An RCT said the reform worked. What did the voters say?
At its best, democracy gives voters the chance to hold politicians accountable for the public services they want. But in many of the democracies of the developing world, public services are persistently really bad. Why? Is it because patronage interests scuttle reform? Are poor voters unable to perceive or unwilling to reward school quality? Perhaps politicians just don’t know how to improve services? These are vital questions about how democracy works. But getting credible empirical answers is tough because usually neither quantity nor quality of public services are randomly assigned.
My job market paper uses a randomized Liberian school reform in an election year to answer these questions. The reform was a pilot test of a public-private partnership in primary schools, called Partnership Schools for Liberia (PSL). Because the PSL schools were randomly selected from a set of eligible government schools, it’s possible to measure PSL’s effect on school quality. The program worked better in some schools than others, but on average it raised test scores and teacher attendance significantly (as my previous paper in the AER with Mauricio Romero and Justin Sandefur shows).
Improved schools came at the price of alienating important political brokers
So far, so good — researchers might call this policy a success. But would politicians? This depends on whether the policy gets votes. Because it was randomized, I can measure how this school reform affected electoral outcomes in the subsequent election.
Surprisingly, the school reform caused a 10% reduction in vote share for the responsible presidential candidate. Most models of political agency would have predicted the opposite. Why was a politician punished for a policy which improved test scores?
One possible answer is that the policy did not only improve test scores — it also angered teacher unions. In many countries, civil servant teachers play a key role in mobilizing voters. The National Teacher Association of Liberia stridently opposed the reform. Did this opposition translate into an electoral impact? To measure this, I surveyed teachers in treatment and control schools. The school reform sharply diminished teachers’ support for the incumbent government’s candidate — especially among union members, where voting intentions for the candidate fell by 25 percentage points. The reform also reduced (by nearly a third) teachers’ participation in political activities typically associated with political patronage, like staffing polling booths and campaigning for candidates. This suggests union opposition had a hand in the negative average electoral effect of the policy.
Voters do perceive — and reward — school improvements
But does this mean voters are blind to school quality? Not at all. Thanks to the experiment’s unique setup, I can compare voter responses in places where the policy worked well against places where it worked poorly. Before randomization, schools were sorted into pairs of schools with similar geography and resources. A treated school was then randomly selected within each pair. This means that each treatment school has an experimentally valid comparison school. This is a very unusual situation even among RCTs, and it allows me to test whether voters reacted differently to big vs. small school improvements. (It should be noted that while each of these pair-level treatment effects are internally valid, the distribution of treatment effects is not random, so this analysis involves stepping away from experiment land.)
Figure 1 shows that the reform was a big vote-getter where it worked well — and a vote-loser where it didn’t. In school pairs where reform had a big, positive effect on test scores (above about 0.5 sd) it also caused voters to favor the responsible candidate. This local positive electoral effect more than counteracted the negative level effect of teacher opposition. However, in places where treatment’s effect on test scores was negative, voters punished the responsible party. This shows that voters do notice changes in school quality, and reward or punish them accordingly.
It may be surprising that parents in Liberia are able to perceive school quality. After all, even better-educated parents in the developed world lack full information about school quality. But one of the strongest predictors of test score increases in Liberia was teacher attendance increases. At least in this context, parents can infer a lot about test scores through easily observable inputs such as teacher attendance.
Voters and candidates are well-informed
Prior to the election, I suspected that poor information might hamper accountability for services. I conducted two additional surveys: one of over 600 legislative candidates, and the other of over 400 household members of schoolchildren. I embedded information experiments in the surveys to measure the effect of policy evidence on credit-claiming and support. But huge majorities of both candidates and households already knew what the evidence showed: PSL increased children’s learning. The information therefore had no effect on support for the policy or for candidates. Furthermore, nearly all households in the survey who attributed credit (or blame) for the reform did so correctly, leaving little scope for credit-claiming — and candidates in the candidate survey also expected their voters to attribute credit correctly. This survey evidence of a well-informed electorate complements the electoral results I mentioned above. Indeed, the electoral impacts of the reform only appear for the presidential candidate whose administration created the policy, and not for any other level of government. Voters cared about this policy, and rewarded or punished the correct politician depending on how well the policy worked for them.
Conclusion: effective policies can empower politicians who want to appeal to voters by improving services
Reform often brings incremental improvements to a broad swath of society while harming a narrow subset, who fight against the reform. This happens even in venerable democracies, but we might especially worry about it in a poorly-educated young democracy like Liberia. My paper shows that this concern is valid: a school reform which raised test scores also galvanized opposition from teacher unions, making it an electoral liability on net. But I also show that voters were well-informed and sophisticated. Where the reform dramatically improved school quality, voters rewarded the government, which more than counteracted the opposition of teacher unions. This shows that an important piece of democratic accountability is the salience of service improvements. If research can play the role of identifying which policies are likeliest to work well, that knowledge may empower politicians who want to bypass patronage networks and appeal directly to voters with improved public services.
