Friday, February 03, 2012

The Silver bullet for election prediction may need lengthening

Nate Silver argues that job growth during a presidential election year is the best single indicator of whether a president will be reelected.  After a series of calculations, he pegs Obama's"magic number" for job growth in 2012 at 150,000 jobs gained per month.

I have neurotically tracked Obama's trajectory -- approval ratings and national unemployment numbers -- against Reagan's for 2-3 years, and my intuition is in sync with Silver's conclusions (and partly shaped by his prior analyses, no doubt). Far be it from me with my sixth grade math skills to quarrel with the intricacies of his model. But there's one basic assumption that I would quibble with: that the relevant periods are necessarily defined by the calendar year.

Silver argues that the actual year of election matters a lot, and the three years prior, not so much. When conditions are bad early in a president's term, he may even get extra credit for a late-term surge. More specifically, Silver finds
that job growth during the third year of a president’s term has a positive effect on his re-election odds, while the coefficients attached to the first two years are negative.

But none of these results are statistically significant or particularly close to it; only job growth during the fourth year of a president’s term has a clear effect.
True, perhaps, as far as it goes. But why break the term into years? Why assume that voters' relevant perceptions track the movement of the earth around the sun?

As Silver indicated, the ur-case for unemployment figures tracking closely with election results is lucky Ronald Reagan, who in Silver's model benefited from the 2012 equivalent of 487,000 jobs created in 1984.  Yet my own personal break point for comparing Obama's home-stretch performance to Reagan's has been July in the year prior to election -- because July 1983 is the month in which measured employment surged most dramatically for Reagan, from 10.1% to 9.4%. While that particular break point may be arbitrary, the broader point is that the unemployment rate shrank most dramatically for Reagan in the second half of 1983, from 10.1% in June to 8.0% in January 1984. It fell further, to 7.4% in October 1984.  But morning in America shone brightest at dawn, in late 1983.

While the 16-month starting line may be arbitrary, I found six months ago that it seems to work pretty well on an admittedly small data set.  Here's what I posted then:
Might it be postulated that the direction in which employment moves in the 16 months prior to an incumbent president's reelection bid is a fair predictor of his political fortunes?  The postulate works for six of the seven incumbents since Nixon, the earliest president for whom I have monthly unemployment numbers. Based on yearly unemployment figures, it seems also to work for Roosevelt and Eisenhower; and while the data for Truman is neutral, the rate was extremely low to begin with, and recent research indicates that economic growth was very strong.

Here are the figures for each incumbent, with the one exception to the 'rule' highlighted:

Nixon:    July 1971 - 6.0%; Oct 1972 - 5.3%

Ford:      July 1975 - 8.6%; Oct 1976 - 7.7%
Carter;   July 1979 - 5.7%; Oct 1980 - 7.5%
Reagan:  July 1983 - 9.4%; Oct 1984 - 7.4%
Bush Sr: July 1991 - 6.8%; Oct 1992 - 7.3%
Clinton:  July 1995 - 5.7%; Oct 1996 - 5.2%
Bush Jr: July  2003 - 6.2%; Oct 2004 - 5.5%

Roosevelt:   1933 - 19.8%; 1936 - 16.6%; 1940 - 14.6%
Truman:       1947  - 3.9%;  1948 - 3.8%
Eisenhower: 1955 - 4.4%;   1956 - 4.1%------

Perhaps Silver's 'election year' needs a bit of a stretch. If so, the drop from 9.1% in August 2011 to 8.5% in December may be relevant. On the other hand, since the job gains in the second half of 2011 were close to  Silver's 'magic number' for Obama's needed monthly job growth, any adjustment along these lines wouldn't have much predictive value.

I wonder, too, how the unemployment rate and other key economic measures register with voters. Is the overall perception of how the economy is doing the sum total of people's personal impressions, based on their own fortunes and those of people they know?  Do the top-line numbers reported each month filter into general consciousness, either as news points or as shapers of the perceptions of opinion shapers? Some combination of all of the above?

No comments:

Post a Comment