Is there a sophomore slump, or just probability?

[Czech Your Math]

One change to suggest - I'd remove the games played restriction in year two. If you keep it, then you'll miss players who disappeared entirely after their rookie season (the ultimate sophomore slump).

That may be a good idea, but I tend to think there should be some minimum number of games for the second year. If a player has substantially reduced number of games in year 2, it's probably due to:

A) injuries
B) being traded to a different team that has more depth at his position- this isn't a "slump" in the traditional sense
C) his team acquiring player(s) at his position which are better than him- this would seem to be more related to his performance in his first season, and his expected performance in his second, than his actual second season performance (except that it likely gives him a lesser role on the team)
D) the player is sent down to minors- like "C", it's very possible this is related to his first season as much or more than his second

What's the maximum number of games a player can play before he's considered an NHLer for the year? Isn't it like 10 or 12? I would set the minimum for season two above that... perhaps twice that number.

 

[charlie1]

One change to suggest - I'd remove the games played restriction in year two. If you keep it, then you'll miss players who disappeared entirely after their rookie season (the ultimate sophomore slump).

Ignoring the small sample size problem, what is your metric of performance? The only ones that make sense to me are rates (e.g., ppg), which would be undefined for players who disappeared in year two.

 

[Caeldan]

So here's my thought to still look at rate - but adjust for people dropping out over the years:
% contribution to total league point totals by cohort.

So we use PPG (with minimum of 20 games played - 2x the slide rule amount for a rookie) to establish the 'overachiever' cohort per year. Criteria for overachiever to still be exactly defined but somehow based on a top percentile of the rookie group for that season.

Then take that cohort through the seasons, and just do a simple (points by cohort)/(total league points) for each season.

See if there's any trend that occurs?

 

[charlie1]

So here's my thought to still look at rate - but adjust for people dropping out over the years:
% contribution to total league point totals by cohort.

So we use PPG (with minimum of 20 games played - 2x the slide rule amount for a rookie) to establish the 'overachiever' cohort per year. Criteria for overachiever to still be exactly defined but somehow based on a top percentile of the rookie group for that season.

Then take that cohort through the seasons, and just do a simple (points by cohort)/(total league points) for each season.

See if there's any trend that occurs?

When people talk about a slump, they're talking about players who played poorly in their second year, not players who dropped out. So I don't even think we need to consider the dropout effect. Let's just take the mean ppg of all the players that played at least N games in their 1st and 2nd years and see if there is any slump. Split that up into overacheivers and underachievers however you'd like.

I think we can drop the whole dropout discussion.

 

[Doctor No]

If someone's out of the league entirely in year two, wouldn't you consider their performance a slump?

 

[Caeldan]

If someone's out of the league entirely in year two, wouldn't you consider their performance a slump?

That's sort of what I'm thinking, and why I thought looking at the overall performance of the cohort vs the league would be a nice simple way to account for those players without losing anything else.

Now the question is. Anyone know a database where it's relatively simple to build the cohorts to get the information?

Also. Secondary question... for goalies, is there a stat we could use for them to work similarly? Just define their performance by wins and look at how many wins their cohorts contribute? I'm assuming the sample size though would be ridiculously low by year especially if a GP min is instituted.

 

[charlie1]

If someone's out of the league entirely in year two, wouldn't you consider their performance a slump?

Personally no I wouldn't. Whenever I hear people talk about sophomore slump it is always in regard to a player that actually played in their sophomore year.

Also, Caeldan-- Are you planning on identifying overachievers by ppg, but then measuring the slump based on points? That's what it seems like based on your formula. My issue with that (if that is the case) is that it does not control for ice-time in the second year, which I expect will increase.

I would do this like so: Identify all the players that played in a first AND second year. Track their pp60 in both years. For each player, determine if they "slumped" or not. Quantify the extent of the slump by delta pp60 (change in pp60), then plot delta pp60 vs. rank of player in year 1 to see if there is a correlation between rookie rank and delta pp60. Negative correlation suggests the higher your rank the more likely you are to slump.

You could even put in a little asterisk to denote how many players dropped out .

 

[Doctor No]

Personally no I wouldn't. Whenever I hear people talk about sophomore slump it is always in regard to a player that actually played in their sophomore year.

Just to clarify, I'm not referring to the odd and rare situation where a player just doesn't play at all in year two.

If a player has a great rookie year, and then spends the entire year two in the AHL, you wouldn't suggest that their play regressed?

 

[charlie1]

Just to clarify, I'm not referring to the odd and rare situation where a player just doesn't play at all in year two.

If a player has a great rookie year, and then spends the entire year two in the AHL, you wouldn't suggest that their play regressed?

First off, I think it is relatively rare to have a player with a great rookie year who then is sent down to the AHL for the ENTIRE year 2.

It does happen however, and this is my thought on that:

You're assuming that a player who was sent down to the A was done so because they regressed. When really there are other reasons why a player could be sent down that have nothing to do with their performance. For example, the Islanders' Anders Lee had an excellent rookie season last year, putting up ~0.6ppg. Despite that, he will likely not play much this season because the Isles picked up three new forwards who would have to go through waivers before seeing the AHL. Lee is waiver ineligible so he will likely be in the A for the majority of the season.

Add in the fact that a player could be injured while in the A, which would end their season.

So no, i'm not willing to call that a sophmore slump. I'd rather just leave those players out of the analysis, since we don't have any way to measure their performance.

Also, the most common use of "sophmore slump" IMO is in reference to very good rookies who played poorly the following year. Not very good rookies who disappeared the following year.

So I would get at this question as follows: A simple plot of (pp60 in year 1) vs (pp60 in year 2) for everyone that played some threshold number of games in both year 1 and year 2. Each dot on the graph is a player, and I would add a one-to-one line so we can look for trends relative to whether a player was an overachiever or underacheiver, e.g., fit a quadratic function.

 

[mkwong268]

That's sort of what I'm thinking, and why I thought looking at the overall performance of the cohort vs the league would be a nice simple way to account for those players without losing anything else.

Now the question is. Anyone know a database where it's relatively simple to build the cohorts to get the information?

Also. Secondary question... for goalies, is there a stat we could use for them to work similarly? Just define their performance by wins and look at how many wins their cohorts contribute? I'm assuming the sample size though would be ridiculously low by year especially if a GP min is instituted.

From my really basic query freshmen get about 6.7% of the points and sophomores get about 8.8% of the points from 1930-2000.

 

[mkwong268]

Years since 1st year in NHL and percent of league points from 1930-2000
0 0.067145858
1 0.087794993
2 0.096505916
3 0.101761387
4 0.100062921
5 0.093906049
6 0.086135402
7 0.076540662
8 0.066934199
9 0.055057492
10 0.045678277
11 0.036181961
12 0.02606681
13 0.018950137
14 0.013682421
15 0.00987415

 

[Caeldan]

Years since 1st year in NHL and percent of league points from 1930-2000
0 0.067145858
1 0.087794993
2 0.096505916
3 0.101761387
4 0.100062921
5 0.093906049
6 0.086135402
7 0.076540662
8 0.066934199
9 0.055057492
10 0.045678277
11 0.036181961
12 0.02606681
13 0.018950137
14 0.013682421
15 0.00987415

Is this the entire class, or just the set of overachievers?
Also find the distribution interesting over the years in general

 

[mkwong268]

Is this the entire class, or just the set of overachievers?

Entire class.

Edit: It'd be a lot more work to only grab players who overachieved in their first year.

 

[charlie1]

Is there any chance someone could make a table with the following columns:

Season|Player|Years since first year|Points|TOI

If so I could do the analysis I described (which would have individual player trajectories).

 

[Caeldan]

Entire class.

Edit: It'd be a lot more work to only grab players who overachieved in their first year.

Out of curiousity, what'd you use to grab the data?
Also - was this forwards only, or all players?

 

[Brainiac]

There's probably a lot of confirmation bias associated with the concept of sophomore slump.

 

[mkwong268]

Out of curiousity, what'd you use to grab the data?
Also - was this forwards only, or all players?

I have access to a SQL server with all the NHL.com data scrapped.

All players.

 

[Caeldan]

I have access to a SQL server with all the NHL.com data scrapped.

All players.

Yeah I can see how trying to set up overachiever cohorts would be a bit trickier... almost would have to build a second table to then pull data off I'd think?

Just to see if it skews the distribution, could you do a query that only shows forwards who had 20 or more GP in their rookie season?

 

[wgknestrick]

So here's my thought to still look at rate - but adjust for people dropping out over the years:
% contribution to total league point totals by cohort.

So we use PPG (with minimum of 20 games played - 2x the slide rule amount for a rookie) to establish the 'overachiever' cohort per year. Criteria for overachiever to still be exactly defined but somehow based on a top percentile of the rookie group for that season.

Then take that cohort through the seasons, and just do a simple (points by cohort)/(total league points) for each season.

See if there's any trend that occurs?

Why would you ever look at PPG? You are assuming that every player's "games" are equal when we all know that players vary significantly in ice time and PP time per game. PPG is probably a worse stat than +- for comparisons.

5v5 pts / min are where you start ANY evaluation IMO. You can add further adjustments (ie 5v4 pts/min, ZS%, QoT) from there if you'd like, but you can NEVER have a "game" as a denominator and think you are comparing apples to apples.

 

[wgknestrick]

Years since 1st year in NHL and percent of league points from 1930-2000
0 0.067145858
1 0.087794993
2 0.096505916
3 0.101761387
4 0.100062921
5 0.093906049
6 0.086135402
7 0.076540662
8 0.066934199
9 0.055057492
10 0.045678277
11 0.036181961
12 0.02606681
13 0.018950137
14 0.013682421
15 0.00987415

This is just confirming NHL peak production age (24-26yo) graphs. We need to know what % of 2 year players fall below their 1st year production.