IFPRI Discussion Paper 02132 
July 2022 
Gender Bias in Consumer Perceptions 
The case of agro-input dealers in Uganda 
Anusha De 
Caroline Miehe 
Bjorn Van Campenhout 
Development Strategy and Governance Division 
INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE 
The International Food Policy Research Institute (IFPRI), a CGIAR Research Center established in 1975, 
provides research-based policy solutions to sustainably reduce poverty and end hunger and malnutrition. 
IFPRI’s strategic research aims to foster a climate-resilient and sustainable food supply; promote healthy 
diets and nutrition for all; build inclusive and efficient markets, trade systems, and food industries; 
transform agricultural and rural economies; and strengthen institutions and governance. Gender is 
integrated in all the Institute’s work. Partnerships, communications, capacity strengthening, and data and 
knowledge management are essential components to translate IFPRI’s research from action to impact. 
The Institute’s regional and country programs play a critical role in responding to demand for food policy 
research and in delivering holistic support for country-led development. IFPRI collaborates with partners 
around the world.  
AUTHORS 
Anusha De (anusha.de@kuleuven.be) is a PhD Student at LICOS, KU Leuven, Belgium. 
Caroline Miehe (caroline.miehe@kuleuven.be) is a PhD Student at LICOS, KU Leuven, Belgium. 
Bjorn Van Campenhout (b.vancampenhout@cgiar.org) is a Research Fellow at the Development 
Strategy and Governance Division at the International Food Policy Research Institute, Leuven, Belgium. 
Notices 
1 IFPRI Discussion Papers contain preliminary material and research results and are circulated in order to stimulate discussion and 
critical comment. They have not been subject to a formal external review via IFPRI’s Publications Review Committee. Any opinions 
stated herein are those of the author(s) and are not necessarily representative of or endorsed by IFPRI.  
2 The boundaries and names shown and the designations used on the map(s) herein do not imply official endorsement or 
acceptance by the International Food Policy Research Institute (IFPRI) or its partners and contributors. 
3 Copyright remains with the authors. The authors are free to proceed, without further IFPRI permission, to publish this paper, or any 
revised version of it, in outlets such as journals, books, and other publications
Abstract
Faced with incomplete and imperfect information, economic actors rely predominantly on perceptions and
often base decisions on heuristics prone to bias. Gender bias in perceptions favoring men has been found in
a wide variety of settings and may be an important reason why some sectors remain dominated by men and
gender gaps persist. Using ratings of agro-input dealers provided by smallholder farmers in their vicinity, we
test if farmers perceive male-managed agro-input shops dierently than agro-input shops managed by women.
After controlling for observable characteristics at the input dealer level and including xed eects to account
for farmer-level heterogeneity, we nd that farmers rate male-managed agro-input outlets higher on a range of
attributes related to the dealership in general, as well as when farmers are asked to consider the quality of inputs
sold by the dealer. Our results suggest that consumers' biased perceptions continue to be an important entry
barrier for women in the subsector, and we conclude that policies and interventions designed to challenge gender
norms and customs are needed to correct bias in perceptions.
1
1 Introduction
In the context of imperfect information, economic actors rely predominantly on perceptions and use mental
shortcuts to make decisions using limited data (Kahneman, 2017). Reliance on instinct and emotions becomes
dominant if it is dicult for economic actors to objectively assess the value of a commodity or service being bought
and sold. However, perceptions and decision heuristics may suer from a variety of cognitive biases such as stereotype
thinking and availability bias, and may be inuenced by social and cultural phenomena such as homophily eects
and prevailing norms and customs.
Agricultural inputs such as inorganic fertilizers or improved seed varieties lie somewhere on the continuum
between experience goods and credence goods. When farmers inspect goods at the agro-input shop, the quality can
only be assessed supercially from readily observable characteristics such as homogeneity of the seed or by checking
if the fertilizer package is intact. Even after the commodity has been used and yields can be observed, it may still
be dicult for the farmer to learn about the quality of the input, as many other factors in addition to the input
aect yield. As such, when farmers decide to acquire agricultural inputs or not, perceptions and emotions often
take the upper hand.
In addition to the diculty of judging quality of agricultural inputs, several studies note that there is considerable
heterogeneity in the quality of inputs in the market. For instance, in Uganda, Bold et al. (2017) tested agricultural
inputs purchased in local markets, and found that 30 percent of nutrient is missing in fertilizer, and hybrid maize
seed is estimated to contain less than 50 percent authentic seeds. Also in Uganda, Ashour et al. (2019) tested
herbicides and found that the average bottle in their sample was missing 15 percent of the active ingredient and 31
percent of samples contained less than 75 percent of the ingredient advertised. While it remains unclear if quality-
related issues are due to deliberate adulteration or poor storage and handling, and at what point in the value chain
quality starts to deteriorate, the resulting uncertainty makes reliance on perceptions and decision heuristics more
likely (Barriga and Fiala, 2020).
In traditional societies with strong gender norms and customs, small businesses at dierent nodes along food
supply chains are often one of the few options open to women to earn some money independently from their husband.
While rapid urbanization has led to the emergence of fast-food restaurants, informal food vendors, who tend to
be self-employed women, are still the main source of food for the majority of households (Giroux et al., 2021).
And while supermarkets are emerging throughout the developing world, wet markets where mostly women sell
produce continue to account for the majority of expenditure on fresh produce in many countries (Gorton, Sauer,
and Supatpongkul, 2011). In Uganda, we nd that a surprisingly large share of agro-input shops are operated
and/or managed by women.
However, the same gender norms and customs may also mean that perceptions are stacked against women if
2
they venture into certain areas such as agro-input provision. Farmers, both male and female, may be of the opinion
that modern agricultural technologies are in the male domain. Furthermore, agro-input shops primarily deal in
seed for semi-commercial crops such as maize or rice, as opposed to food security crops such as beans or cassava.
Again, commercial crops are often considered to be the responsibility of the men, while women are assumed to take
care of the household food supply (Orr et al., 2016; Dolan, 2001). As a result, we conjecture that female-managed
agro-input shops may be disadvantaged when farmers form opinions about the quality of services rendered or goods
sold.
In this paper, we test if farmers perceive agro-input shops managed by women less favorably than agro-input
shops under male management using a unique dyadic dataset of farmer-dealer combinations. To operationalize
perceptions, we asked farmers to rate agro-input dealers, on a scale of 1 to 5 on a range of characteristics. We then
make between-dealer comparisons, accounting for observable dierences in the quality of male- and female-managed
shops. Furthermore, we use the fact that a farmer has generally rated more than one agro-input dealer. If the
same farmer rates both male and female-managed agro-input shops, we can exploit this within-farmer variation and
control for farmer specic observable and unobservable characteristics.
We nd that farmers generally rate male-managed agro-input shops more favorably than agro-input shops
managed by women. The dierence in ratings is highest when farmers are asked to rate the agro-input dealership
in terms of price competitiveness and in terms of reputation. We also nd that seed from male-managed agro-input
shops is rated higher than seed from shops managed by women. As the dierences in ratings cannot immediately
be explained by dierences in the quality of the dealerships or the services and products they provide, we conclude
that biased perceptions create an important disadvantage for female-managed agro-input shops.
In the remainder of this article, we rst situate the research question in the wider literature. We then provide the
context for the study and describe the main economic actors: agro-input dealers and smallholder maize farmers. We
also describe how we measure perceptions, the key variable in this study. Next, we lay out the empirical strategy,
followed by the results. A nal section concludes and oers some policy guidance.
2 Research question and relation to the literature
We aim to test if gender equity biasbehavior that shows favoritism toward one gender over anotheris present
in the way smallholder maize farmers in southwestern Uganda perceive agro-input dealers in their neighborhood.
Gender equity bias has been conrmed in a wide range of contexts, usually when people are asked to assess the
performance of another person. Stereotyping and role congruence are often catalysts for distorted perceptions and
false beliefs about the abilities of particular groups of people. We highlight some of the most important studies
that search for systematic bias related to the gender of the person being assessed.
One area where gender equity bias has been studied extensively is in scientic publishing using peer review. For
3
instance, Card et al. (2019) look at dierences in rejection rates at four top economics journals. They compare
male-authored papers to female-authored papers, using citations as a noisy measure of quality to account for
potential sources of divergence between the two. They nd that editors largely follow the referees, resulting in
a 1.7 percentage point lower probability of a revise and re-submit verdict for papers with female authors relative
to a citation-maximizing benchmark. However, evidence on gender biases in the evaluation of economic research
remains mixed. For example, Chari and Goldsmith-Pinkham (2017) nd no disparity in the acceptance rates of
female- and male-authored papers for National Bureau of Economic Research (NBER) conferences; Hospido and
Sanz (2021) do nd a signicant advantage for male authors being accepted at three dierent European conferences.
Gender equity bias has also been studied extensively in student evaluations of teaching. For instance, Mitchell and
Martin (2018) nd that the language students use in evaluations regarding male professors is signicantly dierent
than their language used in evaluating female professors. They also show that a male instructor administering an
identical online course as a female instructor receives higher ordinal scores in teaching evaluations.
Gender equity bias is studied most in the context of peer review processes such as those mentioned above.
However, the same perceptions surface when individuals decide on who to engage with, be it who to work with,
who to elect as leaders, or who to consult. Labor markets and hiring decisions involve situations where managers
decide based on limited information. Discrimination in labor markets, including discrimination related to gender,
has been documented in several studies. Wu (2020) uses data from an online forum for economists called Economic
Job Market Rumors to measure gender bias in discussions about women versus men. Gender equity bias is also
often studied in the context of the wage gap, that is, when women appear to make substantially less money for the
same work than their male counterparts. Often, this is also tied to gender equity bias in performance appraisals,
where (often male) managers' beliefs creep into their evaluations of workers (Correll et al., 2020).
Gender equity bias is also pervasive in politics. Pair et al. (2021) use natural language processing (NLP) to
search for gender bias in Kenya's leading newspaper and sentiment analysis to predict quantitative sentiment scores
for sentences surrounding female leader names compared to male leader names. They nd evidence of improvement
in gender equality but also a backlash from increased female representation in high-level governmental leadership.
Le Barbanchon and Sauvagnat (2021) nd that female candidates obtain fewer votes in municipalities with higher
gender earnings gaps. Klein, Shtudiner, and Zwilling (2021) nd that a nancial advisor's gender is one of the most
important factors inuencing a consumer's choice of investment advisor. Female advisors' gender was found to have
a negative eect on the desire to invest, and this negative attitude was found to be signicantly higher among male
respondents.
Gender bias features so prominently in areas such as scholarly peer review, teaching assessments, or labour
markets partly because perceptions are made explicit in the process, for instance through review reports, student
feedback, or hiring committees. However, in economic transactions, gender biases remain hidden as perceptions are
4
never measured. As a result, dierences in outcomes are often attributed to various other causes, such as dierences
in education or ability between men and women.
3 Study Context
3.1 Sample
Our study area comprises 11 districts in southeastern Uganda, which roughly corresponds to the Busoga kingdom.
We include agro-input dealers located in trading centers and villages as well as maize farmers that live in the
catchment areas of these key market sheds. The dealer sample was obtained by listing all agro-input shops in the
area during a census, which resulted in 348 dealers. We collected information on their characteristics in September
and October 2020.
After the census, these agro-input shops were grouped in catchment areas based on their location. A catchment
area is dened as the area that is served by a dealer, the area where this dealer's customers live. If catchment
areas of two or more dealers overlap because these dealers operate in the same town, street or right next to each
other, they are assigned to the same catchment area. This is done on the basis of geographical location. Using GPS
coordinates of the input dealers, the haversine function is used to construct an adjacency matrix, and shops that
are less than 5 kilometer apart are assigned to the same catchment area. The 5 kilometer threshold was selected
based on visual inspection of the map, the size of an average village and reported distance between farmers and
dealers. The 348 agro-input dealers in Busoga were assigned to 130 catchment areas. In some catchment areas,
there is a high density of shops, while in others there are only one or two dealers. On average, there are 2.7 dealers
in an area, with a minimum of 1 and a maximum of 18.
In each catchment area, we also sampled farmers in proportion to the number of agro-input dealers in the area.
We connected shops to villages by asking every dealer from which village most of his/her customers come from1.
Enumerators were sent to these villages and instructed to randomly sample ten households that meet the inclusion
criteria. Consequently, we sampled 3470 farmers and collected information about their characteristics in April 2021.
It is important to note that only farmers in half of the catchment areas rated dealers2. This means that 1931
farmers rated 193 dealers in April 2021. To have a sucient number, ratings were collected a second time in January
and February 2022. While these ratings were provided by the same farmers, only 1893 of them were found during
the second wave. These two rounds of surveys constitute the key sources of data for the study.
1If this village was already named by another dealer, the village with the second most customers was chosen.
2These ratings are part of the intervention of a dierent study, where a computer algorithm allocated the rating treatment randomly
at the level of the catchment area.
5
3.2 Descriptive statistics
There is substantial heterogeneity among agro-input shops. Some are small informal shops located in rural areas,
that sell other goods and only stock seed during the planting season. Others are located in towns or trading centers
and specialize in farm inputs and tools. About 60 percent of the agro-input shop managers are male. The average
shop has been in operation for about ve years and serves about 40 customers a day, half of who come to buy
maize seed. A shop stocks on average three maize seed varieties. Table 1 provides some descriptive statistics on the
agro-input shops included in our study. More descriptive statistics describing seed handling and storage practices,
eorts, and services of agro-input dealers, are presented in the appendix in Table 12.
The average farmer in our sample works on a small farm, with about 3.4 acres of land for crop production. Half
of our sampled farmers indicate that they used improved maize seed on at least one plot in the season preceding the
survey, and of the farmers that used improved seed, two-thirds obtained this seed from an agro-input shop. However,
fertilizer use is very low. As a result, productivity is also low, with the average farmer harvesting only about 500
kg of maize per acre. Almost 70 percent of farmers are of the opinion that maize seed sold at the agro-input shop
is counterfeit. Table 2 provides additional descriptive statistics of the farmers included in our study.
6
7
Table 1: Descriptive agro-input dealer statistics.
mean min max SD nobs
Dealer's age in years 31.46 15 59 10.02 193
Dealer is male 0.591 0 1 0.493 193
Dealer nished secondary education 0.388 0 1 0.489 188
Dealer owns shop 0.565 0 1 0.497 193
Dealer received training on maize seed handling/storage 0.580 0 1 0.495 193
Shop's distance to nearest tarmac road in km 5.850 0 40 9.20 192
Distance between farmer and shop in km 5.214 0.19 12.561 3.21 193
Shop only sells farm inputs 0.751 0 1 0.433 193
Number of customers per day 44.55 2 300 48.34 192
Number of customers per day buying maize seed 23.01 0 250 28.25 192
Number of years since shop's establishment 5.430 0 33 6.106 193
Number of maize varieties in stock 3.057 0 10 1.948 193
Number of hybrid maize varieties in stock 1.803 0 8 1.476 193
Number of OPV maize varieties in stock 1.301 0 5 0.717 193
Sales price of maize seeds (per kilogram) 4353.674 2500 12000 1291.825 186
Cost of maize seed for dealer (UGX per kilogram) 3522.765 2000 8500 959.380 179
Amount of maize seed dealer bought (in kilograms) 969.978 0 52500 4683.849 184
Shop's cleanness/professionality rating by enumerator 3.503 1 5 1.142 193
Shop received seed-related complaint from customer 0.674 0 1 0.470 193
Shop's UNADA registration 0.489 0 1 0.501 182
Shop has trading license from local government 0.789 0 1 0.409 190
Note: "Distance between farmer and shop in km" is calculated using the haversine function based on the GPS coordinates obtained during data collection. Farmers were asked to rate
agro-input shops they know (one farmer can rate multiple shops) and these shops are not necessarily located in the immediate vicinity.
8
Table 2: Descriptive farmer statistics.
mean min max SD nobs
Homestead's distance to nearest tarmac road in km 8.850 0 100 9.39 1844
Homestead's distance to village headquarters in km 0.744 0 12 0.899 1914
Homestead's distance to nearest agro-input shop in km 3.826 0 52 4.894 1858
Farmer's age in years 48.51 20 97 13.34 1923
Household head is male 0.789 0 1 0.408 1931
Farmer is married 0.881 0 1 0.323 1931
Farmer nished primary education 0.525 0 1 0.500 1913
Number of people in household (incl. respondent) 8.651 1 25 4.029 1931
Years since farmer started growing maize 22.85 0 82 13.00 1931
Farmer is member of (maize) farmer group/association/cooperative 0.132 0 1 0.338 1927
Farmer's land for crop production in acres 3.350 0.25 80 3.980 1915
Land productivity in kg/acre (yield/area) 504.1 0 28000 842.5 1921
Farmer used improved maize seed (OPV/hybrid) for any eld last season 0.499 0 1 0.500 1929
Farmer bought maize seed at agro-input shop (if he/she used improved seed) 0.671 0 1 0.470 932
Farmers thinks seed at agro-input shops is counterfeit/adulterated 0.684 0 1 0.465 1512
Farmer is satised with maize seed used on eld 0.666 0 1 0.472 1931
Note: "Homestead's distance to nearest agro-input shop in km" is the distance reported by the farmer. Respondents could only report one answer for the nearest agro-input shop.
4 Measuring perceptions
Quantifying perceptions of the quality of services provided by agro-input dealers and of the products they sell
improved maize seed varieties in particularis central to our analysis. To do so, we asked farmers to rate agro-input
dealers in their catchment area on a range of attributes. We broadly categorized the attributes into two families of
indicators. A rst set of indicators attempts to measure overall quality of agro-input dealers and the services they
provide, while a second set of indicators has a much more narrow focus and asks about a particular product sold
by the agro-input dealer (maize seed). Note that as agro-input dealers are often clustered in towns and trading
centers, farmers often rate several agro-input dealers.
To measure the perceived quality of agro-input dealers, farmers were asked to rate these dealers on a scale of
1 to 5 on their general quality, location (convenience, accessibility, closeness to clients), price (competitive pricing,
discounts), seed quality, stock (availability of seed, number of varieties in stock), and reputation (what do other
farmers think about the dealer). We also compute an average of these six dealer level ratings. For these indicators,
farmers were asked to rate the shop as a whole, which also includes the person who operates the shop.3
To measure the perceived quality of seed, farmers were asked to rate the improved maize varieties that dealers
sell on a scale of 1 to 5 on their general quality, yield, drought tolerance, pest/disease tolerance, crop dura-
tion/maturation period, and germination reliability.4 We also compute an average of these six seed-level ratings.
For these variables, farmers were asked to rate seed, the product itself. Farmers were also allowed to indicate that
they could not rate seed on a particular dimension (for instance because they never bought seed from the agro-input
dealer).
We asked farmers to rate agro-input dealers twice, a rst time in April 2021 and a second in January 2022. The
average farmer in our dataset provided ratings for about two agro-input dealers, with some farmers rating up to 15
dealers. The average agro-input shop received ratings from almost 12 farmers, while one shop received ratings from
almost 50 farmers. Table 3 provides descriptive statistics of the ratings used in our study.
3It may be that the shop is owned by a particular person, but the owner employs another person to manage the shop. This is
especially the case for larger shops in towns. In villages and trading centers, the shop owner is generally also the manager.
4Some of these attributes are variety specic. For instance, some hybrid seeds may be particularly drought tolerant, while other
seeds may be higher yielding. Therefore, we specically asked farmers to rate varieties relative to what is advertised for these attributes.
9
10
Table 3: Descriptive agro-input dealer ratings.
mean min max SD 1st quartile 3rd quartile nobs
Dealer's maize seed rating on general quality by farmers 3.799 1 5 0.873 3 4 992
Dealer's maize seed rating on yield by farmers 3.562 1 5 0.921 3 4 972
Dealer's maize seed rating on drought tolerance by farmers 3.018 1 5 0.882 2 4 937
Dealer's maize seed rating on pest/disease tolerance by farmers 2.464 1 5 0.945 2 3 949
Dealer's maize seed rating on speed of maturing by farmers 3.833 1 5 0.738 4 4 952
Dealer's maize seed rating rating on germination by farmers 3.676 1 5 0.898 3 4 967
Dealer's rating on general quality by farmers 3.734 1 5 1.006 3 4 1022
Dealer's rating on location by farmers 3.811 1 5 1.232 3 5 1022
Dealer's rating on price competitiveness by farmers 3.255 1 5 1.183 3 4 1022
Dealer's rating on seed quality by farmers 3.797 1 5 1.074 3 5 1022
Dealer's rating on maize seed stock by farmers 3.910 1 5 1.093 3 5 1022
Dealer's rating on reputation by farmers 4.162 1 5 0.937 4 5 1022
Note: The descriptives convey the dyadic nature of the dataset.
5 Empirical Strategy
Our empirical strategy exploits the particular nature of the data, that is, that each farmer in our dataset rates
several agro-input dealers (and each dealer is rated by several farmers). A useful starting point is the following
specication:
yf,d = α+ βgd + εf,d (1)
Here, yf,d represents the rating, on a scale of 1 (poor) to 5 (excellent), given by farmer f to agro-input dealer d.
gd is the main variable of interestthe gender of dealer d. α and β are parameters to be estimated, and εf,d is a
residual.
Because the same farmer may rate several agro-input dealers, we cannot assume that the ratings yf,d in equation
1 are independent. For example, the ratings that a farmer provides may be aected by a (potentially unobservable)
characteristic of the farmer (e.g., a poor experience with an agro-input dealer in a previous year), which may aect
the ratings received by all agro-input dealer rated by this particular farmer. Furthermore, the same agro-input dealer
may be rated by several farmers, leading to interdependence in the other dimension. For example, the ratings that
an agro-input dealer receives may be aected by a (potentially unobservable) characteristic of the dealer (e.g., dealer
friendliness), which may aect the ratings given by all farmers that rated this particular dealer. To account for
this two-way interdependence in equation 1, we dene a composite error term, (εf,d) that can be decomposed into
a farmer specic component (νf ), an agro-input dealer specic component (ωd), and a residual (ϵf,d) that varies at
the level of the farmer-dealer interaction.
εf,d = νf + ωd + ϵf,d (2)
Equation 2 shows that the dyadic nature of our data leads to two-way clustering in the error term. As long
as the error term is uncorrelated with the explanatory variable(s) included in equation 1, Ordinary Least Squares
(OLS) remains consistent. However, not taking into account within-cluster error correlation generally leads to
standard errors that are biased downward, leading to under-rejection of the null hypothesis that gender does not
aect ratings. In our case, it should be noted that clustering is non-nested. As traditional cluster-robust inference
can only deal with clustering in one of the dimensions, our strategy will consist of including sucient regressors to
minimize concerns about error correlation at the agro-input dealer level, and then cluster standard errors at farmer
level (Cameron and Miller, 2015).
To test for an agro-input dealer gender eect, we can simply compare average ratings received by male-managed
agro-input shops to average ratings received by female-managed agro-input shops. Equation 3 shows how this can
11
be done using a simple OLS regression on dealer-level averages.
1∑ 1∑ 1∑ 1∑ 1∑
yf,d = α+ β gd + νf + ωd + ϵf,d (3)
F F F F F
f f f f f
yd = µ+ γgd + εd (4)
In equation 3, identication of the gender equity eect (β) relies on dierences between agro-input dealers. As
dealer gender is constant for all farmers that rate a particular dealer, the average gd is also a binary indicator of
the gender of that particular dealer d. The farmer-specic component νf is absorbed in the intercept term µ, while
the dealer specic-component ωd is now included in the error term ε. ∑
It is important to note that in equation 4, the error component 1
F ωd in the error term εd may be correlated
f
with the independent variable gd. This would be the case if, for example, female agro-input shop managers are on
average less educated than male agro-input shop managers, and less educated agro-input dealers are rated lower
by farmers. In this case, dierential ratings are not caused by gender in itself, but rather driven by dierences
in education. We will control for a range of agro-input dealer-level potential confounders by including them as
additional regressors (xd) in equation 4:
yd = µ+ γgd + φxd + εd (5)
As we are interested in explaining gender equity bias in perceptions in dierent dimensions, the set of control
variables used will also dier. For example, when farmers are asked to rate agro-input dealers in terms of price
competitiveness, it seems reasonable to include prices charged by dealers as controls. Similarly, for perceptions
related to the quality of seed sold, we are particularly interested in testing if the coecient on the gender of the
agro-input dealer changes after adjusting for various observable dealer characteristics that are directly related to
quality, such as storage technology, infrastructure such as leak-proof roof or insulation, etc. In this way, we attempt
to dierentiate between situations where farmers perceive female-managed agro-input shops less favorably and
situations where dierences in ratings reect real dierences between male- and female-managed shops.
Farmer-level characteristics could also confound the relationship between an agro-input dealer's gender and the
rating that the farmer provides. For example, it may be that farmers who are better educated generally provide
higher ratings. At the same time, it may be that better-educated farmers are more inclined to shop at male-
managed agro-input dealerships. This makes it dicult to dierentiate a gender equity eect from an eect arising
from dierences in education of the farmer. Fortunately, we often have instances where the same farmer rates both
male- and female-managed agro-input shops. This allows us to exploit within-farmer variation for identication.
12
While we would be able to control for a farmer's education level by simply including it in an OLS regression, a
within-farmer transformation also controls for characteristics that would be dicult or impossible to measure and
to control for, like motivation, kindness, locus of control, norms and values, etc. In other words, the within-farmer
(xed eects) estimator removes all farmer-level heterogeneity.
∑ (
1∑ ) (
1∑ )
yf,d −
1
yf,d = β gf,d − gfd + εf,d − εf,d (6)
D D D
d d d
( )
yf,d − yf = γ gf,d − gf + εf,d (7)
Finally, we will also run a xed eects model that, in addition to controlling for farmer heterogeneity through
xed eects, also controls for dealer-level observable characteristics. We will again do this by including additional
regressors (xd) in equation 7, which leads to:
( )
yf,d − yf = γ gf,d − gf + φ (xf,d − xf ) + εf,d (8)
6 Results
Table 4 reports dierences in perceived agro-input dealer quality between male- and female-managed agro-input
outlets using an OLS regression based on equation 4. We nd that on all but one dimension, male-managed agro-
input shops are rated higher than female-managed shops, and the dierence in rating is signicant for ve out of the
seven comparisons (at a 10 percent signicance level). The largest dierence is found when farmers are asked to rate
price competitiveness. Here, female-managed agro-input shops are scored only 3.237 out of 5, while male-managed
agro-input shops receive a score of 3.437 out of 5. Equally large dierences exist when farmers are asked to rate an
agro-input dealer in terms of product availability. Here, male-managed agro-input shops receive an average score
of almost 4 out of 5, while female dealers get 3.792. Interestingly, female-managed agro-input shops are not rated
signicantly worse with regards to the quality of seed sold. They also appear to be equally rated with respect to
location.
Table 5 repeats the between-dealers analysis, but compares ratings for quality attributes of seed sold by the
dealers. While we still nd that on most characteristics, male-managed agro-input shops get a higher rating than
female-managed agro-input shops, the dierences are never signicant. Note that these results are consistent with
what was found in Table 4, where we also failed to nd a signicant dierence when asking about seed quality. This
suggests that when farmers are asked to think about a particular product, they make abstraction of the person
selling it, and the gender eect becomes less important.
13
14
Table 4: Between-dealers model focusing on dealer ratings (control variables not included).
Dependent variable: Average rating received by dealer
Average Dealer's Dealer's Dealer's Dealer's Dealer's Dealer's
dealer general location price seed stock reputation
rating quality quality
(1) (2) (3) (4) (5) (6) (7)
Constant 3.764 3.738 3.89 3.237 3.878 3.792 4.051
(0.05) (0.06) (0.086) (0.071) (0.064) (0.077) (0.064)
Dealer is male 0.109∗ 0.129∗ −0.085 0.2∗∗ 0.052 0.199∗∗ 0.161∗∗
(0.063) (0.075) (0.109) (0.089) (0.081) (0.097) (0.081)
Number of obs. 152 152 152 152 152 152 152
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the dealer is a dummy variable where 1 is male and 0 is female. The ratings received by the dealers are averaged at the dealer level. Standard errors are presented
in ().
15
Table 5: Between-dealers model focusing on seed ratings (control variables not included).
Dependent variable: Average rating received by dealer
Average Seed's Seed's Seed's Seed's Seed's Seed's
seed general yield drought pest/disease speed of germination
rating quality tolerance tolerance maturing
(1) (2) (3) (4) (5) (6) (7)
Constant 3.417 3.87 3.592 3.005 2.683 3.716 3.651
(0.044) (0.051) (0.059) (0.061) (0.062) (0.055) (0.066)
Dealer is male 0.053 0.058 0.066 −0.03 0.011 0.081 0.104
(0.056) (0.064) (0.075) (0.077) (0.078) (0.068) (0.083)
Number of obs. 152 152 152 150 152 151 152
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the dealer is a dummy variable where 1 is male and 0 is female. The ratings received by the dealers are averaged at the dealer level. Standard errors are presented
in ().
Results of between-dealer regressions with added control variables (see equation 5) are presented in Table 6 for
the more general ratings of dealer characteristics and in Table 7 for ratings directly related to seed quality. In
all regressions, we add dealer age and dealer education level as general control variables, and additional controls
depending on the attribute being rated.
In the rst column of Table 6, we again explain the overall dealer rating. As this is an average of all the other
attributes, we also include most controls in this regression, nine in addition to dealer age and education. The
rationale for including each control is provided below in the discussion of the regressions for each rating attribute.
We nd that, even after controlling for a range of observable indicators for overall quality dierences between male-
and female-managed agro-input shops, male-managed shops are rated signicantly higher by consumers.
The second column of Table 6 corresponds to the second column of Table 4, which compares average general
quality ratings given to male- versus female-managed agro-input shops. We have three relatively objective observable
proxies for general dealership quality. First, we asked enumerators to provide an overall cleanness and professionality
rating for the agro-input shop for which they collected data. Second, we construct an index that measures dealer
eort and a range of services that dealers oer to clients.5 In particular, this index accounts for whether an agro-
input dealer 1) always explains how seed should be used (seed spacing, seed rate, complementary inputs); 2) always
recommends complementary inputs such as fertilizers and chemicals; 3) provides extension/training to clients on
how to use improved seed varieties; 4) provides discounts to clients who buy large quantities of seed; 5) sells small
quantities; 6) provides seed on credit; 7) has received a seed-related complaint from a customer; and 8) accepts
mobile money as a payment modality. Descriptive statistics for the variables which constitute this eort and service
index are shown in the appendix in Table 12. Third, we asked enumerators to carefully observe and note down
a range of capital-intensive seed handling and storage practices, which were also summarized in an index. In this
index, we account for whether 1) the roof is leak-proof; 2) the roof is insulated to keep the heat out; 3) the walls
are insulated to keep the heat out; 4) the area where seed is stored is properly ventilated; 5) any ocial certicates
are on display in the shop (e.g., inspection certicates, training certicates, registration with association, etc).
Also for these variables which constitute the capital-intensive practices index, descriptive statistics are shown in
the appendix in Table 12. We see that after controlling for these three groups of variables, the male premium
on general quality ratings increases from 0.13 to 0.16. Note that the index of capital-intensive seed handling and
storage practices observed by the enumerator is signicant and has the expected sign, as input dealers who score
better on this index also receive higher scores on general dealership quality.6
When farmers were asked to assess agro-input dealers in terms of their location, the average distance between
5Indices were created by weighing each component by the inverse covariance matrix, constructed following Anderson (2008).
6However, caution should be taken when interpreting control variables, as control variables do not necessarily have a structural
interpretation. For instance, it may be that the relationship between the control variable and the outcome variable is confounded by a
third (potentially unobservable) variable (Hünermund and Louw, 2020).
16
dealers and their customers,7 capturing some indication of centrality of the dealer, provides an obvious candidate
as a control variable (third column in Table 6). We do not nd a gender equity eect on ratings concerning
location in Table 4, nor do we nd a dierence after controlling for centrality. It should also be noted here that the
control variable is signicant in the expected direction, as dealers for whom the average distance between dealer
and customer is higher (or centrality is lower) also are scored lower in terms of location.
In the fourth column of Table 6, we look at price competitiveness again. To account for the possibility that the
dierence in price ratings between male- and female-managed agro-input shops is driven by actual price dierences
between these shops, we control for the average price charged by the dealer, as well as for the cost at which the
dealer obtains seed (which is an important determinant of the price). The analysis conrms that there is a dierence
in perception of male and female dealers, and that this dierence is not due to actual price dierences. The gender
equity eect is similar in size to the one found without controlling for actual price dierences in Table 4. Note
again that one of the control variables is signicant and suggests that dealers who charge higher prices also receive
signicantly lower price competitiveness ratings.
In the fth column of Table 6, we control for another index, one that reects all seed handling and storage
practices observed by the enumerator. This index includes the ve capital-intensive practices mentioned above, but
also accounts for whether the agro-input dealer 1) destroys seed that has exceeded shelf-life; 2) stores seed in a
dedicated area, away from other merchandise; 3) has no problem with rats, insects, or other infestations; 4) stores
seed in ambient light conditions as recommended; 5) stores seed on pallets or shelves; and 6) does not store seed
in open bags or open containers. This index also includes the shop's overall cleanness and professionality rating
provided by the enumerator. Again, descriptive statistics for the variables which constitute this all (i.e. capital-
and labor intensive) practices index can be found in the appendix in Table 12. As in column 5 of Table 4, we do
not nd a gender equity eect regarding the seed quality rating after controlling for observable quality indicators.
In the sixth column, we repeat the analysis for perceptions related to a dealers' stock, now controlling for the
number of hybrid maize varieties that this dealer has in stock and the quantity bought by the dealer from seed
producers or wholesalers, the former being signicant and having the expected sign. The male premium on the
rating persists, although the eect becomes slightly weaker as compared to a regression without controls (column
6 in Table 4).
The analysis regarding reputation is repeated in column 7, now controlling for the number of years the shop
has been in business, and whether the shop is registered with the Uganda National Agro-input Dealer Association
(UNADA), as we expect both to have an impact on a dealer's reputation. Here we also see that male dealers receive
7The Haversine formula calculating the arc distance between two points on Earth is used. The latitudes and longitudes are extracted
from the GPS coordinates for both farmers and agro-input shops and inserted as paired values in the Haversine formula. This formula
then calculates the distances between these paired latitudes and longitudes in meters, following which we obtain the distances in
kilometers and standardize the variable.
17
higher scores even after controlling for experience and UNADA registration, with the eect becoming slightly weaker
(as compared to column 7 in Table 4).
Table 7 repeats the between dealer analysis for quality attributes of maize seed sold by the agro-input shops
reported in Table 5, but controls for practices that are expected to improve seed quality. As all ratings in Table 5
concern quality, we include the same control in all regressions. We use the most elaborate index of all seed handling
and storage practices as observed by the enumerator that was also used in model (5) of Table 6. Recall from Table 5
that we did not nd a gender equity eect for any seed-quality-related dimensions, and adding the index to control
for quality does not change this. Note that the index is generally positively correlated with the rating, but only
signicantly so when farmers are asked to assess the seed's yield.
Overall, comparing Table 6 to Table 4 and Table 7 to Table 5, we notice that results, both in terms of pa-
rameter estimates for β and their signicance, are very similar. This suggests that dierences between male- and
female-managed agro-input shops reect structural dierences in perception of the two genders, rather than actual
dierences in the dimension being rated (quality, price competitiveness, stock, and reputation).
Tables 8 and 9 show parameter estimates for an indicator variable that takes the value of one if the agro-
input dealer is male, estimated using a model that includes farmer xed eects (using the within transformation
of equation 6).8 In Table 8, we use the general agro-input dealer ratings as outcome variables, similar to Table 4.
Table 9 estimates the same model, but now for the more specic seed-quality-related ratings, similar to Table 5.
8As errors are also correlated within agro-input dealers, we report standard errors that are robust to clustering in this dimension.
18
19
Table 6: Between-dealers model focusing on dealer ratings (control variables included).
Dependent variable: Average rating received by dealer
Average Dealer's Dealer's Dealer's Dealer's Dealer's Dealer's
dealer general location price seed stock reputation
rating quality quality
(1) (2) (3) (4) (5) (6) (7)
Constant 3.79 3.77 3.939 3.326 3.822 3.647 3.921
(0.134) (0.227) (0.185) (0.169) (0.15) (0.204) (0.156)
Dealer is male 0.157∗∗ 0.161∗∗ −0.048 0.219∗∗ 0.057 0.183∗ 0.158∗
(0.063) (0.071) (0.1) (0.091) (0.081) (0.097) (0.082)
Dealer's age in years −0.003 −0.003 −0.002 −0.003 −0.001 −0.001 0
(0.003) (0.004) (0.005) (0.004) (0.004) (0.005) (0.004)
Dealer nished secondary 0.11 0.144 −0.001 −0.01 0.198∗∗ 0.092 0.15
education (0.078) (0.088) (0.122) (0.108) (0.098) (0.122) (0.1)
Shop's cleanness/professionality −0.005
rating by enumerator (0.047)
Index of dealer's eorts 0.181∗ 0.152
and services (0.1) (0.11)
Index of capital-intensive seed 0.29∗∗∗
handling/storage practices (0.079)
observed by enumerator
Standardized distance between −0.089∗∗ −0.302∗∗∗
farmer and shop (0.04) (0.051)
Standardized sales price −0.098∗ −0.176∗∗
of maize seed (0.052) (0.074)
Standardized cost of maize 0.057 0.085
seed for dealer (0.052) (0.072)
Index of all seed 0.093 0.184
handling/storage practices (0.103) (0.118)
observed by enumerator
Number of hybrid maize −0.02 0.07∗
varieties in stock (0.03) (0.038)
Standardized amount of maize −0.013 0.062
seed dealer bought (0.028) (0.041)
Number of years since 0 0.002
shop's establishment (0.006) (0.007)
Shop's UNADA 0.151 0.113
registration (0.097) (0.103)
Number of obs. 149 151 152 151 151 152 152
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the dealer is a dummy variable where 1 is male and 0 is female. The ratings received by the dealers are averaged at the dealer level. Standard errors are presented in ().
20
Table 7: Between-dealers model focusing on seed ratings (control variables included).
Dependent variable: Average rating received by dealer
Average Seed's Seed's Seed's Seed's Seed's Seed's
seed general yield drought pest/disease speed of germination
rating quality tolerance tolerance maturing
(1) (2) (3) (4) (5) (6) (7)
Constant 3.416 3.768 3.584 3.128 2.779 3.644 3.612
(0.105) (0.12) (0.139) (0.147) (0.147) (0.131) (0.157)
Dealer is male 0.053 0.047 0.066 −0.019 0.023 0.075 0.104
(0.057) (0.065) (0.075) (0.079) (0.079) (0.07) (0.085)
Dealer's age in years −0.002 0.002 −0.001 −0.005 −0.005 0.002 −0.001
(0.003) (0.003) (0.004) (0.004) (0.004) (0.004) (0.004)
Dealer nished secondary 0.125∗ 0.122 0.118 0.058 0.188∗ 0.052 0.175∗
education (0.069) (0.079) (0.091) (0.096) (0.096) (0.085) (0.103)
Index of all seed 0.061 0.139 0.194∗ 0.064 −0.089 0.083 −0.016
handling/storage practices (0.083) (0.095) (0.11) (0.115) (0.115) (0.102) (0.124)
observed by enumerator
Number of obs. 151 151 151 149 151 150 151
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the dealer is a dummy variable where 1 is male and 0 is female. The ratings received by the dealers are averaged at the dealer level. Standard errors are presented
in ().
21
Table 8: Farmer xed-eects model focusing on dealer ratings (control variables not included)
Dependent variable: Rating of a particular farmer given to a particular dealer
Average Dealer's Dealer's Dealer's Dealer's Dealer's Dealer's
dealer general location price seed stock reputation
rating quality quality
(1) (2) (3) (4) (5) (6) (7)
Dealer is male 0.116∗∗∗ 0.176∗∗∗ 0.023 0.188∗∗∗ 0.049 0.082∗ 0.18∗∗∗
(0.029) (0.05) (0.048) (0.047) (0.043) (0.05) (0.048)
Number of obs. 3562 3562 3562 3562 3562 3562 3562
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the input dealer is a dummy variable carrying the value of either 0 or 1. Fixed eects or within estimation has been used at the farmer level. Cluster-robust SE
are obtained by clustering at the dealer level (presented in ()).
Table 8 shows that male-managed agro-input outlets receive signicantly higher ratings in the areas of quality
in general, price competitiveness, stock, and reputation. The average rating also shows a signicant dierence
between male and female dealers. Comparing Table 8 to Table 4, the largest dierence can still be found for price
competitiveness, even though the magnitude of the eect decreased somewhat. The eect of gender on ratings
related to stocks reduced sharply after controlling for farmer-level heterogeneity.
For seed-quality-specic ratings, comparing Table 9 to Table 5, we see that, after controlling for farmer-level
heterogeneity, some of the dierences between male and female dealers turn signicant. For perceptions related to
seed germination, male-managed agro-input shops receive a score that is on average 0.11 higher than the germination
rating female-managed shops receive. We further nd signs of gender eects when farmers are asked to rate seed
quality in general and whether seed maturity is as advertised. The gender equity bias in these dimensions is also
reected in a signicant dierence in the average seed rating between male- and female-managed agro-input shops
(column 1).
The fact that we do nd gender equity bias when farmers are asked to assess seed quality if we control for
farmer xed eects suggests that, in the between-dealer regressions of Tables 5 and 7, gender equity bias is obscured
by farmer-level confounders. For instance, it could be that farmers that are better educated also provide higher
ratings and that these better educated farmers are also more likely to shop at female managed dealerships. Not
controlling for dierences in education levels of farmers may then lead to an underestimation of discrimination
against female-managed agro-input shops.
Finally, we run a xed eects model that, in addition to controlling for farmer heterogeneity, also controls for
dealer-level observable characteristics (see equation 8), similar to Tables 6 and 7. Table 10 presents the more general
dealer ratings, and Table 11 presets the seed quality ratings.
We nd that controlling for observable characteristics at the dealer level does not change ndings for the rst set
of ratings, which evaluate the agro-input dealer. The largest gender equity eects are found when farmers are asked
to rate agro-input dealer reputation (column 7) and price competitiveness (column 4). In both cases male-managed
agro-input shops are rated about .2 points higher. The dierence in ratings between male- and female-managed
agro-input shops for the stock attribute has become indistinguishable from zero.
Comparing Tables 9 and 11, the signicant dierences between male- and female-managed agro-input shops with
respect to the seed maturity rating (column 6), germination rating (column 7), and general seed quality ratings
(column 2) found in Table 9 persist after controlling for observable dealer-level dierences in seed quality.
22
23
Table 9: Farmer xed-eects model focusing on seed ratings (control variables not included)
Dependent variable: Rating of a particular farmer given to a particular dealer
Average Seed's Seed's Seed's Seed's Seed's Seed's
seed general yield drought pest/disease speed of germination
rating quality tolerance tolerance maturing
(1) (2) (3) (4) (5) (6) (7)
Dealer is male 0.073∗∗∗ 0.075∗ 0.047 0.053 0.041 0.079∗ 0.11∗∗∗
(0.027) (0.038) (0.043) (0.042) (0.044) (0.039) (0.04)
Number of obs. 3520 3496 3442 3356 3384 3398 3428
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the input dealer is a dummy variable carrying the value of either 0 or 1. Fixed eects or within estimation has been used at the farmer level. Cluster-robust SE
are obtained by clustering at the dealer level (presented in ()).
24
Table 10: Farmer xed-eects model focusing on dealer ratings (control variables included)
Dependent variable: Rating of a particular farmer given to a particular dealer
Average Dealer's Dealer's Dealer's Dealer's Dealer's Dealer's
dealer general location price seed stock reputation
rating quality quality
(1) (2) (3) (4) (5) (6) (7)
Dealer is male 0.119∗∗∗ 0.149∗∗ 0.028 0.2∗∗∗ 0.03 0.082 0.201∗∗∗
(0.034) (0.058) (0.049) (0.05) (0.047) (0.051) (0.048)
Dealer's age in years 0.003∗∗∗ 0.001 0 0.003∗∗ 0.004∗∗∗ 0.004∗∗∗ 0.001
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Dealer nished secondary 0.016 0.043 0.032 −0.034 0.037 −0.007 −0.003
education (0.025) (0.036) (0.035) (0.036) (0.037) (0.036) (0.032)
Shop's cleanness/professionality 0.021
rating by enumerator (0.015)
Index of dealer's eorts 0.01 −0.004
and services (0.029) (0.043)
Index of capital-intensive seed handling/storage 0.057∗
practices observed by enumerator (0.031)
Standardized distance between −0.02 −0.025
farmer and shop (0.016) (0.026)
Standardized sales price −0.024∗ 0.009
of maize seed (0.013) (0.02)
Standardized cost of maize 0.022 −0.009
seed for dealer (0.014) (0.02)
Index of all seed handling/storage −0.006 0.072∗
practices observed by enumerator (0.023) (0.04)
Number of hybrid maize −0.008 0.001
varieties in stock (0.008) (0.011)
Standardized amount of maize −0.002 0.013
seed dealer bought (0.009) (0.013)
Number of years since 0.005∗∗ 0.015∗∗∗
shop's establishment (0.002) (0.003)
Shop's UNADA −0.026 0.002
registration (0.027) (0.031)
Number of obs. 2779 3014 3433 3302 3149 3356 3433
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the input dealer is a dummy variable carrying the value of either 0 or 1. Fixed eects or within estimation has been used at the farmer level. Cluster-robust SE are obtained by clustering at
the dealer level (presented in ()).
25
Table 11: Farmer xed-eects model focusing on seed ratings (control variables included)
Dependent variable: Rating of a particular farmer given to a particular dealer
Average Seed's Seed's Seed's Seed's Seed's Seed's
seed general yield drought pest/disease speed of germination
rating quality tolerance tolerance maturing
(1) (2) (3) (4) (5) (6) (7)
Dealer is male 0.073∗∗ 0.073∗ 0.07 0.032 0.044 0.087∗ 0.084∗
(0.029) (0.041) (0.046) (0.044) (0.049) (0.042) (0.043)
Dealer's age in years 0.001 0.003∗∗ 0.004∗∗∗ −0.001 −0.001 0 0.002
(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)
Dealer nished secondary −0.022 −0.023 −0.029 0.009 0.011 −0.084∗∗∗ 0.011
education (0.02) (0.029) (0.032) (0.031) (0.033) (0.028) (0.032)
Index of all seed 0.023 0.051 0.062∗ −0.009 −0.029 0.035 0.033
handling/storage practices (0.022) (0.033) (0.035) (0.033) (0.035) (0.03) (0.033)
observed by enumerator
Number of obs. 3109 3088 3038 2965 2988 3000 3026
∗∗∗p < 0.01; ∗∗p < 0.05; ∗p < 0.1.
Note: Here, the gender of the input dealer is a dummy variable carrying the value of either 0 or 1. Fixed eects or within estimation has been used at the farmer level. Cluster-robust SE
are obtained by clustering at the dealer level (presented in ()).
7 Conclusion and policy implications
Using survey data from smallholder farmers and agro-input dealers in southeastern Uganda, we tested if farmers
perceive female-managed agro-input shops dierently than male-managed agro-input shops. To do so, we asked a
random sample of farmers to rate agro-input dealers in their immediate neighborhood on a scale ranging from 1
(poor) to 5 (excellent). In addition to asking farmers to rate agro-input dealers on a set of general characteristics
such as accessibility and price competitiveness, we also asked them to focus on a particular product that these
agro-input dealers sell (maize seed), and to rate this product on various dimensions like germination, yield, etc.
Using simple comparisons of average ratings given to male- and female-managed agro-input shops, we found that
female-managed shops are generally rated lower than their male-managed counterparts. However, when farmers were
asked to focus on a specic product, the dierence became insignicant. When adding controls for agro-input dealer-
level observable characteristics, parameter estimates and signicance remained similar, suggesting that dierences
in ratings between male- and female-managed agro-input shops reect structural dierences in perceptions rather
than actual dierences.
However, ratings of agro-input dealers provided by farmers may also be inuenced by farmer characteristics.
To control for farmer heterogeneity, we exploited the fact that farmers often rated several agro-input dealers of
dierent genders, and ran farmer xed eects models. In doing so, we conrmed the existence of gender equity bias
when farmers were asked to rate general characteristics of agro-input dealers, but also found dierences in ratings
of dierent dimensions of seed quality sold by dealers of dierent genders.
Looking into the individual dimensions that were rated, we nd particularly strong gender equity bias when
farmers were asked to rate agro-input dealers in terms of price competitiveness. Furthermore, and especially
after controlling for farmer-level heterogeneity, we nd male-managed agro-input shops to have signicantly better
reputation than female shops. This dierence in reputation is also reected in a signicant dierence between male
and female dealers in the general quality rating. On the other hand, we do not nd that male- and female-managed
agro-input shops were rated dierently when farmers were asked to consider location. This may be because location
is easier to assess objectively. For attributes related to the quality of seed sold by agro-input dealers, gender equity
bias was largest when farmers were asked to assess seed germination rates and whether the seed being rated had
the advertised maturing period.
The prevalence of gender equity bias in the context of agro-input dealers again underscores the importance
of customs and norms in rural societies. Interventions and initiatives that focus solely on increasing women's
empowerment are unlikely to be sucient, and may in some cases even backre (Pair et al., 2021). It will be
important to challenge gender stereotypes and role congruence and such interventions should not focus on only one
gender.
26
Gender norms and customs prevent women from aspiring and developing an internal locus of control. To
overcome this, role models have been found particularly eective in raising aspirations. For example, Porter and
Serra (2020) nd that successful and charismatic women who majored in economics at the same university led to
an increase in girls choosing economics as a major. Role model eects may also emanate from ctional characters.
For example, Riley (2022) shows how the female protagonist in a feel-good Disney movie impacts secondary school
students' exam performance in Uganda. This opens the prospect of using mass media to inuence gender roles and
customs (La Ferrara, Chong, and Duryea, 2012).
Gender roles and customs also indirectly prevent women from realizing their full potential, as men may not
support (or may even prevent) participating of women and girls in male-dominated sectors. However, social norms
are also perceived by individuals. Sometimes, men prevent their wives from participating in the economy because
they think it will be frowned upon by their peers. In cases where perception is inaccurate, simply correcting this
perception may be eective. For example, Bursztyn, González, and Yanagizawa-Drott (2020) nd that in Saudi
Arabia, a country with strict norms and traditions on women's mobility, many men overestimate the severity of peer
judgment if their wives work outside of the house. Correcting this false perception led to a signicant increase in
women's employment. Role models can also be important to challenge beliefs and stereotypes about lesser abilities
of one group held by another group: Beaman et al. (2009) show that the appointment of women leaders to Indian
village councils improved men's perceptions of women's leadership abilities.
In addition to underscoring the importance of gender norms and customs in shaping perceptions, our study serves
to draw specic lessons for policy. We restrict ourselves to three areas where our ndings can be operationalized.
First, even though we nd no evidence of actual dierences in quality between male- and female-managed
agro-input shops, existing training and advisory services for agro-input dealers are likely to be biased toward men.
Ensuring that women entrepreneurs have access to, and benet from, training should be an important policy priority.
The eectiveness and inclusiveness of training programs depend on many attributes of the program. This includes
more obvious aspects such as the content of what is taught in the training and who is targeted, but also less obvious
attributes such as the gender of who provides the training, timings of training, etc. However, it is also important
to change the perception that female-managed agro-input shops are likely to receive less training. This could be
achieved by making training attendance publicly visible, perhaps through a register of trained agro-input dealers,
through certicates that are posted in the shops, etc. such that equal capactity between male- and female-managed
agro-input shops becomes more apparent to clients.
Second, the share of women among opinion leaders and experts in the sector needs to increase. In light of the
emerging evidence of the importance of role models, the presence of women among agro-input dealers, inspectors,
and leadership of professional associations such as UNADA needs to increase. For public sector positions, quota's
may be considered.
27
Finally, we nd that biased perceptions exist especially with respect to prices charged by agro-input dealers.
Simply advertising prices may be sucient to make prices objectively veriable, and customers will need to depend
less on perceptions and the use of mental shortcuts that are prone to gender equity bias.
Acknowledgments
This research was partly funded by the CGIAR Research Program on Policies, Institutions, and Markets (PIM),
led by the International Food Policy Research Institute (IFPRI) and carried out with support from the CGIAR
Fund contributors (https://www.cgiar.org/funders/). The authors also acknowledge nancial support from The
Netherlands  CGIAR research programme on Seed Systems Development, grant number W08.240.105, funded
by the Netherlands Organisation for Scientic Research (NWO-WOTRO). We thank Richard Ariong, Leocardia
Nabwire, Wilberforce Walukano and Marc Charles Wanume for excellent eld support. The analysis is the sole
responsibility of the authors, and does not reect the views of any funding agency or organization mentioned here.
28
References
Anderson, M. L. 2008. Multiple inference and gender dierences in the eects of early intervention: A reevaluation
of the Abecedarian, Perry Preschool, and Early Training Projects. Journal of the American statistical Association
103 (484): 14811495.
Ashour, M., D. O. Gilligan, J. B. Hoel, and N. I. Karachiwalla. 2019. Do Beliefs About Herbicide Quality Cor-
respond with Actual Quality in Local Markets? Evidence from Uganda. The Journal of Development Studies
55 (6): 12851306.
Barriga, A. and N. Fiala. 2020. The supply chain for seed in Uganda: Where does it go wrong? World Development
130: 104928.
Beaman, L., R. Chattopadhyay, E. Duo, R. Pande, and P. Topalova. 2009. Powerful women: does exposure reduce
bias? The Quarterly Journal of Economics 124 (4): 14971540.
Bold, T., K. C. Kaizzi, J. Svensson, and D. Yanagizawa-Drott. 2017. Lemon Technologies and Adoption: Mea-
surement, Theory and Evidence from Agricultural Markets in Uganda. The Quarterly Journal of Economics
132 (3): 10551100.
Bursztyn, L., A. L. González, and D. Yanagizawa-Drott. 2020. Misperceived Social Norms: Women Working
Outside the Home in Saudi Arabia. American Economic Review 110 (10): 29973029.
Cameron, A. C. and D. L. Miller. 2015. A practitioner's guide to cluster-robust inference. Journal of human
resources 50 (2): 317372.
Card, D., S. DellaVigna, P. Funk, and N. Iriberri. 2019. Are Referees and Editors in Economics Gender Neutral?
The Quarterly Journal of Economics 135 (1): 269327.
Chari, A. and P. Goldsmith-Pinkham. 2017. Gender representation in economics across topics and time: Evidence
from the NBER summer institute. Tech. rep., National Bureau of Economic Research.
Correll, S. J., K. R. Weisshaar, A. T. Wynn, and J. D. Wehner. 2020. Inside the Black Box of Organizational Life:
The Gendered Language of Performance Assessment. American Sociological Review 85 (6): 10221050.
Dolan, C. 2001. The 'Good Wife': Struggles over Resources in the Kenyan Horticultural Sector. The Journal of
Development Studies 37 (3): 3970.
Giroux, S., J. Blekking, K. Waldman, D. Resnick, and D. Fobi. 2021. Informal vendors and food systems planning
in an emerging African city. Food Policy 103: 101997. Urban food policies for a sustainable and just future.
29
Gorton, M., J. Sauer, and P. Supatpongkul. 2011. Wet Markets, Supermarkets and the Big Middle for Food
Retailing in Developing Countries: Evidence from Thailand. World Development 39 (9): 16241637.
Hospido, L. and C. Sanz. 2021. Gender Gaps in the Evaluation of Research: Evidence from Submissions to
Economics Conferences. Oxford Bulletin of Economics and Statistics 83 (3): 590618.
Hünermund, P. and B. Louw. 2020. On the nuisance of control variables in regression analysis. arXiv preprint,
arXiv:2005.10314 .
Kahneman, D. 2017. Thinking, fast and slow. Macmillan.
Klein, G., Z. Shtudiner, and M. Zwilling. 2021. Uncovering gender bias in attitudes towards nancial advisors.
Journal of Economic Behavior & Organization 189: 257273.
La Ferrara, E., A. Chong, and S. Duryea. 2012. Soap Operas and Fertility: Evidence from Brazil. American
Economic Journal: Applied Economics 4 (4): 131.
Le Barbanchon, T. and J. Sauvagnat. 2021. Electoral Competition, Voter Bias, and Women in Politics. Journal
of the European Economic Association 20 (1): 352394.
Mitchell, K. M. and J. Martin. 2018. Gender bias in student evaluations. PS: Political Science & Politics 51 (3):
648652.
Orr, A., T. Tsusaka, S. H. Kee-Tui, and H. Msere. 2016. What do we mean by `women's crops'? Commercialisation,
gender and the power to name. Journal of International Development 28 (6): 919937.
Pair, E., N. Vicas, A. M. Weber, V. Meausoone, J. Zou, A. Njuguna, and G. L. Darmstadt. 2021. Quantication
of Gender Bias and Sentiment Toward Political Leaders Over 20 Years of Kenyan News Using Natural Language
Processing. Frontiers in Psychology 12.
Porter, C. and D. Serra. 2020. Gender Dierences in the Choice of Major: The Importance of Female Role Models.
American Economic Journal: Applied Economics 12 (3): 22654.
Riley, E. 2022. Role Models in Movies: The Impact of Queen of Katwe on Students' Educational Attainment.
The Review of Economics and Statistics 148.
Wu, A. H. 2020. Gender bias among professionals: an identity-based interpretation. Review of Economics and
Statistics 102 (5): 867880.
30
Appendix
Table 12: Descriptive agro-input dealer statistics: Variables in indices
mean min max SD nobs
Capital-intensive seed handling/storage practices observed by enumerator
Shop has leak-proof roof 0.539 0 1 0.500 193
Shop has insulated roof 0.617 0 1 0.487 193
Shop has insulated walls 0.813 0 1 0.391 193
Shop is ventilated 0.834 0 1 0.373 193
Shop displays ocial certicate 0.503 0 1 0.501 193
Shop always handles expired seed correctly 0.941 0 1 0.237 185
Labor-intensive seed handling/storage practices observed by enumerator
Shop stores seed away from other products 0.404 0 1 0.492 193
Shop has problem with pests 0.637 0 1 0.482 193
Shop's light is ambient (not direct sunlight/dark) 0.819 0 1 0.386 193
Shop stores seed on pallets/shelves (not directly on wood/oor/cardboard) 0.697 0 1 0.461 185
Shop stores maize seed in open containers 0.192 0 1 0.395 193
Shop's cleanness/professionality rating by enumerator 3.503 1 5 1.142 193
Dealer's eorts and services
Shop always explains to customers how seed should be used 0.472 0 1 0.500 193
Shop always recommends complementary inputs to customers 0.549 0 1 0.499 193
Shop oers extension/training 0.523 0 1 0.501 193
Shop oers discounts for large quantities 0.772 0 1 0.421 193
Shop's smallest seed bag is 1 kg (not larger) 0.728 0 1 0.446 184
Shop provides seed on credit 0.648 0 1 0.479 193
Shop received seed related complaint from customer 0.674 0 1 0.470 193
Shop accepts mobile money as payment 0.399 0 1 0.491 193
31
 
 
ALL IFPRI DISCUSSION PAPERS 
 
 
All discussion papers are available here  
 
They can be downloaded free of charge 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
INTERNATIONAL FOOD POLICY RESEARCH INSTITUTE 
www.ifpri.org 
 
IFPRI HEADQUARTERS 
1201 Eye Street, NW 
Washington, DC 20005 USA 
Tel.: +1-202-862-5600 
Fax: +1-202-862-5606 
Email: ifpri@cgiar.org