Trollminator wrote:
Rigged for Hillary wrote:
His model is based on primary results and it has been an accurate metric to predict presidential elections. Norpoth has a much better batting average than 538.
Norpoth > 538
So the guy is intentionally ignoring new data for the sake of staying invested in a losing position... ok
To be more precise, his primary model is based on results from February. People were aware of COVID back then, but it had not started to get out of control yet and obviously we all know what happened after that. Trump's handling of COVID is without a doubt a massive turning point in his term. Norpoth's model ignores all that and pretends nothing has changed since February.
Not saying it doesn't work this way with other models, but his takes a few basic indicators and looking back it has "predicted" most elections correctly (that's what most models achieve btw). When you consider that an election result is binary, that's not very impressive. Let's peel the onion a bit.. for 2016, the model predicted a Trump win, but with a popular vote win (that part obviously didn't happen). For 2008, the model predicted it would be a tight contest and it wasn't even close. You see the glaring flaws... the less variables and flexibility a model has, the weaker it's predictive power will be. What happens with models like Norpoth is that people start to look at them like it's some sort of witchcraft because the results seem impressive, when it's just basic stats that are bound to go it terribly wrong at some point. After the election I bet the model changes.. they'll say they would have gotten it right IF COVID didn't get out of control. They add one of two new factors to try to correct for the blunder. Here's another way to look at it. If trump passed away in May would you believe any model that predicts a win for him? It's kind of insane not to take all relevant information into consideration.
The other thing about Norpoth is that it uses the primaries to gauge unity and enthusiasm of the candidates. This is why Biff keeps banging the table about enthusiasm. But lets see, Trump was the uncontested incumbent and most presidents win re-election. Not a very revealing set of factors. Where the model fails miserably is that it doesn't take into consideration trump's unique ability to get more than half the country to despise him. Sure, that wasn't reflected in the Democratic primary until the later contests and indeed we saw a Biden landslide, which when you consider the robust support for Bernie that is extremely impressive and indicative.
So why trust 538's predictive power over Norpoth's? Pretty simple, the underlying polls have been enhanced and Nate's aggregation covers more than before. Also, 538 rationally doesn't see a reason to hang on to some probabilistic outcome when the data is changing drastically. Anybody who thinks COVID hasn't impacted trump's chances is nuts.
Last but not least, Biff you're an idiot.
