«WORKING DRAFT, DO NOT CITE OR CIRCULATE WITHOUT PERMISSION Abstract Can government programs that fail to deliver still inﬂuence citizen behavior? ...»
Do Good Intentions Matter? Experimental Evidence on how
Citizens Respond to Promises of Government Service Delivery∗
Ali Cheema† Asim I. Khwaja‡ Farooq Naseer§ Jacob N. Shapiro
This version April 8, 2014.
WORKING DRAFT, DO NOT CITE OR CIRCULATE WITHOUT PERMISSION
Can government programs that fail to deliver still inﬂuence citizen behavior? Large literatures in political science and economics study the eﬀect of various government programs on how citizens engage with the state. A tacit assumption in many of these papers is that citizens value government programs proportionally to the amount of money spent. Yet there is tremendous heterogeneity in the mapping between spending and how much value citizens actually get from a program. And even programs that do not work may still reﬂect substantial government investments, thereby informing citizens’ beliefs about how much weight the government places on their welfare. Using a large-scale randomized evaluation of a vocational training program in southern Punjab we provide evidence that good intentions might matter; citizens oﬀered a program that almost no one used voted for the ruling party at higher rates in subsequent elections if oﬀered multiple training vouchers than if only oﬀered one. Men who received the training oﬀer became more socially engaged and used government services at higher rates. Women had the opposite reaction. These results have implications for theories of civic engagement.
∗ We thank Arqam Lodhi and Sameem Siddiqi for leading the PEOP team at CERP. Minahil Asim, Mathilde Emeriau, Jonathan Furszyfer, Kunal Mangal, Sahaab Sheikh, Osman Siddiqi, Anam Sohaib, and Gabriel Tourek all provided critical research assistance at various stages of the project. Sarah Khan, Kunaal Sharma, Macartan Humphreys, and the members of the 2013 Columbia University Seminar on the Study of Development Strategies provided excellent feedback. This work was supported by grants from the UK Department for International Development - Pakistan, the IZA/DFID Growth and Labour Markets in Low Income Countries Programme (GLM — LIC), and the Institute of Development and Economic Alternatives (IDEAS).
† Senior Research Fellow Institute of Development and Economic Alternatives, Associate Professor of Economics LUMS, and Director Center for Economic Research in Pakistan; Email: email@example.com.
‡ Professor, John F. Kennedy School of Public Policy, Harvard University; E-mail: firstname.lastname@example.org.
§ Assistant Professor, Lahore University of Management Sciences; E-mail: email@example.com.
Assistant Professor, Woodrow Wilson School and Department of Politics, Princeton University; Email:
firstname.lastname@example.org. Corresponding author.
Introduction Can government programs that fail to deliver value still inﬂuence citizen behavior? Large literatures in political science and economics study the eﬀect of various government programs on how citizens engage with the state.1 A tacit assumption in many of these papers is that citizens value government programs proportionally to the amount of money spent.2 Yet there is tremendous heterogeneity in the mapping between spending and how much value citizens actually get from a program.And even programs that do not work may still reﬂect substantial government investments, and thereby inform citizens’ beliefs about how much weight the government places on their welfare. From a policy perspective evidence that citizens reward eﬀort even when it does not yield tangible beneﬁts would imply that government oﬃcials should be willing to take risks and innovate as even wellintentioned failures could win votes.
The theoretical literature is ambiguous on whether good intentions should matter. Classic arguments about retrospective voting and turnout suggest that citizens incentivize politicians by rewarding good performance or honesty and turnout to vote when they believe they can inﬂuence the ultimate election outcome (Downs, 1957; Ferejohn, 1986; Fearon, 1999). For these mechanisms to work a number of supporting conditions must be in place: there must be meaningful competition among politicians; citizens’ judgments must not be too badly clouded by irrelevant factors (Healy, Malhotra and Mo, 2010); and people must be able to learn about politician’s competence by observing them in oﬃce.3 Much of this work presumes that citizens learn about government competence through observing policies and prefer more competent candidates over less, but still does not resolve the paradox of voting. Myatt (2012) provides a potential solution to the paradox in a model where turnout emerges in equilibrium due to aggregate uncertainty about the popularity of each candidate plus citizens’ beliefs about the long-run value of having their preferred candidate elected. In a setting like Pakistan where making good policy is hard and opportunities for corruption manifest, the Recent contributions which include good literature reviews include Healy, Malhotra and Mo (2010); Finan and Schechter (2012); Gallego (2012).
In papers which include a linear term in the value of spending this assumption is explicit (see e.g. Levitt and Snyder, 1997). In papers exploiting discontinuities in spending the implied functional form assumption is much weaker (see e.g. Litschig and Morrison, 2011).
On this last point Meirowitz and Tucker (2013) show that when citizens learn about both government competence and the nature of the problem environment by observing past performance then repeated poor performance may reduce participation as citizens realize that no government can help much in their environment.
lifetime value to citizens of any given politician has to be a function of both their competence at responding to shocks (the parameter that varies over agent types in most principal-agent models of voting) and the value they put on citizens’ welfare when making policy decisions. In such settings it seems possible that policies which fail but reveal good intentions would inﬂuence beliefs about how much politicians value serving their constituents, and therefore move voting behavior.
To understand whether good intentions can matter in a developing country context we exploit a large-scale ﬁeld experiment oﬀering vouchers for vocational training to a representative sample of households in four of the poorest districts of Punjab, Pakistan. The experiment was part of the Skills for Employability (SFE) program run by the Punjab Skills Development Fund (PSDF). After a large-scale baseline household survey vouchers for vocational training were oﬀered to 973 individuals living in households that had expressed previously expressed interest in having a household member acquire greater skills; a standard encouragement design. Households were randomly assigned to slightly diﬀerent voucher oﬀers. Some received a voucher for a male trainee, some for a female trainee, and some received two vouchers (one male and one female). The vouchers were ﬂexible, they could be transferred between household members within a gender, allowed them to enroll in any of the courses being oﬀered by PSDF, and came with a stipend of that varied by course and training provider, but that averaged around 900 rs/month, about 33% of the mean monthly household expenditures per capita in our sample.
Even though this population systematically overestimated the returns to acquiring skills almost no one took advantage of the training.4 Of the 973 individuals oﬀered vouchers only 43% accepted a voucher, 70% of those who accepted a voucher failed to enroll in their chose course, and a fair number of those who did dropped out before the course was ﬁnished. Ultimately of 973 people oﬀered a training voucher only 46 completed training.
Despite the exceptionally low uptake for this program, there is some evidence that it had an eﬀect on voting, at least among men. While receiving the voucher oﬀer had little impact on turnout or vote-choice, living in a household oﬀered two-vouchers had a clear impact. Men who lived in households oﬀered two vouchers were 14 percentage points more likely to report voting for the ruling party in Punjab, the Pakistan Muslim League Nawaz (PML-N), in the May 2013 National The 359 women in our baseline sample who wanted to acquire tailoring skills, for example, estimated they would make between 2,775 rs/month and 4,834 rs/month on average if they had better good skills in that area, yet female tailors in our sample made only 1,415 rs/month on average.
Assembly election than control households and 12 percentage points more likely to report voting for the PML-N candidate than those living in households oﬀered only one voucher. Women in dual-voucher households were 10 percentage points more likely to report voting PML-N than those in single-voucher households.
While there was no major diﬀerence in attitudinal measures of faith in government across the treatment conditions, the voucher oﬀer does appear to have had gender-speciﬁc eﬀects on a range of more behavioral measures of civic engagement, however. Women systematically became less engaged; reporting lower rates of pro-social community behavior, less charitable giving, and less use of government services. Men systematically became more engaged; reporting higher rates of pro-social community behavior, more charitable giving, and more use of government services. We suspect this diﬀerence across genders may be related to diﬀerences in the value of the oﬀered menu of courses to each gender—the male course menu had a number of relatively high-prestige options while the female course oﬀerings were limited largely to various forms of tailoring and handicrafts—but cannot yet test that hypothesis.
The remainder of this paper proceeds as follows. Section 1 reviews previous work relevant to this piece. Section 2 provides background on the institutional context. Section 3 describes our research design. Section 4 presents the results. Section 5 summarizes our results and discusses implications for research.
1 Previous Work There are three relevant literatures for this research. The ﬁrst is the theoretical literature on why people turn out to vote at all. The second is the empirical literature on turnout and vote choice.
The third is the literature on the impact of vocational training programs.
The question of why people vote at all given the vanishingly small probability any individual has of inﬂuencing an election is one of the deep enduring questions in political economy (Downs, 1957; Fiorina, 1989; Good and Meyer, 1975; Edlin, Gelman and Kaplan, 2007). Classic arguments frame the situation as a principal-agent problem in which voters seek to select a principal who can eﬀectively deal with unexpected events (typically modeled as random shocks to which the agent can respond) (see e.g. Ferejohn, 1986; Fearon, 1999). Much of this work presumes that citizens learn about government competence through observing policies, prefer more competent candidates over less, and ﬁnd turning out worthwhile when they can get a better candidate by doing so. In more a recent contribution Myatt (2012) provides a model where turnout emerges in equilibrium due to aggregate uncertainty about the popularity of each candidate plus citizens’ beliefs about the long-run value of having a given candidate be elected. In his model turnout is higher when citizens expect the long-run value of their preferred candidate to be higher.
The empirical literature on turnout and vote choice has two strands, experiment and observational. On the experimental side most of the work has been on understanding how to get citizens in developed countries to vote at higher rates. A broad range of studies explore the impact of Get-Out-the-Vote (GOTV) campaigns in the United States (Gerber and Green, 2000b,a; Green, Gerber and Nickerson, 2003). While these studies typically ﬁnd that face-to-face canvassing stimulates political participation, recent work by Michelson and Nickerson (2011, 234) points out that such experiments might have limited external validity for a number of reasons. Recent work in developing countries has found similar results from canvassing and voter education programs.5 On the observational side recent work has provided insights into what drives voting behavior.
Fujiwara, Meng and Vogl (2013) use county-level data on U.S. presidential elections from 1952to show that past shocks to the costs of voting (election day rainfall) reduce future voting.
Looking at the relationship between distributive beneﬁts and political participation Chen (2012) shows there is a substantial political response to disaster-related spending in Florida, with aid distributions driving up vote share among voters identifying with the governor’s party and providing an incumbency advantage to all politicians. Fair et al. (2013) show that the 2010-11 ﬂoods in Pakistan led to a robust and sizable increase in voter turnout as well as changes in citizens’ behavior and attitudes which reﬂect greater civic engagement.
Of course in order to understand the political impact of government programs, a key question is whether it matters if the programs actually work. Since Vocational Training Programs (VTP) are targeted to very speciﬁc individuals one might expect participants would reward the government’s spending on such programs, particularly if they improve economic outcomes. The evidence on that score, however, is quite mixed.
There is modest evidence that VTP can increase labor skills and employability in the U.S., See, for example, Guan and Green (2006) on China and Gine and Mansuri (2010) on Pakistan.
though their impact is modest and may be transient once selection is taken into account.6 Evaluations in developing countries on VTPs and classroom programs report positive results on employment, wages, and empowerment, but mostly for women. Attanasio, Kugler and Meghir (2011) study the Juvenes en Accion (JA) program in Colombia which provided three months of classroom training and three months of on-the-job training to youths in the two lowest socio-economic strata.