Now that the Best Worst Team in Hockey ™ has won the draft lottery, and will almost certainly draft Connor McDavid, what’s a reasonable RE for young McSaviour next year?

*NOTE: I just updated this to include a link to the Excel file containing the underlying data set. See note at end of post.*

If you use a “classic” equivalency like Desjardins, OHL to NHL is about 0.3. That puts McDavid at 2.55 ppg x 0.3 x 82 games = 62 to 63 points in a full season.

The problem with that equivalency is that it is deliberately broad, and its right to ask the question: how well does it apply to elite players? A quick look suggests that many forwards drafted in the top 6 have outperformed the NHLE.

So I thought it would be interesting to look only at elite forwards drafted out of the CHL and see if that gives us anything interesting.

### Data

Looking back to 2005, forwards drafted in the Top 6 from the CHL who then played at least 10 games in the NHL the next year: Leon Draisaitl, Nathan MacKinnon, Sean Monahan, Nail Yakupov, Alex Galchenyuk, Ryan Nugent-Hopkins, Gabriel Landeskog, Taylor Hall, Tyler “Sequins”, John Tavares, Matt Duchene, Evander Kane, Steven Stamkos, Patrick Kane, Samwise “Hobbitses!” Gagner, Jordan Staal, and of course … Sidney Crosby.

It’s a sparse (17) dataset, but one makes do with what one must.

### Correlations

Let’s take the points (goals, assists, points) rates of the players in that dataset and see how that correlates to their ppg first year in the NHL*:

Goals do a surprisingly poor job of predicting success in the NHL. This surprised me because I’d heard differently. I have a feeling it was because of those four guys in the middle, who fit the pattern perfectly. But the bigger pattern contradicts the smaller one. Keep it in mind though – for high-end but not truly elite draft picks, goals may indeed be an outstanding predictor.

Overall, assists do a better job than goals alone. Then again – that’s usually a larger number, so it may be a spurious correlation reflecting magnitude rather than results.

And lastly, we find that there is a good correlation between scoring in the CHL and scoring in the NHL. Huzzah!

One more thing before we start predicting things … a look at correlation between goals/goals and assists/assists:

Again – a surprisingly poor correlation between goal scoring at the Jr level and at the NHL level. And remember, these are the elite players.

Playmaking skills on the other hand very much seem to carry over.

### Predictions

OK, so knowing that Jr ppg is a good predictor of NHL ppg, let’s run a multiple regression on Gpg and Apg correlated to NHL Ppg and use the resulting model in the prediction process (the gory details of the regression are at the end of the post for those of such inclination).

Just as quick sanity check, we’ll confirm that the model result is in line with what we saw in the ppg correlation, and indeed it is, almost exactly (though if you look closely at the dots, you’ll see they are in fact different).

The prediction model looks like this: NHLppg = draftgpg x 0.2679 + draftapg x 0.3621 + 0.0697

- Connor McDavid: GP 47, G 44, A 76
- Model prediction: 44/47 x 0.2679 + 76/47 x 0.3621 + 0.0697 = 0.2508 + 0.5855 + 0.0697 = 0.906
- The model predicts McDavid will be at 0.906 ppg next year.

- Over 82 games, that’s 74 points.
- Also worth noting that in our dataset, the average number of games played is 71, with 9 players ringing in 79 or more games. Pro-rating the season to 79 games spits out about 72 points.

And lastly, we can consider a couple of possible adjustments to this number:

- The Oilers D sucks at moving the puck. If you’re feeling pessimistic about that, you might scale the number downward.
- The Oilers don’t have a good track record of protecting players. If you’re feeling pessimistic about that, you might scale the number of games played downward.
- The two players closest to McDavid are Kane and Crosby. Kane’s model number is almost dead on (71.7 predicted vs 72 actual). Crosby blew his out of the water (77 predicted vs 102 actual). McDavid’s balance between goals (which are not so sticky) and assists (which are more sticky) are more like Crosby’s numbers than Kane’s numbers. Kane’s assists/goals ratio was 1.34. Crosby’s ratio was 1.55. McDavid’s ratio was 1.73.

**So taking a baseline of 70 to 74, it would be perfectly reasonable (pessimistic) to scale that back to about 65 points (or lower if you’re worried about durability), but it’d also be perfectly reasonable (optimistic) in riding the assists boost and the Crosby comparables and scaling that up to 80 or 90 points. **

As a rookie.

McJesus indeed.

### A Word About the Charts

* You’ve seen these charts before. The two variables are shown in a scatter diagram. The individual distributions of the two variables are shown above and to the right, and include a kernel density estimation i.e. a smooth version that helps with generalization and comparison. The line is the regression line of course, and the shaded area is the 95% confidence interval. The text shows the correlation value and the p (significance) value.

### Gory Regression Details

For those that care, the data file for this study can be found here: Draft Study NHLE Elite Players 2015-04.xlsx.

And here’s the multiple regression parameters using Draft Gpg and Draft Apg to predict NHL Ppg:

————————-Summary of Regression Analysis————————-

Formula: Y ~ <DraftGpg> + <DraftApg> + <intercept>

Number of Observations: 17

Number of Degrees of Freedom: 3

R-squared: 0.3622

Adj R-squared: 0.2711

Rmse: 0.2014

F-stat (2, 14): 3.9755, p-value: 0.0429

Degrees of Freedom: model 2, resid 14

———————–Summary of Estimated Coefficients————————

Variable Coef Std Err t-stat p-value CI 2.5% CI 97.5%

——————————————————————————–

DraftGpg 0.2679 0.2690 1.00 0.3362 -0.2594 0.7951

DraftApg 0.3621 0.1771 2.04 0.0602 0.0150 0.7093

intercept 0.0697 0.2138 0.33 0.7492 -0.3493 0.4886

———————————End of Summary———————————

## One thought on “McPredicting: what does the math say?”