Friday, November 20, 2020

Fool me twice

On Tuesday, November the 3rd 2020, as I sat eagerly to watch the results of the Presidential Election, very quickly my hope of seeing the country get rid of Trump turned into despair. In the first few minutes, the Florida results were encouraging. Then Miami-Dade county reported and the shock I felt will remain forever etched in my memory. Hillary Clinton won that district by 35 points, and still lost Florida. Biden won it by around 15 points or so, and it was immediately clear that he is not going to win Florida.

As they say, it was “deja vu all over again”. Four years ago, it was Florida that delivered the first shock and things got progressively worse throughout the night - for people like me who wanted to see the country rejecting Trump. This year, it seemed like a replay of the same movie. Both times, the pain became worse because I wasn’t prepared for it. The polls had left me no doubt that Trump was going to be defeated, and both times, the polls got it horribly wrong.


In 2016, when Trump won, the polls became a joke in the eyes of the most. I was also surprised how the polls got it wrong. I decided to spend time figuring out why. As I studied the issue, I realized that I didn’t really understand what the polls were actually saying. Of course, the polls got many things wrong, but their error was pretty much in line with their historical performance.


Even in 2012, the polls had underestimated Obama’s chances of winning reelection. But they got the direction right, and hence the polling error did not get too much attention. The polls were actually not as wrong in 2016, but the error margin was bigger than the slim vote margin that went in Trump’s favor. More importantly, Trump’s margin of victory was far less than the “undecided” percentage of voters reported by polls. That was the key to understand why we all though the polls were wrong. As the election neared, there was a clear movement of these “undecided” in Trump’s favor, which many missed, including the media and me. Then there was the last minute bombshell by Comey which sealed the deal, and there were few polls after that announcement. 


Many theories such as “the shy Trump voters” have been debunked in many studies since the 2016 debacle. The real reason, the pollsters argued, was that they didn’t account for the correct proportion of rural voters, people without a college degree etc. In other words, their sampling technique needed refinement to capture a better representation of the electorate.


So I defended the polls and prediction models such as the one at FiveThirtyEight. I argued that a few swing states, going in Trump’s favor by a very slim margin, is definitely within their prediction of a 30% probability of Trump’s win. In life, we shouldn't be surprised by events with 30% probability actually happening. For example, we are not at all surprised to see a pair of dice rolling to 6 on each, which has a much less probability (less than 3%). 


This time, I thought I was interpreting the polls correctly. The polls were indeed much better in 2018 midterms. In 2020 as time progressed, they tended to move more strongly in favor of Biden. That was not the case 4 years ago when the polls were fluctuating a lot. There were hardly any “undecided” in the polls this time, and Biden’s margin was much much higher than Clinton. To top it, the polls were consistent with each other, across different pollsters, across states etc. 


So I made my prediction for a Biden win, and doubled down on it by giving a 50-50 chance for a landslide victory by Biden. I even offered the bets with specific odds. Although I did not lose the bets, and got a few things right, I was wrong about lot more things.


Now the way ballots were counted, made it seem like a close and tough win. Actually it was quite a comfortable victory for Biden. When Trump won it last time, many called it a “clear mandate”. Biden’s 2020 electoral count is going to be the same as Trump in 2016. Biden’s popular vote margin (of close to 7 million votes) is even larger than what Clinton got. By the way, my estimation of the margin was quite right. Biden’s margin in swing states is much much higher than what Trump had in 2016. Finally, Biden won back the Blue states Clinton lost, and flipped 2 Red states in addition. So by any yardstick, this is a very nice victory. Just the way counting was done, made it look like a tough win.


In spite of that, I have to say that my prediction was quite wrong.


Most of the arguments I made were based on eroding support for Trump. That was completely incorrect. Trump got even more votes than he got the last time (close to 10 million more), and was able to get more votes from Blacks and Latinos than he got in 2016. There is a lot for me to understand there.


Mainly, I wasn’t predicting just a Biden win. I was giving a 50-50 chance for a landslide. As it turned out, there was really no chance of a landslide. I made that prediction thinking Florida will go to Biden and Texas will be a close loss. This was based on polls. For example in Florida, the polls continued to shift towards Biden as the election neared. Most highly rated polls gave Biden a +5 chance, when the reality was he was -3. That was an astronomically huge miss by the polls. When most swing states are decided by a margin of less than 2%, any poll that can be wrong by 8-10% is worse than being useless.


Florida wasn’t an exception. Iowa, Texas, North Carolina - none were as close as the polls claimed to be. And wherever Biden won, the polls had estimated a much higher margin.


The common theme in the errors made by the polls was - they underestimated the support for Trump. Just like they did in 2016. Except this time, the pollsters had told us that they have learned from their mistakes and have made the necessary adjustments. If this is how anyone “improves” in their job, they would be fired.


Fool me once, shame on you. Fool me twice, shame on me.


Hence, I am declaring that polling is useless for me. Yes, they will try to explain why they got it wrong this time, or how it wasn’t as wrong as we perceive. Yes, they might get it right the next time. But I am not going to trust them again. Ever. And I know, I am not alone. Even though Biden won as was the prediction, many people understand that polls were really inaccurate this time - especially in the purple swing states where they matter the most. The entire polling industry has lost its trust. Even if they get it right the next time, it won’t increase their trustworthiness any more than throwing a dice.


Nah. I did not make a good prediction. Live and learn. Well, at least I won’t waste any time and suffer heartburn in the next election, by engaging in studying polls and coming up with a prediction model. 




No comments:

Post a Comment

Related Posts Plugin for WordPress, Blogger...