Fiscal Effects of the Voter Initiative Reconsidered: Addressing Endogeneity

Abstract

Several recent studies have investigated the relationship between direct democracy and public policy outcomes, with mixed findings. These inconsistencies may stem, in part, from researchers' failure to recognize that direct democracy institutions are distributed nonrandomly across the American states. That is, certain factors may lead a state to adopt the initiative process and influence other policy choices. We revisit the question of how the initiative influences state fiscal policy using panel data from 1960-2000 and a full-information maximum likelihood estimator that explicitly accounts for the endogeneity of the initiative. Our findings suggest that failure to endogenize the initiative in empirical analyses leads to substantially biased estimates of its effects. In particular, we find that once factors that predict whether a state has adopted the initiative are controlled, the initiative has a positive effect on state revenue generation and spending.Does direct democracy affect the substance of public policy? In particular, does direct democracy influence the level of public spending? These questions have been the subject of much research in the American states in recent years (

Institutions and Fiscal Sustainability

“…However, Marschall and Ruhil (2005) fi nd evidence that contradicts this hypothesis. Using panel data from 1960-2000 and a maximum likelihood estimator, they fi nd a positive relationship between the initiative and spending and taxes.…”

New Directions in Campaigns and Elections

“…It is very difficult to assign causality to the use of direct democracy in a state, given that other features of that place co-vary with use of direct democracy (Marschall and Ruhil 2005). It is possible that states which adopted direct democracy shared some common social context that, a century later, continues to affect levels of efficacy and participation.…”

The Endogeneity of the Initiative: A Comment on Marschall and Ruhil

“…< .10; **p < .05; ***p < .01 Note: Variables are defined inMarschall and Ruhil (2005), and the models follow their specification (except for the particular variables included). Three models are reported, each with 441 observations.…”