Sep

19

I will keep to my promise not to make any predictions before October 9th; but - like the Penguins (who have replaced the Peyton Manning Broncos as my favorite winter team) I have to get ready for the real political prediction season by working through the necessary exercises and drills.

There are two big things for political prediction: (1) getting the sample - like trades - correctly sized and (2) guessing the turnout - what those you who trade call the trend.

Last year's Senate race in Alabama taught me a lesson that I should have already learned from reading what LW and the other pros have said about trends. The only proper times for guessing turnout are when people are going to the polls; anything but recent "history" is, by itself, no guide at all.

What you can do early on, before the political season starts for real, is being figuring out what the proper allocation of the actual voters (Republican, Democrat, Independent) will be. Trump's odds in 2016 were far less of a longshot bet than the "expert" journalists and pundits said, once you at the cross-tabs in the polls. In an national election in which the Republicans controlled both the House and Senate and a clear majority of statehouses and governorships, the respectable polls and pundits were still allocating to the Democrats and Democrat-leaning Independents over 50% of the likely voters.

The drill this week is to read all the recent polls that are NOT about the elections but include cross-tabs that define party affiliation. First up is one from Harris surveying "Blue Collar" workers.

Their unweighted sample shows a break-down of 1049 respondents as 353 Conservative, 444 Moderate and 252 Liberal. This is hardly surprising, given that the sample is heavily weighted toward males (628 Men, 421 women). Yet, even before we get to the pollsters own weighting, the thumb begins pressing down on the scale. Somehow, when the question of party affiliation is asked, 80 respondents disappear from the sample base and the remaining 969 report themselves as being 333 Republicans, 327 Independents, and 309 Democrats.

A sample that was 33.7% Right(Conservative), 42.3% Middle (Moderate) and 24% Left (Liberal) magically becomes 34.3% Right (Republican), 31.9% Middle (Independent) and 33.8% Left (Democrat). The party of segregation, slavery, racial quotas and unConstitutional naturalizations finds its allocation increased by 40%.

Harris' weighting of their sample show comparatively minor bias. The weighted sample for Party affiliation does add back 36 respondents and give them all to the Democrats; it also finds 19 respondents who, in going from unweighted to weighted sampling, have magically changed their affiliation to the Democrats (10 from the Republicans, 9 from the Independents). But these are - compared to the slight of hand in the unweighted sample from political sentiment to party affiliation, relatively minor changes; in percentage terms the Democrats only gain 1 point and the Republicans and Independents shares only lose a half point each.

To their credit, the Harris people do keep the complete survey data for their weighted sample of political sentiment; yet even there the Conservatives find themselves losing 30 respondents, while the Moderates gain 7 and the Liberals 23.

Preliminary conclusion: If 2016's numbers were distorted by sample bias, the ones for this year are trending towards being being actual works of fiction.

anonymous writes: 

Some good thoughts there.

Mr Market has already made a pretty hard-and-fast prediction though, and, rather cryptically says "Nothing changes, it;s the same bull market we've been seeing since early 2016, just a little noise in Feb and March of this year, and back to the same chorus."

Whether or not this means the party in power in the various chambers change is uncertain, but the bass line will remain the same.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search