It's an empirical question, answered by statistical methods---failing to code based on exogenous variables corrupts their evidence
Chiba and Gartzke 21, Associate Professor of Political Science in the Department of Government and Public Administration at the University of Macau, Ph.D. in Political Science from Rice University, LL.M in Jurisprudence and International Relations from Hitotsubashi University, and Dr. Erik Gartzke, Professor of Political Science at the University of California, San Diego, PhD in Political Science from the University of Iowa (Daina and Erik, “Make Two Democracies and Call Me in the Morning: Endogenous Regime Type and the Democratic Peace,” https://dainachiba.github.io/research/make2dem/Make2Dem.pdf)
Before we review our approach in detail, it may be useful to explain why this type of analysis has not been pursued successfully in the past and what makes our effort different from other, broadly related projects. We are not the first to apply an IV framework (more specifically) or multi-equation models (more broadly) to the democratic peace. However, previous attempts suffer from two major problems. First, previous studies have typically used a dyad (country pair) as the unit of observation in analyzing conflict, which requires some summary measure(s) of democracy for a pair of countries rather than the state-level (monadic) democracy measure. 6 Use of a dyadic aggregate to represent regime type creates a discrepancy between the first stage regression (predicting democracy at the country level) and the outcome stage regression (predicting conflict at the dyad level). 7 We avoid this problem by using the directed dyad as the unit of observation in predicting conflict, distinguishing between the potential challenger and target in a dispute. This allows us to connect the first stage equations (predicting the challenger’s and target’s regime types) and the outcome stage equation seamlessly. Doing so has several benefits: the outcome stage model could directly include country-level covariates (such as challenger’s and target’s democracy) without having to convert them to a dyadic summary. This also allows us to estimate the system of equations jointly rather than relying on the “forbidden regression.” 8
Second, a more daunting challenge in applying an IV approach to democratic peace research is the difficulty of finding a plausible instrument for regime type — a variable that is strongly correlated with regime type but is unrelated to war. This is the challenge that has plagued empirical researchers in many fields. For example, a recent study of the effect of regime type on economic growth uses a diffusion-based measure of democracy (i.e., average value of democracies in a given region) as an instrument for democracy (Acemoglu et al. 2019). However, diffusion-based instruments such as this are unlikely to be a valid instrument, due to spatial spill-over, interdependence, and, most importantly, simultaneity (Betz, Cook, and Hollenbach 2018). Recognizing problems with spatial instruments, McDonald (2015) seeks to exploit the very discrepancy between country-level and dyad-level designs as the source of identification. His discussion, however, lacks a clear explanation as to why some determinants of regime type do not influence conflict. 9
We turn to a demographic variable — average female fertility rate in a given country — as a source of variation in regime type that is exogenous to international conflict. As we will argue below, a lower fertility rate is a strong driver of democratization. We will also present theoretical arguments and a series of falsification tests that support the claim that average national fertility rate does not directly influence international conflict.