Following Mutharika’s departure from the UDF in 2005, the president engaged in an aggressive campaign to attract support across ethnic and regional divisions, an effort that was highly successful as indicated by the broad-based electoral support he received in the 2009 election in which he defeated the second-place candidate, John Tembo, by a margin of 66% to 31%. The shifting political landscape, however, meant that traditional ethnoregional alignments were temporarily upended.8 Second, Malawian parties lack the local level networks that allow machine-based parties in other parts of the world to identify core and swing voters at the village or neighborhood level. Political parties in Malawi function primarily at the national level with weak or nonexistent formal structures at the local level . This was particularly true of Mutharika’s DPP, which, having been formed in the aftermath of the president’s 2005 departure from UDF, initially lacked even basic local-level infrastructure. While individual politicians, such as parliamentarians, no doubt maintained their own informal networks at the local level, the challenge of distinguishing partisans from non-partisans based on observable markers would have been formidable. Given these informational limitations we expect the subsidy program to be largely untargeted with respect to ethnicity and party support at the individual level within villages . The analysis of subsidy targeting and its political effects employs data from two waves of a panel survey of rural Malawians in 122 villages clustered in three districts: Rumphi, Mchinji, and Balaka. Each administrative region of Malawi is represented in the study, as are the three major ethnolinguistic groups .
The surveys are part of the Malawi Longitudinal Study of Families and Health ,danish trolley which began in 1998 and aims to understand how villagers cope with health challenges like HIV/AIDS. In each district, the MLSFH used a cluster sampling strategy across selected census enumeration areas. A random one-in-four sample of women of reproductive age and their husbands was drawn from villages to yield a target sample in 1998 of 1,500 women and their husbands . The resulting sample in 1998 included 1,532 evermarried women aged 15-49 and 1,065 of their spouses. In 2004, the MLSFH added a sample of 984 adolescents aged 15-24, and during the 2008 round a sample of 549 parents of respondents in earlier MLSFH rounds was added . New spouses of MLSFH respondents were also added in each wave. Though the original sampling strategy in 1998 was not designed to be representative of the rural population in Malawi, the sample’s characteristics are very similar to those of the rural population interviewed by the Malawi Demographic and Health Surveys that covered nationally representative samples . Between the two rounds studied here, 1,016 respondents were lost to follow-up;9 thus, though the 2008 round included a sample of 3,909 respondents and the 2010 sample included 3,786 respondents, our analysis includes only those 2,851 respondents who were interviewed in both 2008 and 2010. The analytical sample in this paper drops to 1,846 when we remove study participants whose responses on standard questions are inconsistent between the 2008 and 2010 waves and when responses are dropped because of missing information on key variables.We augment the individual-level panel data with village-level data collected through a survey of village headmen in the 122 research villages conducted in 2008.The first step in the analysis is to test for evidence of targeting.
While the results from this analysis are interesting in their own right, the primary goal is to identify factors that might confound the analysis in the next section of the subsidy’s effects on voters’ political preferences. To examine targeting in our survey area, we draw on a question on the 2010 survey that asked respondents whether they had received a voucher for fertilizer or seeds in each of the previous two years. We focus on those who received the subsidy in the 2009/10 growing season, which immediately preceded the second wave of the survey. The data show that 73.7% of respondents received the agriculture subsidy in 2009/10. Respondents in the three district clusters were about as likely to receive the subsidy, with 74.6% receiving it in Rumphi, 78% in Mchinji, and 68.2% in Balaka. The independent variables for the targeting analysis come from the 2008 survey unless otherwise specified. Our key independent variable measures party support according to responses to a question that asked, “Do you feel close to any particular political party?” Those who answered affirmatively were then asked which party. In total, 53.2% of the sample registered support for a party, with the largest share expressing support for the incumbent party, DPP, and smaller shares indicating support for one of the opposition parties . The distribution of party support in our survey area mirrors national-level trends found in the 2008 Afrobarometer survey.We also include a measure of whether respondents are “minority partisans” in their villages, supporting a party other than the party thought to be supported by most people in the village. We test for targeting along ethnic lines, given the centrality of ethnic divisions and ethnoregional favoritism in Malawi’s political history . We include a dummy variable for Lomwe respondents and also include dummies for Malawi’s other major ethnic communities, the Tumbuka, Chewa, and Yao. In our survey area, these four groups make up 93.9% of the sample. We also include an indicator of whether respondents come from minority groups within their villages, using data from the headmen survey on the majority ethnic group at the village-level.
Following Pan and Christiaensen , who found that local elites in Tanzania tended to capture the benefits of a similar subsidy program, the analysis includes several measures of social stature to test whether those in leadership positions within their communities may have been more likely to benefit. We include indicator variables measuring whether respondents were members of the Village Development Committee, the Chief’s Council, or the District Development Committee in 2008. To test whether the program benefitted the most needy, we include multiple measures of socio-economic status. First, we include a measure of wealth constructed as an index of household asset ownership.We include three measures of whether respondents experienced negative economic shocks in the year prior to the 2009/10 growing season. These relate to: 1) loss of income, 2) poor crop yields, or 3) the death or serious illness of an adult member of the household. Though only the second measure explicitly references an agriculturally related loss, all three measures capture severe shocks that make households particularly vulnerable to food insecurity. We also include a measure of farm size and standard demographic measures: age, education, and whether the household was headed by a female. We use logistic regression to examine individual-level subsidy targeting,vertical aeroponic tower garden and include village fixed effects to account for village-level differences that might affect access to coupons, including politically-motivated targeting across villages. We cluster standard errors by household because in some cases multiple respondents were interviewed in the same household. Table 1 shows the results and Figure 1 plots the marginal effects of each variable holding other covariates at their mean values. Figure 1 indicates that, conditional on village, the subsidy was not targeted with respect to party preferences or ethnicity in our survey area. Supporters of the incumbent party in 2008 were no more or less likely to benefit from the subsidy program in 2009/10 . Likewise, opposition party supporters were no less likely to receive the benefit, nor were individuals who were minority partisans in their villages. The results also show that, relative to members of smaller ethnic groups , respondents from Malawi’s major ethnic communities were no more or less likely to receive the subsidy during the 2009/10 season. We also entered each ethnic dummy variable individually in additional tests and found no evidence of ethnic targeting in these specifications. Likewise, we find no evidence of discrimination against individuals from minority groups at the village level. Consistent with findings by other scholars , we fail to find evidence that Malawi’s AISP effectively targeted those with greatest need, despite the program’s stated goal of reaching those most at risk for food insecurity. Our results show that neither poorer respondents, nor those with smaller land holdings were more likely to receive the subsidy. Moreover, respondents who experienced an economic shock related to loss of crops, livestock or income in the previous year were no more likely to benefit from the AISP than those unaffected by such income shocks.
Further, households that suffered the death or serious illness of an adult in the previous year were actually less likely to receive the input subsidy. Together, these results suggest the program did not successfully target those with greatest need – poor, smallholder farmers threatened by food insecurity. The results on demographic factors show that age and education were unrelated to receiving the subsidy. We find negative and significant gender effects. In our survey area, female headed households were 10.4% less likely to receive the subsidy. Given that the models control for a wide range of factors that might affect levels of need, social stature, and economic shocks, the finding reported here suggests that female heads of households were less likely to benefit from the program as a result of gender discrimination rather than other factors that might be correlated with gender.Finally, the results indicate that members of district and village development committees or chief’s councils were not statistically significantly more likely to benefit from the program. To answer this question we examine trends in party support among recipients and non-recipients across the two survey rounds. Our measure of political preferences – taken from questions on the 2008 and 2010 surveys that asked respondents whether they “feel close” to any party – sets a high threshold for finding a positive effect. Similar studies typically use measures of vote choice , which are likely to be more fluid and potentially subject to influence by anti-poverty programs. Our measure probes deeper connections between voters and parties, and is therefore less likely to be influenced by short term changes in government policy.Moreover, Zucco’s study of the political effects of a conditional cash transfer program in Brazil found evidence of an effect only on vote choice and not on partisanship. Thus, the measure of party support available in our survey data in all likelihood biases against finding a positive effect of the subsidy on political preferences. A second challenge relates to the nature of the treatment effect we seek to estimate. Sociotropic theories of economic voting have found that voters may punish and reward incumbents based on the overall performance of the economy . If voters in Malawi base assessments of the incumbent on aggregate outcomes – rather then their own personal welfare – the subsidy program may represent a treatment that was received by all Malawians. Studies have shown that the AISP contributed to reduced food prices in Malawi, indicating that Malawians may have benefited indirectly even if they did not receive subsidy coupons . These factors also bias the analysis against finding a connection between individual measures of subsidy reception and party support. Despite these challenges, the data suggest that the subsidy did affect political preferences. Table 2 compares the increase in support for the DPP between 2008 and 2010 among those who did and did not receive the 2009/10 subsidy. Among non-recipients, the percent expressing support for the DPP rose by 3.4%, from 34.8% to 38.2%, while for those who did receive the subsidy, the increase was approximately three times larger: 9%, from 35.3% to 44.3%. Thus, receiving the subsidy in 2009/10 is associated with a 5.6% increase in DPP support. While these results are suggestive of a treatment effect, there is, of course, the possibility that subsidy recipients differed from non-recipients in some important ways, and these underlying differences – not the subsidy – account for the greater increase in DPP support among recipients, relative to non-recipients. We are reassured by the results from the previous section, which showed that the 2009/10 subsidy was largely untargeted in our survey area, particularly with regard to prior party support and the primary demographic factor – ethnicity – that has traditionally been associated with party support in Malawi. Nonetheless, differences might remain that could confound the treatment estimation.