Regime transitions are contagious according to the diffusion-of-democracy literature: a country's regime is affected by others' through various predefined networks (e.g. geographical proximity), as well as by the country's own political, economic and social attributes (e.g. GDP levels). My account departs from the existing diffusion theory by allowing for countries' self-selection into peer regime networks based on their democracy levels in the past. For example, a country can form stronger dependency ties with countries that demonstrated similar democracy levels in the past (homophily). In the longitudinal setting, the traditional diffusion mechanism with the presence of self-selection generates the "co-evolutionary dynamic" between country networks and democracy levels. With this recursive feedback process between tie formation and democracy levels, it becomes extremely difficult to evaluate empirically how each country's level of democracy is determined, because we need to distinguish the following three processes statistically. First, country-specific attributes determine the level of democracy as in the earliest democratization studies. Second, other states' democracy levels also predict a country's regime as demonstrated in the conventional diffusion studies. Finally with my theory of endogenous network formation, the seeming diffusion effect is partially a consequence of their self-selection into peer networks. A newer spatial econometric model, an "M-STAR + Co-Evolution" model, is one of the first that allows us to test for all of these three dynamics behind democratization. In my first-cut analysis, I find that all three processes indeed exist.