One of the biggest conversations going on in NHL front offices today is not just about analytics. It's also about how to apply statistical analysis to the world of prospects. These discussions were brought up with prospects like Lawson Crouse and Daniel Sprong last season, and they will apply to top prospects like Matthew Tkachuk and Alex DeBrincat this season.

The proper approach is synergy in this area -- and for the NHL to enter the new age of decision-making at the draft, as other sports have already done. Some NHL teams have faulted in their processes in recent seasons, including passing up players whose greatness a simple computer model could have foreseen.

The issue with a scouts-only approach

Calgary Flames president Brian Burke is famous for saying, in less eloquent terms, that stats at the draft could be a helpful tool to supplement the scouts but that this is still an eyeball business. Other hockey people have echoed the phrase, opining that stats cannot be a primary tool until they leap the accuracy rate of the scouts.

The scouts provide a lifetime of experiences in the game that they can apply to player evaluations. Their abilities to scout a tournament, a weekend or even just one game and be able to roughly gauge the caliber of a player is not something that should be downgraded. The best ones bring attention to detail, a furious work ethic and a library of experience with which to identify trends. When you think about what scouts have done for decades -- subjectively evaluating teenage hockey players and having a pretty high hit rate-- it is really quite remarkable compared to many other industries in the world that rely on subjective experts.

The issue with relying solely on scouts can be seen in the work of psychologist Paul Meehl, whose extensive research showed that simple formulas (in the field of psychology) often out-predict experts. This is due to the fact that experts tried to be too clever in attempting to spot the exceptions to the numbers and could not summarily comprehend all the information they were trying to evaluate.

Statistics and algorithms can solve these problems. Algorithms are not clever, they are consistent; numbers do not get overwhelmed by more numbers.

Of course, given how well scouts have performed in the many years without high-end statistical analysis available, there is obviously something to their expertise, showing that exceptions are needed in the field of sports. Numbers struggle to spot the guy playing hurt, the player with questionable work ethic, the player performing in a small sample way above or below where his skills dictate he should be.

Stats vs. scouts: By the numbers

I attempted to quantify the skill of scouts versus statisticians at a basic level. I did this by taking a data set consisting of the 1990-2010 drafts, of just first-year draft-eligible prospects and all CHL forwards drafted. I then created three categories, all relative to an individual draft class.

Industry rank: This ranking is simply the order in the draft the player was picked. While not all scouts agree on all players, an industry consensus definitely emerges.

Stat rank: This is a very simplistic way of approaching this. It is simply the player with the most points in his CHL draft season that is available. There is no adjustment for missed games or age/month discrepancies or team strength or height/weight or anything else.

Production: The amount of NHL points scored by the player. While this ignores defensive value, we’re focusing on forwards in this column, so it should capture much of the value. NHL scouts tend to agree that scoring is mostly what they want in forwards.

For example, Brandon Dubinsky was the 12th CHL forward selected in 2004, so he gets an industry rank of 12. But his 78 points in the WHL would rank him No. 2 in the stats rank, and his rank among CHL alum forwards in scoring from his draft class is No. 2, as well.

Here is a chart showing how well each of the groups did in predicting the production of the players in each class, as well as what happened in the case where I took a simple average of the scores -- in the Dubinsky example, it would be a 7.

In terms of trying to find a best method, there is an ideal weight you could put on both scores, but the maximum relationship didn’t differ much than simply taking an average of the two.

The height factor

What explains the strength of the stats rank? Part of it is the fact that a cold look at the numbers alone does not take into account a player’s height.

I found that NHL scouts in the 1990-2010 time span improperly rated height for both forwards and defensemen. Height was a poor predictor of NHL success and was a far inferior indicator than just simple scoring. This height problem goes to the aforementioned issues Meehl discussed with experts not being able to examine all the information precisely. From my experiences, every time a small guy lights up the CHL, Rob Schremp -- or someone of his ilk -- is bound to be brought up by someone in the discussion. This line of thinking isolates an example, rather than looking to see if there is a general rule on height that an algorithm would detect.

Another interesting takeaway here is that the stats rank correlated at a 49 percent rate to draft pick order, meaning it was achieving roughly the same results as the scouts while selecting players in a substantially different order. By achieving the same thing using different routes, it shows the two sides have something to offer each other.

The key takeaway here is not whether either score did better but rather that the two sides should have spent more time working together during the past several years, which would’ve resulted in significantly better results. The margin might not seem huge, but given the inherent giant error rate in drafting, such a small improvement in 20 years could yield huge differences for an NHL organization.

Stats vs. scouts in the 2015 and 2016 drafts

One doesn’t have to go back far to remember conflicts. The Lawson Crouse debate was well-documented last season; Crouse was a player without great point totals but whom scouts universally praised. That debate continued to rage even after a solid performance at the 2016 IIHF World Junior Championship and nearly making the Florida Panthers this season.

Despite fantastic statistical production, Daniel Sprong dropped to the second round in 2015, due to concerns from scouts about his all-around game. However, he made the Pittsburgh Penguins’ NHL roster out of camp.

There are major scouts versus stats debates already ongoing for the 2016 NHL draft. Here are a few examples:

• Matthew Tkachuk: Endorsed by most NHL scouts as a top-5 pick, some have Tkachuk as high as No. 3. Statisticians are worried that a top-5 pick should not have his production buoyed by so many secondary assists.

• Adam Mascherin: He has huge scoring and shot totals in the OHL, but scouts worry about his lack of speed hindering him as a pro.

• Tyler Benson: Widely hailed in the scouting community for years as being among the top players in the 1998 age group, he has been under a point-per-game pace this season, with several injuries.

• Tyson Jost/Dante Fabbro: Widely believed to be top-20 prospects, if not top-15 talent, in the scouting community, a lack of BCHL comparables worries statisticians. This is particularly true for Fabbro.

• Logan Stanley: He is a big, rangy defender who has not put up many points in the OHL this season, but scouts believe he has a fair amount of puck-moving skill and could be a first-round pick.

• Alex DeBrincat/Vitaly Abramov: These are small players who have put up giant numbers in their respective junior leagues, but they are being hedged on by scouts due to their size.

Maybe one side has more convincing arguments in particular cases than others, but in general, one should be able to see the conflicting evidence and try to find reasonable middle ground in these player evaluations.

Takeaways

There needs to be cooperation in the field of player evaluation. That information is power, since both the scouting and statistical disciplines have shown they have unique things to offer and together can help produce a greater result.

I’m not advocating that statisticians should be running the draft table. But in front office 2.0, the statisticians should have a fair opportunity to be a respected, if not equal, voice at the table. If there is disagreement with their findings, the scouting side has an equal burden to display why scouts think the quantitative evaluation is incorrect.

Together, maybe they can make fewer errors while getting better at finding the best NHL players.

Thanks to St. Lawrence University professor of statistics Michael Schuckers for help with this article.