So the Lichtman Test so far

[SteveRogers]

The Lichtman test is overrated.

It's worked on all but one election in the past 100 years.

But you understand that it didn't predict all those elections, right?

[SteveRogers]

The Lichtman test is overrated.

It's worked on all but one election in the past 100 years.

Well, sure it does great when you know what result you're aiming for when you retroactively decide the subjective ones.

It was created in the 80's and has been used to predict every election from 1984 to the present. Again, all but one time it was correct. Then it was retroactively applied to elections going all the way back to 1860 and it still worked.

Um, nope. Nope nope nope. Lichtman looked at every election from 1860 to 1980 and identified common factors in order to create the model. The model cannot predict the data points that were used to derive the model in the first place. When you plug in an election from 1860 to 1980, the model spits out the right answer because it was designed (overfitted) in the first place specifically to account for each and every one of those elections.

As for whether or not the model got 2000 right, doesn't that just kind of go to show the inherent uselessness of a model for U.S. presidential elections that doesn't even purport to tell you what percentage of the popular vote a candidate will win by?

[SteveRogers]

The Lichtman test is overrated.

It's worked on all but one election in the past 100 years.

Well, sure it does great when you know what result you're aiming for when you retroactively decide the subjective ones.

It was created in the 80's and has been used to predict every election from 1984 to the present. Again, all but one time it was correct. Then it was retroactively applied to elections going all the way back to 1860 and it still worked.

Um, nope. Nope nope nope. Lichtman looked at every election from 1860 to 1980 and identified common factors in order to create the model. The model cannot predict the data points that were used to derive the model in the first place. When you plug in an election from 1860 to 1980, the model spits out the right answer because it was designed (overfitted) in the first place specifically to account for each and every one of those elections.

As for whether or not the model got 2000 right, doesn't that just kind of go to show the inherent uselessness of a model for U.S. presidential elections that doesn't even purport to tell you what percentage of the popular vote a candidate will win by?

The only reason it got 2000 wrong was because 2000 was stolen by Bush.

That's not actually a response to any of my points.

[SteveRogers]

Key 4: There is no significant third-party or independent candidacy.
Licthmann defines "significant" as earning 5%, so FALSE.

Um, [Citation Needed]

Though I guess there's really no way to prove that one true until after the election. Man this test is dumb.

[SteveRogers]

Key 4: There is no significant third-party or independent candidacy.
Licthmann defines "significant" as earning 5%, so FALSE.

Um, [Citation Needed]

Though I guess there's really no way to prove that one true until after the election. Man this test is dumb.

I kept that from the original post, so I don't have a citation.

Ah I see. That one should be true then though.

[SteveRogers]

Isn't the charisma one so subjective as to render it useless?

[SteveRogers]

Key 4: There is no significant third-party or independent candidacy.
Licthmann defines "significant" as earning 5%. Webb isn't going to do that and it's probably too late for Trump to go indy, so FALSE.

Shouldn't that be listed as true, then?

Key 11: There has been a major military or foreign policy success during the term.
TRUE. The opening of Cuba relations and Iran deal would definitely qualify.

If you say so. Most people don't care about either of those things or even have a negative view of them.

The real problem on Key #4 is whether the Third Party nominee draws significant support from the incumbent's Party. So if some conservative independent runs and gets a significant support largely from Republicans, then the test would say FALSE when the effect is basically TRUE.

Third-Party nominees hurt

Democrats in 1948
Democrats in 1968
Democrats in 1980
Republicans in 1992
Republicans in 1996 (but the test would have shown Clinton being hurt).

This ^^^^
More to the point, Key 4 seems useless as a predictive tool since in elections where there is a noteworthy third party candidate this will be the one key that cannot be called until after the election. Look at 2000: In June and July of that year, Nader was polling at 5-6%, so at that point that key could be considered turned against the Democrats. And in the last 2 months of the election Nader's polling still fluctuated between 2 and 4%. But of course Nader finished with only 2.7%, so Key 4 resolves as a TRUE after the fact and the model is praised for correctly "predicting" Gore's victory in the popular vote.  Ugh.

Same thing in 1980. If you were looking to the Keys at this point in the 1980 election you'd confidently declare Key 4 turned to FALSE since John Anderson was polling in the mid-20s. But of course he only finished with 6.7%, a photo finish for Key 4.

Same thing in 1996. You wouldn't have known until election night whether Perot was really going to get over 5%.

Ugh.

[SteveRogers]

Stop.