Democratizing countries have been shown to engage more often in interstate conflict. While these findings (particularly those of Mansfield and Snyder) have faced criticism, continuing references to these findings in both academia and policy circles suggest that the argument is still found to have merit and that the criticisms have not served to discredit the argument. One reason might be that the criticisms have been mainly limited to statistical model specification and sample issues. We argue instead there are other issues with the argument, data selection, and more centrally with concept misformation in a classic Sartorian sense that need attention. Fundamentally, we believe that Mansfield and Snyder (and others), by relying on a Politybased operationalization of regime change, fail to capture processes of democratization and instead agglomerate several processes of regime change into their measure of "democratization" and that they are engaging in concept stretching. Moreover, we highlight the importance of considering regime volatility. We test the democratization-conflict relationship statistically with several data sets that capture both regimes and interstate conflict. Our tests demonstrate that Mansfield and Snyder's findings in their work on interstate conflict are a product of concept misformation rather than just model specification. In addition, we inspect the conditioning role of institutional strength on democratization and conflict nexus. Simulation-based analyses show that the neither incomplete democratization nor other types of regime transitions exert a statistically significant effect on war onset at monadic level. On dyadic level, the most optimistic estimates that account for the zeroinflation in the dataset show that incomplete democratization is statistically significant at some levels of institutional strength, yet this time substantively insignificant. In the process, we raise new questions about the overreliance on a single dataset and exclusive use of quantities of interest such as relative risk that hides baseline point estimates in trivial magnitude, hence, misguides the discipline about the true location of causal effect attributed to a covariate in the model. [ABSTRACT FROM AUTHOR]