The Importance of Correct Measurement: A Response to Palmer, et al

Abstract

Palmer, D’Orazio, Kenwick, and McManus (PDKM) review a minority of changes we made to the CoW MID data, agree with most of those changes, but then argue that no scientific findings will be influenced by these or the thousands of changed values they ignore. We respond in two ways. First, we point out that principles of good science dictate that we discard dispute cases that either cannot be substantiated by the historical record or that do not meet dataset coding rules; we also discuss the ten-thousand-plus changed values PDKM still have not examined. Second, contrary to PDKM, we demonstrate how data differences greatly affect previous studies by returning again to the replications. There were substantial errors in both of PDKM’s replications, and these errors led them to incorrect inferences.

The Militarized Interstate Confrontation Dataset, 1816-2014

“…Indeed, over half of the cases CoWMID codes as attacks-without-reciprocation included targeted state forces returning fire and many times repelling the attack. As we highlight elsewhere, recip should not be used as evidence of lack of fighting by Side B in studies of audience costs (Gibler, Miller and Little 2020). It often reduces to a marker for whether the initiator of an attack could be identified by the coder.…”