Login

Beaver · 02-13-2018, 11:44 PM

Code:
1518 words

During last season's SMJHL playoffs we in Montreal were crunching the numbers trying to get any advantage we possibly could ahead of our series against Prince George and happened to notice something unusual: we had about a 10 point spread between our home win percentage and our away win percentage. I had never seen home ice mentioned around the forums except in passing (in stark contrast to NSFL where it's all anyone talks about) so I'd never considered that there might be an advantage baked into the sim. We had a pretty solid sample size (around 3000 games) so it was unlikely that this was just random variance.

So after putting those thoughts away for awhile I decided to circle back to them on this fine Sunday afternoon. (actually I'm posting this on <day I actually post this> cuz lazy but Sunday was the day I did most of the testing)

<div align="center">~scroll until bolded text for the tl;dr~</div>

What I did was:
1. Import the NHL Ratings file into STHS
2. Create a 4 team league and perform an automatic fantasy draft
3. Mass Edit every player on a roster to be 6' 200 lbs with the following stat line:
[Image: gQ9Volp.png]

4. Turn off injuries, suspensions, coaches, morale, etc. (only did this after some hiccups to start out)
5. Let auto-lines do its thing for each team and make sure all their settings and such were identical

In theory this created 4 identical teams that then would play a 750 game schedule against each other. If there's no home ice advantage, home teams should have around a 50% win rate.

I did two seasons (3000 total games) with overtime turned off and I found that home teams went 1544-1456 overall, a 51.47% win rate, and 1221-1143 in regulation games, a 51.65% win rate.

I then did another two seasons with overtime turned on and home teams went 1536-1464 overall, a 51.20% win rate, and 1210-1141 in regulation, a 51.47% win rate. Home teams scored 50.81% of goals.

Home teams won 50.89% of overtime games (86-83) and 50.45% of shootouts (563-553) which tracks with the prevailing thought that each of those is more random than regulation results (obligatory sample size warning, especially for overtime)

In addition, when you break down each individual team's win percentage versus their home win percentage there's not a whole lot of difference. Each team (except Team 4 (OT)) was slightly better at home than overall but typically only by a few percentage points:
Team 1 (No OT): 51.33% Win%; 52.13% Home Win%; 49.10% Regulation Win%; 49.41% Home Regulation Win%
Team 2 (No OT): 48.60% Win%; 49.34% Home Win%; 50.60% Regulation Win%; 50.86% Home Regulation Win%
Team 3 (No OT): 51.40% Win%; 53.47% Home Win%; 51.26% Regulation Win%; 54.05% Home Regulation Win%
Team 4 (No OT): 48.67% Win%; 50.94% Home Win%; 49.04% Regulation Win%; 52.25% Home Regulation Win%

Team 1 (OT): 49.80% Win%; 52.00% Home Win%; 49.37% Regulation Win%; 51.37% Home Regulation Win%
Team 2 (OT): 50.67% Win%; 53.06% Home Win%; 51.10% Regulation Win%; 53.52% Home Regulation Win%
Team 3 (OT): 50.07% Win%; 50.40% Home Win%; 50.60% Regulation Win%; 51.45% Home Regulation Win%
Team 4 (OT): 49.47% Win%; 49.33% Home Win%; 48.93% Regulation Win%; 49.49% Home Regulation Win%

This alleviates one of my concerns which was that through any myriad of possibilities the teams wouldn't actually be equal and the imbalance would skew results. Since pretty much every team was better at home than overall I think it's a stronger case.

Now, pretty much all of what little statistics knowledge I have is self-taught and therefore spotty at best and it's been years since I've had to find a p-value so take all of this very skeptically but it seems to me that there likely is a home advantage but that it's so small it's nearly negligible.

I also tested Version 2.1, which I believe there has been some talk of switching to, which saw home teams go 1783-1217 overall, a 59.43% win rate, and 1637-1051 in regulation games, a 60.90% win rate.

PS: holy shit poor V2.1 goalies. League average Sv% was 81% and there were over 14 goals per game, down from 88% and 4.6 respectively.

I do think that even at that sample size some luck was involved in the V2.1 simulation - home teams only scored 53.51% of goals so I'm not sure home ice advantage is quite 60/40. Something closer to 55/45 seems more likely and lines up with 538's findings in the NHL. It makes sense that if Simon were to implement home ice advantage he'd try to mimic the actual NHL advantage.

If we accept that there is a home advantage, then, how is it implemented? In what facet of the game does it appear? Do home players get a minor attribute buff? Do their attributes decay slower? Are visiting teams more likely to get penalties called on them? Are visiting goalies less likely to make any given save? This is particularly difficult to figure out because we don't have a lot of "process" stats. For example, home teams scoring more goals than away teams is great but doesn't tell us about how those goals are scored or where the home advantage is implemented. To try to tease out any disparities I edited the schedule so that:
1 team played all of their games at home
1 team played 2/3 of their games at home
1 team played 1/3 of their games at home
1 team played none of their games at home

A few possibilities emerged:
-Shots For:
[Image: KL42Ev7.png]

-Shots Against:
[Image: TwteBSK.png]

-Faceoff Percentage:
[Image: py1zmmW.png]

-Shooting Percentage:
[Image: MmEG6Se.png]

All of these (plus Goals For and Against, obviously) see better results at home compared to on the road whereas things like Penalty Minutes, Power Play Attempts, Penalty Kill Attempts, and Opponent Shooting Percentage saw much weaker correlations.

I ran the same thing but on V2.1 where there is a clear home ice advantage and the only stats that had pretty good relationships were:
-Faceoff Percentage:
[Image: IOazwam.png]

-Shooting Percentage:
[Image: vKeE07P.png]

-Opponent Shooting Percentage:
[Image: PhiDXbM.png]

Whereas in the first test where every team played an equally balanced schedule the starting goaltenders put up nearly identical save percentages and in the V1.5 unbalanced test where save percentages were similar, we saw some disparity in the V2.1 unbalanced test. The starting goalie for the team that played all their games at home finished with a .817 compared to the starting goalie for team that played all their games on the road's .794. Not a huge difference, but noticeable. The 4 goalies finished in exactly the order you'd expect.

I think it makes sense that if there's a home advantage in V1.5 it would be implemented in the same way as V2.1 so Faceoff and Shooting Percentage are the common threads. If there were a minor buff to all attributes across the board you'd expect to see stronger relationships in Shots For and Against, PIM, etc so the fact that the strongest relationships are in stats that see lots of variation makes me think that whatever luck mechanic is under the hood gets dialed up a touch for home teams (at least in V2.1).

That's not to say that Shooting and Faceoff Percentage aren't skill-based - you'd expect someone with 99 FO to beat someone with 40 FO more often than not - but if you used the FO and SC attributes to try to predict players' season-end percentages you would probably be wildly unsuccessful. For example the top 10 Sh% last season had the following SC stats, in order: 83, 90, 96, 76, 84, 82, 82, 88, 88, 97. And the top 10 FO% last season had the following FO: 40, 89, 86, 90, 90, 85, 87, 85, 90, 90. Simulating the equivalent of 15 seasons smooths some of that randomness, of course.

Stop here for the tl;dr

In conclusion, I don't really know. I was unable to reproduce the home/away split we saw last playoffs and I could definitely be way off track here but it seems to me like there's a negligible home advantage in V1.5 and there is an advantage in V2.1, where it's obtained by making home teams get "luckier" (though this is more or less a glorified guess).

There's definitely not enough here to draw many meaningful conclusions and there are tons of issues a critic could raise about my methodology to poke holes in any findings, not to mention all the assumptions I made that may or may not be accurate, but I do feel somewhat confident in drawing those conclusions.

If an industrious SHLer out there wanted to build on this I'd do it by perhaps having more teams that play varying numbers of home games in order for any relationships that emerge to be more believable. With only 4 teams random variation could've wrecked something that should've been a strong relationship or produced a relationship where there should not have been one. It also might be worth testing at different attribute levels. I have the data that I transcribed into a spreadsheet for anyone who wants it but I didn't save the actual sim file data between each test. But I really don't think it's worth anyone's time to dig more into V1.5 since there's not much there.

Wasty · 02-14-2018, 04:08 AM

"PS: holy shit poor V2.1 goalies. League average Sv% was 81% and there were over 14 goals per game, down from 88% and 4.6 respectively."

I didn't really read much but I did notice the 2.1 comment. The Sim engine is actually really fun to play with and there is a way to have goalies numbers better than 81% etc. We use the 2.1 in the PHL and most of our goalies are in the .915-.899 range and scoring is balanced but still high for the players. I wouldn't say the SHL is ready for the 2.1 unless there was some serious settings. Did you mess with any of the sim settings in Pro Simulation?

#40 Niclas Wastlund - W - VANCOUVER WHALERS Whalers

/ MINNESOTA MONARCHS

Beaver · 02-14-2018, 04:27 PM

Quote:Originally posted by Wasty@Feb 14 2018, 03:08 AM
"PS: holy shit poor V2.1 goalies. League average Sv% was 81% and there were over 14 goals per game, down from 88% and 4.6 respectively."

I didn't really read much but I did notice the 2.1 comment. The Sim engine is actually really fun to play with and there is a way to have goalies numbers better than 81% etc. We use the 2.1 in the PHL and most of our goalies are in the .915-.899 range and scoring is balanced but still high for the players. I wouldn't say the SHL is ready for the 2.1 unless there was some serious settings. Did you mess with any of the sim settings in Pro Simulation?

It could be just the attributes I decided to use. I didn't mess with any settings so it's definitely possible that adjusting some sliders, etc will get goalies to a good place but the striking difference in Sv% and goal scoring caught my attention.

Cyanide · 02-15-2018, 08:50 AM

I think we already tried the 2.1 sim, but couldn't get the sliders to work for the goalies. It was S9-11, and looking at the SHL records it was not pretty for goalies. When 0.900 save % was only reached 4 times (3 times by 1 dude), you can only imagine how pissed goalies were.

Credit to Wasty
[Image: wolfpack3.png]

Credit to Sulovilen

Navigation

Extra Menu

About us