Duration Models in Politics
Also known as: Event history models, Survival models in political science, Hazard models, Time-to-event models in politics
Duration models — also called event history or survival models — analyze the time until a political event occurs: how long a cabinet lasts before it falls, how long a war runs before it ends, how long a policy takes to be adopted, or how long a regime survives. Rather than asking only whether an event happens, these models ask when, modeling the hazard rate as a function of covariates while correctly handling censored cases that have not yet experienced the event. The Cox proportional hazards model and parametric alternatives such as the Weibull, popularized in political science by Box-Steffensmeier and Jones, form the core toolkit.
Key highlights
- Correctly incorporates right-censored cases, using information from units that have not yet experienced the event rather than discarding or mismodeling them.
- Directly models the timing and the time-dependence of events, answering when as well as whether, which is often the substantive political question.
- The Cox model estimates covariate effects without assuming a baseline hazard shape, giving robustness to misspecification of how risk evolves over time.
- Connects naturally to binary panel data via the Beck-Katz-Tucker result, so widely used logit models can be turned into proper event-history analyses.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use duration models whenever the dependent variable is the time until a political event and some observations are censored — cabinet survival, war and dispute termination, the timing of policy adoption or diffusion, regime breakdown, leader tenure, or the duration of peace. Choose the Cox model when you want covariate effects without assuming a shape for the baseline hazard and when proportional hazards is reasonable. Choose a parametric model such as the Weibull when duration dependence is itself the question, when you need to predict survival times, or when the hazard shape is theoretically motivated. Use the Beck-Katz-Tucker approach with splines when your data come as binary time-series cross-section observations rather than as explicit durations. Duration models are less appropriate when events recur in complex ways without a clean clock, when censoring is informative and unmodeled, or when the proportionality or distributional assumptions are clearly violated and unaddressed.
Strengths & limitations
- Correctly incorporates right-censored cases, using information from units that have not yet experienced the event rather than discarding or mismodeling them.
- Directly models the timing and the time-dependence of events, answering when as well as whether, which is often the substantive political question.
- The Cox model estimates covariate effects without assuming a baseline hazard shape, giving robustness to misspecification of how risk evolves over time.
- Connects naturally to binary panel data via the Beck-Katz-Tucker result, so widely used logit models can be turned into proper event-history analyses.
- The proportional hazards assumption — constant hazard ratios over time — is frequently violated for political covariates and must be tested and, if necessary, relaxed with interactions or stratification.
- Parametric models impose a distributional form on the baseline hazard that, if wrong, biases both duration-dependence and covariate estimates.
- Unobserved heterogeneity (frailty) can masquerade as negative duration dependence, confounding inference about whether risk truly declines over time.
- Repeated events, competing risks, and time-varying covariates require extensions that complicate estimation and interpretation.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
Should I use a Cox model or a parametric model like the Weibull?
Use the Cox model when you want to estimate covariate effects while remaining agnostic about how the baseline risk evolves over time; its semiparametric partial likelihood is robust to misspecification of that time path. Use a parametric model such as the Weibull when duration dependence itself is theoretically central, when you need to predict actual survival times, or when you have a strong prior about the hazard shape and want the efficiency gains of a correctly specified model. Box-Steffensmeier and Jones recommend letting the substantive question and diagnostics guide the choice rather than defaulting reflexively to either.
How is a binary time-series cross-section logit related to a duration model?
Beck, Katz, and Tucker showed that a binary dependent variable observed repeatedly over time (e.g., conflict yes/no each year) is grouped-duration data: each unit accumulates time at risk until the event occurs. A logit that ignores how long it has been since the last event omits the baseline hazard and produces biased, overconfident estimates. Including a flexible function of time since the last event — originally cubic splines, later time polynomials — recovers the duration dependence and corrects inference, effectively turning the logit into a discrete-time event-history model.
What is the proportional hazards assumption and how do I check it?
Proportional hazards means each covariate multiplies the baseline hazard by a constant factor that does not change over time, so hazard ratios are time-invariant. When it fails, a covariate's effect strengthens or fades as duration elapses, and a single hazard ratio is misleading. The standard check uses Schoenfeld residuals, testing whether their correlation with time is zero; if violated, analysts can add interactions of the covariate with a function of time, stratify the baseline hazard, or move to a model that allows time-varying effects.
Sources
- 1.Box-Steffensmeier, J. M., & Jones, B. S. (2004). Event History Modeling: A Guide for Social Scientists. Cambridge University Press.ISBN 9780521546737
- 2.Cox, D. R. (1972). Regression Models and Life-Tables. Journal of the Royal Statistical Society: Series B (Methodological), 34(2), 187–202.
- 3.Beck, N., Katz, J. N., & Tucker, R. (1998). Taking Time Seriously: Time-Series-Cross-Section Analysis with a Binary Dependent Variable. American Journal of Political Science, 42(4), 1260–1288.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). Duration Models in Politics. ScholarGate. https://scholargate.app/political-science/duration-models-politics