James Dowd Breaks Down Our "unoffical" just-for-fun completely non-scientific poll (Surprise Our Margin Of Error Was Better Than Gallup)
You guys aren’t going to believe this-
But your average margin of error for your “unofficial” just-for-fun completely non-scientific poll was 2.48.
That’s better than Rassmussen
That’s better than Gallup
The only four errors of the poll were, and I predicted two of these:
-undershooting Romero: anybody knows that the Godmother is Queen. A huge portion of her people don’t take online polls.
-overshooting Lundberg- He still did well, but your sample favoured him higher than reality.
-Kimberly was a fluke, possibly due to her association with GMG. Known online but not IRL perhaps
-Favazza fizzled. The surprise of the night. Don’t know what happened there.
BUT OTHERWISE THIS THING WAS SO FUCKING SPOT ON
And the two big errors were easy to spot.
Here is the spreadsheet where I take the raw GMG numbers and compare them on the left to the outcome. On the right, for fun, I took the GMG numbers and then try and “correct” them- I did this the day before the election to see how accurate I could be, if I could improve on the raw. It turns out, if you were betting- GMG came in at a 2.48 margin of error, and my “corrected” numbers at 3.78 in the candidate-based questions. You would have been better off going with the GMG raw even thought it had some obvious errors like the Romero whiff and over-reporting/under-reporting support for some school committee candidates. GMG also loved it some Lundberg more than the rest of Glocuester, but it didn’t matter in the outcome.
My hypotheses as to why this is revolves around the fact that GMG probably accurately reflects the opinions of a large class of “likely voters” and is therefore more likely to push through the noise with it’s large sample size. It’s limits might be that it underepresents SoRo (south of the Rotary) to the favor of Eglo, Magnolia, Wheeler’s and outsiders. But still- 2.48? Amazingly close predictor.