pushes etc

[newbee]

Hi,
I have (probably stupid) question to all:

Why in sport betting people are going for "push" distribution, rather than actual over/under probabilities? For instance, if game line is -6 and we need to estimate what is ML of -4 using push probabilities we we need to know what is the probability of a -5.5,-5,etc pushes. If we have the subset which was used in estimating these numbers, why don't we just estimate what is the probability that favorite covers the -4 spread given -6?

Also just to get a vote/opinion on 1 vs 2 factor models (i.e. spread vs spread/total)? Though 2 factor models might give much better estimated results, but one has to be mindful of sample size issues. What are people using?

Thanks,

[djiddish98]

I'm not going to get into the statistical descriptions (because I'll mess it up), but when you cross the even number, the side you're betting on essentially gains that push probability in their chance of winning.

So if you have Team A +8 and Team B at -8, and both are at 100, and the push prob is 8%, the probability of either team winning is 46%, with and 8% push chance.

From what we know about the spread at 8, If you buy Team A up to +8.5, their odds are now (46% + 8%), so 54%. We don't need to know the odds at 8.5, because using the push prob, we can derive the odds based on what they are at 8.

I'm at work so I can't confirm my thinking on the half point calc - someone correct me if so.

[newbee]

obviously that makes sense, but I was not really asking why that is so (I know what you are saying is true ). I guess I should have put my question differently: what are the pros and cons of the two methods or they are virtually the same. Intuitively they should not be since we are working on different samples, but I might be wrong.

[djiddish98]

Ok - so you are asking about whether to use just a spread as a variable as opposed to spread + total.

the spread + total model seems more accurate as a methodology, since if you have a total that is higher than the mean total for a sport, then the probability of pushing on a specific spread would intuitively seem lower than the average push prob.

I'm not sure what the minimum sample size is for constructing one of these, but I would think there would be enough data out there to implement totals - I would imagine the calculation would be a lot more complex though.

I get the sense that a lot of people just use the lovely half point calc as opposed to constructing their own. If someone has created their own (and I'm sure some have here), feel free to chime in.