Identifying Unit-Dependency and Time-Specificity in Longitudinal Analysis: A Graphical Methodology

Longitudinal analysis in criminology and other social sciences has become an important research tool because it allows us to draw conclusions from observing how multiple units change over time. Unfortunately, its results are more vulnerable to potential influences of unusual observational units or periods of time. Current leverage diagnostics are designed for cross-sectional analysis and are fallible when applied to longitudinal models. This article introduces a graphical diagnostic methodology to systematically examine the sensitivity of longitudinal results to extreme observational units and periods of time—unit-dependency and time-specificity. Further the article illustrates its use with an example testing policy effects on black and white female victimization of intimate partner homicide. Results are displayed in an easily understood graph that provides a snapshot of the results' time-specific patterns and robustness to unit-dependency. Currently, comparable tests for panel analysis are tedious and cumbersome. With this new illuminating methodology, researchers and policy-makers can easily decide whether a time-specific or unit-dependent pattern is consequential.