So if I'm correct and I can't use poisson to calculate individual player r+h+rbi or individual player hit o/u 0.5, any hints or help as to what I can use if I have projections? Or is poisson appropriate after all?

And now I'm thinking poisson doesn't work for pitcher strikeouts in a game because it fails the rare event test, right?

Yes, that is how you would use that calculator to answer the specific question posed above. When I used the inputs you provided I got 0.65 as the answer. Note that in Excel you would input it as =1-poisson.dist(0, 1.05, TRUE) to get the same answer.

Well you could argue that the heuristic that says a random variable "needs to have a low probability of occurring and a large number of opportunities to happen" (or however it usually gets worded) to be modeled using the Poisson Distribution actually gets satisfied here; if you consider all the opportunities to hit a run or throw a K a player encounters over the course of an entire season.

Whether or not it is actually the best fitting distribution to model those particular variables would still need to be verified statistically though. See https://www.statisticshowto.com/goodness-of-fit-test/ for a quick explanation. My subjective guess is that the Poisson or Negative-Binomial distributions will probably get you close though. Poisson is a good start because it is easy to use. Whether or not it is good enough to beat the market is a separate question that needs to be considered too. Try pulling the stats into Excel and looking at the histograms as well. Getting a visual intuition of the data never hurts.

Thank you so much! Gonna fiddle around and hopefully come back with daily plays.

Awesome, I look forward to seeing some plays.

If you just care about HIT YES or NO, you can just do 1 - Probability he doesn't get a hit in the game. I'll leave the exact calculation for you to work out. Hint: You will need to know his expected PAs and the probability of hit in each PA (I assume you know otherwise how did you calculate your projection of 1.05 hits?).

Thanks for the help, very much. Sorry one last follow up, I'm back on the poisson doesn't work for h+rbi+r train, this time because I think it fails the independent event test, right? Hits, runs, rbi are not independent of each other.

Here's what I have as the fair lines for the 1st day of snooker qualifications on Monday 4/05. I'm not putting up any units because except for White and Hendry, I don't even have a sample size of 30 for half those guys. It looks like the market currently has the White and Hendry matchup at the right price so no play there either.

hicks v evans -+419, no bet (small sample size for evans)
jianbo v muir +-115, no bet (small sample size for jianbo)
lines v emery -+140, no bet (small sample size for emery)
surety v zhengyi -+196, no bet (small sample size for zhengyi)
patrick v fernandez -+193, no bet (small sample size for fernandez)
pinches v wilson -+202, no bet (small sample size for wilson)
yang v davison +-245, no bet (small sample size for yang)
cahill v maddocks -+472, no bet (small sample size for maddocks)
castle v benzey +-108, no bet (small sample size for benzey)
lilley v amine -+234, no bet (small sample size for amine)
white v hendry -+116, no bet

If a guy hits a homerun and you got at least 3 HRR. HRRs are heavily correlated with each other so Poisson won't work

Thank you TomG, at least I am sorta understanding how to use Poisson. I lied before I have one more I have to ask about I'm sorry to have to be spoon fed but I realized one more market I think Poisson can help me, mlb team totals for a game? I think Poisson can be used for mlb team totals but would love confirmation. After this I am set. I appreciate the help more than I can convey, I know asking to be spoon fed this stuff is annoying.

No same issue. Runs can be scored in clumps so Poisson won't work.

On BetOnline a lot of the lines are repeated with different players in them, which makes me think theyre just using one number (Elo? Glicko?) for each player and generating a line off of that. Anyone more experienced able to give some insight into that?

I.e. if Player A vs Player B is -377/+297 and Player C vs Player D is also -377/+297 it probably means A and C have the same rating as do C and D, so its more likely just that one number being used to generate lines.

Also I'm probably going to lolnomodel take White at -130 vs Hendry. Apparently Hendry was very good and then retired for like 10 years, then unretired and hasnt been great.

Thanks for your patience.

For the HIT YES OR NO, if I want to calculate probability of no hit and I have projected PAs and hits, then I would divide hits by PA to get probability of a hit in a PA. 1 - probability of a hit in PA = the probability of no hit in a PA, and this is where I'm unsure of my bad math. Can I simply take my probability of no hit in a PA and ^ projected PAs?

For example, 1.05 pHITS in 4.4 pPA would be 23.8% chance of hit/pa so 76.1 chance of no hit/pa. Can I then do .761^4.4 to get chance of no hit in the game?

Nicely done. You got the thought process correct and the answer right as well. Now as long as you're confident in your projections for hits and PAs you now have the tools to beat this prop. Keep us posted on your results.

Welp my stuff can’t beat frame based ELO. Good luck rabbit and skelts in this last tournament for the season.

Watched the first 4 frames, Hendry looked absolutely awful at the start, playing far too many safeties when chances were there. Got going later and potted the pink near the end while snookered, then potted the black to win the 3rd frame by a slim margin. Frame 4 Hendry looked very sharp.

White started ok to win the 1st frame, but later made too many mistakes and seemed to lose interest.

I think I have a reasonable approach for improving the frames based Elo, but it will take a few days to get it up and running.

Its a good reminder to myself that me, as a person, is bad at betting, and should stick to making models that might be 1% less bad

Finally had some time today to look at the snooker models again. Here's the link to a Google Sheet with what I have as the current frame-based Elo rating, and Bradley Terry rating for each player in my database.

https://docs.google.com/spreadsheets...it?usp=sharing

There seems to be a weak positive correlation between the ratings, I get 0.40 whether or not I log transform the Elo ratings. But there are definitely some players on which the models disagree, as can be seen in the scatterplot.




Missed the start of the matchups today, but didn't have enough of a sample size for most of the players anyway so probably only lost 1 or 2 betting opportunities. I'll take a look at tomorrow's matchups in a bit.

In the Elo calculation it's the difference in points that matters. A difference of 400 points will return a 90%/10% win probability regardless of the ratings.
player A Elo = 1400
player B Elo = 1000
player A win probability =1/(1+10^((1000-1400)/400)) = 0.9091

player A Elo = 1800
player B Elo = 1400
player A win probability =1/(1+10^((1400-1800)/400)) = 0.9091

Maybe they just sort everyone by win rate and then make 5 buckets from best to worst. Then just compute a win rate matrix for every possible bucket matchup. For example, take all matches between a tier 3 bucket guy vs tier 2 bucket guy and use that win rate to set the line. That would pretty hilarious, but would explain why so many games have the same exact opening lines.

It's an interesting thought experiment. Although Glicko is kind of a pain to work with compared to Elo, I do really like how the Glicko model increases the uncertainty of a player's rating as more time goes on without them playing in any matchups which naturally forces any projected matchup win probabilities to converge towards 50%.

Who knows, maybe worth a try coding it up to see how close it comes to the market lines.

That trader who made a few posts here the other day mentioned that on the game he checked the snooker lines at his book were originated by a provider. Since snooker isn't a big betting market I wouldn't be surprised if one linesmaking service is the main originator of snooker lines, and one of the quirks of their custom rating method is that it spits out the same win probabilities for different matchups under similar conditions, giving more credibility to the idea that one system is being used to make the opening lines.

One play for tomorrow:
si v hendry +-132, hendry +102

I think this thread can use some "Game Within The Game Theory" by TomG.