Change Comes with Time: Substantive Interpretation of Nonproportional Hazards in Event History Analysis

Abstract

While methodologists have provided us ample notice of both the problem of non-proportional hazards and the means of correcting for them, less attention has been paid to the post-estimation interpretation. The suggested inclusion of time interactions in our models is more than a statistical fix: these corrections alter the substantive meaning and interpretation of results. Framing the issue as a specific case of multiplicativeinteraction modeling, I provide detailed discussion of the problem of non-proportional hazards and present several appropriate means of interpreting both the substantive impact and the significance of variables whose effects may change over time.

Diabetes, hepatocellular carcinoma, and mortality in hepatitis C‐infected patients: A population‐based cohort study

“…For the violation of proportional hazards assumption, we directly modeled the interaction between the variable and natural logarithm of time and rechecked the proportional hazards assumption . To facilitate the interpretation of time‐varying coefficients, we conducted post‐estimation simulation techniques and examined graphs with visual weight to demonstrate the results for the reasons that follow: Using this simulation technique allowed us to estimate full time‐interactive effects, evaluate how they changed over time, evaluate the effects, and show the uncertainty surrounding the estimates. For visual weighting presentation, more visual weight could be created by using more “graphical ink.” Thus, areas of the distribution with many simulations would be darker, whereas areas with fewer simulations, often near the edges of the distribution, would be lighter.…”

Non-conformable, partial and conformable transposition: A competing risk analysis of the transposition process of directives in the EU15

“…The heterogeneity of member states' interests exerts an even larger effect: an increase in the heterogeneity of states' interests from the first to the third quartile increases the hazard rate by almost 130 percent. Figure 2 shows the effect of an increase from the first to the third quartile in the two PHA-violating covariates on the hazard rate over time (Licht, 2011). Substantively, the model reveals an almost 40 percent lower hazard rate for directives that delegate substantial transposition authority to the member states.…”

Working through the issues: how issue diversity and ideological disagreement influence coalition duration

“…Given the large number of interactions, the combined effect of C-ENMI in each of the models is difficult to interpret using coefficients alone (Licht 2011 By around the end of the third year of a coalition with disagreement at one standard deviation above the mean, C-ENMI reduces the risk of a replacement cabinet relative and early elections relative to the baseline hazard by just over 50%. Broadly, this finding is consistent with the explanation that C-ENMI reduces the effect of ideological disagreement once easy policy compromises have been reached for the risk of both replacement cabinets and early elections.…”