use can have a
direct impact on health.
Second, it can also affect
health by reducing the number of times that farmers
are exposed to pesticides through spraying. These two
channels are likely to be closely correlated, but it may
be possible to measure their separate impacts. A
number of studies have modeled farmer’s health as a
production function. Health production-function
theory states that health is commodity usually repre-
sented by mortality and morbidity. Here, incidence of
poisoning is used as a proxy variable for morbidity. The
health-production function is described as:
Poisoning = f [pesticide use, adoption of Bt cotton,
farmer human capital (education, age), farmer
income, the environment (provincial dummies)]
Poisoning represents the number of times a farmer
reports sickness from applying pesticide. Pesticide use
is measured as the quantity of formulated pesticide
TABLE 2. Environmental and Health Impact of Bt Cotton, 2000
Poisonings† Reported by Farmers in 2000 Season
Required visit to:
% Farmers
No. of Quantity* _______________________ Self Reporting
Farmers (Kg/Ha) Hospital Doctor Treatment Total Poisoning
Bt 316 18 2 3 23 28 9
Both Bt and non-Bt 61 29 0 6 10 16 26
Non-Bt 30 46 2 2 6 10 33
*Total pesticide (active + inert ingredients).
†Farmers were asked whether they had had headache, nausea, skin pain, or digestive problems when they applied pesticides.
TABLE 3. Average Quantities (kg/ha) of Farmers’
Pesticides Use by Type of Pesticide, 2000
Average Quantity
Non-Bt Decline
Bt Varieties Varieties in Use
(n = 377) (n = 90) (%)
Organochlorines 1.6 3.9 58
Organophosphates 8.8 21.0 58
esters 0.3 0.4 25
Pyrethroids 5.2 13.0 60
Organosulfates 2.8 6.0 53
Other insecticides 0.8 2.1 64
Fungicide 0.1 0.3 62
Herbicide 0.8 1.2 32
TOTAL 20.5 48.0 57
applied per household. The characteristics of the
farm households are from our survey. The income
variable may be related to health services in China,
since the health system is privatizing. We control for
environmental variables only by using provincial
dummy variables.
The quantity of pesticide use in the above equation
is a decision variable that depends on a number of vari-
ables. Farmers’ decisions regarding how much pesti-
cide to apply is a function of total area under cotton,
area under Bt cotton, price of pesticides, severity of
pest attack, the human capital of farmers, extension
advice, and environmental variables. Specifically, the
quantity of pesticide used is modeled as:
Pesticide = f [total cotton area, area under Bt cotton,
Pesticide price, severity of pest attack, farmer’s
human capital (education and age);extension
advice; environmental factors (provincial dummies)]
The dependent variable is the quantity of pesticide
used by a farm. Pesticide price is obtained as the total
cost of all the pesticides applied to cotton divided by
the quantity used. Severity of attack is the farmer’s per-
ception of the severity of the attack of bollworms in that
year (defined as a dummy variable). Education and age
of the farmer are from our survey. Data on extension
advice were not available for all years so it was dropped.
The dummy variable for the provinces gives a crude
measure of the different environments. Cotton area is
each farmer’s total area of cotton, and Bt cotton is
farmer’s area under Bt cotton.
Two different empirical frameworks were used to
model the relationship between farmer health and the
adoption of Bt cotton. In the first approach, we used a
logistic model to examine the impact of Bt cotton
adoption on the likelihood of a farmer’s becoming sick
from pesticide poisoning. In the other approach, we
used the Poisson regression model to analyze the
impact of Bt cotton use on the number of pesticide-
related poisoning incidents.
The Logit Framework
The logit framework models the relations between the
likelihood of an individual farmer’s becoming sick
from pesticide poisoning and the socioeconomic and
farm characteristics of the respondent. In order to
implement the model, a binary dependent variable poi-
soning is defined as follows:
Poisoning = 1 if the farmer i became sick at least
once from pesticide during the year, and 0 otherwise
The empirical model assumes that an individual
farmer’s probability of becoming sick from pesticide
use, P
, depends on a vector of independent variables
) associated with farmer i and variable j, and a
vector of unknown parameters :
= F(Z
) = F(X
) = 1/[1+exp(–Z
)= the value of logistic cumulative density func-
tion associated with each possible value of the
underlying index Z
= the probability that a farmer becomes sick from
pesticide use, given the independent variables
In the above equation, X
is a linear combination of
the independent variables so that
+ . . . +
=unobserved index level or the logarithm of the
odds ratio of the i
attribute of the i
= parameters to be estimated;
= random error or disturbance term.
For empirical analysis, the following equation is
Poisoning = f [pesticide use, adoption of Bt cotton,
farmer human capital (education, age), farmer
income, the environment (provincial dummies)]
Since some of the variables on the right-hand side of
the above equation also influence the quantity of pesti-
cide used, the estimated coefficient of pesticide is likely
to be biased. To control for this problem, we first use
OLS (ordinary least square) to estimate the pesticide
equation model. The predicted pesticide quantity
obtained from this model is then used as an explana-
tory variable in the logistic equation. In addition, in
some specifications of the health production model we
used the Bt cotton variable again because the number
of times sprayed might have an impact directly on
farmers apart from the quantity of pesticide used.
VOL 10/NO 3, JUL/SEP 2004 Genetically Modified Cotton and Health 299
TABLE 4. Type and Toxicity Level sof Pesticides
Causing Farmer Poisonings, 1995–2000
Category Poisoning
Toxicity Cases
Chlordimeform I 94
Parathion-methyl I 65
Acephate I 19
Carbofuran (furadan) I 9
Phorate I 9
Parathion III 8
Monocrotophos I 5
Cypermethrin II 12
Killingthrin 39 III 6
After eliminating the statistically insignificant vari-
ables from the right hand side, the following pesticide
use equation was estimated using OLS data:
Pesticides =
Henan +
cotton +
Bt cotton +
Pesticide price
The predicted pesticide quantities were then used as an
explanatory variable in the logit model.
Poisson Regression Model
The logistic model above is unable to differentiate
between single and multiple incidences of pesticide-
related poisoning. Some farmers were victims of poi-
soning more than once. It is entirely possible that as a
result of adopting Bt cotton, a farmer was able to
reduce the number of times he or she became sick
from poisoning. The dependent variable in the logistic
model is unable to capture this reduction in the inci-
dence of poisoning associated with Bt cotton adoption.
In order to account for multiple poisoning incidences,
a Poisson regression model was estimated to examine
the impact of Bt cotton adoption on poisoning inci-
dence. In the Poisson regression framework, the
expected number of poisoning incidences is modeled
as a function of pesticide quantity, adoption of Bt
cotton, and other farm/farmer characteristics. Specifi-
cally, the Poisson model stipulates that the number of
poisoning incidence for farmer i follows a Poisson dis-
tribution with a conditional mean (µ
) that depends on
pesticide use, adoption of Bt cotton, and other
farm/farmer characteristics. Formally, the model is
specified as follows:
= E(number of poisoning cases | X) = exp(X
where X
represents the same set of explanatory vari-
ables used in the logit model and corresponds to
farmer i, and is the parameter vector.
The object of this study was to examine how pesticide
poisoning is affected by the adoption of Bt cotton. In
areas where only Bt cotton is grown, we do not get any
variation in the Bt cotton area to examine its impact on
the pesticide-related incidence of poisoning. There-
fore, in the econometric analysis we exclude those
provinces when only Bt cotton was produced. For
instance, farmers in Hebei province produced only Bt
cotton during 1999–2001. Therefore, we excluded all
observations from Hebei province for those years. For
the same reason, 2000 and 2001 observations from
Shandong province were also excluded from this study.
As a result, the sample used in the study consisted of
cotton growing farm households from four provinces
as follows: for 2001, 79 farmers from Anhui, 121 farm-
ers from Jiangsu, and 81 farmers from Henan; for 2000,
147 farmers from Henan; and for 1999, 183 farmers
from Shandong. In total, there were 611 observations
in the sample used for this study.
The estimated coefficients of the pesticide use equa-
tion, standard error, and p values are presented in
Table 5. All independent variables are statistically sig-
nificant at the 0.05 level except one. The adjusted R
was 0.47, which is reasonable given that the results are
based on cross-section data. As we expected, the sign of
Bt cotton was negative and statistically significant,
which means that an increase in the area under Bt
cotton, holding total cotton area constant, reduces the
quantity of pesticides used. On the other hand, the sign
of cotton is positive, which implies that an increase in
the total area under cotton leads to increased use of
pesticides. Average mixed pesticide price has the
expected negative impact on quantity of pesticides use,
which indicates that higher pesticide prices lead to less
pesticide use. The variable for the severity of the pest
attack is positive and significant at the 0.10 level.
In this model, it is possible that the area under Bt
cotton is endogenous in the sense that it depends on
other independent variables and is correlated with the
model error term. We explored this possibility by
regressing the area under Bt cotton on pesticide price
and farmers’ economic and demographic attributes.
However, none of the explanatory variables was found
to be statistically significant. We also computed Haus-
test for endogeneity of area under Bt cotton. To
implement this test, we regressed area under Bt cotton
on pesticide price and farm/farmer characteristics. The
predicted area under Bt cotton was then used as an
additional independent variable in the pesticide use
equation. The t ratio of the predicted Bt cotton area was
–1.13, which gave an F statistic (the square of the t ratio
TABLE 5. Estimated Coefficients of the Pesticide-use
Model: First-stage Regression
Variable Coefficient Error t-Ratio
Intercept 72.20 7.55 9.56
Total cotton area 5.64 0.82 6.91
Area under Bt cotton –5.57 0.81 –6.87
Pesticide price –0.75 0.17 –4.38
Heavy Incidence of
pest attack 13.08 7.31 1.80
Dummy for Henan
Province –23.59 7.58 –3.11
Dummy for Shandong
Province –29.22 7.71 –3.79
Dummy for Anhui
Province 90.64 8.72 10.39
Adjusted R
F-Statistic of model significance 79.39
Degrees of freedom 7
Note: Dummy variable for Jiangsu Province is excluded from
this estimation.
VOL 10/NO 3, JUL/SEP 2004 Genetically Modified Cotton and Health 301
in this case) of 1.28. This F statistic (with appropriate
degrees of freedom) is equivalent to the Hausman’s test
Using appropriate degrees of freedom, we could
not reject the null hypothesis that area under cotton is
not endogenous (i.e., there is no evidence of endo-
geneity for the variable area under Bt cotton).
The estimated coefficients along with t ratios from
the logit model are presented in Table 6. The goodness
of model fit can be evaluated by the chi-square statistic
of overall model significance. Given the estimated
values of the unrestricted and restricted (i.e., all coeffi-
cients other than the intercept) log-likelihood func-
tions (–358.08 and –438.26, respectively), the estimated
log-likelihood ratio–based chi-square statistic is 160.35,
which clearly exceeds the 95% critical value of the test
statistic with 8 degrees of freedom. This indicates that
that the model has significant explanatory power.
Another goodness-of-fit measure is McFadden’s R
which is obtained as 1 – LL
, where LL
and LL
are values of the unrestricted and restricted (i.e., all
slope coefficients are zero) log-likelihood functions.
The estimated value of McFadden’s R
is 0.143, which is
quite reasonable for cross-sectional data.
The coefficient of the (estimated) pesticide quantity
is positive and statistically significant at the 0.05 level.
This confirms our hypothesis that pesticide use is
strongly and positively associated with poisoning. It
therefore follows that adoption of Bt cotton, which
reduces the quantity of pesticide use, also reduces the
likelihood of getting sick from pesticide poisoning. Fur-
ther, the coefficient of Bt cotton is negative and signif-
icant at the 0.10 level. This implies that increases in Bt
cotton adoption reduce the probability of getting sick
from pesticide poisoning. Since the effect of Bt cotton
through the quantity of pesticide sprayed is picked up
by the pesticide quantity, this variable is picking up the
added impact of Bt cotton on the number of times
farmers have to spray. Farmers who have to spray more
times, holding quantity constant, will be exposed more
to pesticides and thus be more likely to be poisoned.
The positive sign on the dummy for poor health indi-
cates that these farmers have a greater likelihood of
getting sick from pesticide poisoning than farmers in
good health. Education, however, reduces farmers’
probability of poisoning. The statistically insignificant
coefficient of the age variable suggests that probability
of poisoning is not related to variations in age.
The marginal effect of Bt cotton adoption on the
probability of poisoning incidence works through two
channels. First, adoption of Bt cotton reduces pesticide
use and thereby lowers the chances of poisoning. This
is given by
Pr(poisoning) pesticide
_____________ _________ .
pesticide Bt
The other effect works through the coefficient of Bt
cotton in the logit equation. This is given by the direct
partial derivative of the dependent variable with
respect to Bt cotton, holding pesticide quantity
unchanged. Therefore, the marginal effect of Bt cotton
adoption on the probability of poisoning is given by:
Marginal effect =
Pr(poisoning) pesticide Pr(poisoning)
______________ _________ + _____________ .
pesticide Bt Bt
Using the above formula, the marginal effect of Bt
cotton adoption can be obtained as follows. The deriv-
ative Pr(poisoning)/pesticide, evaluated at mean
values of other variables, is 0.0023. From Table 5, pes-
ticide/Bt is –5.57. Finally, the derivative of poisoning
with respect to Bt cotton in the poisoning equation is
–0.0011. Combining, we obtain the marginal effect of
Bt area is –0.014. This result can be interpreted as fol-
lows: as a farmer devotes one more mu (1/15 of a
TABLE 6. Predicting Probability of Poisoning: Logistic Model
Estimated Standard Marginal
Variable Coefficient Error t Ratio p Value Effect
Constant –1.0408 0.7209 –1.44 NA
Pesticide quantity (predicted) 0.0096 0.0047 2.05 0.04 0.0023
Area under Bt cotton –0.0048 0.0026 –1.85 0.07 –0.0011
Age of farmer –0.0063 0.0096 –0.65 0.51 NA
Education –0.0793 0.0322 –2.46 0.01 –0.0196
Poor health condition 0.7518 0.3684 2.04 0.04 0.1859
Province dummy (Henan) 0.6714 0.3012 2.23 0.03 0.1703
Province dummy (Shandong) 0.6961 0.3267 2.13 0.03 0.1722
Province dummy (Anhui) –0.5812 0.5817 –1.00 0.32 –0.1437
Log likelihood (LL) function –358.08
Restricted LL function –438.26
Likelihood ratio test for model significance 160.35
Degrees of freedom 8
McFadden’s R
Note: Marginal effects are computed for coefficients that are significant at 10% or lower level.
hectare) of area to Bt cotton, the probability of that
farmer’s becoming sick from poisoning decreases by
1.4%. Since all coefficients involved in the estimation
of this marginal effect are statistically significant, this
estimated marginal effect is also statistically significant.
The results of Poisson modeling of the poisoning
incidence are listed in Table 8. The likelihood ratio
based the chi-square statistic of the overall model signif-
icance is 167.16, which far exceeds the 95% critical
value of the test statistic with 8 degrees of freedom. This
indicates that the model has significant explanatory
power. Further, the Pearson R
(0.15) and the deviance
(0.14) are quite reasonable for cross-sectional data.
As in the case of the logit specification, the independent
variables have the expected signs and all coefficients
except age are statistically significant at the 0.10 level.
Consistent with our expectations, coefficients of pesti-
cide quantity and poor health increase the incidence of
poisoning, while the area of Bt cotton and education
levels reduce the incidence of poisoning.
Using reasoning similar to that used for the logit
model, the marginal impact of Bt cotton adoption on the
expected poisoning incidence can be obtained as follows.
Marginal effect =
E(Y | x) pesticide E(Y | x)
_________ _________ + ________
pesticide Bt Bt
where Y be the number of poisoning cases.
Using E(Y | x)/pesticide equal to 0.0022 (from the
estimated model), pesticide/Bt equal to –5.57, and
the partial derivative of the Poisson regression equation
with respect to Bt area equal to –0.0007, the marginal
effect of the Bt cotton area is obtained as –0.013. This
result may be interpreted as follows: every 100 mu
increase in area devoted to Bt cotton reduces the
expected number of pesticide poisoning by 1.3 cases.
These econometric results provide clear evidence
that the adoption of Bt cotton has a very important
impact on reducing pesticide poisoning in China. It is
consistent with the evidence from earlier studies, which
indicated the mean pesticide use of adopters of Bt
cotton was lower than that of users of conventional
cotton and that adopters of Bt cotton reported less pes-
ticide poisoning.
This study provides some of the first econometric evi-
dence of the link between the adoption of a GM crop
and improvements in human health. We modeled the
linkage as a health-production function with farmers’
reports of poisonings as the dependent variable and
pesticide use, farmers’ characteristics, and environ-
ment as independent variables. We hypothesized that
the main impact of Bt cotton on poisoning would be
through its impact on pesticide use. Therefore, we first
estimated the impact of Bt cotton adoption on pesti-
cide use with data from three surveys of cotton farmers
in northern China. Our estimates showed that Bt
cotton use reduced the quantity of pesticide applied.
We then estimated the health-production function and
found that predicted pesticide use had a positive
impact on poisoning. In addition, we included a Bt
cotton variable to try to capture any separate impact
that it had on poisoning. Our hypothesis was that the
TABLE 7. Model Fit: Prediction Success
Actual 0 1 Total
0 278 59 337
186188 274
Total 364 247 611
TABLE 8. Predicting Number of Poisonings: Poisson Model
Estimated Standard Marginal
Variable Coefficient Error t Ratio p Value Effect
Constant –0.9799 0.3166 –3.10 NA
Pesticide quantity (predicted) 0.0028 0.0010 2.73 0.01 0.0022
Area under Bt cotton –0.0009 0.0050 –1.72 0.08 –0.0007
Age of farmer –0.0029 0.0050 0.58 0.56 NA
Education –0.0201 0.0116 –1.73 0.08 –0.0163
Poor health condition 0.3608 0.1750 2.06 0.04 0.2917
Province dummy (= 1) 0.1146 0.0411 2.79 0.01 0.0901
Province dummy (= 2) 0.06019 0.3262 0.18 0.85 NA
Province dummy (= 4) –0.1444 0.1843 –0.08 0.94 NA
Value of likelihood function –773.31
Restricted likelihood function –856.89
Likelihood ratio test for model significance 167.16
Degrees of freedom 8
Pearson R
Deviance R
Note: Marginal effects are computed for coefficients that are significant at 10% or lower level.
VOL 10/NO 3, JUL/SEP 2004 Genetically Modified Cotton and Health 303
pesticide-use variable, which is the quantity of pesticide
used, might not pick up all of the exposure impacts. Bt
cotton did have a small negative impact on poisonings
in addition to reducing pesticide use.
Taken together these results indicate that the adop-
tion of Bt cotton will significantly reduce the risk and
the incidence of poisonings. We first calculated the
marginal impact of one more mu of Bt cotton use on
the probability of farmers’ having pesticide poisoning
using the logit specification of the health function. We
found that an increase in one mu reduces the proba-
bility by 1.4%. Using data from the Poisson specifica-
tion, we calculated that the marginal decrease in the
number of poisonings due to increasing Bt cotton
adoption by one mu was 0.15.
This evidence strongly suggests that when assessing
the risk of GM crops policymakers should put more
weight on reducing the well-documented risks of pesti-
cide poisoning than on the speculative but so far unsub-
stantiated health risks to consumers and farmers from
the consumption of GM crops. It also suggests that the
recent decisions of Indonesia and India to adopt Bt
cotton technology are justified in terms of reducing
farmers’ health risks and that other cotton-growing
countries should also consider using this technology.
The evidence also suggests that policymakers should
put high priority on other ways to reduce the use of
pesticides, for example, integrated pest management
(IPM), which could also reduce farmers’ pesticide use.
IPM is clearly needed in China, because even with Bt
cotton varieties, farmers spray far more pesticide than
they need for good pest control.
Thus, it appears that
IPM should be encouraged by the government to com-
plement farmers’ decisions to adopt Bt cotton.
1. Rola AC, Pingali PL. Pesticides, Rice Productivity, and Farmer’s
Health. World Resources Institute, Washington, DC, 1993.
2. Crissman CC, Cole, DC, Carpio F. Pesticide use and farm worker
health in Ecuadorian potato production. Am J Agr Econ. 1994;
3. Huang J, Qiao F, Zhang L, Rozelle S. Farm Pesticides, Rice Pro-
duction and the Environment: A Final Project Report Submitted
to EEPSEA. Center for Chinese Agricultural Policy, Beijing,
China, 1999.
4. Gianessi LP, Silvers CS, Sankula S, Carpenter JL. Agricultural
biotechnology: current and potential impact for improving pest
management in U.S. agriculture: an analysis of 40 case studies.
National Center for Food and Agricultural Policy, Washington,
DC, 2002.
5. Huang J, Hu R, Rozelle S, Qiao F, Pray CE. Transgenic varieties
and productivity of smallholder cotton farmers in China. Aust J
Agr Resource Econ. 2002; 46:367-87.
6. Pray CE, Huang J, Ma D, Qiao F. Impact of Bt cotton in China.
World Development. 2001;29(5).
7. Hausman J. Specification tests in econometrics. Econometrica.
8. Green H. Econometric Analysis. Upper Saddle River, NJ: Pren-
tice Hall, 2000.
... A survey of cotton farmers in northern China in 1999-2001 showed that reduced pesticide use following adoption of Bt cotton adoption led to decreased incidence of acute poisoning (35). Positive impacts of Bt cotton on the health of farmers also have been observed in Pakistan (2) and India (52). ...
... By reducing the number of pesticide applications, adoption of IRGE crops has led to improvement of farmer health. As mentioned above, reduced use of insecticide on IRGE cotton in China resulted in fewer cases of pesticide toxicity (35,41,91). In farm-level production trials, Chinese farm households adopting IRGE rice lines reported no adverse health effects, compared to 3.0-10.9% ...
... reporting adverse health effects among similar households planting non-IRGE rice lines and using higher input of pesticides (40). Adoption of IRGE crops will not only lead to decreased incidence of acute poisoning (35), but also reduce adverse impacts on farmers' neurological, hematological, and electrolyte systems (39). NASEM (86) provides a thorough review of socioeconomic impacts of GE crops. ...
Full-text available
With 20% of the world's population but just 7% of the arable land, China has invested heavily in crop biotechnology to increase agricultural productivity. We examine research on insect-resistant genetically engineered (IRGE) crops in China, including strategies to promote their sustainable use. IRGE cotton, rice, and corn lines have been developed and proven efficacious for controlling lepidopteran crop pests. Ecological impact studies have demonstrated conservation of natural enemies of crop pests and halo suppression of crop-pest populations across a local scale. Economic, social, and human health effects are largely positive and, in the case of Bt cotton, have proven sustainable over 20 years of commercial production. Wider adoption of IRGE crops in China is constrained by relatively limited innovation capacity, public misperception, and regulatory inaction, suggesting the need for further financial investment in innovation and greater scientific engagement with the public. The Chinese experience with Bt cotton might inform adoption of other Bt crops in China and other developing countries. Expected final online publication date for the Annual Review of Entomology, Volume 65 is January 7, 2020. Please see for revised estimates.
... Even if GM technology has achieved multiple benefits (See Pray et al. 2001;Huang et al. 2002aHuang et al. , 2002bHuang et al. , 2003Hossain et al. 2004;Brookes and Barfoot 2005;Qiao 2015;Qiao et al. 2016;Qaim 2003;Kathage and Qaim 2012 on the discussion of economic benefits, the benefits in terms of human health and the environment), there are few studies in the literature that document its adoption process from the farmers' risk management perspective. From the farmers' perspective, only if the potential benefit of new technology overcomes traditional technology, will farmers choose the new technology (Chavas and Nauges 2020). ...
Full-text available
Although the benefits of genetically modified (GM) crops have been well documented, how do farmers manage the risk of new technology in the early stages of technology adoption has received less attention. We compare the total factor productivity (TFP) of cotton to other major crops (wheat, rice, and corn) in China between 1990 and 2015, showing that the TFP growth of cotton production is significantly different from all other crops. In particular, the TFP of cotton production increased rapidly in the early 1990s then declined slightly around 2000 and rose again. This pattern coincides with the adoption of Bt cotton process in China. To further investigate the decline of TFP in the early stages of Bt cotton adoption, using aggregate provincial-level data, we implement a TFP decomposition and show that the productivity of GM technology is higher, whereas the technical efficiency of GM technology is lower than that of traditional technologies. Especially, Bt cotton exhibited lower technical efficiency because farmers did not reduce the use of pesticide when they first started to adopt Bt cotton. In addition, we illustrate the occurrence of a learning process as GM technology diffuses throughout China: after farmers gain knowledge of Bt cotton, pesticide use declines and technical efficiency improves.
... Pray et al. (2001) and Huang et al. (2003) showed that the frequency of pesticide poisonings was significantly lower among Bt cotton adopters than among nonadopters in China. Hossain et al. (2004) used econometric models to establish that this observation is causally related to Bt technology. Bennett et al. (2003) made the same observation for Bt cotton in South Africa, and there is first evidence that similar effects can also be expected for other Bt crops in smallholder agriculture, such as Bt rice in China (Huang et al. 2005(Huang et al. , 2008. ...
In 2010, the White House announced the goal of eradicating food deserts—low-income neighborhoods without nearby supermarkets—in seven years. The efficacy of this initiative is premised on the presumption, mostly untested in 2010, that food deserts significantly contribute to health disparities in low-resourced communities. We synthesize the post-2010 line of research that seeks to establish causality in the relationship between food access and nutrition/health. All things considered, there is so far little evidence that food deserts have a causal effect of meaningful magnitude on health and nutrition disparities. The causes of diet quality disparity lie more on the side of food demand than on supply. Therefore, from the public health perspective, policies that lower the relative price of healthy food or change the “deep parameters” of preferences in favor of healthy food would be more appealing than eliminating food deserts. Expected final online publication date for the Annual Review of Resource Economics, Volume 13 is October 2021. Please see for revised estimates.
... The probability of poisoning is inversely proportional to the education level. 28 The education status of patients in this study also showed that the poisoning incidence is high among the uneducated or less educated population ( Figure 1B). Therefore, increasing the awareness on poisonous substances may be effective in reducing the number of poisoning cases in future. ...
Full-text available
Introduction: Morbidity and mortality associated with pesticide poisoning is a major public health issue, especially in lower and middle income countries, including India. Timely understanding of poisoning trends is required for improved prevention. The objective of the present study was to analyze the trend of poisoning cases in Ahmedabad, India in the period of 2015-2017. Methods: Detailed history, including demographic data, risk factors, poisoning history, agents involved, and occupational influence were collected for poisoning cases reported to the Poison Information Centre in Ahmedabad. Cholinesterase activity and HPTLC method for detection of sanguinarine in urine were used to investigate the agents of poisoning. Non-parametric tests, such as Chi-square test and Mann-Whitney U Test were applied to test statistical significance between the groups. All statistical analysis was carried out using IBM SPSS Statistics for Windows, Version 26.0. Armonk, NY: IBM Corp. Results: A total 1373 poisoning cases were investigated. The incidence and fatality rate was found to be higher in males compared to females (M/F ratio 1.89:1). About 91.62% of the poisoning were through the oral route. Erythrocyte cholinesterase activity assay results indicated that 41.29% of the cases were due to organophosphorus/carbamate poisoning. Insecticides were found to be the agent of poisoning in 26.29% cases, and 11.07% of all the cases were agricultural workers. Poisoning with medications, household pesticides and chemicals were also reported. Few cases of food poisoning with sanguinarine were detected. Conclusion: The data presented here suggest that pesticides used for agriculture are the major source of poisonings. Implementation of usage guidelines, educating farmers and vulnerable population, and finding novel alternatives for highly toxic chemicals may be helpful in decreasing the number of poisoning cases.
... If the GM seeds would raise the profit margin of the farmers, they may be well disposed to it. In India and China, farmers quickly adopted Bt. cotton because of its high yield, greater productivity, and reduced occupational hazard due to less exposure to pesticide as the modified cotton required little or no pesticide applications (Pray et al., 2003;Hossain et al., 2004). By this, the farmers see as leverage on the initial high cost of the GM seeds. ...
... Other consequences of cotton production include the exposure of cotton growers to pesticides, which causes health and fertility issues (e.g. Rupa et al., 1989Rupa et al., , 1991 for which the only alternative seems to be the use of transgenic cotton (Hossain et al., 2004;Morse et al., 2006), a remedy that is also subject to controversy (Liu et al., 1999;Dhurua and Gujar, 2011). Ironically, cotton production is mostly distributed in areas where water abstraction and pollution (e.g. ...
There is increasing recognition that functional bioindicators are needed for ecosystem health assessments. In this perspective, cotton strip assays are widely considered as a standard method to account for organic matter decomposition in streams. However, cotton cultivation and manufacture raise both environmental and societal dramatic issues that are - in our opinion - irreconcilable with the objectives of bioindication. In this study, we assessed the relevance of four alternative - eco-friendly - textiles (made of organic cotton, hemp and linen) by comparing their chemical composition and degradation rates in six streams. Chemical composition exhibited low variations among textiles, but contrasted sharply with the expectation that cotton is mostly composed of cellulose. Moreover, surprisingly high nutrient (0.49% N) contents occurred in the conventional cotton strips compared with the organic textiles (N < 0.12%). All textiles provided similar degradation rates across the six streams, meaning that they could be interchangeably used as alternatives to conventional cotton strips. We thus call for the adoption of such ethical and eco-friendly tools as 'next-generation' indicators for the functioning of stream ecosystem.
Sustainability is the development which meets the needs of the present without compromising the ability of future generations to fulfill their needs. Environmental sustainability respects and cares for all kinds of life forms existence without affecting the sustenance of natural resourcesNatural resources. The best method of sustaining the environmentEnvironment is paying back all the components of ecosystem services in a recyclable mode. Where in biotic and abiotic harmony of environmentEnvironment restores aesthetic values and ecosystem services of the nature. This in turn maintains intricate equilibrium required for resurrecting the natural ecosystems. Environmental biotechnologyBiotechnology is the branch of biotechnologyBiotechnology that addresses environmental issues removal of pollutants, renewable energy generation or biomass production, by involving biological entities and their process. Environmental biotechnologyBiotechnology has its greatest contribution to agriculture, especially by improving crop yields for environmentEnvironment sustenance. It offers opportunities to create designer crops of specific environmentsEnvironments and to make crops more efficient producers of food and energy. Thus, biotechnologBiotechnologyy can manipulate primary energy flows; it can also reduce fossil-fuel energy inputs into agricultural systems. Moreover, it contributes to the mitigation of environmental problems such as deforestation and soilSoil erosion. Green energy methods/biofuels are urgently needed to replace fossil fuelsFossil fuels in order to battle pollution and the threat of global warmingGlobal warming. Biotechnology constitutes a vehicle for the improved manipulation of biogeochemical cycles, wherein bioremediatioBioremediationn and biodegradatioBiodegradationn alleviate conditions of polluted soilSoil and degraded water ecosystems. Industrial biotechnologyBiotechnology aims to alter the manufacturing process by reducing wastes generation-conserving natural resourcesNatural resources, trimming costs, and speeding new “greener” market products. Emerging biotechnologiesBiotechnologies having low-input techniques involving microbes, plants and animals offering novel approaches (genetic manipulation or ‘engineering’) for striking a balance between developmental needs and environmental conservation. This chapter reviews the issues relating to the use of biotechnological methods vis-à-vis biotools in solving the problems of environmental degradation and sustainable developmentSustainable developmentDevelopment.
Full-text available
Transgenic crops are the revolutionary outcomes of genetic engineering. These crops are currently being cultivated on a commercial scale in many countries. But they have remained a matter of debate since they were first introduced. The debate over their environmental impact is growing increasingly complex and intense. The benefits and risks of any particular transgenic crop depend on the interactions of its ecological functions and natural history with the agro-ecosystem and ecosystems within which it is embedded. Several concerns related to consequences of gene escape, adverse impact on biodiversity, natural enemies, pollinators, soil organisms, decomposers and various non-target organisms have been raised. In addition, corporatization of agriculture raised several concerns. On the other hand, many positive impacts of transgenic crops are also praised like reduced environmental impact from pesticides and insecticide, increased yield, soil conservation, phytoremediation etc. These evolutionary and ecological factors must be considered when assessing transgenic crops. A critical analysis of this controversy is the main concern of the following discussion. Along with that, attempt has also been made to highlight the possible impact of corporatization of agriculture through transgenic crops, especially in developing countries.
Full-text available
As agricultural production in African countries intensifies; pesticide utilization becomes more widespread and the users are extremely exposed to these pesticides due to lack of pesticide registration scheme; importing highly toxic pesticides; no national plan for pesticide residue; involvement of children and women. The purpose of this systematic review was to review adverse effect of pesticide among top ten imported African countries. In this review, top ten importers African countries were selected based of imported amount for ten years were considered from imported period of 2002 to 2017. The articles were searched from PUBMED, GOOGLE SCHOLAR, and MEDLINE and EMBASE engines. The first leading three continents for pesticides exported were European (48.2%), Asian (33.7%) and North America (12.7%), while the countries were China (14.3%), Germany (11.8%) and United States (11.5%) at the end of 2017. The first three leading importer of African countries were South Africa shared (25.7%), Nigeria (15.8%) and Ghana (14.5%). The three major imported pesticides were Fungicides, herbicides and insecticides. In this review, Ethiopia (827), Kenya (801), and Morocco (542) are the main importers of pesticides until end of 2017. The review also found that farmers were faced with endocrine disruption, carcinogenicity, mutagenicity, teratogenicity, cardiovascular, dermatitis and birth defects. The main associated factor for these problems were low awareness, improper handling of pesticide, and lack of training, and careless disposal of empty pesticides containers. The study concluded that more than one billion US$ of pesticides sales was carried out into ten African countries. The farmers within these country were faced different health problems due to different determinant factors. Proper training and education should be advised for farmers.
Full-text available
This review was originally prepared in 2012 as a background paper, but never published. I was asked to gather and review a broad range of academic research and grey literature on the status and impacts of transgenic crops and other agricultural biotechnologies worldwide, with special attention to the ‘developing world’. While the work was under way, I was asked to include some information about transgenic fish and about alternative agroecological approaches to agricultural improvement. Although the contents of the document are now out of date, the large body of literature and materials gathered and reviewed here may still be useful to others. I am therefore publishing the document online, so that it may be freely available to readers around the world.
Two adjacent watersheds totaling about 150 kilometers in Montufar Canton in Carchi Province in a cool moist highland zone in northern Ecuador served as the case study site. The authors find that with current technologies, reducing pesticide use would improve health but reduce yields. This tradeoff presents a dilemma for policy makers. Using pesticides has a positive effect on yields and farmers are not irrationally using pesticides; but farmers and their families are suffering health and economic impacts from the use of toxic pesticides. Acute poisonings cause loss of labor and considerable private health care costs. One short-term policy option is to substitute less dangerous but similarly effective chemical products for the more highly toxic ones. Education on the proper management and handling of these dangerous products with clear discussion of the long-term health effects may also be useful. However, unless there is a greater supply of personal protective equipment and incentives for their use, such interventions may be ineffective. A longer term solution is to change potato production technology. -from Authors
Genetically modified cotton varieties have greater production efficiency for smallholders in farming communities in China. We also find that the adoption of Bacillus thuringiensis (Bt) cotton varieties leads to a significant decrease in the use of pesticides. Hence, we demonstrate that Bt cotton appears to be an agricultural technology that improves both production efficiency and the environment. In terms of policies, our findings suggest that the government should investigate whether or not they should make additional investments to spread Bt to other cotton regions and to other crops.
A sample of 283 cotton farmers in Northern China was surveyed in December 1999. Farmers that used cotton engineered to produce the Bacillus thuringiensis (Bt) toxin substantially reduced the use of pesticide without reducing the output/ha or quality of cotton. This resulted in substantial economic benefits for small farmers. Consumers did not benefit directly. Farmers obtained the major share of benefits and because of weak intellectual property rights very little went back to government research institutes or foreign firms that developed these varieties. Farmers using Bt cotton reported fewer pesticide poisonings than those using conventional cotton.
This paper investigates the impact of international migration on technical efficiency, resource allocation and income from agricultural production of family farming in Albania. The results suggest that migration is used by rural households as a pathway out of agriculture: migration is negatively associated with both labour and non-labour input allocation in agriculture, while no significant differences can be detected in terms of farm technical efficiency or agricultural income. Whether the rapid demographic changes in rural areas triggered by massive migration, possibly combined with propitious land and rural development policies, will ultimately produce the conditions for a more viable, high-return agriculture attracting larger investments remains to be seen.
Using the result that under the null hypothesis of no misspecification an asymptotically efficient estimator must have zero asymptotic covariance with its difference from a consistent but asymptotically inefficient estimator, specification tests are devised for a number of model specifications in econometrics. Local power is calculated for small departures from the null hypothesis. An instrumental variable test as well as tests for a time series cross section model and the simultaneous equation model are presented. An empirical model provides evidence that unobserved individual factors are present which are not orthogonal to the included right-hand-side variable in a common econometric specification of an individual wage equation.
Pesticides, Rice Productivity, and Farmer's Health
  • Ac Rola
  • Pingali
Rola AC, Pingali PL. Pesticides, Rice Productivity, and Farmer's Health. World Resources Institute, Washington, DC, 1993.
Econometric Analysis
  • H Green
Green H. Econometric Analysis. Upper Saddle River, NJ: Prentice Hall, 2000.
Farm Pesticides, Rice Production and the Environment: A Final Project Report Submitted to EEPSEA
  • J Huang
  • F Qiao
  • L Zhang
Huang J, Qiao F, Zhang L, Rozelle S. Farm Pesticides, Rice Production and the Environment: A Final Project Report Submitted to EEPSEA. Center for Chinese Agricultural Policy, Beijing, China, 1999.
Pesticides, Rice Productivity, and Farmer's Health. World Resources Institute
  • A C Rola
  • P L Pingali