Modeling Heterogeneity in Pooled Event History Analysis

Abstract

Pooled event history analysis (PEHA) allows researchers to study the effects of variables across multiple policies by stacking the data and estimating the parameters in a single model. Yet this approach to modeling policy diffusion implies assumptions about homogeneity that are often violated in reality, such that the effect of a given variable is constant across policies. We relax this assumption and use Monte Carlo simulations to compare common strategies for modeling heterogeneity, testing these strategies with increasing levels of variance. We find that multilevel models with random coefficients produce the best estimates and are a significant improvement over other models. In addition, we show how modeling similar policies as multilevel structures allows researchers to more precisely explore the theoretical implications of heterogeneity across policies. We provide an empirical example of these modeling approaches with a unique data set of 29 antiabortion policies.

The Spreading of Innovation: State Adoptions of Energy and Climate Change Policy

“…Because our unit of analysis is an adoption for a given state‐policy‐year combination, we use a mixed effects logistic regression that accounts for state‐level variation and policy heterogeneity (Kreitzer & Boehmke, ). Our initial analysis employs a crossed random effects model that includes varying intercepts for policy type and state so that we can effectively model heterogeneity that could otherwise cause significant bias in nonlinear models like a logit (Rabe‐Hesketh & Skrondal, ) .…”

“…The new wave in diffusion research is to consider a larger variety of policies within one model (see Kreitzer & Boehmke, ), though there are relatively few scholars who have taken such an approach. One exception includes work by Nicholson‐Crotty et al () who conduct a pooled event history analysis on interstate compacts.…”

“…Model 3 is a policy random effects logit model where policy heterogeneity is captured instead. Model 4 is a more complex crossed random effects logistic regression where both state and policy variation is allowed (Kreitzer & Boehmke, ).…”

“…We move the literature forward by examining several policies with a variety of multilevel stacked hazard models. We take a general approach to estimating policy and state heterogeneity that expands on current work in policy diffusion literature by estimating a variety of fixed effects and multilevel models (Kreitzer & Boehmke, ). Here, we are able to look at the average effect of factors across policies and identify factors that do vary across the policies we use.…”

Introduction to SPPQ Special Issue on Policy Diffusion

“…We concur with Mallison that a continuous measure of adoption speed holds promise for our understanding of the mechanisms of policy diffusion as well as how the political or policy environment influences the spread of policies. Kreitzer and Boehmke (2016) highlight the importance of choosing the correct estimation strategy for studying the policy diffusion process across multiple issues. Using Monte Carlo simulations, the authors show that multilevel models with random coefficients produce better estimates than strategies using fixed effects or clustering to model heterogeneity across policies.…”