Goalies and the Playoffs Part 3: How to Win a Series

Updated: May 24, 2018 at 2:37 am by Joel Short

We know that goalie performance is a major factor to playoff success. I wanted to get a better sense of just how big a factor it is, so I looked at shot volume and Sv% for every best of 7 playoff series since 1984. (I excluded the first round for 84 to 86, since it was then a best of 5.)

The chart below shows every best of 7 series winner in the past 33 years.  The x-axis is the winners’ SOG (the percentage of total shots in the series taken by the winners).  The y-axis is the winners’ PDO (Save Percentage + Shooting Percentage). Series in which the winner was outscored are shown in red.1


You can see that what I’ll call outsaving – i.e. which team has a higher Sv% – is a much better predictor of success than outshooting.  For series in which the team with fewer shots had a better Sv% (the upper left and lower right quadrants), the outshot team won 167 and lost 67 – a 71% win rate.  Teams that both outshot and outsaved their opponents were 203 and 16 – 93%.

As a rule of thumb, teams win about:

  • 40% of series when outshot (183W-270L)
  • 20% of series when outsaved (82W-374L)
  • 10% of series when outscored (34W-409L)

I also looked at only the years since 2006, to see if anything has changed significantly.  It hasn’t:

  • Since 2006, when a team is outshot but has a better Sv%, they win 68% (60W-28L)
  • Teams that outshoot and outsave win 93% (70W-5L)
  • Outshot teams win 40% (65W-98L)
  • Outsaved teams win 20% (33W-132L)
  • Outscored teams win 8% (13W-148L)

But that means…

If you’re familiar with PDO, you’re aware that it’s commonly understood to be mostly an indicator of luck.  PDO can swing wildly over short periods (like, say, 4 to 7 games), and numbers more that about 2% off par generally aren’t sustainable.  But because playoff series are short, a run of bad luck can sink even the best team.  Maybe it’s not surprising that teams with PDOs below .965 are winless in playoff series since 1984. What is surprising is that this represents 33% of all series: matchups in which the percentages were tilted to an insurmountable degree.

There are a few reasons that the common understanding of PDO (as ≈ luck) may be less true in the playoffs.  A key difference is that you’re only playing one other team.  Each team faces roughly average goalies and shooters over the course of a season, but in a playoff series, you may get a steady diet of Washington Capitals.I’m also receptive to the argument that certain teams “match up well” against certain others, and could (say) reduce the quality of that team’s shots in a sustainable way.  And to the extent that psychology affects the fluctuations Sv% and S%, those pressures are surely amplified in the playoffs – it can’t be fun to keep facing a goalie (or shooters) who have been eating your lunch for 3+ games.

But probably the biggest factor here is score effects.  The average series lasts 5.76 games,3 so series winners collectively win 69% of their games (against other playoff teams!) – an exceptional rate.  You’d win 57 of 82 games at that pace – the most since the ’06 Wings. The average series-winner spends significantly more time leading than trailing, which will boost their PDO, but drag down their share of shots.  In other words, winning teams’ high average PDO is a major cause of their success, but also partially an effect of success.  On the other hand, while outshooting your opponent is obviously beneficial, playing with the lead tends to suppress your shot share.  So the striking fact that you’re twice as likely to win a series while being outsaved than outshot is not as surprising as it initially seems.  Again, the impact of score effects is unusually strong in this case, because we’re looking at a sample of teams with a collective 69% win rate, over stretches of 4-7 games.  (In the future, I’ll look into PDO scores for teams that play in all 4 rounds.)

I also broke down the above chart by series length:

PDO vs SOG #Games

As expected, winners are more dominant in shorter series.  It’s interesting to see how some of the win percentages noted above shake down into series lengths:

Series length


Winner’s PDO <1.000

Winner’s SOG <50

PDO/SOG split*

PDO beats SOG

Win without either





























* How often the team with the better PDO takes fewer shots

† When PDO and SOG are split, how often does the team with higher PDO win

‡ When PDO and SOG are not split, how often does the team with lower numbers win

There’s little difference in most categories between 4 and 5-game series, but a lot of things become more likely if you can hang around for 7 games.  Also notable:  The R2 between series length and PDO is .32 (so high PDOs predict short series); the R2 between series length and SOG is basically nil.

But who cares about SOG

It makes sense to start by comparing SOG to PDO, because they’re the two variables that contribute directly to goals.  But not a lot of analysis is done using shots, because we have other stats that are more predictive of success/better indicators of which team is outplaying the other.  Puck on Net is a key resource for advanced stats, and while I was working on this article, they released adjusted shot metrics for the playoffs.

I decided to compare raw SOG to Puck on Net’s most heavily adjusted measure: Event, Score and Venue Adjusted Corsi (ESVA Corsi).  (An additional adjustment is that my SOG is for all situations, whereas ESVA Corsi is for even-strength only.)  The chart below shows series winners’ SOG, ESVA Corsi, and the difference between the two (green when the difference is positive, red when negative).



You can see that a positive adjustment is correlated with higher PDOs, and (less strongly) with lower SOG.  To check the strength of these relationships, I subtracted each team’s SOG from their ESVA Corsi, and compared the difference (the length and direction of the lines above) to their PDO and SOG.  The relationship with PDO has an R2 of .37; SOG has a negative R2 of .20.

I’m cautious about proclaiming ESVA Corsi (or any other stat) to be the true measure of which team outperformed the other, and Puck on Net’s data suggests that there’s not much difference in predictivity between different adjusted metrics over 4-7 games.  But it’s surely a better measure of team performance than raw shots.

Returning to the win rates above, teams win:

  • 31% of series with a negative ESVA Corsi (51 and 114)
  • 35% when ESVA is positive but PDO is negative (22 and 40)
  • 11% with negative ESVA Corsi and PDO (11 and 92)

Encouragingly (if you want the better team to win playoff series), ESVA Corsi is a better predictor of success than SOG – series go to the better team by this metric 69% of the time, vs. 60% for SOG.  Less encouragingly, ESVAC winners fare scarcely better than SOG winners when they lose the PDO battle: 35% vs. 32%.  The team with the better Sv% will beat the team with more possession 2 times out of 3.

If we remove SOG and the difference lines from the above chart, we can see a couple things more clearly:



Of note:

  • In contrast to SOG (the first chart), there’s essentially no relationship between ESVA Corsi and PDO.  Teams that score well on ESVA Corsi are no more or less likely to post high PDOs, relative
    to other series winners
  • But teams that win series are more likely than not to post positive ESVA Corsis and positive PDOs – 56% of winners are in the upper right quadrent.  (Whereas 43% of winners had positive PDOs and SOG.)  Add in the lower left quadrant, and 62% of series had the same team come out ahead in both PDO and ESVA Corsi.  Suggesting, perhaps, that short term PDO hot streaks – although not sustainable – are more likely to come to teams that are playing better hockey.
  • Only one winner with a PDO below .990 has a green dot (ESVA Corsi > SOG).  And only one winner with a PDO above 1.050 has a red dot (ESVA Corsi < SOG).  So teams with high PDOs tend to have more missed/blocked shots than their opponents, and/or play a lot with the lead.  Which suggests:
  • It’s fair to say that a team with a poor PDO had bad luck, but it’s a good bet that they didn’t play as well as their shot total suggests.

Again, I’m not arguing that ESVA Corsi is the best possible measure of which team played better.  I’d be interested in comparing PDO to other available metrics, if anyone can make a strong case that they better capture team performance.

I would also like to see how an adjusted PDO would compare to the raw PDO above.  Sv% is clearly affected by score, but mostly for the trailing team.  So if team spends a lot of time with the lead, their own Sv% won’t be meaningfully affected, but their opponent’s Sv% will be depressed.  It would be interesting to see how much the extreme PDOs above were boosted by taking a lot of shots against teams that were trailing and gambling for offence.  But as far as I can tell, no one is tracking any form of adjusted PDO.

A few takeaways

  • Goalie performance is a huge factor in playoff success.  If you know which team had the higher Sv%, you can predict the winner about 4 times out of 5.  But it’s hard to disentangle short-term
    performance from luck.
  • The role of luck in the playoffs is large, due to the high degree of parity in the league, and the relatively short length of a playoff series.  Goalies and shooters can get incredibly hot or cold over short stretches, and a low PDO is extremely hard to overcome.
  • In fact, teams with PDOs below .965 are 0W-151L.  This represents 1 in 3 series since 1984.  Historically, you’d be better off losing your first 3 games than having a PDO below .965.
  • ESVA Corsi suggest that more often than not, teams with positive PDOs also outplayed their opponents.  However, when PDO and ESVA Corsi diverge, the team that wins PDO usually wins the series.

Up next

I’ve focused on individual series in this article, but of course you need to win four of them to claim the Cup.  I’m planning to look into the impact of goaltending on teams that play a full four rounds.   I’d also like to compare the career postseason performances of goalies throughout history.

1 A note on methodology:

I manually removed empty net goals from the series totals, since they a) don’t reflect the performance of the goalies, and b) disproportionally affect series losers, inflating winners’ PDO.  Unfortunately, the box score data I relied on was incomplete for a few years in the 80’s.  So a few of the dots are a bit lower or (more often) higher than they should be.

2 Although strong regular season PDO teams are not necessarily strong on PDO in the post-season.

3 The average length of best of seven series if the results of each game were totally random (e.g. a coin toss) would be 5.8125, if I’ve done my math correctly.  The fact that the average playoff series is only ~.05 games shorter than that is another indication that the luck is a major factor in playoff success.  To put in another way, there’s a high degree of parity in the NHL, and seven games isn’t much time for the minor disparities in skill between playoff teams to assert themselves.

As a comparison, the average length of a 7-game NBA series since 1984 is 5.62 games – almost .2 games shorter than a best of 7 coin toss, and .14 shorter than the NHL.

Goalies and the Playoffs: Series Home


, , , ,