Most MID Changes Hold — But Replication Choices, Not Data, Explain Many Reversals

🔎 What GML Claimed and What Was Rechecked

Gibler, Miller, and Little (2016) conducted a wide-ranging review of the Militarized Interstate Dispute (MID) dataset covering 1816–2001, identified numerous possible inaccuracies, and recommended a substantial set of drops and merges. GML reported that analyses using their revised data sometimes yield substantively different inferences. The current review evaluates GML’s drop-and-merge recommendations and reassesses their claimed substantive impact.

🧾 How the MID Recommendations Were Evaluated

• Examined each of GML’s suggested drops and merges to assess whether the proposed changes are warranted.
• Recreated GML’s replication exercises and compared outcomes to determine whether altered inferences stem from the revised data or from the replication strategies employed.

📊 Key Results

• Agreement: about 76 percent of GML’s recommended drops and merges are accepted as appropriate.
• Attribution: several of the reported overturned findings in GML’s replications are attributable not to the altered MID data but to choices made in replication procedures.
• Overall impact: after reexamination, the remaining differences in substantive inference that can be traced to variations in the MID data are uncommon and modest in scope.

💡 Why This Matters

These findings indicate that many of GML’s specific data corrections are reasonable, but caution is warranted before attributing changes in empirical conclusions to dataset revisions alone. Replication strategies can produce apparent reversals that are not due to the underlying MID data. Scholars using the MID dataset (1816–2001) should check both data coding and replication choices when interpreting shifts in results.