The Advanced Stats Thread Episode VI: RIP To Our Databases

[SnowblindNYR]

So does anyone know why I couldn't attach my excel spreadsheet and it said I had an incorrect file extention? I used XLS, XLSX, and CSV.

 

[silverfish]

I did an analysis and I'd appreciate those that are better at statistics and data than me to audit whether my analysis and more importantly conclusion makes sense.

I wanted to compare in all situations expected goals scored and goals scored and see whether teams can consistently over or under-perform their xGF over a 5 year period (12-13 through 16-17). In order to do that I created an index (GF/xGF). I needed to compare that index to something and I decided to compare it to actual goals scored. I looked at the standard deviation of this index and compared it to the one for goals scored. I figured that goal scoring should be relatively consistent year to year. (Maybe that's not a correct assumption.) If the GF/xGF index standard deviation is smaller than it should be relatively consistent. The standard deviation (standardized using the coefficient of variation or "STD/Mean") was smaller for the GF/xGF than for goals scored 0.0673 vs. 0.0751.

So now that I know that the GF/xGF index is relatively consistent I checked to see whether it was consistent for good predictions (where GF/xGF is close to 1) only or if it was also consistent for teams who over or under-perform the xGF. I created another metric, which was "distance from expected" to see how far each index is from 1 where xGF perfectly predicts GF. This metric was done using the absolute value of GF/xGF-1. I then compared the average of that metric over 5 years with the standard deviation of GF/xGF to see whether the teams that consistently were close in GF to xGF were more likely to be consistent (and thus the prediction was good) or if those that were not close in GF to xGF were consistent thus showing that teams can consistently over or under-perform xGF. Theoretically, if the closer a team was to xGF the more consistent they were over 5 years (thus xGF is a better predictor of GF), the line would slope upwards. The x intercept was positive (~1.25) but the p-value is 0.0849 which gets rejected under the most common test of 0.05. However, 0.0849 is close enough that it's clear there's a relationship and it would be considered significant using less strict tests. All in all it appears that most of the lower standard deviation may in fact be coming from well predicted teams, thus showing that xGF more usually predicts GF well consistently, but it stands to reason that some teams still over or under-perform their xGF consistently.

CV GF/xGF Index (Avg)	CV GF/60 Stand (Avg)	CV GF/xGF Index (STD)	CV GF/60 Stand (STD)
0.0673	0.0751	0.0251	0.0221

[TBODY] [/TBODY]

Variable	Coefficient	Std. Error	t-Statistic	Prob.
DIST_FROM_EXPECTED__AVG_	1.251059	0.700298	1.786467	0.0849
C	0.23897	0.061498	3.885827	0.0006

[TBODY] [/TBODY]

P.S. I tried attaching the spreadsheet using XLSX, XLS, and CSV, and I keep getting an error message that the file extension is wrong. Can someone help me figure out how to upload my spreadsheet? I never had a problem under the old format. Thanks!

Conducting YoY analyses for teams is tricky. Often, there are way too many outside factors that can greatly impact xGF, GF, or really any standard metric.

1. Did they change coaches?
2. How much roster turnover did they experience?
3. Specifically for xGF vs GF, did their goalie change? (I maintain that GSAE (xGA - GA) is a goalie based metric more than a team based metric.

And many, many more that I am just probably forgetting right now.

Another way I'd focus this analysis is on the coach, rather than the team. Can coaches directly impact over or undershooting?

 

[SnowblindNYR]

Conducting YoY analyses for teams is tricky. Often, there are way too many outside factors that can greatly impact xGF, GF, or really any standard metric.

1. Did they change coaches?
2. How much roster turnover did they experience?
3. Specifically for xGF vs GF, did their goalie change? (I maintain that GSAE (xGA - GA) is a goalie based metric more than a team based metric.

And many, many more that I am just probably forgetting right now.

Another way I'd focus this analysis is on the coach, rather than the team. Can coaches directly impact over or undershooting?

Thanks. So what you're saying is that I wasted a perfectly good Saturday afternoon nerding out about something useless.

 

[silverfish]

Thanks. So what you're saying is that I wasted a perfectly good Saturday afternoon nerding out about something useless.

No analysis is useless. I'd have just done it differently.

 

[SnowblindNYR]

No analysis is useless. I'd have just done it differently.

Have you every tried to attach an excel file in the new board? I couldn't.

 

[silverfish]

Have you every tried to attach an excel file in the new board? I couldn't.

Nope, but I don't really mess in Excel for hockey stuff anymore

 

[SnowblindNYR]

Nope, but I don't really mess in Excel for hockey stuff anymore

Someone's too good for excel. :p

 

[silverfish]

Someone's too good for excel. :p

Excel is cool. It's what I use at work. But I'm trying to make my coding game strong, so I make sure to do all my hobby stuff (hockey) in R.

 

[SnowblindNYR]

Excel is cool. It's what I use at work. But I'm trying to make my coding game strong, so I make sure to do all my hobby stuff (hockey) in R.

Excel has its limitations. I had to get p-values from EViews. I don't know if excel even provides that, it just gives the equation as far as I can tell.

 

[silverfish]

lol Vesey

 

[DanielBrassard]

Worst game all season?

 

[SnowblindNYR]

What was the most lopsided Corsi game since the stat has been tracked and how did this game compare?

 

[Pavel Buchnevich]

How can someone play less than 9 minutes and be -21?

I usually don't care about our corsi, but I found today's corsi stats to be hilarious. McDonagh -34, Vesey -21 in less than 9 minutes.

 

[DanielBrassard]

How can someone play less than 9 minutes and be -21?

I usually don't care about our corsi, but I found today's corsi stats to be hilarious. McDonagh -34, Vesey -21 in less than 9 minutes.

He's just so useless, and so damn soft. McDonagh has been so bad, he looks so slow, that explosive skating is missing. I hope he isn't declining, otherwise we are in trouble.

 

[SnowblindNYR]

Anyone else annoyed how each player is talking about the first and second period? They stunk in the third too, Dallas just didn't attack as much. Got a fluke goal but no chances all period.

 

[Leonardo87]

Looks like Nash was the best player on the ice from what i can gather. Trying to comprehend the numbers but the eye test makes me think he was one of the very few silver linings in this really bad game tonight.

 

[Bob Richards]

 

[Kakko]

Does anyone really need to look at the advanced stats for this game?

 

[Mac n Gs]

McDonagh hot takes might give me an aneurysm.

 

[DanielBrassard]

McDonagh hot takes might give me an aneurysm.

Which ones?

 

[silverfish]

McDonagh hot takes might give me an aneurysm.

Vesey is so lucky McDonagh shit the bed as much as he did last night.

 

[Mac n Gs]

Vesey is so lucky McDonagh **** the bed as much as he did last night.

I don’t even care about Vesey because he’s depth piece that should never leave the bottom-6. How anyone can’t see McDonagh has been good this year, minus the hiccups that every dman has, is mind boggling.

 

[NickyFotiu]

For some reason I get a kick out of a portion of Larry Brooks column revolving around "advanced" stats.

 

[silverfish]

Who is Tucker Poolman? Four games for WPG so far this year and crushing the relCA60 game. Obviously the smallest of small sample sizes, but something to keep an eye on. Would also love to find a way to get TML to give us Connor Carrick. I'd put him right with McDonagh. I think pairing McDonagh with a RHD that's a pure suppressor would be the easiest way to upgrade this team right now.

 

[Mac n Gs]

Who is Tucker Poolman? Four games for WPG so far this year and crushing the relCA60 game. Obviously the smallest of small sample sizes, but something to keep an eye on. Would also love to find a way to get TML to give us Connor Carrick. I'd put him right with McDonagh. I think pairing McDonagh with a RHD that's a pure suppressor would be the easiest way to upgrade this team right now.

Kid they drafted out of the USHL, which has been cranking out good two-way dmen. Spent some time with Boeser at NoDak clowning fools and won a national championship. The good dmen we need to be drafting are all USHL and Euro kids. Slavin, Pesce, and the list goes on and on.

f***ing yes please to Carrick or Colin Miller.