# Expected Goals - Forwards

There exists a few formulas for expected goals out there, but with a few more seasons worth of data, I was wondering if we could do any better. My favorite one currently runs here at hockeyanalytics.com.

That post ran in 2012, and we've got some more data to look at now.

To see if something new can be developed, I ran a correlation for all statistics to goals to decide which variables to use in my regression analysis. The data set that was used was forwards since 05-06 minimum of ten games played in a respective season, all 5v5.

Variables:

• Individual goals per 60
• Individual high danger scoring chances per 60
• Individual Fenwick per 60
• Teammate Corsi For per 60
• Takeaways per 60

These four variables prove to be very strong indicators for expected goal totals. Running a regression for the full data set returns an adjusted R^2 of 0.892.

In order to see if this formula has any predictive values, we will randomly select half of our data set, run a new regression, and see if the resulting formula fits well to the half of data that was not included in the analysis.

The random half regression returns an even stronger adjusted R^2 at 0.897.

Here is the random half regression analysis formula applied to the half of players that were not featured in the regression:

Definitely a very strong relationship.

This can prove to be a very useful stat. Hockey, when you boil it down to its pure core is a simple game: score more goals than your opposition. A metric like expected goals, that clusters very closely to a line of best fit, is a 'moneypuck' metric. It is with these kind of metrics that you find your hidden gems.

As we enter the 'dog days' of the off-season, we may be able to find some "defective" players. Players who, for one reason or another, are not being given their chance by NHL teams. Jeremy Giambis, Scott Hattebergs, David Justices.

Of remaining UFA's, here are the top 10 in terms of expected performance:

And of the top 50 remaining UFAs (in terms of last year's salary), here are the bottom 10, the guys GMs may want to think twice about overpaying for:

The really interesting aspect of this analysis, at least to me, is that it is another metric that Bergenheim excels in, or in this case, proves that he is underrated in (more on Bergenheim in our next post).

Resulting equation:

Expected Goals = -2.10 + (g60*7.29) + (iHSC60*0.122) + (iFF60*0.064) + (tCF60*-0.025) + (TK60*0.097)

Other notes:

Among players who appeared in at least 25 games in 14/15:

1. Nathan Gerbe had the highest differential; scoring 7 goals while expected to score 11.8
2. Alex Tanguay had the lowest differential; scoring 17 goals while expected to score 10.59.
3. Rick Nash had the highest expected goals number at 24.46
4. Jordan Caron had the lowest expected goals number at -2.16
1. 10 players ended the season with a negative expected goals number:
1. Jordan Caron, Brian Gibbons, Pat Kaleta, Cody McCormick, Michael Latta, Colin Greening, Chris Neil, Scott Laughton, Joakim Nordstrom, and Jared Boll.
2. Of these ten players, only Chris Neil and Jared Boll would score this season (1 goal each)