A Rebuttal In Praise

As Micah published his posts over the past couple of weeks, I was obviously irked that his projections had me squarely in second place. There were a couple of other things that didn’t quite sit right with his model, and so I set out to improve on it, but more importantly settle on something that would crown me the preseason favourite. After toiling with Micah’s points above replacement allotment for draft picks, however, I realized that I couldn’t really do much better—either in translating draft position to predicted points from the draft OR in predicting an overall first place finish for the Patrik Stefans. You’ll see I inevitably lean heavily on Micah’s approach; I think he’s done well to quantify each team’s relative strength to one another, and as you’ll also see, he’s done pretty well in building a pretty damn good team.

As a first point of departure, I simply used Left Wing Lock’s projections instead of Dobber’s for each team’s projected keeper roster. There have also already been four trades that have affected Micah’s rankings, and I have incorporated those into my analysis. Below is a ranking of KL teams by their projected keeper roster points.

Patrik Stefans	673
Milan Micahleks	671
Los Amjawors Kings	647
Mackhawks	629
Valeri Nickrooshkins	628
Quebec Rordiques	616
Phillipsadelphia Flyers	611
Teeyotes	594
Winter Claassics	591
Fylanders	576
Powder Rangers	571
Joshfrey Krupuls	560
WBS Parkers	545
Toilers	533
Schizzarks	528
Dicklas Lidstroms	481

That’s a much better start. But now the tricky part. How can we meaningful look at a team’s complement of draft picks and assign it some value that captures its absolute strength for the purposes of our format?

It initially struck me as incorrect that Micah focussed on points above replacement for his entire analysis. For starters, there didn’t seem to be much reason to obfuscate season ending totals, when you could simply add replacement value of 560 points (9*40 + 4*30 + 80) to his totals and put things in more intelligible, apples to apples, terms for GMs. But I see the need for his approach when we turn our minds to the draft and the uncertainty of a forward, defenseman, or goalie getting nabbed at each draft pick.

The second thing that seemed wrong about Micah’s approach to the draft, though, was that he totalled all 10-12 picks’ points above replacement. Our scoring rosters cap out at 14. What does my 6^th round pick matter if I can fill out my scoring roster with my earlier picks? But then, of course, every year we see GMs hitting homeruns in late rounds, and so those picks actually do matter #allpicksmatter.

I was inspired by Micah’s attempt and also this paper by some oyster lover to come up with a truer way of capturing pick value, based on probability. http://myslu.stlawu.edu/~msch/sports/Schuckers_NHL_Draft.pdf

Taking draft data from 2013-2016, and using Micah’s fungible thresholds of 30, 40, and 80, I assigned probabilities* to each draft pick 1 through 160 of picking a player to surpass those thresholds (table below). I then multiplied those probabilities by the average points* above replacement for each draft position. The result is the graph below, after and about which I provide further comment.

Draft Pick	P(scoring roster)
1	0.667
2	0.563
3	0.55
4	0.536
5	0.5
6	0.429
7	0.464
8	0.464
9	0.536
10	0.536
11	0.536
12	0.536
13	0.571
14	0.607
15	0.607
16	0.5
17	0.5
18	0.464
19	0.5
20	0.464
21	0.464
22	0.464
23	0.536
24	0.536
25	0.571
26	0.5
27	0.536
28	0.571
29	0.571
30	0.536
31	0.607
32	0.571
33	0.643
34	0.607
35	0.571
36	0.643
37	0.607
38	0.536
39	0.607
40	0.5
41	0.464
42	0.429
43	0.393
44	0.393
45	0.357
46	0.357
47	0.464
48	0.464
49	0.464
50	0.464
51	0.5
52	0.571
53	0.464
54	0.357
55	0.429
56	0.393
57	0.393
58	0.393
59	0.357
60	0.464
61	0.5
62	0.5
63	0.536
64	0.5
65	0.536
66	0.5
67	0.464
68	0.5
69	0.464
70	0.429
71	0.429
72	0.357
73	0.357
74	0.286
75	0.214
76	0.143
77	0.214
78	0.25
79	0.286
80	0.321
81	0.357
82	0.429
83	0.5
84	0.464
85	0.393
86	0.321
87	0.25
88	0.25
89	0.214
90	0.179
91	0.107
92	0.214
93	0.214
94	0.25
95	0.25
96	0.214
97	0.179
98	0.179
99	0.107
100	0.143
101	0.179
102	0.179
103	0.214
104	0.286
105	0.321
106	0.393
107	0.429
108	0.357
109	0.393
110	0.393
111	0.321
112	0.321
113	0.214
114	0.143
115	0.179
116	0.143
117	0.179
118	0.179
119	0.143
120	0.179
121	0.179
122	0.143
123	0.143
124	0.071
125	0.071
126	0.107
127	0.071
128	0.143
129	0.143
130	0.143
131	0.179
132	0.214
133	0.179
134	0.179
135	0.107
136	0.143
137	0.107
138	0.107
139	0.071
140	0.143
141	0.179
142	0.214
143	0.25
144	0.25
145	0.25
146	0.286
147	0.25
148	0.214
149	0.214
150	0.179
151	0.214
152	0.214
153	0.214
154	0.214
155	0.214
156	0.214
157	0.214
158	0.179
159	0.179
160	0.179

*probabilities and averages were calculated using picks at or within 3 of each given draft pick for the last 4 drafts (2013-2016). For example, to calculate the probability of selecting a player that would make your scoring roster with the 7^th overall pick, I looked at picks 4 through 10 for each season. Likewise, in calculating the average, I considered only those picks at or within three draft picks that made a scoring roster.

I think the graph tells a compelling and somewhat accurate story on the one hand, but doesn’t help all that much in forecasting points that GMs will get out of the draft, on the other. It captures that very real steep drop off of talent available at the draft over the course of the first round. Things seem to level off as GMs mine for solid scoring roster depth over the next three rounds or so, and then there’s a steady decline that approaches complacency in the middle to late rounds, before some GMs feel inspired to hit those late round jacks; or, you know, that tail could be all a function of luck.

To the “doesn’t help all that much in forecasting points” remark, clearly there are going to be players drafted between the 10^th and 70^th pick that score more points than the 48 or so predicted by the graph. And clearly, you would want the 10^th pick over the 70^th pick, right?

I have some possible explanations. I think the graph is revealing that draft “strength” may not actually count for as much as we’d like to think. At least that’s how the data bears out. I believe it’s a combination of this all being a clusterfuck of chance as well as some GMs just preparing a lot better than others, and showing up on draft day with a clear plan and sticking to it a lot better than others (me). Further, I’d think that GM performance on draft day bears no correlation with draft strength going in, or if anything may even be negatively correlated; I can see a GM with poor draft position thinking that he needs to prepare better for some late round steals.

Where does that leave us with getting back to this pre-season forecasting? I have one other observation. Back to the point about only needing to fill out scoring rosters, and the implication that only #4picksmatter, guess how many KL teams had all of their keepers on their scoring roster at season end in 2016-17 (or would have had them but-for trades)? Just two: the Joshfrey Krupuls and the Dicklas Lidstroms. Allow me to go off on this tangent further, if you haven’t already concluded from the mention of those teams that having all 10 of your keepers make your KL is no great achievement.

24 keepers didn’t make scoring rosters last year, either due to injury or shitty performance. I included Loui Eriksson in this, because even though he would have made the Parkers’ SR, all that took was 17 points!! (it’s like Rome actually tried to build his team in a day). It’s further interesting to note that the list of failed keepers included 5 goalies but only 1 defenseman, Hampus Lindholm (who would have made Rome’s SR as a forward). I point all this out for GM consideration as we approach the keeper roster submission deadline tomorrow.

Back to the task at hand. 24 failed keepers means an average of 1.5 not making a team’s scoring roster, which means that #5.5picksmatter. In recognition of that, and completely ignoring what the data says from past drafts (!), I’ve used Micah’s model, but just factored in the top SIX draft picks for each team in forecasting their points from the draft.

What I also sought to capture was the positional mix of keepers and the differing needs of GMs to fill out their positional scoring rosters at the draft. I have the (Edenton?) Toilers as the only team not keeping a goalie, so they have the highest draft points “base” of 250 on which to build. I have my team and the ‘sics keeping only one dman each, and so we have the lowest base of 200 on which to build. After whatever mix you needed to fill out your SR, I just added 70 pts to your base, representing the replacement value of an extra d and an extra forward (and again, completely ignoring that it is much rarer for a kept defenseman to miss an SR than it is for a kept forward). After adding pts above replacement to the base, I’ve multiplied everything by 5.5/6 to predict the contribution that the draft will make to each team’s scoring roster.

Team	Pts above Rep	Base	SR Draft Pts
Toilers	78	250	301
Powder Rangers	91	220	285
Schizzarks	78	230	282
Dicklas Lidstroms	87	220	281
Joshfrey Krupuls	69	230	274
Fylanders	79	210	265
Valeri Nickrooshkins	75	210	261
Milan Micahleks	61	220	258
Teeyotes	70	210	257
Quebec Rordiques	69	210	256
Phillipsadelphia Flyers	67	210	254
Mackhawks	67	210	254
WBS Parkers	64	210	251
Winter Claassics	70	200	248
Patrik Stefans	68	200	246
Los Amjawors Kings	49	210	237

Lastly, putting it all together, I’ve multiplied each team’s projected keeper points by 0.85 (1.5 out of 10 keepers don’t make a scoring roster) and added that to the draft points above. Below are my predicted standings at the end of the 2017-18 KL season!

Team	Pts
Milan Micahleks	828
Patrik Stefans	818
Valeri Nickrooshkins	795
Mackhawks	789
Los Amjawors Kings	787
Quebec Rordiques	780
Phillipsadelphia Flyers	773
Powder Rangers	770
Teeyotes	762
Fylanders	755
Toilers	754
Winter Claassics	750
Joshfrey Krupuls	750
Schizzarks	731
WBS Parkers	714
Dicklas Lidstroms	690

For a whole host of reasons, the standings won’t look like this. Some teams are going for it, while some aren’t. Some GMs will intentionally draft players early, who they know won’t play a single game this season. Prospects are not factored in whatsoever. And trades and free agent signings will shake things up. While I think the above standings are reflective of where we sit going into our 8^th campaign, nothing is predetermined; that’s why we play the games! Wait…