clock menu more-arrow no yes mobile

Filed under:

Clear Victories and One-Goal Games (Part Two)

It's time for Part Two.
It's time for Part Two.

Yesterday, I wrote about the fact that success in games decided by two goals or more tends to carry over into the next season much better than success in games decided by one. There were some good points made in the comments about how this analysis might be improved. Some of those improvements will take me some time (I'll need to learn how to use a program or two), but another was easy enough to look up. One of our commenters, Dessert1111, asked the following:

How big is the difference between teams repeating their 1-goal victories year to year and their 2-goal victories (not 2+ goals) year to year? And how big is the difference between teams repeating their 2-goal victories year to year and their 3-goal victories year to year?

After the jump, I'll take a look at the data to come up with an answer.

The NHL keeps track of two-goal games and games decided by three or more as separate categories, so it's pretty easy to gather the winning percentage data. Here's what the Pearson correlations for our various season pairings look like:


I found these results pretty interesting. The results for two-goal games alone are pretty poor, but including them in the overall numbers for "Clear Victories" helps because it increases the sample size. This led me to wonder whether or not stripping away the one-goal games is actually helpful. Here's what the chart looks like with an "All Games" column added:


It's an interesting result. Whenever you cut down on the sample size (and that's what we're doing by cutting out games decided by one or games decided by one or two), you end up losing some accuracy. The result is that whether we're looking at all games, 2+ goal games or 3+ goal games, the correlation is very similar over a large sample of seasons, and yet, in any individual season, the three groups perform quite differently.

So what might help these results? I think the answer might be a larger sample. That being the case, the next place to look is whether or not two seasons worth of results (e.g. 2005-06 and 2006-07) do a better job of predicting the third season (e.g. 2007-08), which is something I'll take up tomorrow.