A little more than a week ago I was talking to a friend who was panicking about Clinton's slipping numbers in the polls. I poo-pooed his concerns, noting that Trump had no plausible scenario to win if you consider the electoral map. "Look at 538" I said. "I am looking at 538," he replied, "Clinton's chances of winning are good, but they keep going down. Where is that trend line going?"
I claimed that Clinton's decline will eventually level off. Also, I said, even if you project the current lines forward to election day, she still would end up over 50%. My friend insisted I was wrong on the latter point. While he also thinks that Clinton's declining percentage will not continue all the way to election day, if it did, the line would definitely cross 50%.
Neither of us looked at the 538 site as we debated where the trend line would go. It wasn't that serious of a discussion. In the end, we didn't disagree that much. We both think that Clinton will win in November, but we are both concerned that it won't be a crushing victory as it should be to really discredit Trumpism. But we never reached an agreement about what would happen to the trend lines on the 538 chart.
I have been thinking of that conversation ever since. This morning on the train to work I realized I could resolve the trend line question. 538 actually has three election projection models, each with its own chart. The default when you load the model is the "polls only" projection (see here for a description of the methodology behind each of the three models). If I project Clinton's current trend forward you get this:
In other words, if the current decline continues under the "polls only" model, Trump would have a significantly higher percentage of winning than Clinton. Conclusion: I was wrong.
Here is what happens if I project Clinton's current trend forward under the "polls plus" projection:
So if Clinton's decline continues under the "polls plus" model, Trump would have a slightly higher than 50% chance of winning. Conclusion: I was wrong.
Here is what happens if I project Clinton's current trend forward under the "now cast" projection:
If Clinton's decline continues under the "now cast" model, Trump would have a significantly higher chance of winning. Conclusion: I'm wrong again.
So there you have it, I was completely and utterly wrong. (Slightly less wrong under the "polls plus" model, which I think is the best model of the three, but still I was wrong.) Except the part about Clinton's downward trend won't continue. I still stand by that. It hasn't been proven wrong yet!