Technical Appendix – Will the race tighten as the election approaches?

Applies to Will the race tighten as the election approaches?

I compiled all the polling data from RCP for the 2008 and 2012 Presidential elections with field work ending on or before October 2nd.  I then created daily averages, aggregating the polls by the date fieldwork was completed.  Most days have only one poll, but as the election nears multiple polls begin to appear.  So far in 2012, the day with the most poll releases was September 30th with five.  In 2008, there were three days in October with as many as nine polls reporting with a common final fieldwork day. I then used these common end-of-fieldwork days to measure the number of remaining Days Before the election in both 2012 and 2008.

I measure “closeness” in the polls by taking the simple absolute value of the difference in support between the two major-party candidates.  An Obama lead of three over Romney and vice versa both generate a closeness score of three.

Finally, I restricted the time period to the final 120 days before the election.  Some of the polling data for both elections dates back nearly two years.  I prefer to focus on the campaign itself and used the four months before the election as the basis of comparison.  Shortening the period to 90 or 60 days did not have a big effect on the results.

Here are the results from regressing the absolute margin on the number of days before the election, a “dummy” for 2012, and an “interaction” term multiplying the dummy with the number of days before.  Adding these two variables lets us estimate a separate relationship for each of the two elections.

OLS; 120 days or less until Election Day (N=134)
Dependent variable: Absolute margin between the Candidates

               coefficient   std. error   t-ratio   p-value 
  2008 Estimates:
  Constant      6.57763      0.534344     12.31     1.46e-23 ***
  DaysBefore   −0.0350667    0.00826569   −4.242    4.17e-05 ***
  2012 Estimates:
  2012 Dummy   −3.08490      1.19412      −2.583    0.0109   **
  2012*DaysB4   0.0334960    0.0158528     2.113    0.0365   **

Mean dependent var   4.156866   S.D. dependent var   2.638088
Sum squared resid    771.5533   S.E. of regression   2.436191
R-squared            0.166442   Adjusted R-squared   0.147206

Let us start first with the results for 2008. The constant term measures the average margin between the candidates on Election Day, when “DaysBefore” is by definition zero. The model predicts than an average poll taken on Election Day would show a 6.6% margin between the candidates. That’s about a point tighter than RCP’s final estimate of 7.6%, but it is certainly not far off the mark.

Now consider the DaysBefore variable which measures the number of days a poll is taken before the election. This variable actually has a negative effect in 2008, meaning that as the election got nearer the outcome became more assured, not less.  Arithemetically, if we plug in 120 for the DaysBefore variable, we predict an average poll showing a margin of about 4.2% (= 6.58 – 0.0351×120). The 2008 race became less competitive as time went on, not more.

Now let us take a look at the results for the current election.  The most striking difference is that both estimated effects for 2012 have the opposite signs from their values for 2008.  The “2012 Dummy” measures the average difference in competitiveness between 2012 and 2008.  Broadly speaking this election is nearly twice as competitive as the one four years’ ago.  The model predicts that a margin between the candidates of about 3.5% on Election Day 2012, compared to 6.6% four years ago.

Even more remarkable is the evidence that the 2012 election is basically stuck in neutral.  To see the total effect of closeness to the election on the candidate margin for 2012, we must add together the estimated effects for both “DaysBefore” overall and “DaysBefore” in 2012.  Because the estimated coefficient for DaysBefore in 2012 is about equal and of opposite sign from the overall coefficent for DaysBefore, we find there is no trend in competitiveness in 2012