A Political Economy Model of Congressional Careers (March 2005)

from all strategic considerations. 28 We specify that the vector (INC, ES) evolves according to a Markov process with transition probabilities P(INCi,t! 1 , ESi,t!1!Cycleit, INCit, ESit). Note that INC and ES could be predicted perfectly using lagged Cycle, INC, and ES if incumbent senators always ran for reelection and never left office due to death, appointment to other offices, or early retirement. Thus, these are the natural variables to use in predicting INC and ES. Of the 768 elements in this transition matrix, only 240 are feasible and, within this subset, only 56 are positive. We estimate these elements using empirical frequencies from the data, and then treat them as known in the solution and estimation of the model. 29 Thus, in our model, a phenomenon like the shift of the South from being solidly Democratic to being more Republican in recent years is captured by changes over time in SOS, since SOD is measured relative to SOS. Also note that, while we assume that SOD remains constant over a representative’s time in office, it is allowed to change when the identity of the district’s representative changes. 30 SOSit and SOWt also influence the chances of winning a bid for higher office if a Senate seat is up for election. And, even with no Senate election at time t, SOSit and SOWt help forecast the chance of winning a Senate bid in the future. 355 VOL. 95 NO. 1 DIERMEIER ET AL.: CONGRESSIONAL CAREERS