May 2020 probabilistic senate prediction competition

I have a Twitter robot that produces a probabilistic prediction for senate races of every day at 4:42 MT.

My prediction is derived directly from the prediction market PredictIt which has markets for 13 of the 35 senate races this fall where the ones there aren’t markets for are pretty unlikely to be interesting. I perform a small adjustment to debias the market prices for favorite-longshot bias that I can go into more detail on should anyone request it.

Arizona (special)
Georgia (special)
North Carolina

Note: the Alaska race is between an incumbent Republican (Sullivan) and a Democratic Party independent (Al Gross). For the purposes of this contest I’m considering Gross a Democrat, which I’m sure would bother him to no end.

The goal is for each race to assign a value, D, between 0 and 1 that represents the probability that a Democrat wins the seat, and value, R = 1 - D, that represents that a Republican wins the seat.

A sample entry looks like this:

Race			D	R
Alabama			2.8%	97.2%
Alaska			5.7%	94.3%
Arizona (special)	83.4%	16.6%
Colorado		95.6%	4.4%
Georgia			24.2%	75.8%
Georgia (special)	32.4%	67.6%
Iowa			27.1%	72.9%
Kansas			22.6%	77.4%
Kentucky		8.6%	91.4%
Maine			72.3%	27.7%
Michigan		84.2%	15.8%
Montana			49.3%	50.7%
North Carolina		69.4%	30.6%

Entries will be submitted by posting in this thread. The most recent post by a user with a timestamp on or before May 31 will be that user’s entry (i.e. you can change your entry until the end of May). I plan to post mine by on May 31, but the above sample will be my entry if I don’t post an update.

After the election results are known, I will compute the Brier score for each entry. Lowest Brier score wins.

Can you, using polls, pundits, or pig entrails, outperform PredictIt this far out from an election?

Let’s find out.

Throw your predictions out the window until the second week of October, at the earliest.

The country is in the midst of a crisis that has affected the life of every single voter. We don’t know how or when this will be resolved, and new heroes or villains my arise at any time.



(1) How will you rule if election(s) are postponed or ruled invalid?

(2) Depending on the answer to (1), I think most submissions will have p(D) + p® = 1. Will we have to submit both those probabilities or just p(D) with p® = 1 - p(D) implied?

(1) I’m not worried about it, but if that does come to pass, we’ll have a lot more to worry about than this thread.

(2) As long as your intent is clear, I’m not going to be a stickler for any particular exact format.

My submissions for p(D); p® = 1 - p(D) assumed.

Alabama 0
Alaska 0
Arizona (special) 0.8
Colorado 0.9
Georgia 0.2
Georgia (special) 0.3
Iowa 0.2
Kansas 0.2
Kentucky 0
Maine 0.55
Michigan 0.8
Montana 0.35
North Carolina 0.45

Excellent. Thank you for your submission. Keep in mind that you may change your entry as many times as you like until submissions close on May, 31.

I just realized now that I should probably specify a time zone so let’s make it Chicago time. Thus entries are valid with a timestamp 11:59:59 PM CDT May, 31.

septimus, is there some reproducible method behind your numbers, or is that mostly your gut? You are not required to go in to any detail.

My prediction: you aren’t listing enough North Carolina senate races.

Could be, but I feel like Burr can hang on long enough so that his seat isn’t up for grabs in 2020. He may step down the day after that deadline which is in August I think.

There is no legitimate reason to postpone elections. We can have valid elections using mail-in voting. If elections are postponed or invalidated, that will simply mean that we no longer have elections and there is really not much use predicting who will come out ahead.

I believe that confidence in the November elections proceeding smoothly is misplaced, given the criminality of this Administration and Republican organizations more generally, and the fact that Covid-19 quarantining (or related effects) provides pretexts to postpone or invalidate elections.

The chance of such an electoral catastrophe may be only 10% or even less, but those who think it to be very close to Zero seem Pollyannish to me.

Nothing special. I started with your numbers but shaved them all to be more pessimistic. (I didn’t check: were your numbers posted here the ones computed by your robot?) I thought pessimistic numbers were appropriate for two reasons:

(1) I am pessimistic. The GOP will create obstacles to voting for D voters. If Covid-19 deaths remain lowish outside blue states like New York, Trump’s approach will be seen as successful by low-information voters. The Kremlin-GOP Lie Machine will drone on and on: After Tara Reade there will be something else. If anything, the numbers I submitted weren’t pessimistic enough.

(2) I expect most Dopers to be more optimistic than your bot. Therefore simple game theory (cf. Ice cream vendor problem) implies that my best win chance will be in the “pessimistic” part of the beach.

I guess the Dems in Alabama need to run against a child molester to win.

That’s what I thought. The example I posted were computed by the bot at around the time of my post. It seemed like your were just a little more pessimistic version of mine, but I also thought you may be using Betfair or something else to inform your decision making.

Thanks again for submitting an entry. I hope some other posters are up to the challenge. In particular there are a handful of posters who are very dismissive of prediction markets. I’d like to see what they can do with a polls only approach. Or any other approach really.

Show me what you got. Can you get schwifty?

How is this useful?

Title edited per OPs request.


Burr is not going anywhere before August unless someone releases a big smoking gun on his stock sale. And I still say 63% is way high for Tillis to lose. All the people ticked off about the stay home orders are likely to blame that on Dems since gov Cooper is a Dem.

I’ll shoot for a perfect score:

Alabama 0
Alaska 0
Arizona (special) 1
Colorado 1
Georgia 0
Georgia (special) 0
Iowa 0
Kansas 0
Kentucky 0
Maine 1
Michigan 1
Montana 0
North Carolina 0

You can keep on talking about how much smarter you are than prediction markets or you can do something about it. This thread is the perfect opportunity for you to show your stuff.

*But some recent Senate polls have been so eye-popping that we’re compelled to write about them — if only to sound a note of caution.

Those polls imply a Democratic wave of truly epic proportions. In the space of just two days last week, we got:

Ultimately, it’s hard to know at this point if these polls are outliers or early indicators of an overwhelming Democratic electoral environment. But the fact that they are even remotely plausible reflects a vulnerability for the GOP in the age of the coronavirus. Americans are souring on President Trump’s handling of the crisis, and congressional Republicans are reportedly worried that it will drag them down too. The pandemic has also devastated the economy, which has historically been bad news electorally for the party in the White House. In the worst-case scenario, the outbreak could lead to a Democratic wave à la 2008, when an unpopular Republican president and a tanking economy helped elect President Barack Obama and gave Democrats full control of the federal government.*