26 October 2016

Thought For The Day, 26 October 2016

The latest in The Donald's mindless jabbering is the execution of "oversampling" in polls. As any drug developer knows, with enough observations one can find stat sig in a ham sandwich. That's what oversampling does: narrows the CI/lowers p-value. It doesn't change the point estimate one way or the other systematically. Oversampling won't make The Donald look like a bigger loooooser or Hillary a bigger winner. It will only solidify the accuracy of the point estimate. Moreover, the technique isn't, on the whole, used in public facing polls, but by campaign's internal polls where it wants more accurate estimates of sub-groups known/suspected of being heterogeneous.

So, I just did a simple search, and lo and behold, Pew Research (The Donald's source of oversampling evil) had this to say:
Oversampling is the practice of selecting respondents so that some groups make up a larger share of the survey sample than they do in the population. Oversampling small groups can be difficult and costly, but it allows polls to shed light on groups that would otherwise be too small to report on.

Even the non-brained dead rightist organs (or some anyway) know the truth.
The conservative blogosphere is lighting up again with accusations of polling bias against Republican presidential candidate Donald Trump in his race against Democratic opponent Hillary Clinton. However, Trump supporters should avoid giving into this temptation to assume unfavorable results must be biased results. Clinton really is leading Trump, and by nearly 6 percentage points.
[June, 2016]

No comments: