xtreg y x1 x2, re estimates store random Perform the test: hausman fixed random Interpretation: If the p-value is small (typically
Already reported after xtreg, fe .
When endogeneity is a concern, you can use instrumental variables. The xtivreg command fits IV models for panel data using two-stage least squares. It supports several estimators, including:
To see how many periods of data you have for each entity, use xttab : xttab industry Use code with caution. Line Plots stata panel data
The Fixed Effects model explores the variation. It controls for all time-invariant, unobserved characteristics of the entities (e.g., a country's culture, an individual's genetic traits) by subtracting the group-level means from each variable. xtreg y x1 x2 x3, fe Use code with caution.
), reject the null hypothesis. This means you should use the model. If the p-value is large ( >0.05is greater than 0.05
reg wage hours tenure age i.year, robust xtreg y x1 x2, re estimates store random
An introduction to Stata panel data analysis requires understanding how to manage and model data that tracks the same units over multiple time periods. Panel data—also known as longitudinal data—combines the characteristics of both cross-sectional and time-series data.
Choosing blindly between Pooled OLS, FE, and RE can invalidate your empirical findings. Stata provides specific post-estimation tests to guide your selection.
| Feature | Pooled OLS | Fixed Effects (FE) | Random Effects (RE) | | :--- | :--- | :--- | :--- | | | reg y x | xtreg y x, fe | xtreg y x, re | | Assumption | No individual effects | $\alpha_i$ correlated with $x$ | $\alpha_i$ NOT correlated with $x$ | | Time-Invariant Vars? | Yes | No (Dropped) | Yes | | Efficiency | N/A | Low | High | | Best For | Preliminary analysis | Causal inference (observational) | Efficiency / Random sampling | It supports several estimators, including: To see how
Example:
: Before running any panel-specific models, you must tell Stata which variable identifies the entity and which identifies the time using the xtset command. xtset entity_id time_var Use code with caution. Copied to clipboard 2. Core Estimation Models
) as a predictor, standard FE estimators are biased (Nickell bias). Use the Arellano-Bond framework:
The RE model treats the entity-specific effects as random draws from a larger population, uncorrelated with the independent variables. This allows it to include time-invariant variables in the model, which is a key advantage over FE. To run this model, you use the re option:
reg D.wage D.hours D.tenure D.age, noconstant