Panel Data Propensity Score Weighting
Panel Data Propensity Score Weighting Estimator · Also known as: panel PSW, panel IPW, longitudinal propensity score weighting, panel inverse probability weighting
Panel Data Propensity Score Weighting (panel PSW) extends inverse probability weighting to longitudinal settings where the same units are observed across multiple time periods. It reweights observations by the inverse of each unit's time-varying probability of receiving treatment, creating a pseudo-population in which treatment is balanced on observed covariates at each period, and then estimates causal effects from repeated-measures data.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use panel data PSW when you have longitudinal data on the same units across multiple periods and want to estimate the causal effect of a treatment that is not randomly assigned. It is well suited when treatment uptake depends on time-varying covariates that can be measured, and when the treatment effect is of interest for the full population (ATE) rather than only those who received treatment. Do not use it when important confounders are unobserved or unmeasured, because PSW only balances on observed covariates. Avoid it with very few repeated periods or very high treatment variability that leads to extreme weights; consider doubly robust or difference-in-differences alternatives instead.
Strengths & limitations
- Corrects for selection bias due to observed time-varying covariates in longitudinal settings.
- Creates a pseudo-population that is balanced on measured confounders at each time point without requiring explicit matching.
- Can target the average treatment effect (ATE) for the full population, unlike matching which typically targets the ATT.
- Compatible with a wide range of outcome models once weights are constructed, including linear, logistic, and survival models.
- Stabilized weights reduce variance and improve numerical stability in long panels.
- Assumes no unmeasured confounding — if key covariates are not observed, the weighted estimates remain biased.
- Extreme propensity scores near 0 or 1 produce very large weights and highly variable estimates; truncation is a workaround but introduces bias.
- The propensity model must be correctly specified; misspecification in the logistic regression propagates directly into biased weights.
- Long panels with time-varying treatments can require multiplying weights across periods (as in marginal structural models), which compounds variability rapidly.
- Cluster-robust or bootstrap inference is mandatory, adding complexity compared to standard OLS.
Frequently asked
How does panel PSW differ from ordinary (cross-sectional) PSW?
In cross-sectional PSW each observation is independent and a single propensity score is estimated per unit. In panel PSW each unit is observed multiple times, so the propensity score can vary over time with changing covariates, observations within a unit are correlated, and inference must use cluster-robust or bootstrap standard errors.
What is the positivity assumption and why does it matter?
Positivity requires that every covariate profile has a non-zero probability of both receiving and not receiving treatment (0 < e < 1). If some profiles occur only in treated or only in untreated units, the weights become infinite and causal identification fails. Weight distributions near 0 or 1 are a warning sign.
Should I use stabilized or unstabilized weights?
Stabilized weights are generally recommended. They have the same expectation as unstabilized weights under correct model specification but lower variance, making estimates more numerically stable. Unstabilized weights can be extremely large when propensity scores are close to 0 or 1.
Can I combine panel PSW with regression adjustment?
Yes. Combining weighting with regression adjustment yields doubly robust estimation: the causal estimate is consistent if either the propensity model or the outcome model is correctly specified. This offers additional protection against model misspecification.
When should I prefer difference-in-differences over panel PSW?
Prefer DiD when there is a clear pre/post structure with a policy or event affecting one group, and you are willing to assume parallel trends rather than ignorable treatment selection. Panel PSW is preferable when treatment assignment is driven by observed time-varying characteristics and no clean before/after structure exists.
Sources
- Hirano, K., Imbens, G. W., & Ridder, G. (2003). Efficient Estimation of Average Treatment Effects Using the Estimated Propensity Score. Econometrica, 71(4), 1161-1189. DOI: 10.1111/1468-0262.00442 ↗
- Robins, J. M., Hernan, M. A., & Brumback, B. (2000). Marginal Structural Models and Causal Inference in Epidemiology. Epidemiology, 11(5), 550-560. DOI: 10.1097/00001648-200009000-00011 ↗
How to cite this page
ScholarGate. (2026, June 3). Panel Data Propensity Score Weighting Estimator. ScholarGate. https://scholargate.app/en/causal-inference/panel-data-propensity-score-weighting
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Marginal Structural ModelCausal inference↔ compare
- Panel Data Difference-in-DifferencesCausal inference↔ compare
- Panel Data Inverse Probability WeightingCausal inference↔ compare
- Propensity Score MatchingResearch Statistics↔ compare
- Propensity Score WeightingCausal inference↔ compare