Polls: Will We Never Learn?

The results of last night’s Iowa caucuses were yet another stake in the heart of the polling business. After epic disasters in the 2014 Senate races, the 2015 leadership races in Britain and Israel, the industry covered itself in shame yet again. The final average of the polls collected by Real Clear Politics said Ted Cruz would do far worse than he did—4 points worse—and Donald Trump would do better than he did by 4. Marco Rubio overperformed the polls by 6 full points. Most important, every one of the 13 final polls in Iowa had Trump winning, when of course Cruz prevailed by a comfortable margin. On the Democratic side, Bernie Sanders outdid his polls by 6 points, and Hillary Clinton outdid hers by 2.

  • DavidinNorthBurnaby

    As Dief the Chief put it “polls are for dogs” 😉

  • 2maxpower

    the Iowa case is not normal since you need to know who is polled. only registered republicans were allowed to vote. …on top of all the other games pollsters play.

  • k

    Under a classical frequentist interpretation of probability (still the most common among statisticians) if one flipped a fair coin 100 times, there is a miniscule but non-zero chance of getting 100 heads – and nothing wrong with the coin. Only if one flipped it for ETERNITY would we see exactly 50 percent heads, 50 percent tails.

    Similarly, one hears “within plus or minus X percentage points, 19 times out of 20”. That last part is the important part that folks tend to gloss over: One never actually knows whether one is dealing with that “unlucky” 1 out of 20 poll or survey (and, unfortunately, most statisticians do not condition on the single poll or survey that they have access to, but instead resort to long run, convergence arguments). Therefore, individual polls are only meaningful as indicators, if one considers them in comparison to other polls, and averages them out over a long period (strictly speaking, an infinite period).

    • Well put. The margin of error is always distorted once you start isolating subsets of data as well which papers rarely report.

      So “within plus or minus 3 percentage points” can end up being “within plus or minus 15 percentage points”

      • Justin St.Denis

        Never let these pollsters handle a budget. Math seems somewhat beyond them.

        • Clausewitz

          Their margin of error is huge. 19 out of 20 times.

  • Everyone Else

    social science is an oxymoron