Analyzing Policy Impact: Selection of a Linear Trend Model

Abstract

Policy analysts, as well as politicians, have shown great interest in assessing both short‐term and long‐term consequences of public policies in recent years. Recent time‐trend studies have attempted to depict the time dimension of policy consequences through extensions of regression techniques. This study examines three linear trend models which have been used to depict policy impact through time‐series analyses, and identifies the relative advantages and disadvantages associated with the use of each model. The three models are applied in a quasi‐experimental time‐series design to time‐series of per capita state expenditures for large cities, in eight states, over a twenty‐year period. Differences in degrees of multicollinearity and autocorrelation inherent in the three models are discussed, and the model providing the most conservative coefficient estimates is identified.

Previewing Policy Sciences: Multiple Lenses and Segmented Visions

“…Regression models, such as Schneider and Jacoby’s (1996) study of state Medicaid adoptions include variables for environmental factors, political factors, and organizational factors and isolating the effects of individual variables while holding other variables constant. Ordinary least squares models have been supplemented by non recursive causal models (Dean 1980), time series models (Newcomer and Hardy 1980), and substantively weighted least squares (Meier, Gill, and Waller 2000). While regression methods estimate the overall effects of multiple variables, the methodology does little to estimate whether the configuration of variables produces particular outcomes.…”