Difference in totals based on the spread

[davopnz]

I'm trying to take into account how much affect the spread has on the total.

From the data I have, the league average is 52, when the spread is between -1.5 & -3.5 the average score is 46, when the spread is above -10.5 the average score is 56.

My question is, do I need to divide the number in half like you do with home advantage?

For example if I'm trying to model a total for a game where the spread is between -1.5 & -3.5, should it be 52-46 = 6 points lower than league average or should it be 52-46 = 6/2 = 3 points.

With home advantage you usually take the difference between the home and away scores and divide by two, but I'm not sure if I should be doing that here.

Hopefully this makes sense.

Cheers!

[Zocki]

I'm trying to take into account how much affect the spread has on the total.

Is this what you are looking for?

league avg total	52		52
avg total@spread	46		56
base average total w/o taking spread into account		new total with spread adjustments
48	42.46		51.69
49	43.35		52.77
50	44.23		53.85
51	45.12		54.92
52	46.00		56.00
53	46.88		57.08
54	47.77		58.15
55	48.65		59.23
56	49.54		60.31

In example for average games the base avg would be 52 (= league average). Then you would expect 46 (Spread -1.5 to -3) or 56 (spread < -10.5).
If your matchup is expected to outperform the average, lets say 54 instead of 52, then your new adjusted total would be 47.77 or 58.15.
And vice-versa for underperforming, lets say 50 instead of 52: 44.23 or 53.85

The model now needs your assessment if you expect more or less points than "normal" or "average" for any matchup.
Making it depending on just the spread without a prior could be too simplified (even this method of course could be)

But at least if the totals are distributed normally you would expect to see on average 46 for even games and 56 for games with a big fav.



new adjusted total =

base avg total w/o taking spread into account * avg total@spread / league avg

[davopnz]

Thanks for your reply!

Yes your formula is what I thought was the correct way to do it.

What inspired this was I noticed a bunch of close games going under where my model had predicted them going over. Often, it would be in the 2nd half where things got "cagey" and the scoring dropped off.

For example, I had over 54.5 in a match (model predicted 60) where the cap was -5.5. The dog was up 24-12 at halftime, early in the 2nd half the score was 27-19, and that was the final score.

Conversely, I noticed in a bunch of blowouts, the winning team would concede what my model would judge as below-average defensive performances, but it was in the context of a blowout. For example, a team was up 33-7 at halftime, and the final score was 54-28.

[Optional]

Maybe you just need to ignore the outliers for modelling purposes.

trying to account for them might ruin the overall accuracy.