Popular Error‑Correction Model Misused in Political Time Series, Study Warns

Why This Matters: Political scientists increasingly rely on a General Error Correction Model (GECM) to analyze time series data. Grant and Lebo show that routine applications of the GECM can produce misleading inference when practitioners mix series with different statistical properties or ignore key model constraints. The result is unreliable hypothesis tests and overstated evidence for so‑called "error correction."

What Grant and Lebo Investigate: The authors revisit the recommendations of DeBoef and Keele, who promoted the GECM even for stationary data. They ask whether common GECM practices in political science—especially pooling series with different orders of integration and relying on standard t‑tests—are valid across realistic data settings.

How the Study Was Done: The paper evaluates GECM practice using a broad empirical and simulation strategy:

• Six types of data-generating processes that vary in integration properties (stationary, non‑stationary, explosive, near‑integrated, fractionally integrated, etc.).
• 746 Monte Carlo simulations to trace estimator behavior across sample sizes and covariate counts.
• Five replications of published political science papers that used GECM approaches to assess real-world consequences.

Key Findings:

• The GECM is not the flexible, catch‑all tool it is often treated as; it requires care because mixing series of different orders of integration leads to model misspecification.
• When series of differing integration orders are analyzed together (i.e., without balancing equations to match integration properties), estimated long‑run multipliers and hypothesis tests can be unreliable.
• The sampling distribution of the error‑correction term changes substantially with the order of integration, sample size, number of covariates, and the boundedness of Yt—so standard t‑tests frequently overstate evidence for error correction.

What This Means for Researchers: Using a GECM without checking the integration properties of each series, ensuring equation balance, and accounting for finite‑sample behavior risks incorrect substantive conclusions. Grant and Lebo’s diagnostics and simulation evidence encourage more cautious application of error‑correction methods and suggest that analysts should verify integration orders and consider alternative inferential approaches when sample sizes or variable properties make standard tests unreliable.