Previous research by Goldstone et al. (2010) generated a highly accurate predictive model of state-level political instability. Notably, this model identifies political institutions – and partial democracy with factionalism, specifically – as the most compelling factors explaining when and where instability events are likely to occur. This article reassesses the model’s explanatory power and makes three related points: (1) the model’s predictive power varies substantially over time; (2) its predictive power peaked in the period used for out-of-sample validation (1995–2004) in the original study and (3) the model performs relatively poorly in the more recent period. The authors find that this decline is not simply due to the Arab Uprisings, instability events that occurred in autocracies. Similar issues are found with attempts to predict nonviolent uprisings (Chenoweth and Ulfelder 2017) and armed conflict onset and continuation (Hegre et al. 2013). These results inform two conclusions: (1) the drivers of instability are not constant over time and (2) care must be exercised in interpreting prediction exercises as evidence in favor or dispositive of theoretical mechanisms.