Vote buying with intergovernmental grants (my paper published in Public Choice)

When I started working in the academia a few years back, my friend and co-author Josip Glaurdić asked me which journal would I like to be published in the most? Without hesitation I said: Public Choice! 

Well, that goal has now been accomplished. I have a publication in one of my all time favorite political economy journals! You can read the paper on this link, it's been published online first. Next big goal: Quarterly Journal of Economics (I will also accept American Economic Review, Journal of Political Economy or American Political Science Review). 

Our paper is on the political bias in the allocation of intergovernmental grants in Croatia. Here's the abstract: 

"Instead of alleviating fiscal inequalities, intergovernmental grants are often used to fulfill the grantors’ political goals. This study uses a unique panel dataset on more than 500 Croatian municipalities over a 12-year period to uncover the extent to which grant distribution is biased owing to grantors’ electoral concerns. Instead of the default fixed effects approach to modelling panel data, we apply a novel within-between specification aimed at uncovering the contextual source of variation, focusing on the effects of electoral concerns on grant allocation within and between municipalities. We find evidence of a substantial political bias in grant allocations both within and between municipalities, particularly when it comes to local-level electoral concerns. The paper offers researchers a new perspective when tackling the issue of politically biased grant allocation using panel data, particularly when they wish to uncover the simultaneous impact of time-variant and time-invariant factors, or when they cannot apply a quasi-experimental approach because of specific institutional contexts."

Basically, we have taken a new spin on a well-researched topic in the field of political economy: does central government allocate local government grants based on selective political criteria? There is a multitude of papers on this for various countries (just check out our references), with the overreaching conclusion being: yes, there is a political bias in the allocation of intergovernmental grants (intergovernmental meaning the flow of funds from the central to the local government). It happens for two main reasons: 1) central government helps its local co-partisans (mayors from the same party as the national government) retain office by giving them more money to buy votes in local election years, and 2) the central government helps itself (increases its own chances of re-election) by giving more money to important districts in national election years. An important district can be either a swing district, where voters often switch from one party to the other, or a core district, where voters always vote for the same party. The literature has found evidence of both. We find that money mostly goes to core districts. Politicians thus want to get as many votes as possible in districts where they are already strong. 

So what makes our paper special? The standard literature approach was mainly to uncover the within unit variation of grant allocation over time. This means that they wanted to see which factors' changes over time affect how much money does a local unit of government get. When uncovering the effect this way the literature usually discards any between-unit variation, i.e. it cannot make any inferences between local units. To clarify here is a sentence from the paper: "For example, finding that larger vote shares for the government within counties result in more allocated grants over time—clearly a within effect—often is misinterpreted as the between effect and generalized into a cross-sectional conclusion that counties received more grants because the government garnered a larger share of the votes in a previous election."

A few clarifications before moving on: A panel dataset means having observations on multiple units over time. This is opposed to a cross-section where you just have observations on multiple units in one fixed time period. Having panel data is great because it allows you to eliminate any changes across units that stay fixed over time (like gender, geography, demographics, or any slow-changing variable like institutions), and focus only on estimating the effect of the changing independent variables on your outcome of interest. It is a very neat way of making correct inferences in the social sciences. 

What we wanted to do is to use our panel dataset to explore the variation both within and between our units of interest. So not only the standard within effect in a municipality over time, but also the cross-sectional effect of the differences between units to see which non-changing factors also could affect our outcome. In our own words:

"We test how the effects of political considerations on grant allocation change over time within each entity and how they vary across them. The within-between approach thus allows for the inclusion of potentially influential time-invariant variables, which the fixed effects approach eliminates, as a separate between-entity effect, in addition to keeping all the benefits of the fixed effects estimation. Disentangling the within- and between-entity effects is important as it not only provides a more substantive interpretation, but also enables the researcher to correctly identify the source of variation by not confusing which of the two effects is driving the estimated relationship. By utilizing this particular approach our goal is to offer researchers a new perspective on tackling the issue of grant allocation when one wishes to test for the simultaneous impact of time-invariant and time-variant variables, and when a quasi-experimental setting is unfeasible owing to specific circumstances of the observed political system."

The within-between approach is a new method referenced to a great paper by Bell and Jones (2015). 

Results

What do we find? As I've said before, there is a clear conclusion that there is a significant political bias in the allocation of intergovernmental grants. The national government favors municipalities that support them in the national elections, and those that were won over by their co-partisan mayors. They give more money during election years (both national and local), and they support core municipalities rather than swing municipalities. 

The within-between approach was most helpful in examining the interaction effect of votes for government and turnout. This is best seen on the figures below:

In our own words: 

"...in Fig. 1 it is obvious that higher national turnout is conditioning only the within-municipality changes in grants in a positive way, whereas the between effect goes in completely the opposite direction (and also is insignificant). In other words, the government rewards only those municipalities wherein they gain support through higher voter turnout rates across time. 

In Fig. 2, representing local level estimates, the conditionality of turnout on a between-municipality level is shown to be crucial for concluding that mayors who win on higher voter turnouts are likely to receive larger grants. The within effect plays no role here, so the conclusion regarding the effect of mayoral alignment and turnout on grant allocation is valid only on a between-municipality level. In other words, aligned mayors who win their posts with high voter turnout rates do not get more intergovernmental grants (they do get more such funds, but not conditioned on turnout), while aligned mayors already holding power do get more money if they can increase voter turnout. Both findings make sense, since winning over a new municipality is good for the national party regardless of turnout, while for existing incumbents establishing their dominance with even more support is likely to be rewarded. None of these conclusions would have been possible without the use of the WB approach."