Yeah, I'm not so sure about that model. He isn't very transparent about what his inputs are, or how the model is constructed. He first started publishing his predictions in the 1996 election, and his "model" was right about 85% of the time. The claim that his model predicted the last 25/27 elections correctly is pretty meaningless. The reason is because he used 21 of those results in building his model in the first place (all elections prior to 1996). It's how models work, generally. You use
historical data and results to develop a model that will use
current data to predict a future result.
Here are some of his claims:
The Coronavirus outbreak appears not to have damaged Trump’s re-election prospects. To the contrary, his job approval rating ticked up and still holds at 49% in the May Gallup Poll.
Trump's approval now sits at 38%.
In fact, the battle against the virus may help his re-election prospects by casting Trump as a wartime president, as happened with other presidents.
This is a forgone conclusion, I think. It did not.
This [wartime president shtick] may also help mitigate electoral damage from the economic fallout such as over 30 million losing their job.
Again, I think not. If anything, it made it worse.
The Media-ite interview says the following:
The model calculates a candidate’s chance of winning based on their success in early presidential nominating contests, putting former Vice President Joe Biden at a severe disadvantage because of crushing losses in his party’s first two presidential nominating contests. Norpoth said his model succeeds by placing an emphasis on how much enthusiasm candidates are able to generate early in the nominating process, and by discounting public opinion surveys.
If this is true, then the crux of the 2020 prediction is based on the fact that Biden didn't do well in early contests. This seems dubious on its face.
Also, I note that in 2016 he said the following:
“Trump beats Hillary 54.7 percent to 45.3 percent [of the popular vote]. This is almost too much to believe.†Norpoth said, with a few members of the audience laughing nervously. “The probability of that [outcome] is almost complete certainty, 97 percent. It’s almost ‘Take it to the bank.’ â€
Think about that for a moment. He predicted with 97% certainty that Trump would get 54.7% of the vote to Hillary's 45.3%. That's Trump +9.4%. The final result was 48.2% for Clinton and 46.1% for Trump. That's Clinton +2.1%.
While his topline prediction for 2016 happened to be right, he was exceptionally far off on the popular vote.
Here's how the polls have fared over the last 30 years:
The polls seemed to miss the mark for a few reasons. First, they
were biased in favor of Clinton by about 2-3%. Second, he had an unusually high electoral college advantage of ~3%. This effectively resulted in a 5-6% swing from what the polls showed. Trump would likewise need a series of things to go well for him for him to win this time around.
The 2020 section is there for illustration. Obviously there isn't a final result, but what we can note is that polls were 2.2% better for Trump throughout the year than they are now. And, if we compare the current national polls with the current tipping point state polls, we see an electoral college advantage of about 2.3%, which is roughly in line with what it was last time, with a modest reversion to neutral.