[Ed Law]

Originally Posted by Citrus Dad

Ed
Thanks for doing this. I've used it extensively for pre-scouting coming events. A few questions/comments:

- I noticed a few cases where the OPR for specific segments (i.e., Assists & TeleOp give out-of range results). For example 2156 shows Assists = -520.5 and TeleOp = + 417.9. The net is -103.4 for that aspect. The Auto = 74.2 and the Truss & Catch = 42.4, so the overall net gets down to +14.0. Another team has a similar result. I don't remember seeing such oddities before. Is this a bug in your results, or an artifact of the analysis?

- Which brings me to the second point. Since this is a linear regression should be able to produce two fundamental statistics that tell about goodness of fit. (There's other stats that also can tell us about potential bias as well, but those are more difficult in a spreadsheet.) First is to compute the standard error around each estimate so we can see the probability that the parameter estimates are statistically significant. The second is the r-squared that tells about the overall goodness of fit. Can you produce either or both of these in future versions?

- And one last observation. For the statistically minded, this looks like a random effects model. (http://faculty.ucr.edu/~hanneman/linear_models/c4.html). Probably too difficult to implement in a spreadsheet, but there's some important differences in the statistical properties.

Thanks for pointing that out. Others have pointed it out also recently. I need to look into it. This happens last year when I did it the old way. Every time a team plays a surrogate match, things can get mess up. I added a shift so the total of the sub OPR score will add up to OPR. It seems to work in all the events from last year. However this year there are instances when it will happen. I am not sure if I can fix it quickly as my team is competing week 4, 5 and 6. And also 7 if we make it to Michigan State Championship. However I know the program is okay for most of the other teams since it does the shift independently of each other. I will try it without the shift and see if it is the shift that mess it up.

(Update)
I looked into the macro and found the problem. I came up with the shifting instead of scaling over the summer. Unfortunately I used the wrong version to start this year's spreadsheet which is using the old way. I updated the macro and reran the data of all the 5 weeks of events. The problem should not come up any more. Sorry about the confusion.