The Gartner-Siverson war initiation model's focus on selection bias and ini
tiators selecting vulnerable targets is attractive, but it demands further
specification. A predictive model of war initiations might not retain stati
stical significance after 1945. The authors examine 16 possibly related var
iables and develop a three-variable initiation model, adding information on
target regime type and war duration to the original lone initiator-lone ta
rget variable for war initiations between 1816 and 1992. The instability of
the initiation model before and after 1945 can be traced to changes in who
initiates wars. The predatory initiator model does not predict minor power
dyads very well, and these dyads have come to dominate the post-1945 war l
andscape. Several reasons for this phenomenon are suggested, including the
argument that the context in which weaker states initiate wars renders oppo
rtunistic predators less likely and less successful.