Ordinal logistic regression

**brettd** · 09-02-14, 10:24 AM

The most appropriate approach for the Brownlow is Monte Carlo simulation.

**antonyp22** · 09-02-14, 04:47 PM

My plan is to run the Monte Carlo Simulation after I have the predicted total number of votes for each player. In order to get the predicted total number of votes I need to deduce the probability of them polling 3, 2 or 1 vote in each individual game.

**donkeyshark** · 09-12-14, 09:35 PM

Could you be more specific about what your model outputs and what sort of help you're looking for? Logistic regression models should yield probabilistic estimates as their outputs.

...I know nothing about ARF, but I do build predictive models for a living.

**Stratesfaction** · 09-13-14, 09:47 PM

Nothing to do with the topic, but OP, I'm looking for another sport to get into next season (I primarily follow union) as I feel the season for it in Aus is too short for my liking. I'm tossing up between NRL and AFL (maybe both), but what is the learning curve like for AFL in relation to understanding the game, its statistics and finding value? it seems like information and statistics are a lot more in-depth and are easier access than league? But it also seems that bets require more research?

Also, donkey, do you do much betting yourself? I'm curious as to whether predictive model builders by trade start betting or stay away.
GL OP with the Brownlow Medal!

**jizay** · 09-14-14, 09:00 AM

Originally posted by antonyp22

Guys I have a stats question

I'm performing an ordinal logistic regression on the Brownlow medal (MVP for Aussie Rules Football). The three best players in each match are awarded with 1, 2, or 3 votes.

I have carried out an ordinal logistic regression with the dependent variable being whether a player polled 1, 2, or 3 votes and my independent variables being various player statistics.

I have obtained my parameters and odds ratios for each variable, I am now stuck as to how to use this information to deduce the probability of a player polling 1, 2 or 3 votes in each match.

I know how to work this out for binary logistic regression but am struggling to do it here. Any help would be greatly appreciated.

Any reason you're using logit instead of probit? The latter is easier to interpret in terms of how each variable contributes to the probability of winning.

**antonyp22** · 09-14-14, 09:14 PM

Guys problem solved I worked it out myself

Stratesfaction, this was my first season modelling/betting on AFL it has been quite successful but there are other members of the forum who are far more experienced than I am in betting on AFL markets. In terms of the learning curve of the sport, I have followed AFL from a very young age so my knowledge of the game was quite good. To an outsider the game is quite hard to get a grip on but you should be able to determine which variables are significant based on your statistical analysis.

**Stratesfaction** · 09-14-14, 11:28 PM

Thank you!