Today will see the Electoral College confirm Donald Trump as the new President of the US.
I predicted an Electoral College landslide for Clinton.
Boy was I wrong! Why?
The vagaries of the Electoral College gave Trump a win even with him losing the popular vote by a significant margin.
Perhaps FBI Director Comey’s reopening of the email investigation was a factor. Perhaps Russia was a factor.
The race narrowed as polling approached. Was that Republicans “coming home”? Who knows? But the gap between them in many polls approaching Election Day did have Trump within the margin of error, i.e., his numbers were within that +/- 3% margin.
But I’m not alone in still wondering “Why?”
Maybe pollsters are too hung up on margins of error. It’s no use having a small margin of error if your sample is unrepresentative. E.g., if you’re a polling company that only phones landlines then you’re missing all those people who only have mobiles. And perhaps people with only mobiles tend to be younger?
Having an unrepresentative sample can happen by accident or by design. When it happens it’s often because of a sampling bias, e.g., only calling landlines.
Or let’s go back to the Presidential Election of 1936 in America – when telephones were a luxury. The Literary Digest ran a poll that predicted Franklin D. Roosevelt would get 43% and that his opponent would get 57%.
The actual result? Roosevelt got 62%.
Yep – the poll was off by 19%.
And one of the problems was that, even though they had a tiny margin of error, they used mainly telephone directories as well as club membership lists and magazine subscriber lists.
Large swathes of the population were excluded from the sample.
A small but representative sample is better than a large but unrepresentative one!
Who did FDR beat? Lost to history! And when did the Literary Digest meet its demise? Just 2 years later – 1938. And who did correctly predict FDR’s win? Using a much smaller sample? Gallup – and they’re still going strong!