I have a few comments on yesterday’s post from Prof. Wilson. Those on ELS who are not on the lawcourts listserv will have missed the comments from Bert Kritzer on this issue, so after I post this, having obtained Bert’s permission, will post his response as well.
Social scientists frequently and appropriately run regressions on aggregate data, even when individual-level data are available. We ask why the president’s aggregate approval rating is high or low when we have individual level responses. We examine Congressmen’s aggregate voting behavior, either across the board (as scaled through NOMINATE-style procedures), or on narrower issues (through ADA, ACU, or NARAL support scores), even though we have their individual-level roll calls.
William Rehnquist voted in the liberal direction about 21% of the time during his career; William Brennan about 80% of the time, and most justices fall somewhat in between. It is a legitimate question to ask what explains the variation amongst all the justices in those totals. It ends up that an exogeonous measure of the justices’ attitudes explains about 60% or so of the variation in those aggregate votes, whether the aggregates are of civil liberties cases, or an expanded set that covers virtually all cases.
Of course, the fit will be lower when one moves from the justice level (the aggregate votes of justices) to the justice-case level, i.e., the individual-level votes of justices in specific cases. Methodologically, this is because we now have to explain not only variance across justices, but variance within justices that are caused by case level factors. Needless to say, measures of justices’ attitudes that do not vary within the justice by the issue area cannot pick up variance within a justices’ votes. Theoretically, at this level, the attitudinal model calls for the interaction between attitudes and case stimuli. Any model that includes only attitudes at this level is incomplete. And it is at this level, not at the aggregate level, that multi-level modeling would be called for (and indeed would even be possible).