It’s been about two weeks since Solid Purple formally launched, and in that time, it feels like a lot has happened and nothing has happened simultaneously. FiveThirtyEight released its forecast of the Presidential race and much to my surprise, for being a significantly more robust and sophisticated model, it came to most of the same conclusions. The Economist also released its forecast and, well...it sees this election very differently.

In the interest of full disclosure, I realize that the market for statistical analysis of the upcoming election is a saturated market and thought it was very important to get a first mover advantage. As I talked about in my post two weeks ago introducing this website, my background and expertise in this area is marginal at best. I started tinkering around one day and the whole thing just kind of snowballed from there.

So naturally, despite thinking I had crossed every t and dotted every i prior to releasing the inaugural outputs of the model, it turns out there were at least a couple glaring bugs in my code that impacted the topline results to such a tiny degree that it caused me to overlook some pretty egregious red flags in individual races.

I won’t go into too much detail only because I am frankly kind of burned out after dedicating most of my free time to getting all this in order, but essentially there were two problems with the forecast that have since been fixed:

1. The House forecast was only taking incumbency into account for some of the races, underestimating Republicans on net by about half a point on the generic ballot and causing some Likely/Solid D/R seats to behave like Lean/Tossup seats.

2. Both the House and Senate forecasts were not calculating correlations between states and districts properly, underestimating Democrats’ chances in the Senate slightly.

As a result, this shifted the odds in the House from an even 50/50 split to Republicans having a 53% chance of controlling the chamber after November, while Senate odds went from Republicans having an 85% chance of taking over to a 74% chance. I cannot promise there won’t be more revisions to come but I will do my best to document them in real time.

That said, I have decided because I’m clearly some kind of masochist, to accelerate the frequency of forecast updates over the next five months. I had initially planned to do an update monthly until Labor Day, and then every two weeks after that. The plan now is to update every week on Monday mornings starting July 1st. On my end, I can run the model whenever I want, and I do pretty much every time a bunch of polls get released all at once, but the problem is that I don’t really know how to programatically update the site and the interactives. I know it can be done, and I’m sure it’s actually quite easy, but I haven’t been able to figure it out yet. So if anything on the site looks sloppy, it’s because I’m manually going in and updating stuff every single time.

As I said...masochist.

With that, how have things shifted in the last two weeks? Honestly, not a lot. Either Trump being a convicted felon truly has not moved public opinion one iota, or any erosion in his standing among the electorate has been counteracted in surveys by partisan response bias caused by his strongest supporters rallying around him and being more likely to respond to polls. Hard to say one way or the other and I’m not sure we’ll find out since we are supposed to be having the first debate in a week or two!