REPERCUSSIONS OF FERTILIZER SUBSIDY PROGRAMS ON PRIVATE SECTOR
INPUT RETAILERS:
EVIDENCE FROM MALAWI AND PROPOSAL FOR FURTHER RESEARCH
By
Natalie Bess Fitzpatrick

A THESIS
Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of

MASTER OF SCIENCE
Agricultural, Food and Resource Economics
2012

ABSTRACT
REPERCUSSIONS OF FERTILIZER SUBSIDY PROGRAMS ON PRIVATE SECTOR
INPUT RETAILERS:
EVIDENCE FROM MALAWI AND PROPOSAL FOR FURTHER RESEARCH
By
Natalie Bess Fitzpatrick

This paper looks at the experiences of agricultural input retailers in Malawi as a result of
the Agricultural Input Subsidy Program. In particular, it focuses on the impact of
excluding private sector retailers from participation in the subsidy program. Using a
difference-in-difference approach to modeling fertilizer sales, it attempts to evaluate how
retailers who were allowed to participate for part of the duration of the program but were
then excluded experienced this policy change in comparison to retailers who were not
allowed to participate at any point in the program. Furthermore, by looking at survey
attrition over a two-year period, this paper looks at the evidence for and against the
case that the subsidy program is driving private sector retailers out of business entirely.
Due to known problems with the data and the results of testing done in the process of
this research, this paper goes on to discuss the ideal methods that could produce the
data needed to more accurately and comprehensively address the questions discussed
in the first parts of the paper.

TABLE OF CONTENTS

LIST OF TABLES

IV

LIST OF FIGURES

V

ACRONYMS

VI

1. INTRODUCTION
1.1 MALAWI’S AGRICULTURAL INPUT INDUSTRY
1.2 MOTIVATIONS AND IMPLEMENTATION OF AISP
1.3 MOTIVATION, IMPLEMENTATION AND RESULTS OF 2006/07 STUDY
1.4 MOTIVATION, IMPLEMENTATION AND RESULTS OF 2009 STUDY
1.5 ADDITIONAL RESEARCH
1.6 OBJECTIVE

1
1
4
6
12
14
16

2. IDEAL ANALYSIS AND MODEL DEVELOPMENT
2.1 THE DIFFERENCE-IN-DIFFERENCE APPROACH AND THE TOTAL SALES
MODEL
2.2 DESIRED VARIABLES AND SAMPLE

17
17
18

3.DATA AVAILABILITY AND LIMITATIONS
3.1 ATTRITION BETWEEN 2007 AND 2009
3.2 MISSING DATA POINTS
3.3 INCONSISTENCIES AND UNRELIABILITY OF DATA
3.4 OTHER FACTORS INVOLVED

20
20
21
22
24

4. ANALYSIS CONDUCTED
4.1 TESTING FOR BIASES IN THE DATA
4.2 PRELIMINARY EVIDENCE FROM THE DATA
4.3 TOTAL SALES MODEL
4.4 ADDITIONAL EVIDENCE ON THE INDEPENDENT AGRODEALER SECTORS

26
26
30
31
36

5. DESIRED DATA AND DATA COLLECTION METHODS
5.1 SAMPLING METHODS
5.2 TRAINING
5.3 SURVEY INSTRUMENT
5.4 IMPLEMENTATION

41
41
43
44
46

BIBLIOGRAPHY

48

iii

LIST OF TABLES

Table 1: Composition of Retailer Sample by District and Type of Retailer .................... 13
Table 2: Attrition Among Survey Participants, 2006/07 to 2008/09 ............................... 20
Table 3: Missing Sales Data.......................................................................................... 21
Table 4: Survey Attrition Bias Test Results ................................................................... 27
Table 5: Missing Value Bias Test Results ..................................................................... 30
Table 6: This year do you expect the percent of your sales revenue coming from your
agricultural input business to be higher or lower than in 2007/08? ......................... 31
Table 7: Random Effects Results .................................................................................. 32
Table 8: Fixed Effects Results....................................................................................... 33
Table 9: Expanded Missing Value Bias Test ................................................................. 35
Table 10: Importance of Sales of Small, Intermediate and Large Fertlizer Bags........... 39

iv

LIST OF FIGURES
Figure 1: Structure of the Malawi Fertilizer Industry ........................................................ 2
Figure 2: Conceptual Framework of the 2006/07 Input Supply Assessment ................... 8
Figure 3a, 3b: Discrepancies in Reported Sales Data ................................................... 23

v

ACRONYMS

AD Agrodealer
ADMARC Agricultural Development and Marketing Corporation
AISP Agricultural Input Subsidy Program
SFFRFM Smallholder Farmers’ Fertilizer Revolving Fund of Malawi

vi

1. Introduction
Beginning in 2005/06, Malawi’s government began a series of changes in their
agricultural input subsidy policies. In 2005/06, the government implemented a new
subsidy program to replace and improve upon their previous policies. In subsequent
years they significantly changed how the new subsidy program was implemented.
These changes, which affected subsidies for seed and fertilizer, had a significant effect
on the viability of private sector input supply. This paper looks first at the structure and
composition of Malawi’s agricultural input industry, then at the policy changes that were
enacted and finally at the effects on the private sector input retailers.
1.1 Malawi’s Agricultural Input Industry
Malawi’s agricultural input industry has four general categories of retailers:
government retailers, major distributers, cooperatives, and independent agrodealers.
These are illustrated in figure 1, which shows the general structure of the industry and
the relationships within the industry.

1

Figure 1: Structure of the Malawi Fertilizer Industry

Source: Kelly, Boughton and Lenski 2010

2

The first category is in the public sector, which consists of ADMARC and
SFFRFM outlets. SFFRFM imports fertilizers and also does some distribution; however,
ADMARC is the major distributor of fertilizers for the government, selling supplies that
are imported by SFFRFM as well as products imported for the government by private
sector importers. In the early 1990’s ADMARC was the only seller of fertilizer to
smallholders, however the private sector entered the market during the 1990’s with
market liberalization and a removal of fertilizer subsidy programs. In 2005/06 ADMARC
and SFFRFM were responsible for 45% of fertilizer sales (Dorward et al, 2008). The
extent of government participation is highly variable, but these outlets are important for
farmers who live in areas that do not have a sufficient private presence.
The second category is the major distributors (e.g., Rab, Farmers’ World, Agora).
Most of the distributors are also major importers. All distributors have a network of retail
outlets that not only sell fertilizers but also conduct other commercial functions such as
buying crops from farmers or selling food products, farm equipment, seeds, and crop
chemicals. However, as of 2008/09, RAB has withdrawn from selling fertilizer as a
result of uncertainty about government policy (Kelly, Boughton and Lenski 2010).)
A third category is the independent agrodealers. These are independently
owned small outlets that usually sell some combination of agricultural inputs (improved
seeds, fertilizers, and crop chemicals) and other products such as agricultural
equipment, groceries, hardware, clothing and housewares (Kelly, Boughton and Lenski
2010). They typically have smaller stocks and less variety than the major distributor
outlets. Many of the agrodealers concentrate on seeds and crop chemicals because
the capital required is lower than for fertilizer and the margins tend to be higher. Among

3

those who do sell fertilizers, they often differentiate themselves from the distributors by
selling in smaller quantities than the 50 kg bags sold by the distributors.
Two organizations provide support to the independent agrodealers, CNFA and AISAM.
These organizations provide support at the retail level including training programs that
improve dealers’ ability to provide advice to customers on the appropriate use of inputs,
to manage finances, and to access market information. CNFA also facilitates AD
access to supplier credit by offering the suppliers a 50-75% guarantee on any credit
extended to ADs in the CNFA network. Thus far, only a small percent of CNFA
agrodealers have met the eligibility criteria for getting the credit. Several of the early
recipients, however, have “graduated” and now receive credit directly from the suppliers
without the CNFA guarantee.
A fourth category is the network of cooperatives. These are part of the NASFAM
network, and commercially supply inputs to members. These cooperatives benefited
from donor support, which provided substantial funding for training and other types of
assistance through 2006/07. Although a major source of input supply for their members
in the past, including having been authorized to distribute subsidized fertilizers in
2006/07 and 2007/08, NASFAM, due to an increasingly difficult financial situation,
withdrawn from fertilizer sales almost entirely before the 2008/09 decision to limit
distribution of subsidized fertilizers to government outlets.
1.2 Motivations and implementation of AISP
During 2002/03 and 2003/04, when Malawi was using a Targeted Inputs
Program, only 40% of smallholders bought commercial fertilizer, indicating that more
than half of smallholder households could not afford commercial fertilizer. This would

4

result in lower maize productivity and contribute to food insecurity problems (Kelly et al.
2008). Therefore, in 2005/06 the government began the AISP program. The objectives
were to increase smallholder productivity as well as increase food and cash crop
production in order to reduce vulnerability to food insecurity.
In its first year, 1.3 million farmers received a total of 130,000 tons of fertilizer
(World Bank, n.d.). By the 2007/2008 season, 1.7 million farmers received 170,000
tons of fertilizer. The program uses vouchers, which are distributed to farmers and are
redeemable for fertilizer at a discounted price at retail stores. The price farmers pay
with the vouchers is about one-fifth of the market price for fertilizer (IFPRI, 2009). In
2008/09, this program cost 80 percent of the budget of Malawi’s Ministry of Agriculture
and Food Security, and 15% of the national budget.
The four types of retailers in Malawi’s agricultural input sector have had distinct
experiences through the course of AISP. In 2005/06, only government structures were
allowed to participate in the voucher program. However, in 2006/07 and 2007/08,
vouchers were redeemable at both government retailers and at government authorized
private sector distributors; independent agrodealers were excluded from direct
participation, but a few managed to arrange with the distributors to redeem vouchers on
their behalf. In 2006/07 and 2007/08, the private sector sold around 25% of the
subsidized fertilizer. However, in 2008/09 the inclusion of the private sector was
discontinued, and since that time the vouchers are only redeemable at government
retail stores. According to the government, this was necessary to prevent fraud and
improper use of the vouchers. This claim is not supported by evidence which showed
that fraudulent vouchers were less prevalent, in proportion to the number of vouchers

5

accepted, among the private sector than at ADMARC/SFFRFM shops (Logistic Unit,
2009).
The effect of this exclusion on fertilizer sales and on the viability of the private
sector fertilizer retailers has not yet been fully determined. Understanding the effect of
the fertilizer subsidy program on private sector retailers is important for several reasons.
First, other countries in Africa are developing or considering programs that are similar.
Therefore, a deeper understanding of the effects of this program may provide lessons
for others. Second, the subsidy program is quite expensive for Malawi, creating
questions about the sustainability of the program. If the subsidy program cannot be
sustained long term, then maintaining a strong private sector would be important.
However, some (especially government retailers) argue that the public sector is better
suited to providing agricultural inputs compared to the private sector, which was part of
the justification of private sector exclusion in Malawi’s program in 2008/09.
Understanding the effect of policy changes within the general subsidy program is also
important. Kelly, Boughton and Lenski (2010) reported, “In 2007, [private sector] input
suppliers were unanimous in stating that the most important thing the government could
do to assist the development of a viable input supply network was to ensure policy
stability.” Two large-scale input supply assessments that included the perspectives of
the retail sector have been done, one in 2006/07 (reported in Dorward et al. 2008) and
one in 2008/09 (Kelly, Boughton, Lenski, 2010).
1.3 Motivation, implementation and results of 2006/07 study
The first of these input supply assessments was done in 2006/07 as part of a fullscale assessment of the subsidy program conducted for Malawi’s Ministry of Agriculture

6

and Food Security (see Dorward et al. 2008). The goal of this evaluation was to
“assess the impact and implementation of the AISP in order to provide lessons for future
interventions in growth and social protection” (Dorward et al, 2008). The overall study
was broad in nature, looking at a wide range of actors and factors hypothesized to have
an effect on the subsidy program implementation and impacts. Figure 2 is a diagram of
the conceptual framework used for the overall study.

7

Figure 2: Conceptual Framework of the 2006/07 Input Supply Assessment

Source: School of Oriental & African Studies, Wadonda Consults, and Michigan State University, 2008.

8

The part of the initial study of most relevance to the current paper is that dealing
with the input supply system (top left corner of Figure 2). The general objectives of the
input supply component were to examine the performance of the program, describe its
impacts on different types of actors (ADMARC/SFFRFM, private sector importers,
private sector distribution chains, cooperatives, independent agro dealers), and solicit
recommendations for improvements. The indicators of interest were sales volume
trends, displacement of commercial sales, and changes in costs of doing business,
confidence in the sector, investment, and numbers of competitors.
Information about the input supply sector came primarily from formal interviews
with the major importers and distributors (both public and private sector), a survey of
input retailers in 6 purposively selected districts, secondary data on fertilizer imports
provided by government services, and retail outlet sales data provided by
importer/distributors for selected outlets in the 6 districts covered by the retail survey.
Quantitative data on input access and use collected at the household level through a
nation-wide survey of 2,491 farm households was used to supplement the input sector
study (particularly the analysis of displacement). Information collected through focus
group discussions carried out with farmers and community leaders in the same 6
districts as the retailer survey were also used to supplement the input sector analysis.
The data of most interest to the present study is that collected by the retailer survey.
The sampling frame for the 2006/07 retailer survey was created by combining complete
lists of AISAM dealers participating in the seed subsidy program or who expressed
interest in the fertilizer voucher program, a complete list of individuals who had been
trained to be agrodealers by CNFA, lists of distributor retail outlets provided by the

9

home office of each distributor, and a list of ADMARC/SFFRFM outlets provided by the
logistics unit. One problem with the sampling frame was the lack of a complete census
of agrodealers, which meant that any who were not members of CNFA or AISAM were
not included in the initial sampling frame.
Retailers who were interviewed were selected randomly from each category
(distributors, government, independent agrodealers). Agrodealers were intentionally
over-sampled relative to others because of their heterogeneity compared to distributors
and government outlets.
There were several constraints affecting the survey of retail shops, and what
analysis it could be used for. The first was that because of costs and logistics, the
retailer survey was done in only six districts, meaning that the results could not be
considered nationally representative. Two districts were selected in each region, one
district that had relatively dense input supply, and one that had less dense coverage.
Consequently, the results of the retailer survey are not nationally representative but
reflect the situation in each of the selected districts. A second constraint was the fluidity
of the independent agrodealer part of the retail sector within those districts. Some of
the shops from the CNFA and AISAM lists could not be found or were not selling inputs,
and a number of independent AD shops not on the list were identified and used as
substitutes. 1

1

CNFA listed as members all individuals who had received CNFA training to become
agrodealers, whether they had actually started selling inputs or not. Many individuals were
encouraged to participate in the training by the expectation that they would receive credit and
other financial support to start a business; when this was not forthcoming, those trainees
without their own financial resources did not start selling inputs or sold seeds and agricultural
chemicals only, which are less capital intensive than fertilizers.

10

In addition to the sampling issue, there were data challenges. During the
questionnaire testing it became apparent that collecting accurate quantitative data on
sales and profits was difficult, with very few of the independent agrodealers able to
provide reliable information over time and most of the distributors reporting that only
their home offices were able to provide sales data. To address these problems, the
home offices of the distributors (including ADMARC/SFFRFM) were contacted and
agreed to provide the sales data for the selected outlets. The quantitative questions
about sales and profits were retained for the independent agrodealers, but a number of
more qualitative questions about perceptions of trends in sales and profits were added
to supplement the weak quantitative data.
One component of the 2007 analysis was an estimate of fertilizer
displacement, using both national sales data and household survey data. There are two
relevant types of displacement in the case of the fertilizer subsidy. The first is on the
consumer side of the subsidy. In this case displacement can be defined as the percent
decrease in purchases of un-subsidized fertilizer per unit of subsidized fertilizer
purchased. The second type is on the retailer side. In this case displacement can be
defined as the percent decrease in sales of un-subsidized fertilizer per unit of
subsidized fertilizer sold. In both cases these are aggregate measures, not individual
consumer or retailer measures.
The results of the 2007 study consumer side displacement analysis show
displacement between 30 and 40%. Displacement is higher among the less poor
farmers. The evidence of displacement suggests that the program is not being as cost
effective as it could be, with each sack of subsidized fertilizer the government is offering

11

bringing about less than a full sack of additional fertilizer use. The study also discussed
the retailer side displacement. A regression of private sector sales from 1997/98 to
2004/05 on parastatal sales showed a negative relationship, with parastatal sales in
2005/06 leading to a 32% reduction in private sector fertilizer sales, and a 26%
reduction in 2006/07.
However, there were some issues affecting the ability to use national sales data
to look at the retail side, making the estimate less ideal. One of these weaknesses was
because looking only at a simple regression of sales ignores the effects of other
relevant variables that affect unsubsidized sales, but there was not sufficient data to
include variables such as fertilizer and crop prices, income, and density of sale outlets
and there were limited degrees of freedom as well.
The study used the retailer survey to look at suppliers’ views about the strengths
and weaknesses of the program, as well as their general opinions about the concept of
the program. Despite implementation problems, most retailers supported continuation
of the program; only 10% of retailers surveyed wanted to end the program or return to
an earlier subsidy policy. Retailers who supported the program cited many reasons for
continuing the program including that it assisted the poor and the small farmers, that it
reduced hunger and increased yields, and that it increased business.
1.4 Motivation, implementation and results of 2009 study
In 2009, a follow up study to the 2007 input supply study was conducted, and
results reported in Kelly, Boughton and Lenski (2010). The goal of this study was to
evaluate how the AISP, and in particular the changes to the program in 2007/08 and
2008/09, affected the structure and performance of the agricultural input sector, with an

12

emphasis on the program’s effect on commercial input supply. This analysis primarily
relied on descriptive statistics. The sources of data used were reports from the
Logistics Unit, interviews with key informants, a retailer survey, focus groups with
farmers and interviews with community leaders.
As in 2006/07, there were several constraints on implementation of the retailer
survey. First, the intent was to re-interview each of the retailers interviewed in 2006/07.
However, only 65% of those retailers were located and re-interviewed. In the process of
adding new retailers to the sample, a few problems occurred. First, the lack of an
updated sampling frame meant that the replacement process was inconsistent.
Replacement retailers were selected by each survey team using a combination of
information from other retailers and community leaders to identify candidates for the
survey. Second, there were significant changes in geographic distribution of the
retailers surveyed and in composition of the retail sample. These changes are shown in
table 1.
Table 1: Composition of Retailer Sample by District and Type of Retailer
ADMARC/
Coops
Distributors SFFRFM
2007 vs.
2007 vs.
2007 vs.
09
09
09
District
Region
North
Mzimba
10 - 21
5-6
4-2
Rumphi
2–6
24 - 20
6-2
Center Kasungu
14 - 16
8-6
4-2
Lilongwe
10 - 11
6-5
7-3
South
Blantyre
5–5
15 - 16
0-1
Machinga
6–7
13 - 12
0-0
TOTAL Number
47 - 66
71 - 65
21 - 10
Percent
17 - 29
26 - 28
8-4
Source: Kelly, Boughton and Lenski (2010)

13

Independent
agrodealers
2007 vs. 09
31 – 10
8–6
24 -14
27 -22
22 – 23
19 – 14
131 – 89
49 – 39

One important issue to note is that these changes in composition are not necessarily an
accurate reflection of changes in population of retailers, especially in terms of the
number of distributors which were substituted for independent agrodealers when there
were not enough agrodealers to survey. This is primarily a result of a lack of a census
of input retailers, and therefore a lack of a sampling frame.
A second constraint on implementation was the time spent training the
interviewers. The shorter training period relative to the prior survey meant that
interviewers were sometimes confused about the process of conducting the survey and
did not get adequate training with respect to sample replacement procedures.
Despite these limitations, the 2009 study found several interesting results. The
first set of these results looked at changes in composition of the agricultural input
sector. Among these was an analysis of the change in number of competitors that input
retailers were facing. Retailers were asked to report the number of competitors within
their marketing area, which was defined as the average distance customers traveled.
The results show a 15% increase in number of competitors from 2005/06 to 2008/09 for
independent agrodealers and a 3% increase for distributors over the same time period.
For ADMARC/SFFRFM, there was a 7% increase in number of competitors from
2005/06 to 2007/08, followed by a decrease in competition to below 2005/06 numbers in
2008/09, following the exclusion of the private sector from the fertilizer subsidy program.
1.5 Additional Research
Considerable work has been done on evaluating input subsidy programs, both in
Malawi and in other countries in Africa. Much of this work has focused on how to
maximize the effectiveness of such programs, especially in terms of increasing fertilizer

14

usage, but there has also been substantial research on how subsidy programs do and
should relate to commercial market development. This section will summarize some of
the most relevant and recent research that has been done.
Substantial research has been done on “smart” subsidies. The principles for this
are outlined by Minde and Ndlovu (2007) “as those involving (S)pecific targeting to
farmers who would not otherwise use purchased inputs…(M)easurable impacts,
(A)chievable goals, a (R)esults orientation, and a (T)imely duration of implementation”
(Minde et al, 2008). These principles are further outlined in Morris et al. (2007), and
Minde et al. (2008) discussed the applications of the principles of smart subsidies to
Malawi, Zambia and Kenya in assessing those countries subsidy programs. Dorward et
al. (2008) discussed the implementation and design of Malawi’s program in context of
this approach.
Xu, Burke, Jayne & Govereh (2009) studied the effects of a subsidy program in
Zambia. Using rural household data, they measured the “crowding in” and “crowding
out” of the commercial fertilizer sector as a result of Zambia’s government programs.
They conclude that the effect depends on local characteristics such as the existing level
of private sector activity and average wealth. However, Zambia’s program continues to
use private distributors as a channel for distributing subsidized fertilizer, so the recent
experience of the private sector in Malawi is substantially different.
Rickert-Gilbert and Jayne (2009) looked at how the fertilizer subsidy program in
Malawi affected demand for commercial fertilizer. Using panel household survey data,
they estimated demand for commercial fertilizer. They found that while the

15

implementation of the subsidy did result in an increase in total fertilizer use, there was
some displacement of commercial fertilizer sales.
1.6 Objective
While the 2006/07 and 2008/09 input market assessments in Malawi provided
valuable insights into the experience of agricultural input retailers with the AISP, the
objectives of the study do not include a formal econometric analysis of the sector.
Although the data collected is not ideally suited for this type of analysis, econometric
results could provide useful insights that cannot be achieved through the use of
descriptive statistics. One ideal measure would be to calculate a formal measurement
of displacement based on retail data as a counterpart to the calculations made from the
household data. In this case displacement would be defined as the decrease in
commercial fertilizer sales as a result of subsidized fertilizer sales. However, there is
not sufficient aggregate data on a national level to estimate this. Instead, using the
sales and price data provided in the 2008/09 survey, this study will attempt to
supplement the results from the existing 2008/09 assessment with an econometric
model for fertilizer sales in the private sector. Additionally, since the experiences of
independent agrodealers have differed significantly from those of distributors, this study
will use data available on survey attrition to assess experiences of agrodealers over a
three-year portion of the subsidy program.

16

2. Ideal analysis and model development
2.1 The difference-in-difference approach and the total sales model
In order to understand the short-term effects of the exclusion of the private sector
from the fertilizer subsidy program, one approach is to model total fertilizer sales. Since
independent agrodealers and distributors had different experiences, with agrodealers
being excluded since the beginning of the program and distributors being newly
excluded in 2008/09, this allows the use of a difference-in-difference approach to model
the effect of the policy. This method can be applied to repeated cross-section data or to
panel data. Since the data available for the proposed regression analysis is quasipanel, this section will only consider that case.
A generalized two period version of the difference-in-difference model using
panel data is:
(1) yit=βo + δod2t + δ1policyit + ai + µit
In this model, d2t is a dummy variable equal to 0 in the first time period and 1 in the
second time period, policyit is a variable equal to one if case i is affected by the policy in
period t and zero otherwise, and ai are observable fixed effects. Using a first difference
yields
(2) (yi2- yi1)= βo + δ1(policyi2- policyi1) + (µi2 - µi1)

This simplifies to
(3) ∆ yi= βo + ∆(policyi) + ∆(µit)
Using fixed effects or variable effects modeling approach for equation 1, this model can
be used to assess a policy impact over a two year period.

17

This approach can be applied to the desired model for total sales, where yit’s are
fertilizer sales. In this case, since the policy change of interest is the exclusion of
distributors from receiving fertilizer coupons, policy it can be represented using an
interaction term between a dummy variable for distributor and a dummy variable for the
year. However, to accurately model sales, other time varying factors need to be
included in the model (including but not limited to prices), so the model for total sales
should be adjusted to
(4) Salesit=βo + δod2t + δ1policyit + βx(cit ) +ai + µit
Where cit is a vector of other time varying factors such as prices.

2.2 Desired variables and sample
To estimate fertilizer sales, there are several desired variables to give the most
accurate model. First, total fertilizer sales need to be provided for at least two years,
but more years are highly desirable to give as detailed and “long-term” data as possible.
The other important variables to consider are the cit’s. The most basic of these is
the sale price of fertilizer for each retailer for each year of the panel or quasi-panel.
Ideally this would be weighted for the varying prices for the different types of fertilizer
sold (urea, DAP, 23.31.0 S4, CAN, and compound D). This would require detailed
sales data on how much they sold of each type of fertilizer in addition to the total
fertilizer sales. It would also ideally be weighted for differentiated prices in the rainy
season and the dimba and dry seasons. Finally, since some retailers sold both 50kg
bags and smaller quantities of fertilizer that were priced per kg, including the
differentiated prices for those would be ideal.

18

Other desirable variables in the cit vector include variables for expected rainfall
and expected maize prices, preferably calculated at the district level. There are several
ways of calculating these expectations. For rainfall using a long-term average is best.
There are several potential models for price expectations. Variables representing fixed
characteristics of each retailer such as business structure and geographic location are
also necessary. Finally, data on fertilizer coupon distribution is desirable to give an
estimate of the approximate effect on retailer sales from the actual distribution process
of the coupons.
The other consideration in the estimation process is the ideal sample to be used.
Maximizing sample size, conditional on maximizing accuracy of the data provided by
each retailer included in the sample, is the most important consideration here.
Therefore, the ideal sample would include all private sector retailers who sold fertilizer in
at least one year of the panel. If using true panel data a balanced panel would also be
desirable. Another consideration is to make sure to exclude any retailers whose
answers are likely to be suspect, including retailers who provided contradictory data or
who were not qualified to answer the questions.

19

3. Data Availability and Limitations
3.1 Attrition between 2007 and 2009
Between 2006/07 and 2008/09, there were significant levels of attrition among
participants. The extent of this problem is shown in table 2.
Table 2: Attrition Among Survey Participants, 2006/07 to 2008/09
Distributors
N surveyed in
2006/07
N re-surveyed
in 2008/09
% attrition

Cooperatives
21

Independent
agrodealers
132

Government
Structures
70

48
40

10

79

61

16.7

52.3

40.2

12.9

While the causes of attrition cannot fully be known, within the private sector it might be
explained by the potential crowding out of the private sector as a result of the fertilizer
subsidy program. If that were the case, the data might be biased towards showing less
of a drop off in sales, as only the businesses that were hurt the least would be included
in the sample. This is more likely to be an issue among agrodealers, since they have
been excluded since the beginning of the program, whereas distributers would have
only been excluded for one year, meaning that they are more likely to have survived
despite any possible crowding out effects. This interpretation is supported by the
relative numbers of retailers that could not be located in 2008/2009, and by the relative
numbers of new retailers interviewed in 2008/09. The high levels of attrition meant that
even if there were no other issues, using a balanced panel between the two years of the
survey would have had a much smaller sample size than either survey alone.

20

3.2 Missing Data Points

In the 2008/09 survey, many interviewees did not answer all of the questions
they were asked. Areas of particular concern for this analysis include data on fertilizer
prices and on the quantity of fertilizer sales. Table 3 demonstrates how extensively
missing data points decreased the useable sample size.
Table 3: Missing Sales Data
Distributors
N interviewed
N providing
sales data for
2007/08
N providing
sales data for
2008/09

Cooperatives
10
7

Independent
Agrodealers
89
50

Government
Structures
65
39

66
43

52

6

54

56

For the case of the desired price data, there were several non-ideal types of
missing data. As described in section 2, ideal price data would be weighted for the
different types of fertilizer and account for differentiated prices between 50kg bags and
smaller quantities of fertilizer purchased. However, weighting the price data was not
feasible since the sales data was not differentiated between types of fertilizer.
Therefore, in terms of including prices for both 50kg bags and smaller quantities of
fertilizer purchased, only 7 of the private sector retailers interviewed in 2008/09 provided
data on what their price was for fertilizer sold by kg for urea (the most common fertilizer
purchased) in 2008/09 and only 11 provided this same data for their 2007/08
operations. A final problem with the price data was that there were some missing
values for prices among retailers who provided sufficient sales data. Instead of

21

excluding all these cases, this problem was resolved by generating an approximation
using the median price within the district where the retailer was located.
3.3 Inconsistencies and unreliability of data
Quasi-panel data on sales were collected in both the 2007 and 2009 surveys. In
2009, data was collected retroactively for several years, including the 2006/07 season.
However, in the process of determining the years to include in the quasi-panel analysis,
a comparison of the reported sales in 2006/07 from the 2007 survey and the reported
2006/07 sales collected in the 2009 survey revealed that there was no clear relationship
between these answers. A simple correlation test found no correlation between the two
sets of answers.
A visual representation of this problem is presented in figures 3a and 3b. Figure
3a shows all of the available data points, whereas figure 3b omits the outliers and
shows the smaller sales points, which are not otherwise visible.

22

Figure 3a, 3b: Discrepancies in Reported Sales Data

23

Reliability of the sales data for 2006/07 provided in 2008/09 is suspect for two
reasons. First, in 2008/09 the survey team had very limited field time and often
interviewed sales clerks instead of waiting to speak with owners or managers. Second,
sales records for previous years were not often kept on site, so even managers would
not be expected to have as accurate recall of sales as they would have in the actual
year of the survey. Therefore, presumably the results from the 2007 survey are more
accurate, however price data was not available from that survey. Since the reliability of
the data for 2006/07 in the 2009 survey is very questionable, the analysis will only look
at the quasi-panel data set of sales in 2007/08 and 2008/09, both of which were
collected in the 2009 survey.
3.4 Other factors involved
In examining the data, it is also important to keep in mind that in addition to the
fertilizer subsidy program, other events in Malawi also affected fertilizer sales and the
experiences of agricultural input retailers. One of these issues deals with the
cooperatives; between 2006/07 and 2008/09, the cooperatives lost some of their
funding from USAID and other donors. This greatly reduced their viability, and explains
why so many cooperatives were not re-interviewed, completely separate from any
possible effects from the fertilizer coupon program. Another issue that affected retailers
was that AISAM network did not work in the Center and North regions in 2008/09,
negatively impacting the agrodealers in those regions. Both of these events are likely to
have had significant effects on the cooperatives and agrodealers respectively, but would
not show up explicitly in analysis and may give a misleading impression about the effect

24

of other events such as the fertilizer subsidy. It is important to keep this in mind during
the subsequent analysis and examination of the results.

25

4. Analysis Conducted
4.1 Testing for Biases in the Data
As discussed in section 3, the quality of the data may have suffered as a result of
high levels of attrition and high numbers of missing values. To assess the extent of
these problems, probit functions were used to examine two separate issues.
The first is an assessment of whether there were systemic biases impacting attrition
from the survey between 2006/07 and 2008/09. Because the final sample is a quasipanel sample rather than a true panel drawing on the first survey, this potential bias is
not directly related to the formal regression analysis. However, since the sampling for
the 2008/09 survey was based on the 2006/07 sample, it is worth considering
separately from the issue of modeling the sales data. This issue will be referred to as
survey attrition bias.
The second issue is an assessment of whether the missing values from the
quasi-panel desired sample created any systemic biases in the data used for the sales
model. Unlike the question of survey attrition, this is directly relevant to the formal
regression analysis and must be considered carefully. This issue will be referred to as
missing value bias.
Survey Attrition
To test for survey attrition bias, a probit function was used. The identified
possible sources of bias were business type (agrodealer, cooperative, distributer and
government), region (north, central and south), participation in the seed coupon
program (in 2006/07), and selling agricultural chemicals. The first two are very basic
business characteristics, which might be likely to influence attrition. The last is included

26

based on the hypothesis that businesses that participated in a subsidy program that
they were not then excluded from might have had more stability in their businesses
despite changes in the fertilizer subsidy program. Because of different levels of
participation in the seed program between independent agrodealers and distributors
(85% of distributors participated in 2006/07 while less than 30% of agrodealers
participated) these effects are separated out into two interaction variables, seed subsidy
participation by agrodealers and seed subsidy participation by distributors.
Therefore, the model for assessing survey attrition bias is:

(5) Survey attrition=f(distributer, agrodealer, cooperative, north, central,
seedparticipation*agrodealer, seedparticipation*distributor, soldchemicals).
The results from this model are shown in table 4.
Table 4: Survey Attrition Bias Test Results
Df/DX
Distributer***
-0.562
Agrodealer*
0.18
Cooperative***
0.353
North***
0.337
Central***
0.250
Agrodealer seed
0.110
participation
Distributer seed
0.95
participation
Sold Chemicals
-0.127
* Statistically significant at the 10% level
***Statistically significant at the 1% level

Z
-15.81
1.71
2.89
4.21
3.31
0.77

P>|z|
0.000
0.088
0.004
0.000
0.001
0.441

.

.

-1.17

0.242

These results point to multiple sources of attrition bias. First, both independent
agrodealers and cooperatives were significantly more likely to fail to be re-interviewed

27

than government retail outlets, whereas distributers were more likely to be reinterviewed than government retail outlets. For the cooperatives, this can be attributed
to a reduction of funding by USAID and other donors that forced them to reduce the
magnitude of their input supply program. For agrodealers, the significant levels of
attrition may have some of the most direct implications on how those retailers were
affected by the 2008/09 private sector exclusion and will be explored in more detail
later. Including the interaction term for seed participation and including the sold
chemicals variable resulted in agrodealer only being statistically significant at the 10%
level, whereas when they are excluded the agrodealer dummy is statistically significant
at the 5% level. This suggests that participation in the seed program and selling
agricultural chemicals may have influenced agrodealers’ abilities to stay in business,
even though they were not statistically significant. Another source of bias was regional;
retailers in the North and Central regions were much more likely to fail to be reinterviewed than retailers in the South. This can largely be attributed to the fact that the
AISAM network did not work in the North and Central regions in 2008/09. By itself,
participating in the seed coupon program was not at all statistically significant, even
accounting for the differences between agrodealers and distributors, and neither was
selling chemicals.
Missing Values
Since all government retailers and cooperatives were excluded from the analysis,
the desired sample would be a quasi-panel sample including all of the agrodealers and
distributors who were interviewed in 2008/09 and sold fertilizer in at least one year,
regardless of whether they were interviewed in 2006/07. The degree to which this

28

sample differs from the actually available sample has already been discussed in section
3.
A modified version of the survey attrition probit model will provide an assessment
of the extent to which there are any systemic biases from missing values. Because all
government retailers and cooperatives were excluded, those variables cannot be
included in the second model because they would be perfect predictors. Additionally,
there is no reason to include participation in the seed program, since this model is
assessing willingness and ability to provide answers, not health of the business. To try
to assess any relationship that existed with the 2006/07 survey, a new variable was
included, “original case” which is a dummy variable equal to one if the retailer was
interviewed in 2006/07 and zero if it was a new retailer selected in 2008/09. Because
agrodealers and distributors had very different experiences, “original case” is in the
model as an interaction term, evaluated separately for distributors and agrodealers.
Therefore, the model for assessing missing value bias is
(6) missing values=f(agrodealer, north, south, originalcase*agro,
originalcase*distributor)
The results from this model are shown in table 5.

29

Table 5: Missing Value Bias Test Results

Distributer
North
South*
Agrodealer
Original case
Distributer
original case

dF/dx
-0.038
-0.116
0.163
-0.182

Z
-0.2
-1.12
1.67
-1.16

P>|z|
0.841
-0.264
0.096
0.246

-0.166

-1.19

0.234

* statistically significant at the 10% level

These results show that there is no strong evidence of any statistically significant
missing value bias. South is significant at the 10% level, with retailers in the south
being less likely to give enough data to be included in the sample. Original Case is not
significant for either agrodealers or distributors, suggesting that there is not a strong
linkage between the two types of potential bias. However, since no variables were
significant at the 5% level, the overall implications of this test indicate that there does
not need to be correction for missing value bias in the main regression models.

4.2 Preliminary Evidence from the Data

Before conducting the formal econometric analysis, looking at descriptive
statistics provides a baseline for assessing the available data. This was done by looking
at retailers’ perceptions of the effects of the subsidy exclusion policy. In 2008/09
retailers were asked how they expected their profits from input retail to change over
time, and why. To maximize sample size, this portion of the analysis will include all
retailers interviewed in 2008/09 who sold fertilizer. These results are shown in table 6.

30

Table 6: This year do you expect the percent of your sales revenue coming from your
agricultural input business to be higher or lower than in 2007/08?
Distributer
Num. %

Government
Num.
%

Higher 27
44
44
79
Lower
32
53
6
11
Same
1
3
6
11
source: Kelly, Boughton and Lenski 2010

Cooperative
Num.
%

Agrodealer
Num.
%

4
5
0

38
26
12

50
50
0

50
34
15

Among agrodealers, the majority of those expecting an increase cited increased
demand for inputs as the reason; other reasons cited were good rains, the subsidy
program, and other business factors. Agrodealers expecting a decrease primarily
attributed the decrease to the subsidy program. Distributers cited these same factors as
well. Unsurprisingly, government structures anticipated increases primarily due to the
subsidy program, but also cited increased demand for inputs. Only a few government
structures anticipated a decrease in sales, but those who did attributed it to the subsidy
program, too much competition, and lower supplies and stock.
These results suggest that distributers and independent agrodealers may have
had mixed effects from the subsidy program, and even those who were negatively
impacted by it may have had this effect partially compensated for by increased demand
for agricultural inputs. This is important to keep in mind as the econometric results are
examined.
4.3 Total sales model
Based on the derivation of the model outlined in section 2, the estimation form for
the total sales model is

31

(7) TSit= β0+δ00809t + β1distributeri + β2distributeri0809t + β3Pmz* it, + β4Pfert it + β5
Northi + β6 Center + β6rainfall*it ,+ eit
To evaluate the specifications for this model, both a fixed effects regression and a
random effects GLS regression were run, and then tested for random effects using a
Breusch and Pagan test had χ2= 57.75, and the Hausman Test had a χ2=8.05. Both of
these tests indicate that a random effects model would be appropriate. However, since
fixed effects is better at dealing with biases that cannot be directly corrected for, results
for both are reported in tables 7 and 8.
Table 7: Random Effects Results
Robust
Coefficient

Z

Rain ex

565.8

0.91

Maize Price ex

2851.9

1.2

North

-5795.7

-0.59

Central

-9799.5

-1.13

D0809

-65459.7

-1.26

*Distributer

18789.64

1.89

D0809*Distributer

-365.8

-0.06

Voucher
Distribution

.074

1.59

FertPrices

4.15

1.13

R2 within

.0569

R2 between

.1289

R2 Overall

.1234

*significant at the 10% level

32

Using fixed effects, the model would simplify to
(8) TSit= θ0+δ00809t + θ1distributerit0809t + θ2Pmz* it, + θ3Pfert it , + θ4 rainfall*it + vit
Table 8: Fixed Effects Results
Robust
Coefficient

Z

Rain ex

-1537.8

-1.29

Maize Price ex

-1465.3

-1.23

D0809

113117.8

1.24

D0809*Distributer

-325.8

-0.05

Voucher
Distribution

.213

1.57

FertPrices

-28.4

-1.24

R2 within

.1238

R2 between

.0615

R2 Overall

.0519

Neither model has very high R2 values. Additionally, in both regressions
combined, only one variable is statistically significant at the 10% level: the coefficient on
distributer in the random effects model. This is not a particularly enlightening response,
as all that it tells us is that distributers sold more fertilizer than agrodealers, which is
completely unsurprising. In neither model are relevant prices statistically significant,
which is very surprising. These results hold up with various modifications to variable
specification and slight changes in the model.
These results do not show any statistically significant negative impact from the
2008/09 exclusion on total sales of fertilizer for either agrodealers or distributers. These

33

results are somewhat surprising given retailers’ highly negative perspectives of the
impact of the policy, however as discussed in section 4.2, increased demand for
agricultural inputs may have at least partially compensated for any negative effects of
the changes in subsidy implementation. They also indicate that commercial sales are
still occurring, sometimes in substantial amounts for distributers. While studies of
demand for fertilizer have found virtually no demand for commercial fertilizer among
small-holder farmers after the subsidy program was implemented (Rickert-Gilbert, etc),
it is possible that distributers are still selling commercial fertilizer to large-holder farmers,
especially for tobacco in 2008/09, compensating for the loss of subsidized sales. Since
there was very little survey attrition by distributers, survey attrition cannot be fully
responsible for these results.
The lack of statistically significant results, especially for prices, and the low R2
values for a fairly standard model suggest that there may be problems within the data,
and may also reflect the inaccurate ways that some of the data points, especially
fertilizer prices, needed to be calculated as a result of insufficient data. The sample size
is also relatively small, creating additional econometric problems.
To investigate these results further, the previously described missing value bias
test was modified to include retailers’ answers to questions about the effect of the policy
change on their business. They were asked whether the policy caused them to have
lower customer traffic, lower profits, lower revenues, and lower sales. When these
variables were added in to the test for selection bias for both distributers and
agrodealers (either looking at only one or a variable representing whether any of the
four were true) the effect was statistically significant (at the 5% level) and negative;

34

retailers who perceived negative impacts on their business were more likely to provide
all the necessary data than those who did not. These results are shown in table 9.
Table 9: Expanded Missing Value Bias Test
dF/dx
Z
Distributor
0.03
0.14
North
-0.099
-0.92
South
0.145
1.45
Agrodealer
0.217
-1.34
Original
case
Distributor
0.083
0.55
original case
Agrodealer
-0.267
-2.15
Negative
experience
**
Distributor
-0.314
-2.35
Negative
experience
**
** statistically significant at the 5% level

p>|z|
0.889
0.358
0.147
0.182

0.58
0.031

0.019

While this bias cannot be directly corrected for in the regression, the implications
of this revised test are important. The results mean that if there is any bias in these
results, it is towards a worse experience by the retailers. Therefore, the total sales
model does not provide statistically significant evidence of decreased sales by
distributors after being excluded from the subsidy program. Again, the discussion
surrounding sales expectations may at least partially explain this lack of statistically
significant results. However, the results are contradictory to previous research and
somewhat surprising, so further investigation would be desirable.

35

4.4 Additional evidence on the independent agrodealer sectors
Unlike distributers, agrodealers were never allowed to participate in the fertilizer
subsidy program, so the 2008/09 shift in policy was less relevant to them, except to
make any competitors who were distributers less competitive, and any competitors who
were government outlets even more competitive. Therefore, to understand how the
exclusion from the subsidy program has affected independent agrodealers, another
framework for evaluating the effects must be found. In addition to the importance of
understanding the effects on this part of the agricultural input sector, this could provide
some insights into how an extension of the exclusion of distributers would affect them.
As discussed earlier, agrodealers were highly significantly less likely to be reinterviewed. Although reports on what happened to the agrodealers who were not reinterviewed is incomplete, at least a large part of this problem was agrodealers who
were no longer in business, rather than retailers who refused to be re-interviewed.
Therefore, understanding the attrition of agrodealers may be a way to gain some
information about the effects of the fertilizer subsidy program on the agrodealers, if
there is sufficient evidence linking attrition among agrodealers and the fertilizer subsidy
program. This section will attempt to informally “test” the hypothesis that the high level
of attrition among agrodealers is a result of their exclusion from the fertilizer subsidy
program.
Since the subsidy program, and the agrodealers’ exclusion from it, started in
2005/06, it would be ideal to be able to look at changes in the agrodealer sector starting
the year before. However, the first survey was conducted in 2006/07, and while it asked
questions about several prior years going back to 2004/05, many retailers were not able

36

to provide that information, often because they had not been in business for that long.
Therefore, while understanding attrition from 2006/07 to 2008/09 is not the ideal way of
assessing agrodealers experience it maximizes the data available and at least gives a
partial picture of agrodealers’ experiences.
While the data cannot actually tell us the causes of attrition, understanding the
characteristics of the shops that were re-interviewed compared to those that were not
can point to some possibilities. Additionally, if those characteristics can be related
directly or indirectly to the fertilizer subsidy program, the preliminary evidence for and
against the hypothesis that the significant agrodealer survey attrition was a result of the
agrodealers’ exclusion from participating in the fertilizer subsidy.
Hypothesis 1: The Exclusion Was Responsible for High Levels of Attrition
There are multiple ways that the fertilizer subsidy could have impacted
agrodealers. The first is direct impact: agrodealers who sold fertilizer would see a
reduction in fertilizer sales, potentially driving them out of business. If in 2006/07 the
agrodealers who were not re-interviewed said that fertilizer sales were more important
to their business than the agrodealers who were re-interviewed, this would provide
some evidence for that claim. However, when analyzing the data from the 2006/07
survey there is no evidence to support this; the difference in importance of fertilizer
sales is not statistically significantly different between the two categories of agrodealers.
Additionally, 64% of the agrodealers who were not re-surveyed did not even stock
fertilizer in 2006/07, whereas 50% of those agrodealers who were re-surveyed carried
fertilizer. This suggests that the agrodealers who were selling fertilizer and closed
between 2006/07 and 2008/09 did not close because of further drops in fertilizer sales,
but since we do not have data on their sales before the program was implemented, the

37

overall effect of the program cannot be determined based on fertilizer sales data alone.
Using a very small sample size, the experiences between 2005/06 and 2006/07 in sales
among agrodealers provide some evidence that the agrodealers who closed were not in
the middle of a large period of decline in fertilizer sales, 35.3% of agrodealers who were
not re-surveyed in 2008/09 said that they experienced a large increase in fertilizer sales
between 2005/06 and 2006/07 and another 17.6% reported a moderate decline.
However, these results are only drawn from a sample size of 16, making them
unreliable for reaching valid conclusions.
One important issue in discussing the possibility of a direct negative impact is
that agrodealers may have had a buffer against those impacts. Agrodealers often sold
more fertilizer in quantities less than the 50kg bags that were used for the subsidy
program. Providing the smaller quantities gave the agrodealers an advantage over
other outlets that only sold 50kg bags, drawing business from farmers who wished to
purchase less fertilizer. In 2006/07, retailers were asked whether sales of less than
10kg, 11-49 kg, or 50kg bags were most important to their business. The results for all
retailers providing that data are shown in table 10.

38

Table 10: Importance of Sales of Small, Intermediate and Large Fertlizer Bags
Business
Structure

Small
Fertilizer
Packs
(less than
10kg)
Distributor
N
4
%
8.9
Government N
1
Structure
%
1.5
Cooperatives N
1
%
5
Agrodealer
N
47
%
58.8
Total
N
53
%
24.9
Source: 2006/07 retailer survey

Intermediate 50 kg
(11-49 kg)
bags of
fertilizer

0
0
0
0
0
0
4
5
4
1.9

41
91.1
67
98.5
19
95
29
36.3
156
73.2

While these results do not give detailed information on who actually sold fertilizer in
quantities less than 50kg bags, they do show that independent agrodealers were the
only retailer type that were more likely to depend on smaller quantities of fertilizer than
the 50kg bags. 63.8% of independent agrodealers had most of their fertilizer sales be in
quantities less than 50kg bags.
A second way that the fertilizer subsidy could have affected agrodealers is
indirectly: potential customers who would have gone to agrodealers to buy fertilizer
without the subsidy went elsewhere both for fertilizer purchases and while buying
fertilizer also bought other goods that they would have bought at the agrodealer. In this
situation, the importance of fertilizer sales is not directly related. Instead, support might
be found for this theory in questions about whether customers who bought fertilizer also
bought other goods. Unfortunately, this information is unavailable. Since participation
in the seed coupon program might at least partially compensate for that effect (because
customers might still come to get seeds even if they weren’t getting fertilizer), any

39

difference in participation between categories of agrodealers would be relevant.
However, as in the original attrition bias testing this variable doesn’t show any
statistically significant differences, so it does not provide evidence either way.
Hypothesis Two: Other Factors Caused the High Levels of Attrition
If the subsidy program cannot be proved to be responsible for agrodealer attrition
there needs to be another plausible explanation for the high levels of attrition among
agrodealers as compared to distributers and government outlets. One compelling
possibility is that independent agrodealers are less stable, and enter and exit the market
with much higher frequency than other types of input retailers. While there is no longterm data on entry and exit of agrodealers, in 2006/07 data was collected on how long
various retailers had been open, and how long they had been selling agricultural inputs.
For distributers, the median length of operation was 6.5 years, and the median length of
input sales was 5 years. For all agrodealers the median length of operation was 5
years, and length of input sales was 3 years. For those agrodealers who were not reinterviewed, length of operation was 4 years and length of input sales was 2 years.
While this data is not conclusive, it does indicate that the agrodealers were on average
open for less long, and selling inputs for a shorter period of time, which might suggest
that they are less stable. However, entry into input sales within that time period may be
partially linked to programs implemented to train agrodealers in selling fertilizer, making
length of input sales a less accurate indicator. This issue does not impact the length of
operation.

Another partial explanation is that the agrodealers in the North and Central

regions who were not re-interviewed were hurt by the fact that AISAM withdrew from
those regions in 2008/09, cutting off a significant source of retail level support.

40

5. Desired data and data collection methods
As discussed in section three, the data collected in the 2009 survey was not ideal
for conducting the econometric analysis that this paper discussed. This is primarily a
result of two factors, the first being that the objectives of both the 2007 and 2009 studies
did not include the ability to conduct these types of analysis, and the second being the
constraints preventing the ability to implement the surveys in ideal ways. While
recognizing these very real constraints, which have already been discussed, this section
will explore the ideal methods for generating the data that would allow for a more
accurate and comprehensive econometric evaluation of the program.
5.1 Sampling Methods
The first set of ideal changes would be to sampling methods. Ideally, districts
would be chosen randomly. However, a modification of this may be more desirable
since some districts have few agricultural input retailers and little data could even be
collected from these districts. Therefore, a list of districts with a certain amount of
agricultural input activity could be generated and the districts chosen randomly from that
list.
To have completely random selection of retailers within each selected region, a
complete sampling frame, including all agrodealers, distributor outlets, cooperatives,
and government outlets would need to be developed. In practice, the costs of doing this
are extremely high, but it would be necessary to conduct the desired econometric
analysis. Getting the data on distributors and government structures should not be
prohibitively difficult; the biggest issue would be getting a complete list of agrodealers.
In some years, AISAM and CNFA have updated lists of their members, which would be

41

a good starting point. However, their lists do include retailers who do not sell fertilizer,
and agrodealers who are not members of either organization would not be included on
those lists, which may generate biases in the data. If the community surveys continue,
these could be used to try to supplement the list of agrodealers. For a more complete
list a very expensive agrodealer census would be necessary. Within the sampling
frame, it may also be desirable to not have complete randomization. For the 2006/07
sampling agrodealers were over-sampled because they are a more heterogeneous
group than distributors, and this practice may be desirable to continue, although exact
numbers should be carefully selected.
In future years, there are a couple of options for how to deal with sampling
replacement cases. One approach would be to randomly replace each unavailable
retailer with a new retailer of the same structure. However, this would not reflect
changes in composition of the retail sector. Therefore, another option would be to
instead compute the relevant proportions of the sector made up by different types of
retailers, and choose the samples to reflect changes in composition. In either case, the
important part is to make sure that this procedure is conducted identically in all cases,
rather than having teams decide for themselves. Doing this would require having a fulltime researcher working in the field. Additionally, the ability to do either of these would
be constrained by the ability to create a complete sampling frame in both the first year
and in all future years. Updating the sampling frame every year is important because of
the frequency of entry and exit of agrodealers, and because of changes in composition
of the market.

42

5.2 Training
One fairly easy change to implement would be having a longer training period for
interviewers. Training was done effectively in 2006/07, so modeling it after that training
would be sufficient. Better training would help interviewers fully understand the
questions in the survey, which would lead to them being more aware of who is
supposed to answer which questions and how to decide whether the person being
interviewed is reliable.
Team leaders also need additional training in how to provide useful summaries
of their experiences during the data collection period. An important piece of this is
training on how to deal with shops that are closed, both during the first year of sampling
and then in future years of follow-up surveys. During the first year, this training would
mostly consist of how to replace a randomly selected retailer. During future years, this
would have two parts. The first would be training in the desired method for replacing
the previously interviewed shop, minimizing sampling bias. The second would be
training team leaders on how to record information about the unavailable retailers, and
how to find out maximum information on why that shop was not re-interviewed. In
2008/09 one team leader provided information about each shop that was not reinterviewed, but since everyone didn’t record that information, those answers cannot be
generalized over the whole sample. The ability to implement either of these methods
would be faced with a time constraint; however this would be the ideal implementation.
A more detailed understanding of the causes of attrition would be highly desirable to
generate accurate econometric results, and as discussed in section four it may also

43

provide an additional way to look at the effect of the fertilizer subsidy. However, in
analyzing the latter, care must be taken because there are other factors affecting
agrodealers, including the generally unstable nature of most agrodealers, which could
cause attrition and are completely unrelated to the subsidy program. Additional data
may be able to help distinguish between causes of attrition.
In addition to improved training methods, budgeting more time in the field for
interviewers would also be helpful. One problem in 2008/09 was that many of the
interviews were with sales clerks instead of owners/managers, which contributed to
inaccuracies in the data given. With more time, interviewers could return on another
day to speak with an owner who was not there during the first attempt.
5.3 Survey Instrument
For the most part, a similar survey instrument to the ones used in 2006/07 and
2008/09 could be used as the baseline for this revised study. Although this could be
combined with questions about the seed coupon program, this section will only discuss
questions about the fertilizer coupon program.
To ameliorate the basic modeling process for sales, the survey needs to consist
at minimum of questions about fertilizer sales history, divided into subsidized and
unsubsidized fertilizer. For distributors and government structures the ideal way of
collecting this would be to get the data from headquarters, where records are usually
most accurate. Asking managers is problematic because they generally do not keep
historical records on site, and because they often have not been at a particular location
long enough to have an understanding of how that location’s sales have changed over
time. It would be useful to have fertilizer sales data divided into different types of

44

fertilizer as well, but the experience in 2008/09 showed that even retailers who can
provide total sales are much less likely to be able to provide that level of detail in the
data. Price data also need to be gathered, in equal categories as for sales data, so that
any sales data can be directly matched with its price data. Because of quantity
discounts there may be some challenges getting comprehensive data on this, but
whatever data can actually be gathered should be, since the sales analysis would be
impossible to conduct without it.
Other data that would be useful would be more subjective questions for the
retailer. As in 2008/09, questions about how the retailer perceived changes between
years would be a useful way of evaluating how opinion reflected the data provided.
Also, any new policy changes should be asked about to assess in detail retailers
perceptions about current subsidy policy. The 2008/09 instrument also included
questions on implementation problems of the subsidy, such as long lines, falsified
coupons, incorrect coupons, corruption, and bribery, which should be retained in the
revised questionnaire.
The 2008/09 questionnaire also asked some questions about business structure
and the importance of various items they stocked. These questions are important and
for agrodealers especially it might be useful for this section to be expanded.
Questions on seed subsidy and sales could also be used in the survey to avoid
duplication of surveying efforts, but it is important to make sure that a sufficient sample
of the retailers actually sells fertilizer or has sold fertilizer in the past.
In terms of creating an effective panel, the instruments and methods need to be
consistent over years. One problematic aspect of the panel using two surveys was that

45

the method for collecting sales data was inconsistent. In 2006/07, agrodealers were
asked to report fertilizer sales at the time of the interview, and for distributor outlets the
sales data was collected from distributor headquarters. In 2008/09 however, all private
sector retailers were asked to provide sales data at the time of collection. As discussed
earlier, in terms of historical sales data, there was no correlation between sales reported
in 2006/07 and 2008/09 for the same years. This problem may in part be due to the
altered methods of sales data collection.
.

Policy-specific questions can be added, subtracted, or modified between years if

there have been specific policy changes that make those questions irrelevant or
inappropriate. However, consistency in wording of questions about policies would be
desirable to be able to compare results from different policies in as accurate a way as
possible. Additional questions can also be added, based on questions raised by past
research or other sources, but subtracting questions, even if they don’t seem useful at
the time, should be minimized because their value may become apparent at a future
time, and having as many data points as possible will be important.

5.4 Implementation
One important part of implementing the study is figuring out methods to minimize
attrition. In 2008/09, at least some portion of the attrition was due to retailers being
unwilling to be re-surveyed, due in part to anger over government policies, and in part
reluctance to answer questions because they were unclear on who was conducting the
survey and who it was for, which was intended to be made clear. While retailers cannot
be forced to participate, there may be ways of maintaining a positive relationship with
retailers and effectively communicating the benefits of participation. Another way of

46

minimizing attrition is to avoid surveying at times that seasonal retailers are unavailable
for surveying. Surveying at regular intervals, and avoiding surveying at time periods
when seasonal shops would be closed can achieve this. In practice, doing this may be
costly but minimizing attrition is a worthwhile goal and even small steps in this direction
could improve the results of the study. The survey instrument could be modified to
include exactly what periods of the year the shop is open, so that interviewers can
prioritize going to those shops during periods when they would actually be available.
This method is not perfect, as this is variable for some shops, depending on rains and
supplies, but it would still be an improvement.

47

BIBLIOGRAPHY

48

BIBLIOGRAPHY
Dorward, A., C. Chirwa, D. Boughton, E. Crawford, T.S. Jayne, R. Slater, V. Kelly and
M.Tsoka (2008). Toward ‘smart’ subsidies in agriculture? Lessons from recent
experience in Malawi. Natural Resource Perspectives 116.
Dorward, A., C. Chirwa, V. Kelly, T.S. Jayne, R. Slater, and D. Boughton. (2008).
Evaluation of the 2006/7 Agricultural Input Subsidy Programme, Malawi. Final
Report. Lilongwe, Malawi.

Dugger, C. (2007, December 2). Ending Famine, Simply by Ignoring the Experts. The
New York Times.

IFPRI. (July 2009). Fertilizer Subsidies in Africa: Are Vouchers the Answer? IFPRI
Issue Brief 6.

Kelly, V.A., D. Boughton and N. Lenski. (2010). Malawi Agricultural Inputs
Subsidy Program: Evaluation of the 2007/08 and 2008/09 Input Supply Sector
Analysis.

Logistics Unit. (2009). Final report, implementation of agricultural inputs subsidy
program 2008/09. Lilongwe, Malawi.

Minde, I. J., and P. Ndlovu. (2007). How to make agricultural subsidies smart. Paper
Prepared for Contribution to the Ministerial Seminar. Regional Strategic Analysis
and Knowledge Support System for Southern Africa (ReSAKSS-SA), Maputo,
Mozambique, 8 June 2007. Mimeo.

Minde, Isaac, T.S. Jayne, Eric Crawford, Joshua Ariga, and Jones Govereh. (2008).
Promoting Fertilizer Use in Africa: Current Issues and Empirical Evidence from
Malawi, Zambia, and Kenya. ReSAKSS Working Paper 13.

49

Morris, M., V. A. Kelly, R. Kopicki, and D. Byerlee. (2007). Promoting increased
fertilizer use in Africa: Lessons learned and good practice guidelines.
Washington, D.C.: World Bank.

Ricker-Gilbert, Jacob and T.S. Jayne. (2009). Do Fertilizer Subsidies Affect the
Demand for commercial Fertilizer? An Example from Malawi. International
Association of Agricultural Economists Conference. August 2009.

School of Oriental & African Studies, Wadonda Consults, and Michigan State University
(2008). Evaluation of the 2006/7 Agricultural Input Supply Programme, Malawi
[PowerPoint slides]. Retrieved from
http://aec.msu.edu/fs2/inputs/power_points/MalawiInputSubsidyFinalMoAMarch2
008PresentationRev.pdf

World Bank, n.d.. Malawi, Fertilizer Subsidies and the World Bank. Accessed
September 21, 2009:
http://web.worldbank.org/WBSITE/EXTERNAL/COUNTRIES/AFRICAEXT/
MALAWIEXTN/0,,contentMDK:21575335~pagePK:141137~piPK:141127~
theSitePK:355870,00.html#3

Xu, Z., W.J. Burke, T.S. Jayne and J. Govereh. (2009). Do Input Subsidy
Programs “Crowd In” or “Crowd Out” Commercial Market Development?
Modeling Fertilizer Use Decisions in a Two Channel Marketing System.
Agricultural Economics, 40 (1), 79-94.

Yawson, David, Frederick Armah, Ernest Afrifa, and Samuel Dadzie. (2010). Ghana’s
Fertilizer Subsidy Policy: Early Field Lessons from Farmers in the Central
Region. Journal of Sustainable Development in Africa, 12 (3).

50