Man fits linear discriminant classifier on small, noisy set of data, finds it has 100% accuracy on entire set. Did not do any cross-validation to figure out robustness of fitting procedure or magnitude of generalization error.
To be fair he claims to have developed the system in 1980 and then used it subsequently to predict elections. But you're right that the data is noisy and small. And he did get it wrong in 2000 by predicting an Al Gore victory. So 7/8 correct on the "test set."
And that's why we're all worried about this thing called "overfitting".
I generally hate it when people post XKCD comics in response to more serious posts, but this one contains a compiled list of presidents who were elected despite some precedent saying they wouldn't: https://xkcd.com/1122/
I have a bachelor's degree in political science* from AU, and Lichtman was my prof for one class. Suffice it to say that prolonged exposure to him is more than sufficient to cure a person of any notion that he is a genius who can see into the future.
There is a tiny kernel of a good observation buried in his model, which is that the everyday "horse race" events that the political media focus on have very little to do with who wins and who loses in a presidential election. In fact it's pretty safe to say that campaigns themselves don't have a whole lot to do with that.
In the modern era at least, presidential elections are generally just a referendum on the performance of the incumbent. If the economy is at least stable, and there are no disastrous wars going on, voters almost always re-elect the incumbent or elect the designated successor from his party. If the economy is in the tank, or a war we're involved in has taken a particularly bad turn, they throw the in-party out and vote the out-party in. That's more or less all there is to it.
Which means that the victor in presidential elections is almost never a total surprise; by looking at the economic indicators, you can make some fairly solid predictions which party will win months before they even get around to choosing their candidate, no elaborate model required.
Lichtman's innovation was take that basic idea and then hang a bunch of bells and whistles on it, the entire purpose of which is to impress credulous reporters by taking a rule of thumb and presenting it as some kind of mechanical fortune-teller. Which, since reporters need things to report on and he provides a kind of pre-packaged story that requires no research and only one interview, invariably gets picked up.
So: every four years he gets to do a round of press, applying his "method" to the current election, which is good for his career and for the profile of the university. And of course none of the reporters circle back after the election to measure how well his Prediction Machine really performed; and even if they did, he's left enough subjectivity in his "keys" to give himself an out if his predictions turn out to have wrong.
Everybody wins, I guess? Except the people who read those stories and think there's actually something more to them than some basic historical knowledge and a dollop of PR razzle-dazzle.
* Well, actually an interdisciplinary degree in "communications, legal institutions, economics and government." But the path I chose through those disciplines was heaviest on the poli sci.
An interesting article and "keys to the White House", but nowhere does he "double down" on a Trump victory. If anything he's saying "anything is possible" for this election.
Glad the headline here is not the same as WaPo's suggestive one.
Not defending his overfitting, but it's worth pointing out that recent polls are suggesting Johnson won't reach his 5% after all. In which case (if Clinton wins) this professor can still claim accuracy.
Statistical bullshit aside, even if his stats were on point, you'd have to say Trump is a Black Swan of a candidate, breaking models generated by past data.
News at 11.