%T Measurement Error and Dynamic Nonlinear Models:
(Over)Estimating the Effect of Habit
%X Estimates from non-linear models are known to be inconsistent when the dependent variable
is misclassified. Although methods have been developed to correct this inconsistency in static
non-linear models, no correction exists for dynamic non-linear models. This is a serious omission
from the literature. Since the lagged dependent variable is an explanatory variable in dynamic
models, any inconsistency that arises from misclassifcation of the dependent variable in a static
non-linear model will be magnifed when that model is made dynamic. Here, we demonstrate
this fact using the habitual voting literature and develop a parametric model to correct for this
inconsistency. We find that, on average, estimates of habitual voting are approximately twice as
large when using survey respondents' self-reports versus official records of their turnout decisions.
When we apply our corrected model to respondents' self-reports, however, the estimates of
habitual voting are significantly closer to those provided by the official records.
%L eprints88
%A Gabriel Katz
%A James Melton
%D 2011