Friday, December 2, 2016

My apology


I've talked in person to multiple people who followed my blog. They tell me that my numbers helped them stay calm.

I wasn't trying to keep people calm. I was trying to get it right.

I failed massively, and for that I am deeply sorry. I have always distrusted the pundits, and after 2000 I wanted to do something on my own to get numbers I could trust. I was trained as a mathematician, but I'd never done statistics. Once I learned stats and learned how fast and loose the field is with methods, I decided to make my own system.

My system told me the electoral college was close in 2004 and it was. Bush needed to take both Ohio and Florida to win. Bush won the popular vote by 2.5%, Florida by 5% and Ohio by 2.1%, so it wasn't that close in terms of the real vote. Obviously, it was nowhere near as close as 2000, but it wasn't a landslide by any means.

The Wall Street Journal found my website in 2004, which was hard to do back in the day as it was a hidden page on my now-defunct website about Pascal's Triangle. They asked me for a prediction by the Sunday night before the election. The polls were close and my system said Kerry was the favorite because Bush's chances to sweep Ohio and Florida were under 50%. I made my prediction, but polls on Monday turned the tide slightly and Bush was a favorite. Still, my Sunday numbers were printed and I screwed up.

Obama's wins weren't close, and my system predicted the states and the district contests in Maine and Nebraska very well. It also did very well on the Senate races in 2012. These positive results made me confident, but I kept looking at where things might go wrong. I had no idea things could go as wrong as they did.

In previous years, I would miss one or two races. In the 56 contests for the electoral college in 2016, I got six wrong and all of the upsets favored Trump. In order from least surprising to most, I missed Maine's District 2, Florida, North Carolina, Pennsylvania, Michigan and Wisconsin. Of those six losses, only Michigan and Wisconsin are truly slender leads, and Pennsylvania is somewhat close.

I honestly don't expect recounts to change anything, much as I don't think Gov. McCrory's hope for a recount in North Carolina is going to go anywhere.

Let me ask myself some questions in public.

Did I have any doubts? Yes, but I didn't have the numbers that made me think the doubts were worth mentioning. Let's look at the three states now looking at recounts.

Wisconsin: 31 polls since May, Trump never lead in even the most optimistic Republican pollster's numbers. The median of the past few weeks said a 7 point lead for Clinton.

Pennsylvania: 54 polls since May, and only Google Consumer Surveys, a new polling company that often disagrees with the consensus, gave Trump a lead. The median said a 4 point lead for Clinton.

Michigan: 28 polls and only one poll even gave Trump a 50-50 chance, another new company named Strategic National. The median was a 2 or 3 point lead.

In simple terms, I had no "what if" I could ask that made these now crucial states look close.

How did the polls screw up so bad? That is a question I can't answer with certainty, but one of the known problems with polling is getting too many "refused to answer". Beyond the fact that Clinton and Trump combined might count for 90% or less in a poll when they averaged about 95% in the actual contest, there are the people who refuse to pick up the phone because they have caller ID, or the people who answer but hang up when they hear the noise of a call center in the background, or even someone like me who might stay on the line but be rejected because I truthfully answer the question about being a public employee. I have no way of knowing how the "refused to answer" crowd will vote or if they will vote at all. This might account for the difference, but I can't say that with any mathematical certainty, as I said before.

Any other factors look important? This was a very strange election and a lot of the strangeness is a matter of public record, not some conspiracy based nonsense. My problem is I have no way to quantify any of them. The oddest things to my mind were Comey's decision to try to sway the vote, the one-sided nature of the WikiLeaks information and the attempts by Russia to assert influence.  Another strong pro-GOP influence were the efforts to discourage voters in many states, whether it was ID laws that were difficult to comply with or allocating minimal polling places in areas with strong Democratic tendencies.

While I know for a fact all these things happened, I don't trust any of the numbers I can get to determine how important these factors were.

And so now we have President-elect Trump. I have no idea how bad things will get, but I have little hope these four years will work out for the best. When it comes to domestic policy, we will have to see how much in lock step the Republicans are on their worst ideas, which for me would have to include the privatization of Medicare. When it comes to foreign policy, my greatest fear is we have a gang who thinks a shooting war with Iran is a good idea. Those are just two of my policy concerns off the top of my head. When it comes to tone of our politics, it will now get worse than it was under Obama, when it was already not good.

If my numbers said this was a serious possibility, I would have made that as clear as I could. But they didn't say that and so I passed out incorrect information, believing it to be reliable.

For that I am truly sorry.

3 comments:

  1. Well, friend, it's hard to condemn you when you were following the best numbers. I too was comforted by your predictions, even though I know that wasn't your intention. I honestly can't even hypothesize reasons without sounding like a nut--that's how this reality has left me feeling. It's been weeks and I still can't believe it. I just can't believe it.

    ReplyDelete
    Replies
    1. I feel the same way, sweetie. I'm not sure how the numbers I counted on went so wrong. I may change my assumptions next time, though right now I think 216 is my last time.

      Delete
  2. Like others have said, I also felt comfort in your analyses, and relayed some of that to friends of mine. But while I will accept your apology if it helps, I don't think it is necessary at all. After all, you were hardly the only one producing similar results, which kind of indicates that the underlying problem was in the data being produced, right? Anyways, I don't feel there is any fault or blame to be laid at your feet.

    From what I am seeing, it looks like the polling data was not representing the fervor among the white supremacists and the aggrieved suburban white voter, and really overestimated how much Trump's hateful and bigoted rhetoric would be turning off minority and women voters. As you say, in States like Wisconsin there was a strong voter suppression factor, and I wish there was a way to get a handle on that number.

    Anyways, I still thank you for your work during this and prior election seasons.

    ....now that I think about it, I also apologized to my friends who I tried to calm using your analyses, so....

    ReplyDelete