Alabama Megathread 3: Results Thread

[Wisconsin+17]

50 - 48.8 Moore with only 4 percent in.

LOL at those calling this now.

[Wisconsin+17]

Moore takes the lead. 5 percent in.

Still TCTC. 51.2-47.7

[Wisconsin+17]

Moore pulling away now.

7% in 53 - 45.

Moore Max is 67. Jones max is 59.

[Wisconsin+17]

Jones max down to 57.

[Wisconsin+17]

My model is already calling it for Moore. Over 75% confidence level.

[Wisconsin+17]

Here come the vote dumps to keep it close. Backing off calling it for now. TCTC.

[Wisconsin+17]

Jones max at 54. Everyone about +4 Jones has lost.

[Wisconsin+17]

Model once again predicting Moore victory at 75% confidence level. Moore has a statistical lead here.

Estimated peak Jones is at 51.56% Moore with a 0.4% advantage.

Callable at 39% in.

[Wisconsin+17]

Callable now down to 31%. Moore with a statistical lead.

[Wisconsin+17]

Moore's lead right at the callable line. About 90% confidence level now.

Jones max still at 50.5 percent here.

[Wisconsin+17]

NYT model is way off here.

If Moore's lead holds at 4% at 50 percent in, this race is called. Holding off for now. Moore still has a statistical advantage.

[Wisconsin+17]

You must be logged in to read this quote.

Stats. Moore's been building on a statistical advantage, hovering between 75-90 percent confidence for awhile now.

After a certain percentage of votes in, the votes outstanding are much less likely to swing.

4% after 50 percent in is callable.

[Wisconsin+17]

Calling it for Moore.

Moore has won this.

[Wisconsin+17]

Jones max now at 50 percent for the first time.

This is a hilarious thread.

[Wisconsin+17]

Comedy relief, perhaps, but so far I've been right.

[Wisconsin+17]

You must be logged in to read this quote.

LOL. Well I had the first call. LOL.

[Wisconsin+17]

You must be logged in to read this quote.

My model assumes that the swing vote is no greater than 2/x where x is the percentage total number of votes counted*100. I've tested it for countless elections, and it's only once been wrong - the PA election last year calling against Trump.

So in this example, 65 percent of the vote is in.

So 2/65 = 3.07. That added to Jones gives us 45.3 + 3.07 = 48

This is under 50 percent so Jones cannot catch Roy Moore.

Within 2% I won't call until 90 percent is in. The model is less good with really close elections.

I also give it an extra percent before I call the election - a 'fudge factor' if you will so that takes care of the 1% off calls, as in PA for Clinton. I also try to call the 'downswings'. Has to have two sections above the callable point.

At 1/x the candidate has a 'statistical lead' meaning that the other candidate is officially behind.

[Wisconsin+17]

You must be logged in to read this quote.

Sure, it's something I've developed, but I've watched a lot of elections and have gradually refined it over time.

The basic assumption has been surprisingly robust, that the election total percentages swing less the more votes are in. It sounds like hocus pocus, but it's really not. It's all stats and error bars.

Bigger samples = less swing. Really that's all it is. It being 2/x was something I suspected and have tested it against every election I've witnessed since. PA was a special case.

Sure it's not as sexy as other models, but it's also turnout independent and precinct independent. It relies on being conservative 2/x vs 1/x

[Wisconsin+17]

Recount baby. Still TCTC and we won't finish the night before we know the results.

Surprised by the swing though. Only the second time it's swung that much over that much over the course of the night. It's unusual. Usually ballots over 70 percent of the electorate won't swing more than 3 percent. Tonight it did.

I'll have to check my bounds. See if it's better with 3/x vs 2/x. Not sure. Anyways, NYT saying it was Jones +10 was crazy sauce. I was right about Jones being under 50, though.

[Wisconsin+17]

You must be logged in to read this quote.

SOP with the Dim's tactics.

When is Franken going to resign?

We're gonna do a recount and keep counting. Just like you guys did to Coleman.

[Wisconsin+17]

You must be logged in to read this quote.

It's just the maths. Bigger numbers just don't shift around as much. You can clearly see the difference between my bounds and the NYT bounds. My bounds were *much* closer to the final results than the NYT predictions at that point in the race. It wasn't until an hour later that I predicted he'd be held under 50 percent.

You must be logged in to read this quote.

That's the reason for the bounds. My model went 49/50 for presidential races, it got Wisconsin, Michigan and VA right but missed on Pennsylvania due to Hillary being up so high early on the race and the votes for Trump catching up.

The bounds are supposed to account for the expected swings. It doesn't do great in close races within recount territory. What my model was trying to predict was whether Moore's gap in votes was sufficient to carry him through 100 percent of the electorate. Doing that at 43% is difficult. Was close, missed by a bit though.

You must be logged in to read this quote.

The swings were *very* exceptional. At the 70 percent mark Moore was 7% up. At the 81% mark, Moore was up by a percent. That means that of that bundle from 70-81 percent, that it was almost 80, 90 percent Jones.

That's pretty much unheard of. Never seen that before. I was saying that the NYT was crazy sauce and that Jones would be held under 50 percent. Both were true.

Basically if I refine the bounds, it should contain even that massive swing. I kept running the numbers, had no idea why the NYT was predicting Jones +10. It made no sense with Moore being up 7 even at that point.

The big assuming that the NYT is relying on is that the composition of early voters = later voters. This is GIGO early on, because the swings for the small numbers will be swingy and exaggerated later on. It's why my numbers were way smaller at the 10% margins. The NYT model is basically useless for most of the race, but it did get it right in the end. Mine was better early on and came close to getting the final result right.

Perhaps it's e/x. That will be interesting to try. I'll try running the new bounds and say what it says.

[Wisconsin+17]

You must be logged in to read this quote.

I'm the quack? LOL.

NYT had the race at 10% Jones at one point. Yeah, sure I missed. But we're in recount territory. I did predict Jones under 50 percent well early.

So obviously I was close. Just gotta tweak a few things.

[Wisconsin+17]

You must be logged in to read this quote.

We'll see what the numbers say when all the ballots are counted. That is what you guys said about Coleman, right?

[Wisconsin+17]

You must be logged in to read this quote.

Why would I do that? It's fun to test these things and hammer them out. I'll give 3/x a try and see how that goes.

[Wisconsin+17]

You must be logged in to read this quote.

We'll count 'em all. I love how counting them all is somehow undignified, but wasn't for Franken.

Also, when is Franken resigning? Since I seem to hear that he was gonna do that sometime.