U.S. Senate races 2020

Gotcha. Thanks.

I don’t think even without Trump stench that Kobach would be a good choice for the GOP in KS. He riles up their primary voters, but his years of gallivanting around the nation instead of minding his knitting at home has made fairly toxic.

After losing the governor’s mansion, I’m surprised he has any real support in any case.


When I’ve heard fellow Democrats complain that Biden’s not being visible and not campaigning, I remind them of Napoleon’s axiom:

Trump is hammering himself - without enough self-awareness to know it - with those COVID briefings. The single best thing Biden can do is stand back and not distract from it.

For the down-ballot Dems they might find opponents who choose to distance themselves from Trump the closer we get to the election to save their seats. I think August is the cut-off point whereby if polls keep showing Trump continually losing ground in swing states and barely above water in some traditionally red states (recent Texas and Georgia polling doesn’t look good for him) then you’re going to see a breakaway.

Not popular? Hasn’t he made Kentucky one of the top three or four richest states when it comes to federal funding? (Heh, despite the evil moron whining last week about federal aid being a “blue state bailout”)

She needs to defeat ol Turts almost as badly as the senate needing to get flipped or a switch in chief executive.

How much of a stretch was that - what I just said? (whoever his replacement finally is, could he/she be any more execrable?) (at the very least I don’t recall Boehner being appointment-happy with garbage judges.)

63% for D in NC is way high. If Tillis loses here it won’t be by more than a few points. He probably starts with 45% at least.

You are misunderstanding what 63% represents in my previous post.

The 63% represents the estimated probability of the Democrat winning the race and not the projected share of the vote. It wiggles around a little bit day to day, but it has been pretty consistently in a range where 63% seems possible.

Yes I know what your 63% means. It is still way high. Right now it’s a tossup or leans R.

What is your methodology for determining this?

Living in NC for 50 years and looking at the results of recent senate races and other races. The metro areas of Charlotte, Triad, Triangle are mostly blue but they also contain a lot of young people who vote Dem but don’t vote nearly as much as older people who tend to vote GOP. Rural areas are pretty much way red.

Tillis might well lose in Nov. but I still see him as the slight favorite. Is anyone else predicting he will lose? Cook report has the race as tossup which I agree with.

Brilliant - getting my House Speakers and Senate Leaders mixed up - apologies. Was Trent Lott any worse than MM?

Bill Perdue (R.) isn’t confident of victory in November in Georgia.

My bold. Yeah? Including mail-in votes, Mr. Republican?

First, 63% D is a pretty much a toss up. It takes at least 40 trials to tell a 63/37 coin from a 50/50 coin.

Second, I’m just reporting what the market is saying. There’s over a hundred shares of Republican ‘Yes’ up for sale at PredictIt for $0.44 per share. If you think that’s underpriced, there’s an obvious course of action for you to take.

I don’t bet on political races. Just the lottery now and then. :slight_smile: I think last time UK bookies were so confident of a Clinton win I believe some actually paid out before the election. I guess they got all that money back. I prefer to listen to political experts about races, not bookies.

Suit yourself.

Also, PredictIt is a prediction market, not a bookie i.e. buyers and sellers set their own prices.

OK I will put $100k on Tillis to lose. Happy?

Don’t do it to make me happy. Do it for yourself.

Lance - What debiasing formula do you use? IIUC Rothschild’s y’ = Phi(1.64 * Phi^-1 (y)) changes 28% into 17%.

I do use f(y) = Φ(1.64 * Φ[sup]-1/sup), and f(0.28) is indeed quite close to 0.17.

Is there good evidence that that formula gives useful results for prediction markets? I ask for two reasons:

(1) The unadjusted percentages inferred from Betfair often sum to about 90 to 100% as they should. I don’t think the adjusted percentages do.

(2) IIUC that formula was derived from horse races, where odds are determined by pari-mutuel. In most prediction markets individual bets are laid; the formula would thus seem to assume long-shot fear, properly adjusted, balances long-shot greed exactly. Does it?

I assume that prediction market players are well aware that inferred percentages should sum to 95%(*) or so, and bid up the costs to lay long-shot tickets, if only to protect their other investments.

Four years ago, in Game Threads, I organized a pari-mutuel prediction market (“Karachi auction”) to predict the GOP nominee, but there was no interest in such a thing this cycle. (The four-years ago Karachi auction turned me into a laughingstock when I advised a foreign Doper to reduce his wager on the joke candidate, Trump. :smack: )

ETA: (* - 95% rather than 100% to leave room for dark horses coming from “out of nowhere.”)

Yes. You have read the same paper as I have. Forecasting Elections: Comparing Prediction Markets, Polls, and Their Biases.

The adjusted percentages for these particular races add up to 100%. They do so because I make them do so.

Here’s the process:

  1. Collect the average of the bid and ask prices from the PredictIt API for both parties for every senate race offered on the site every hour. Call these d_raw and r_raw.

  2. Normalize with respect to the L[sub]1[/sub] norm. Call this d_norm. This means d_norm = d_raw / (d_raw + r_raw). Same goes for r_norm, but that is never used.

  3. Debias for longshot-favorite bias the way Rothschild does as suggested by Leigh, et. al… Call this d_debias. So d_debias = Φ(1.64 * Φ[sup]-1/sup). Due to the nature of Φ, d_norm+ r_norm = 1 implies d_debias + r_debias = 1.

  4. Finally I compute the 24 hour rolling average of the debiased probabilities. Call this d_rolling. with a similarly computed r_rollling we should have d_rolling + r_rolling = 1, but I don’t bother with that. I set r_rolling = 1 - d_rolling.

It seems to based on Rothschild’s results.

I don’t do this, because I haven’t seen any reason to believe that a third party candidate has a greater than zero chance of winning in any of these races, much less 5% chance. If such a candidate should emerge I will adjust my model accordingly.