NATIONAL MEDIA DECISIONS ON ELECTION NIGHT (PART 1)

Featuring guest blogger, Harrison Hickman, President & Founder of Hickman Analytics, Inc.

Election night is just around the corner, and viewers can anticipate some frustration when a media outlet announces that one candidate is the “projected winner” shortly after the polls close. This isn’t new, as the national media have “called” elections before all votes are counted for more than half a century. And although a few mistakes have gotten wide coverage, the reality is that almost all of the networks’ election projections have been accurate.

Missing from most of the coverage though is a good answer to the common question: “How can they do that when the Election Board only shows a few returns?” That’s what I’m going to do in this and a follow-up post next week.

The NBC election unit recently published an article about how they make calls.  In addition to this, I’ll use my experience working for 25 years as an election night analyst for CBS News.  For the last 20 of those, I was the lead consultant for the anchor desk. (For all but the last of those years, the anchor was Dan Rather.) This experience provided an unparalleled opportunity to see “how the sausage is made” when networks make projections about election outcomes.

Calls.  Almost all national media outlets use the same language to characterize their “calls,” which is how they see a race before the votes are counted.  Races are said to be “too early to call” when a candidate seems to have a large lead but the advantage is outside the statistical limits set by the decision desk. In races where no exit poll was conducted, this call is used because the amount of data needed to make a call is insufficient when they want to report on the status of the race. 

Races that are “too close to call” are mostly races within a certain range (e.g. 5 points at NBC) once actual votes are available to be counted. The judgment is not made on the official count that might be reported by the Election Board, but on a statistical analysis of how the candidates are doing compared to what model of the electorate prepared in advance by the decision desk.

Races are said to be “leaning” when most evidence points to one candidate winning, but the levels of the evidence are not yet sufficient for the decision desk to say the candidate is certain to win. Frequent examples of this are when results are slow to come in from areas that swing from one party to another or when the turnout is unusually high or low in a highly partisan area.

Once all statistical criteria are met, the media outlet will declare one candidate the “projected winner.” The actual winner, of course, is not decided until the election board certifies the results.

Exit polls.  A major myth about network election projections is that decisions are based on exit poll results. In reality, exit polls are only used to make a projection when a race is very one-sided, which is to say that the difference between the leading candidates is well outside the statistical margin of error. Warren Mitofsky, the inventor of the exit poll (and the person who hired me to work at CBS), was highly sensitive to constraints on the accuracy of exit polls and exceedingly cautious before making calls based on exit poll results.

Recent developments have reduced the use of exit polls for making projections even more. For example, when the exit poll was invented, almost all voting was done on election day at designated polling places. This made it relatively easy to conduct a random sample of all voters. With increasing numbers of voters taking advantage of “early voting” and “vote-by-mail” (over half the votes cast in some states), election day polls are not reliable.

As a result, traditional exit polls – which are conducted in person at polling places -- must now be supplemented with conventional telephone and on-line polls to account for mail voting.  The inclusion of additional samples increases the total margin of error for the polling, and this means a candidate’s “lead” must be even larger for a call based on the exit poll.

Cost is another factor. The first big cost-saving was to have a research firm (Edison Research) conduct the polls rather than having each network conduct their own. The networks still make their own “calls,” but they rely on Edison for the data.

But even with that cost saving, exit polls remain so expensive that they are only conducted in states with key Senate races or in states likely to make the difference in who wins the most Electoral Votes for President.  Sample sizes of exit polls have also been reduced because of cost. One early consequence of the cost-saving measures was the fiasco in Florida in the Gore-Bush presidential race in 2000.  A reduction in the size of the exit poll sample made it impossible to say whether the counties recounting votes were likely to supply the decisive margin.

In the next post, I’ll try to explain how the calls are made without exit polls.