|Title||Voluntary Audits: Experimental Evidence on a Novel Approach to Bureaucratic Monitoring|
|C1 Background and Explanation of Rationale||
Is it possible to motivate bureaucrats to be honest and exert effort in their work without relying on punitive auditing systems? While state audit institutions are advocated as a solution to waste and corruption, especially in developing countries (Dye and Stapenhurst 1998), they have a mixed record of success. Traditional audits may fail to elicit greater bureaucratic effort (e.g., Litschig and Zamboni 2015) or may displace corruption as officials shift spending away from audited expenses (e.g., Olken 2007). Even where successful, they rely on the strength and autonomy of institutions like the media and judiciary (e.g., Ferraz and Finan 2008), which are often weak in many lower and middle income countries (Helmke and Rios-Figueroa 2011). Other interventions to improve bureaucratic performance employ financial and material incentives (e.g., Duflo, Hanna, and Ryan 2012), but these interventions rarely engage with existing oversight institutions and typically require an ongoing commitment of funds.
We hypothesize that voluntary audits might be an effective tool for eliciting effort. Voluntary audits give street-level bureaucrats a choice and opportunity for self-direction, which existing scholarship suggests can increase intrinsic motivation (Ryan and Deci 2000). Separately, voluntary audits could provide bureaucrats with professional and social recognition, which scholars suggest matters for their performance (Pepinsky, Pierskalla, and Sacks 2017; Wilson 1989). If the opportunity to volunteer for an audit increases either intrinsic or extrinsic motivation, then we expect bureaucratic effort to increase and, potentially, outcomes to improve, as well. Our project aims to contribute to the small but growing literature that explores the role of non-financial incentives on pro-social activities and bureaucratic performance and recruitment (Ashraf, Bandiera, and Jack, 2014; Banuri and Keefer 2016; Dal Bó, Finan, and Rossi 2013).
We partner with the Provincial Auditing Office (PAO) in Chaco, Argentina, to test whether a new system of voluntary audits improves the performance of school principals in their administration of a free meal program. Our field experiment intervenes in the oversight of a school meal program, in which principals are responsible for school-level implementation (including menu planning and overseeing food preparation). We randomly assign principals to receive an invitation to a voluntary audit or to be part of the control group, which is exposed to the status quo situation, where schools face a small (but non-zero) likelihood of a mandatory audit.
|C2 What are the hypotheses to be tested?||
We expect that principals in the treatment group will exert more effort in the implementation of the school meal program compared to the control group. To evaluate this hypothesis, we will measure the following outcomes collected directly from the endline survey of school principals: total number of hours worked, number of “extra” hours worked, and number of hours school principals report devoting to the meal program. We will also check whether any increase in hours comes at expense of time spent in other activities, as well as an index of intrinsic and extrinsic motivation, and the extent to which principals consider the opinion of the provincial auditing office to be important to them. If the intervention changes school principals’ behavior and/or motivation, we expect the school meal program to improve. Our school-level outcomes include student attendance, which interviews with local experts suggest is linked to the number and quality of meals served, the proportion of days schools offer snack/lunch in a given reporting period, basic composition of the meals (for example, number of unique menus in a given reporting period, and number of meals that contain fresh fruits or vegetables), and number of days a school is closed, because unanticipated school closings may reflect lower principal effort to provide meals. Finally, we will explore if treatment interacts with baseline characteristics, including how embedded a school principal is in the school’s community, parental engagement in the school, and socio-economic composition of the school community.
The study will allow us to describe the profile of school principals who accept a voluntary audit, and draw causal inferences on whether (1) voluntary audits generate a direct effect on school principals who receive an invitation to participate in a voluntary audit and, (2) whether the voluntary system generates spillover effects. Our sample includes 188 schools in 13 regions in the province of Chaco. To account for potential spillovers, we employ a two-level randomization. First, we assign the 13 regions to either a high density or low density treatment group. Second, within the high density group, we assign schools to treatment with a probability of 0.5; within the low density group, we assign schools to treatment with a probability of 0.16.
|C3 How will these hypotheses be tested? *||
Because randomization was blocked by region and probability of treatment varies by block, we will estimate the intent-to-treat effect by specifying a regression where the independent variables are a dummy variable measuring assignment to treatment and a set of dummy variables to account for region fixed effects. Observations will be weighted by the inverse of the probability of treatment. We will conduct two tailed hypotheses tests. To increase the precision of our statistical inferences, we will add baseline covariates to our specification. Following Annan, Boyer, Cooper, Heise, and Levy Paluck (2019), we will use an adaptive lasso to identify prognostic covariates to be included in our specification. We will present our results with and without baseline covariates.
For the hypothesis about student attendance, since we will measure this outcome weekly in the period of June to October 2018, we will estimate the effects over time as specified in the pre-analysis plan.
To detect spillovers, we will compare the means of schools assigned to control in regions with high density versus schools assigned to control in regions with low density. Then, we will specify a regression where the independent variables are: a dummy variable measuring assignment to treatment, a dummy variable measuring high versus low density regions, the interaction of the two previous dummies, as well as a set of dummy variables to account for region fixed effects. Observations will be weighted by the inverse of the probability of treatment.
|C5 Scale (# of Units)||188 schools in 13 regions in the province of Chaco, Argentina. This convenience sample includes preschool and primary schools in urban and semi-urban areas. Our sample size reflects the PAO’s capacity to conduct audits.|
|C6 Was a power analysis conducted prior to data collection?||Yes|
|C7 Has this research received Insitutional Review Board (IRB) or ethics committee approval?||Yes|
|C8 IRB Number||Yale: IRB Protocol ID 2000021005; Brown: IRB Protocol ID 1704001741|
|C9 Date of IRB Approval||Yale: June 28, 2017; Brown: May 3, 2017|
|C10 Will the intervention be implemented by the researcher or a third party?||Provincial Auditing Office, Chaco|
|C11 Did any of the research team receive remuneration from the implementing agency for taking part in this research?||No|
|C12 If relevant, is there an advance agreement with the implementation group that all results can be published?||Yes|
|C13 JEL Classification(s)||H41, H75, H77, I28, O31, O35, O38|