In 1968, the average length of a "sound bite" (block of uninterrupted speech) on TV coverage of US presidential candidates was 42.3 seconds.
In 1988: only 9.8 seconds.
Page 4 of Kiki Adato: The incredible shrinking soundbite, Joan Shorenstein Barone Center, Cambridge MA 1990.
In the January-June 1980 pre-convention period, CBS offered 450 minutes of presidential campaign coverage. Of these, 307.8 minutes, or 68.4%, were "horse race coverage." [p64 Adato, citing study M.Robinson & M.Sheehan: Over the Wire on TV, Russell Sage Foundation NY 1983 p149] leaving only 31.6% for actual discussion of their views on issues.
Since 1980, I believe this has become even worse, more like 80-20 or 90-10.
This only makes logical sense when interpreted as a side-effect of our plurality voting system – "who the leading two horses are" is key strategic knowledge for any voter in that system (since without that knowledge, voters would be incapable of voting strategically and would be at great risk of "wasting" their vote by, e.g, foolishly voting for their true favorite). Crazily, "horse race" information therefore actually is among the most important information a voter can have under the present plurality voting system, and information about the actual views of any candidates besides the top two is usually of near-zero true importance.
If we were using a better voting system such as range or approval voting, then voting your favorite top is never strategically unwise, so the candidates' actual views become much more important, and strategy comparatively less important.