The MAE error of the opening line was 9.020 and the closing line 8.983, and the RMSE (which I, not a professional statistician, assume is measured by just taking the square root of the reported Mean Square Error) was 11.486. I don't see that this source has the seasonal summaries for past years but they do have CSV files of every game's line.
Great article Evan! I had similar thoughts about blending the 3 models together, so great to see you analyze that. Another interesting analysis I would find to take it a step further is to see how the models compare throughout the season. I believe you have posted in the past that your model did particulary well towards the beginning of the season, so I'm wondering if that is accurate across KP and Torvik as well. Thanks!
Hi Evan - lovely write up! Any interesting comparisons in the residuals distributions of the different models that would play with the fit statistics? Also in the ensemble is there any meaningful skadicity that would shift the weights throughout the season? I am wondering if any models are stronger for early vs late season by capturing some extra latent factor. Can’t say enough positive things about your work and communication style!
Very interesting and information! Evanmiya's relatively recent jump into analytics compared to the other two just shows how good his algorithms are and how passionate he us to come up with new wsys of looking at data. More helpful to me though is how easily understandable he makes things.
always thought EvanMiya was just a cool name but it was actually a person woah
I'd be curious about how the three individual models and various blends of the three performed relative to vegas spreads and final results.
Bringing in Vegas is worth a whole separate article which I'll have to save for another time.
Here is the performance of the line against various freely-available projection systems for last season: https://www.thepredictiontracker.com/bbresults.php
The MAE error of the opening line was 9.020 and the closing line 8.983, and the RMSE (which I, not a professional statistician, assume is measured by just taking the square root of the reported Mean Square Error) was 11.486. I don't see that this source has the seasonal summaries for past years but they do have CSV files of every game's line.
What a night it was! I’m back in the mix!!! (🙏+🤞=🤲) feelin hella lucky & ready for more!
Great article Evan! I had similar thoughts about blending the 3 models together, so great to see you analyze that. Another interesting analysis I would find to take it a step further is to see how the models compare throughout the season. I believe you have posted in the past that your model did particulary well towards the beginning of the season, so I'm wondering if that is accurate across KP and Torvik as well. Thanks!
Hi Evan - lovely write up! Any interesting comparisons in the residuals distributions of the different models that would play with the fit statistics? Also in the ensemble is there any meaningful skadicity that would shift the weights throughout the season? I am wondering if any models are stronger for early vs late season by capturing some extra latent factor. Can’t say enough positive things about your work and communication style!
-your friendly neighborhood industrial statistician :)
Evan, great article! For the EvanMiya predicted spreads, did you use your ratings including adjustments for injuries?
Very interesting and information! Evanmiya's relatively recent jump into analytics compared to the other two just shows how good his algorithms are and how passionate he us to come up with new wsys of looking at data. More helpful to me though is how easily understandable he makes things.
Was the ensemble model for game predictions fully out-of-sample?
This reports in-sample, but I just added a footnote that shows the out-of-sample results too, which are very similar.