We evaluate pollsters by ignoring the subjects of their polls and examining the patterns hidden in the numbers themselves.

Check our twitter (@imaginarynums)

Advertisements

Parsing the polls

We evaluate pollsters by ignoring the subjects of their polls and examining the patterns hidden in the numbers themselves.

Check our twitter (@imaginarynums)

Advertisements

%d bloggers like this:

Gentlemen, greatly enjoyed your analysis of the R2K poll results. There was another aspect of the parity in the data that struck me as peculiar, which I do not believe was mentioned in your analysis: the numbers always add up to 100%. This could be a clue as to the origin of the improbable correlation.

Consider a poll of 1000 respondents. 494 answered “fav”, 492 answered “unfav”, and 14 answered “huh?” (which was put down as “undecided”). Raw per centages then would be 49.4% fav, 49.2% unfav, and 1.4% undecided; but the rounded figures would be 49%, 49%, and 1% — which add to 99%. In any poll like this there is a very good chance that sum of the rounded figures will not be precisely 100%, even for as few entries as provided in your report, as I am sure you know better than I — yet they all do.

Why is this relevant, besides being yet another anomaly? One consequence of the parity match between the male and female responses is that the sum of these responses will always be even — whether the 2 figures themelves are even or odd. Here’s my conjecture: in order to make the rounded figures always add to 100%, the pollsters “adjusted” the sum of the undecided vote such that it was always even; that is, could be divided in 2. (I am not saying that the undecided sum was always then divided in half, only that it could be; that way, as long as the values for men and women were adjusted away from this central figure by the same delta, the result would be 2 numbers matching in parity.) The result of this one “adjustment” would then cause the two other categories to always match in parity as well. This is a kind of “minimal evil” explanation, akin to “minimal entropy” constraints sometimes imposed on experimental data. Not a very good explanation, I admit, but the best I could come up with on short notice.

Given the incredible 776-out-of-778 matches, one can’t help but wonder at least as much why the few are different, along with why the many are the same. It would be interesting to know what values were reported for those few cases where the parity *did not* match.

Appreciate your efforts. Thanks and regards,

+jtg+

We tried to find a pattern in the 3 answers which violated their gender parity rule. None was obvious.

If you force T, M and F to be integers and set T=(M+F)/2 you then force M and F to have the same parity. That was our immediate first thought. We didn’t think it was necessary to speculate on the initial blog, hence dropped it from the drafts.