FanPost

Predicting Team Winning using PDO and Fenwick

Background and Definitions

Point Percentage (P%) is the percent of total points a team earns (team points divided by total possible points). P% is the measure used to represent team winning. Fenwick % (F%) is the ratio of all shots directed to the opposition net to the all shots directed at their own net and opponents net (excluding blocked shots). A value greater then 50 means a team is getting off more unblocked shots then they allow and is desireable as it correlates to puck possession and thus scoring chances and goals. In addition, Fenwick % Tied Road (FTR%) is used because Fenwick Tied score has been argued to better correlate to winning then the Fenwick of a team that is protecting a lead or mounting a comeback. Fenwick % Tied Road is also preferred because it minimizes (but does not completely eliminate) the bias of the home recorders on blocked shots and other real time data. Tied Fenwick data can be found at behindthenet.

Images_medium

Fenwick Tied Road FTR%

The correlation for P% (team points/total possible points) to FTR% for this season up to Dec 24 was 0.164 and is plotted below with 1, 2 and 3 sigma bands to observe teams that maybe overachieving (lucky) and teams that are underachieving (unlucky). Historically the rsquared has been closer to 0.535. This poor correlation may be a small sample issue but it is bothersome because it is not clear if we can trust the conclusion this statistic provides for intra-season analysis. Most teams are found with the 95% confidence band. (Right clicking the diagram allows the user to open a zoomed view in another tab). MIN, PHI, NYR and BOS are expected to either improve their Fenwick or regress their Point % (or perhaps some combination of both). Conversely CAR, ANA, NYI and CBJ are expected to decrease Fenwick or improve Point %. It also worth noting that Boston with its league leading SV% and CBJ with a bottom SV% is a factor that likely results in the P% results. (That is, the points total maybe pointing to a result of goalie skill rather then puck possession). The Leafs are pretty much in line with their Fenwick expectations and are not significantly lucky or unlucky by the puck possession measure.

Figure 1: P% vs FTR% Fenwick Tied

99_medium

Fenwick Close Road FCR%

Note that some users also prefer Fenwick % Close Road (though as I understand Fenwick % Tied is generally accepted to be provide a better correlation). Below is the same for Fenwick Road Close. It seems to correlate better (0.225) then Tied but the end result of which teams are lucky/unlucky is not much different. Fenwick Close data is also available at behindthenet.

Figure 2: P% vs FCR% Fenwick Close

104_medium

PDO (ES SV% + ES SH%)

Team PDO is even strength save percentage (ES SV%) plus even strength shooting percentage (EV SH%). Generally (as I understand) the measure is not expected to correlate to P% but rather used to determine if a team is lucky or unlucky. There is perhaps a "contentious" observation that PDO tends to drift to 1000 over time so an extreme high or low PDO values are not sustainable because goaltenders and skaters are generally evenly matched but in the short term teams may slump or streak. Of course, there are players like Tim Thomas or Steve Mason that have a longer term impact to a teams PDO based on skill. Team PDO can be calculated from the behindthenet dataset.

The correlation for P% to PDO up to Dec 24 was 0.528 and plotted below with 1 and 2 sigma bands to observe teams that are overachieving (lucky) and teams that are underachieving (unlucky). As can be observed all teams are found within the 95% confidence band. BOS and NYR are examples of teams with PDO that are expected to regress to 1000 but their P% maybe sustainable with the good goaltending they receive though perhaps their shooting will regress somewhat. Similarly CLB and NYI are expected to increase their PDO and also P% but poor goaltending may prevent this (unless they can trade or promote a better goalie). LA and COL are probably examples of teams where goaltending and SH% will drift closer to 1000 over the season as expected by this statistic. If the correlation of PDO to P% is accepted, then CHI and PIT maybe considered lucky and ANA unlucky.

Figure 3: P% vs PDO

108_medium

Equal Weighted PDO

One complaint of PDO is that the PDO is largely influenced by EV SV% which is near 92% and less so by EV SH% typically at 8%. That is, weak goaltending will result in a low PDO whereas poor shooting percentage may not be as significant to the PDO. Below is a Normalized PDO that equally weights team shooting (offence) and team saving (defence). The correlation of NPDO to P% is only 0.295 and the variance is greater with this measure so the 3 sigma band is added. From the chart, PHI with average goaltending and superior shooting percentage has a better NPDO score whereas Florida with average goaltending and lower shooting percentage has a much poorer NPDO. It is curious that teams with good goaltending (BOS/NYR happen to have good shooters and teams with struggling goalies have poor shooters (LA and COL).

Figure 4: P% vs Equal Weighted PDO

112_medium

NPDO is interesting way to observe the effect of shooting % however, it is probably simpler to observe shooting percentage directly. In addition, NPDO does does not better correlate to P% then the original simple additive method. Also changing the weights of SV% and SH% does not improve correlation either and I find it mysterious that the simple PDO addition of SV% and SH% which is more heavily weighted to SV% works so well but it does. Actually a slightly better correlation is achievable by slight adjusted from 1 to 1, but I don't think it is worth the effort.

How well PDO Predict Future Winning?

Below is a plot of the percent change in point percentage and the percent change in PDO from Nov 24 to Dec 24. The correlation is 0.425. Vancouver and LA are great examples of PDO in action where VAN improvement in PDO (SV% in particular) improves Point % and conversely with LA (with poor SV% and SH%). But there are some exceptions like Calgary that rockets in points with little PDO change or EDM that tanks with little PDO change. Also NSH is worth understanding with a decrease in PDO but no change in points. So why does PDO predict P% often but at times fail entirely?

Figure 5: Change in P% vs Change in PDO

123_medium

How well does Fenwick Predict Future Winning?

Well let's look at the same chart above with Fenwick% change from Nov 24 to Dec 24 as presented below. The correlation is very poor at only 0.09 rsquared. We see that Boston's Fenwick improvement and Florida's Fenwick align with the PDO change and that both Fenwick and PDO changes contribute to the Point % change. Also note that Calgary P% change is better explained with Fenwick possession improvement then from the little change in PDO. Also that the Wild's and Tampa's decrease in PDO appears to dominate their significant Fenwick increase. And contrast that to Colorado or Ottawa whose PDO increase dominates their decrease in Fenwick. Along with the Calgary example, this seems to suggest that PDO change is "more important" then Fenwick change. And that the change to Fenwick is important only when PDO does not change or when PDO change is small.

Figure 6: Change in P% vs Change in Fenwick

132_medium

However, EDM remains a mystery with no change in fenwick (and no PDO change) but a significant decrease P%. And ANA increase in both PDO and fenwick does not match with the P% decrease. Also worth considering is NJD Fenwick improvement (with no change to PDO) and no change to P% is this bad luck? Conversely Nashville PDO and Fenwick decrease with a slight improvement to P% - Good Luck?

When results do match PDO/Fenwick predictions is this Chance?

One clue maybe gleaned by going back to the sigma bands on the Fenwick charts (Figure 1) and PDO (Figure 3). ANA, EDM, NJD all show up near the 2 sigma bands of PDO and Nashville in 2 sigma band of Fenwick. There are other "luckier" or "unluckier" teams than these but those team's PDO and Fenwick are somewhat aligned with the luck they are seeing. That is, they are skilled (measured by PDO and Fenwick) and also "lucky" or "unskilled" teams that are "unlucky" (For example, PHI is luckier then their already positive PDO or Fenwick would suggest). These four (ANA, EDM, NSH, NJD) seem to be subject to chance that goes "opposite" to their predicted Fenwick and PDO.

A Few Observations

1. The change in PDO appears to be the dominate factor impacting in change of P% (winning). An improvement in Fenwick is not as important as improvement in PDO. As we know improving PDO (goaltending/defence or SH%) is not a skill that is improved easily. That is, a coach cannot simply ask their goalie to save more shots or a player to score more goals though there is great debate on this point. Perhaps style of play or streaks and slumps effect PDO but PDO seems to correlate well to winning. However, there is little meaningful insight beyond understanding the skill and luck that exists on the team and potential regression.

2. The change in Fenwick appears to have a smaller additive winning effect to the change in PDO and becomes important when the teams PDO does not change (This is not proven statement, rather there is no instance where Fenwick "dominates" PDO. That is there is no example of a high puck possession team and low PDO with a high P% (high winning %) or a low possession team with high PDO and a low P% (losing % team). And it would be interesting to find if PDO also dominates Fenwick for individual players GF/GA ratio as it appears to hold for teams). I'm not sure why Fenwick is used as the predictive measure of team winning when PDO appears to be more important factor (but granted the sample analysis here is small). This data seems to suggest that PDO is responsible for 50% of team winning and Fenwick closer to 25% in the small sample analysis. I will have more on this looking at PK and PP as well.

3. Regression of PDO and changes to fenwick (possession) do not provide the full explanation of why teams win. There is some other factor(s) that may be a measureable skill (shooting skill, PK or PP or skill when the game is not tied) or simply "luck" or "chance" events. If the explanation is chance, then chance can randomly dominate both Fenwick and PDO at times.

If Fenwick (FTR%) and PDO are critical - why not add them?

Being that Fenwick and PDO are important to winning and both add a different dimension to hockey analysis, I had the idea to add Fenwick Road Tied % to PDO (similar to how SH% was arbitrarily added to SV%). The correlation to P% was quite high at 0.638 for this season up to this point. Perhaps someone will check the correlation over past season. Also on a side note, I found by using adjusting the weights to Fenwick + PDO/2, an improved Rsquared was achieved (0.69). From this view it appears NJ, PIT, PHI and CHI are lucky or have higher skill then NHL average and CBJ, ANA and EDM are unlucky or lower skill then the NHL average.

272_medium

Change in P% vs Change in (Fenwick + PDO)

Now plotting the change in P% vs FTR + PDO from Nov 24 to Dec 24 provides a good correlation of 0.470. And the tool provides a visual way to see which teams are lucky (based on the combination of PDO and Fenwick). From this, Calgary's winning seems to be lucky or is not well explained by both Fenwick and PDO. EDM and ANA are somewhat unlucky.

147_medium

Intuitively this measure PDO + FTR includes offence (SH%) and defence (SV%) and puck possession (Fenwick) so it seems to offer a more complete reason for why teams win games and may explain the better correlation observed so far this season. And as shown, it does not offer a complete explanation as in the case of Calgary (and for the remaining 30 to 40% of P% winning). As for the rest of the explanation, perhaps shooting skill needs to be added (Iggy did start scoring more goals for Calgary recently). But more likely there maybe some other factors like PK, PP or play when the games are not tied) or simply "luck" or "chance" events. Using PDO + FCR (close fenwick improves Rsquared by 0.01.

My Takeaways (From The Partial Season Analysis)

  • PDO is more important then Fenwick in regards to predicting winning
  • (Alternatively a high Fenwick (puck possession) will not compensate for poor PDO (SV%+SH%))
  • PDO may persist for some teams (some teams may be better/worse skilled rather then lucky)
  • PDO plus Fenwick Road Tied appears to be a better measure to predict winning
  • PDO and Fenwick do not provide the complete explanation of winning

PensionPlanPuppets.com is a fan community that allows members to post their own thoughts and opinions on the Toronto Maple Leafs and hockey in general. These views and thoughts may not be shared by the editor of PensionPlanPuppets.com.

X
Log In Sign Up

forgot?
Log In Sign Up

Forgot password?

We'll email you a reset link.

If you signed up using a 3rd party account like Facebook or Twitter, please login with it instead.

Forgot password?

Try another email?

Almost done,

Join Pension Plan Puppets

You must be a member of Pension Plan Puppets to participate.

We have our own Community Guidelines at Pension Plan Puppets. You should read them.

Join Pension Plan Puppets

You must be a member of Pension Plan Puppets to participate.

We have our own Community Guidelines at Pension Plan Puppets. You should read them.

Spinner

Authenticating

Great!

Choose an available username to complete sign up.

In order to provide our users with a better overall experience, we ask for more information from Facebook when using it to login so that we can learn more about our audience and provide you with the best possible experience. We do not store specific user data and the sharing of it is not required to login with Facebook.

tracking_pixel_9355_tracker