538 version:

HOR: 5 in 6 the D side wins control

Senate: 2 in 3 the R side maintains control

Stefan Jovanovich writes:

538 offers us their assurance of the accuracy of their 3-headed model by telling us how successful it has been. After all, their accuracy score in actual election predictions has been between 95.7% and 96.9%. Or, to put is another way, their inaccuracy score has been between 3.1% and 4.3%; on average, they have been wrong in only 3.7% of their predictions.

That looks wonderfully impressive until you remember that, in House of Representative elections in the United States, you have to look through the small end of the telescope. Since the passage of the Permanent Apportionment Act in 1929, only 22 of the 44 elections have seen a change in either party's seats that was greater than 538's strikeout average.

If you limit the sample to the current period in which the House changed hands between the two major parties - i.e. 1994-present, there have been only 4 elections out of the 12 that have seen a change greater than 3.7%.

Their astounding "accuracy" is built into the game. Since 1929 the mean change in the number of seats has been 1 (.2%) and the median change has been 3 (.6%). Even when you look only at the "wave" elections - those whose changes are greater than 538's inaccuracy percentage, the mean change has been 44 seats (10.2%), the median gain by the Democrats has been 37 seats (8.5%) and the median gain by the Republicans has been 46 seats (10.6%).

I trust Big Al's and others' math more than my own; but these spreadsheet calculations suggest to me that we are playing the 95% confidence interval game.

In any case, this seems to me a bit like forecasting the pennant based on teams' records in spring training. The primaries are only now ended and even "well-informed" citizens barely recognize the names of the candidates the parties have chosen. The one survey that no political polling firm will ever take is the name recognition one where citizens are asked to match their local candidates with their party affiliations; even among likely voters the results are wonderfully bad. It will not be until 4 weeks before election day that a majority of them will be able to guess the names of the people on whom the fate of the Republic depends.

At that time guessing about the likely results becomes less like propaganda and more like the speculation that we all enjoy.

Back on October 9th.






Speak your mind


Resources & Links