To produce the probabilities 
used in equation
(3), we estimate an individual-level probit regression
equation using ANES data from the preceding presidential election, at time
(Miller and the National Election Studies 1986), based on
, where 
denotes a vector of survey
opinion variables and the disturbance is assumed to satisfy
. To measure vote choices we used
self-reports (1984 ANES var. 787, 788), omitting those validated not to have
voted (var. 1130). contains a constant having the value
1.0 and seven other variables. Four variables count the number of mentions
for each of the open-ended party likes and dislikes items: number of likes
about the Democratic party (var. 266-271); number of dislikes about the
Democratic party (var. 272-277); number of likes about the Republican party
(var. 278-283); number of dislikes about the Republican party (var.
284-289). Two variables measure retrospective economic evaluations,
referring respectively to the nation's economy over the past year (var. 227)
and to family finances now compared to a year ago (var. 139); values of DK or
NA are treated as missing. The seventh variable is the feeling thermometer
for Republican presidential candidate Reagan (var. 290); values of NA are
treated as missing but responses of ``doesn't recognize name'' or ``don't know
where to rate'' are assigned the middle value of 50. An observation missing
data on any variable is omitted from the analysis (i.e., listwise deletion).
We use the estimated parameters with data for the same variables from the
midterm ( 
) ANES to compute 
. The variables from the
1986 ANES data are var. 72-95 (likes and dislikes), var. 372, 355 (economic
evaluations) and var. 130 (Reagan feeling thermometer) (Miller and the
National Election Studies 1987).