1 00:00:03,839 --> 00:00:12,800 Welcome to Fantasy Hockey Life, presented by fan Tracks. Here shus your source 2 00:00:12,839 --> 00:00:16,879 of information and analysis to help you win your fantasy hockey league. Block off 3 00:00:16,960 --> 00:00:22,320 hats, a step hit on, stay lock. Here's your host, Jesse 4 00:00:22,399 --> 00:00:27,120 Sovier and Victor Nunyo. This see Hockey Live back once again. Jesse Severe 5 00:00:27,760 --> 00:00:32,039 coming to you from the vantrack side and over there EP rings side zone. 6 00:00:32,320 --> 00:00:35,600 Victor Nuno. How are you doing today, Victor? I'm doing awesome. 7 00:00:35,679 --> 00:00:39,840 Jesse. I am finally not sick. I'm still dealing with a little remnants 8 00:00:39,840 --> 00:00:42,600 of it, but overall feeling much better. I had my flu game. 9 00:00:42,719 --> 00:00:46,240 Hopefully it went okay for people. Didn't sound too amazingly. But how you 10 00:00:46,320 --> 00:00:51,799 doing, Buddy good Man? Good I am. It's December and that is 11 00:00:52,039 --> 00:00:56,159 shocking to consider, but we are moving through the year. The hockey season 12 00:00:56,280 --> 00:01:00,039 roll so on. You think you think of December as being relatively early in 13 00:00:59,960 --> 00:01:02,479 the hockey season, but it's really not. We're quite a ways in at 14 00:01:02,479 --> 00:01:04,799 this point, aren't we. Yeah, we're quite a ways in. And 15 00:01:04,920 --> 00:01:08,280 at this point you're probably figuring out if your team is legit or not, 16 00:01:08,480 --> 00:01:15,040 whether you are gonna be pushing your tips in or maybe retooling, so those 17 00:01:15,079 --> 00:01:19,200 decisions probably either now or the next few matchups, are going to become a 18 00:01:19,319 --> 00:01:23,920 parent, so I have to make those difficult decisions. But we are definitely 19 00:01:23,480 --> 00:01:27,599 getting pretty close to Fantasy playoff time February ish, which is not that far 20 00:01:27,640 --> 00:01:33,760 away. Yeah, absolutely, and that's that is that's coming up. Yeah, 21 00:01:33,760 --> 00:01:37,079 we like to do our playoffs relatively early, so it is right around 22 00:01:37,120 --> 00:01:40,200 the corner. You need things to be right around the corner. People don't 23 00:01:40,239 --> 00:01:42,200 have attention spans. I don't have an attention span, man, Ever since 24 00:01:42,239 --> 00:01:46,519 I got a phone, ever since you didn't have to wait in a line. 25 00:01:46,959 --> 00:01:49,560 It's difficult for me to go through a six month season before playoffs or 26 00:01:49,640 --> 00:01:53,599 five month season. Come on, man, let's move it on. Is 27 00:01:53,680 --> 00:01:57,319 that, Victor? Have our brains been destroyed? Are we really going to 28 00:01:57,359 --> 00:02:00,640 be able to keep playing something like fantasy sports on the long term if these 29 00:02:00,640 --> 00:02:06,879 things happen. I sure hope so, because I enjoy it. But it 30 00:02:06,920 --> 00:02:09,599 does make some other things difficult with some of the attungon span and some of 31 00:02:09,599 --> 00:02:14,960 the other issues. But hopefully it's just more helpful than harmful. By the 32 00:02:15,000 --> 00:02:17,360 way I throw these questions at Or, Victor has no idea what I'm about 33 00:02:17,360 --> 00:02:21,680 to say, and even if it makes no sense, he reacts professionally and 34 00:02:21,960 --> 00:02:25,439 with great insight all the time. You should appreciate that. About these intros, 35 00:02:25,560 --> 00:02:30,120 We've got a cool topic to talk about today. We're going to get 36 00:02:30,159 --> 00:02:34,400 into some data analysis, and unlike some of our conversations, they don't rely 37 00:02:34,479 --> 00:02:38,560 on pictures. So we're gonna come back right after this. We're going to 38 00:02:38,599 --> 00:02:50,240 blow your minds with a little bit of talk about NHL. Welcome you to 39 00:02:50,280 --> 00:02:53,599 the show. A little bit of a change of pace. Chase thecullum of 40 00:02:53,759 --> 00:02:57,800 EP ringside. He is a data genius. I don't use the word genius 41 00:02:57,800 --> 00:02:59,960 casually. Maybe he wouldn't use it on himself, but I'm going to go 42 00:03:00,080 --> 00:03:04,560 and call him a data genius who's going to have some great insights into maybe 43 00:03:04,639 --> 00:03:08,719 some ways we need to change how we are thinking about NHL equivalenc Chase, 44 00:03:08,719 --> 00:03:13,800 how you doing today? I'm doing absolutely fantastic. Great to be on here. 45 00:03:14,039 --> 00:03:19,400 And how are you guys doing doing great? Doing great? I read 46 00:03:19,560 --> 00:03:24,599 your substack. You have a substack. The substack has the explanation of something 47 00:03:24,639 --> 00:03:30,919 called NHL Z that we want to talk about today and get into and I 48 00:03:30,960 --> 00:03:38,039 guess this is an innovation of going beyond the NHL equivalency that Victor and I 49 00:03:38,319 --> 00:03:44,960 frequently talk about here. Can you explain some of the problems that you're trying 50 00:03:44,960 --> 00:03:49,960 to solve with your new model? Yeah? Absolutely? Should I explain NHL 51 00:03:50,000 --> 00:03:53,199 E first, or do you think everybody will have be fairly well first? 52 00:03:53,560 --> 00:03:55,840 Maybe? NHL E is just a very simple thing where it says, if 53 00:03:55,840 --> 00:04:00,120 you can score this many points in the NHL or if you score the as 54 00:04:00,159 --> 00:04:03,159 many points in the WHL, take it time as a factor. That's how 55 00:04:03,159 --> 00:04:08,039 many points that would be worth in the NHL. Right, it's easypsy you 56 00:04:08,080 --> 00:04:11,039 get a little factor. You guys have seen that chart before out there, 57 00:04:11,560 --> 00:04:15,480 and so it's just a quick shorthand of if you're this good in this league, 58 00:04:15,639 --> 00:04:19,240 how much do we have to water that down to understand how you'd be 59 00:04:19,279 --> 00:04:24,199 in another league? Right? Yeah, yeah, exactly. NHL E I 60 00:04:24,199 --> 00:04:27,279 think it's a genius concept. I've used it a ton in all of the 61 00:04:27,279 --> 00:04:31,800 prospect the research I've done. But there's a couple different problems as it relates 62 00:04:31,800 --> 00:04:36,000 to NHL E, depending on what you want to do with it. So 63 00:04:36,199 --> 00:04:40,680 NHL E is a very good kind of base, right if you, like 64 00:04:40,720 --> 00:04:42,959 you said, if you want to know, hey, this guy played in 65 00:04:42,959 --> 00:04:45,920 the WHL, he scores one hundred points in the WHL, what do we 66 00:04:45,959 --> 00:04:47,600 think he's going to score on the NHL next year. It's great for that. 67 00:04:47,759 --> 00:04:51,439 It's as good as you're going to do. But something I wanted to 68 00:04:51,439 --> 00:05:00,360 look into was league difficulty, which is subtly different than NHL E. E 69 00:05:00,600 --> 00:05:05,040 measures how well points translate from league to league. So your hundred point WHL 70 00:05:05,079 --> 00:05:14,399 player, but that is slightly difficult that they're different than actual league difficulty because 71 00:05:14,399 --> 00:05:16,519 of a couple different things that I wanted to go through in the article. 72 00:05:16,639 --> 00:05:20,240 And the first thing I noted was everything here is ERA adjusted, by the 73 00:05:20,240 --> 00:05:24,120 way, So just to get that out of the way first. That'll come 74 00:05:24,120 --> 00:05:30,680 into play later. But NHL E does not quite measure league difficulty because the 75 00:05:30,839 --> 00:05:36,639 underlying distribution of scoring in the leagues will affect its NHL E translations. So 76 00:05:36,720 --> 00:05:43,040 something like Patrick Bacon has really good NHL estimates and the Czech League ends up 77 00:05:43,120 --> 00:05:46,680 as the third best league in the world and his estimates or second if you 78 00:05:46,720 --> 00:05:49,720 don't count the NHL, and some people are surprised by that. The Chech 79 00:05:49,759 --> 00:05:54,399 League is a good league. But one of the reasons why something like the 80 00:05:54,399 --> 00:05:58,519 Czech League can look so good in NHL E compared to what people think is 81 00:05:58,600 --> 00:06:01,600 people just don't score that much check league. And then in the inverse, 82 00:06:01,800 --> 00:06:06,920 the AHL tends not to look very good because an individual HL point doesn't translate 83 00:06:06,959 --> 00:06:12,639 particularly well, but scoring is really common in the AHL, so the points 84 00:06:12,639 --> 00:06:15,439 don't translate as well because points are more abundant. So that was the first 85 00:06:15,439 --> 00:06:19,800 thing I wanted to adjust for. And then the second thing is positional effects. 86 00:06:19,879 --> 00:06:24,519 So a lot of NHL E models just treat everyone as the exact same. 87 00:06:24,959 --> 00:06:30,120 There are presumably certain leagues where like a defender scoring a point per game 88 00:06:30,199 --> 00:06:35,000 is going to probably translate differently than a forward. So those were a couple 89 00:06:35,040 --> 00:06:39,800 of kids that's not been an NHL E doesn't account for the difference between a 90 00:06:40,480 --> 00:06:46,360 defenseman and a forward in the scoring. It's just the multiplier factor won't necessarily 91 00:06:46,399 --> 00:06:48,920 be equal for forwards as it would be for defenders. Is what you're saying 92 00:06:48,920 --> 00:06:54,319 there right exactly. And then the final thing is NHL E. There's a 93 00:06:54,319 --> 00:06:57,079 lot of math that goes it. It's a genius concept because of how simple 94 00:06:57,079 --> 00:07:00,600 it is, but there's a lot of back end math basically an optimization problem. 95 00:07:00,800 --> 00:07:08,759 But it takes a large sample for that math to work in its best 96 00:07:09,240 --> 00:07:13,319 form. And that's great obviously, and data generally, the larger sample size 97 00:07:13,319 --> 00:07:15,240 you get, the better. But something I also wanted to do was to 98 00:07:15,279 --> 00:07:20,199 be able to look at different time trends. So the example that what all 99 00:07:20,240 --> 00:07:24,480 inspired this was I got a comment when I was writing about Matt ve Mitchikov's 100 00:07:24,560 --> 00:07:29,040 NHLE and people are saying, yeah, because of obviously the geopolitical situation in 101 00:07:29,120 --> 00:07:31,759 Russia, a lot of guys aren't going there. It's like who otherwise would 102 00:07:31,800 --> 00:07:35,839 Maybe some former NHL players and whatnot that are North American based would usually go 103 00:07:35,879 --> 00:07:40,519 to the KHL that aren't so, Like the KHL is easier to score in 104 00:07:40,560 --> 00:07:45,319 now than it used to be. So these time trends are existing, it's 105 00:07:45,319 --> 00:07:48,160 going to really mess within a NHL E based model. So there's certain if 106 00:07:48,160 --> 00:07:53,399 certain leagues are better or worse right now than they currently are, your NHL 107 00:07:53,439 --> 00:07:57,800 E estimates are going to either over underrate the player's production based on that. 108 00:07:58,040 --> 00:08:03,800 And the really difficult thing too, is there's both It's unlikely that leagues are 109 00:08:03,839 --> 00:08:09,040 changing a ton in the short run, but it's also really unlikely that leagues 110 00:08:09,040 --> 00:08:15,600 are completely static over time, and there's pretty much no way to intuit which 111 00:08:15,600 --> 00:08:18,480 of those effects is winning out at any given time. So I wanted a 112 00:08:18,480 --> 00:08:22,680 way to be very explicit about that, plus have error bars. So a 113 00:08:22,839 --> 00:08:26,959 NHL E that estimate will tell you a WHL point is worth one point or 114 00:08:28,199 --> 00:08:31,199 zero point one five NHL points, but there's no kind of error bars on 115 00:08:31,319 --> 00:08:35,120 that, and anything statistical will have, whether you see it or not, 116 00:08:35,320 --> 00:08:37,960 some sort of uncertainty measures. And with my measure, I wanted to be 117 00:08:39,080 --> 00:08:43,480 very explicit about, Hey, these leagues are x close together, so the 118 00:08:43,519 --> 00:08:46,960 difference is negligible. So I wanted to take a couple of the or a 119 00:08:46,000 --> 00:08:50,159 couple of those problems with NHL E and try to solve for them with my 120 00:08:50,440 --> 00:08:56,159 NHL Z. Yeah, so that there's a couple interesting things there. So 121 00:08:56,240 --> 00:09:00,559 yeah, it's this is all intuitive. Leagues change over time. You can't 122 00:09:00,600 --> 00:09:05,440 just take what happened in this year's league, this year's AHL and apply it 123 00:09:05,440 --> 00:09:09,120 to next year's NHL. First of all, we don't know what it next 124 00:09:09,159 --> 00:09:11,759 year's NHL is going to be but more to the point, Yeah, AHL 125 00:09:13,519 --> 00:09:16,120 five years ago, when we're ten years ago, when some of our current 126 00:09:16,200 --> 00:09:22,840 NHLers were in there, the AHL had a different score in environment probably than 127 00:09:22,879 --> 00:09:26,440 it does now. So if you're working with how they translated, you're going 128 00:09:26,519 --> 00:09:30,600 to have some problems. And so that makes sense to me, this business 129 00:09:30,639 --> 00:09:35,440 of ranking league quality. I understand what you're saying is you can't just say 130 00:09:35,039 --> 00:09:39,919 the WHL is worth point twenty seven. It's point two seven, it's twenty 131 00:09:39,919 --> 00:09:45,360 seven percent as good as the NHL. It's that doesn't exactly work. But 132 00:09:45,399 --> 00:09:50,960 I don't entirely understand what's happening here is the translation of points. The only 133 00:09:52,000 --> 00:09:54,759 thing we're going for here, because when you're talking about league quality, I'm 134 00:09:54,759 --> 00:09:58,000 trying to understand what the difference is between seeing this league is better than that 135 00:09:58,120 --> 00:10:05,440 league versus this is the league where you can score more points being the equal 136 00:10:05,480 --> 00:10:11,679 talent. Yeah. Absolutely, So it is still inferring league quality from scoring 137 00:10:11,759 --> 00:10:16,600 rates. It's just taking a couple extra measures to actually to account for things 138 00:10:16,639 --> 00:10:24,080 like those positional differences and which leagues just have like higher lower relative scoring environments, 139 00:10:24,240 --> 00:10:26,960 and they can get pretty extreme. Actually, I think an average AHL 140 00:10:28,159 --> 00:10:31,919 forward scores like fifty percent more than an average KHL forward. So when you're 141 00:10:31,919 --> 00:10:37,519 looking how much the scoring tends to translate, scoring translates way better than the 142 00:10:37,600 --> 00:10:41,200 KHL because points are just way more scarce. So a guy who scores twenty 143 00:10:41,240 --> 00:10:43,159 points, those twenty points are just harder to come by. So if you 144 00:10:43,200 --> 00:10:46,639 want to measure how good the league is, you want to account for how 145 00:10:46,639 --> 00:10:52,039 good the player is within the context of his league. If that makes sense. 146 00:10:52,320 --> 00:10:56,360 Yeah, But so another thing with that, like with a KHL and 147 00:10:56,399 --> 00:11:01,080 I We're going to talk more specifically about the KHL later in the episode, 148 00:11:01,120 --> 00:11:05,399 But any data model you have, you and I I've done enough data stuff 149 00:11:05,399 --> 00:11:09,639 to know the biggest problem is what friggin data you can get and how clean 150 00:11:09,720 --> 00:11:13,799 it is. And one of the big things with some of these overseas leagues 151 00:11:13,919 --> 00:11:18,600 is the number of minutes these guys play and therefore the usage that they get. 152 00:11:18,000 --> 00:11:24,039 How do you translate what somebody who's in the finish or the KHL league 153 00:11:24,240 --> 00:11:28,080 gets as two minutes he gets on the ice as a fourth liner and apply 154 00:11:28,200 --> 00:11:31,320 that to the type of scoring he would get if he was put in a 155 00:11:31,360 --> 00:11:35,759 scoring line and in a higher level quote league. And then can you even 156 00:11:35,799 --> 00:11:39,559 get the time on ice? I can ever find the daying time on ice 157 00:11:39,600 --> 00:11:41,519 for some of these leagues, So I don't know how you're doing that. 158 00:11:41,559 --> 00:11:43,919 But you can say somebody gets five points in the KHL, they might have 159 00:11:43,960 --> 00:11:48,279 played a cumulative thirty minutes to get those five points as opposed to what they 160 00:11:48,320 --> 00:11:52,639 get in a higher level leagues. How do you account for that type of 161 00:11:52,679 --> 00:11:58,240 difference. Yeah, so you're absolutely right. You're entirely at the mercy of 162 00:11:58,279 --> 00:12:01,879 the data you collect, which sucks, but it is what it is, 163 00:12:01,240 --> 00:12:05,600 and I don't have time on ice. I use explicitly points per game, 164 00:12:05,559 --> 00:12:11,200 but there's a number of different things that I think work to the advantage of 165 00:12:11,200 --> 00:12:18,840 that, the first being it's not necessarily just a prospect base, so I 166 00:12:18,000 --> 00:12:24,000 use I apply the model to prospects, but everyone who plays KHL games is 167 00:12:24,000 --> 00:12:26,960 eligible for the model. Everyone who plays LEGA games is eligible for the model. 168 00:12:28,480 --> 00:12:31,519 The small ice time will happen to some of them. But you can 169 00:12:31,559 --> 00:12:35,159 also have the inverse problem, and we're tracking when players go up and down. 170 00:12:35,679 --> 00:12:39,480 So if a guy goes from the LIGA to the NHL, that's in 171 00:12:39,519 --> 00:12:41,799 there. If a guy goes from the Liga and he gets kicked down to 172 00:12:41,559 --> 00:12:46,039 the Maytee I think is the second tier league in Finland that will also be 173 00:12:46,080 --> 00:12:50,200 in there. So the idea is, since there's hundreds of thousands of these 174 00:12:50,240 --> 00:12:54,600 transfers, that the kind of ice time airs will average out and we can 175 00:12:54,679 --> 00:13:00,960 get a generally good idea of the league quality even without ice time. And 176 00:13:01,000 --> 00:13:05,960 then the second thing is, I'm not one hundred percent sure accounting for ice 177 00:13:05,000 --> 00:13:09,360 time would be the best idea even if you want, even if you had 178 00:13:09,399 --> 00:13:15,279 it, because points per game is pretty highly correlated with like war measures or 179 00:13:15,320 --> 00:13:20,639 any sort of total hockey rating value at the NHL level. It's presumably true 180 00:13:20,639 --> 00:13:26,039 at every other level too, but it does obviously miss things, but time 181 00:13:26,080 --> 00:13:30,240 on ice will help capture a lot of what it's missing. So the guy 182 00:13:30,279 --> 00:13:33,159 just sucks defensively, coach isn't gonna play him as much. Stuff like that. 183 00:13:33,240 --> 00:13:37,559 So if you the time on ice ends up, in my opinion, 184 00:13:37,720 --> 00:13:43,720 not being as good of a defense as as possible, because there is signal 185 00:13:43,919 --> 00:13:46,720 in that time on ice too. So the guys, oh, he went 186 00:13:46,759 --> 00:13:50,960 to the KHL and he's not scoring as much because he's not playing as much. 187 00:13:52,039 --> 00:13:54,639 Part of the reason he's not playing as much is because the league's way 188 00:13:54,639 --> 00:13:58,480 better. When you're trying to measure league quality, I'm I don't know because 189 00:13:58,480 --> 00:14:03,440 I don't have time ice, and this could just be me giving a very 190 00:14:03,480 --> 00:14:07,519 self serving explanation of flyay a flaw in my own model is not a flaw, 191 00:14:07,919 --> 00:14:13,759 but I'm not entirely sold accounting for time on ice would help that much 192 00:14:13,960 --> 00:14:18,200 because they're signal in the changes of time on ice as well. The ability 193 00:14:18,240 --> 00:14:22,639 to earn minutes is probably the most important thing, to be honest. So 194 00:14:24,200 --> 00:14:28,240 this is so great, Jason. Obviously, as you're pointing out, there 195 00:14:28,279 --> 00:14:33,320 are some issues with inferring lead quality, and I don't know that there is 196 00:14:33,679 --> 00:14:39,039 a better measure. I think that it's helpful. So I'm wondering what kind 197 00:14:39,120 --> 00:14:43,639 of we always have biases or blind spots. So, assuming we're gonna use 198 00:14:43,720 --> 00:14:50,720 something like NHL E or Z for a league quality inference, even though that 199 00:14:50,759 --> 00:14:54,279 may not be perfect, what kind of issues do you think what kind of 200 00:14:54,360 --> 00:14:58,200 over underestimations might we be falling into? Because I think it's useful to some 201 00:14:58,279 --> 00:15:01,919 extent to look at prospect When you're talking about Mitchikoff as your previous example, 202 00:15:03,919 --> 00:15:07,000 I think you can clearly say that the playing in the KHL as a draft 203 00:15:07,039 --> 00:15:11,519 eligible is certainly more difficult in playing and say the BHL or even the USHL 204 00:15:11,639 --> 00:15:15,720 or probably the CHL. But at some point the line becomes a little bit 205 00:15:15,759 --> 00:15:20,639 more blurred. But it is helpful to have some idea of that, even 206 00:15:20,679 --> 00:15:24,759 if it's not perfect. But what kind of over under estimations or issues might 207 00:15:24,759 --> 00:15:28,639 be running into if we play a little too fast and loose with that as 208 00:15:28,679 --> 00:15:35,320 a comparable. Yeah, that's a very good question. And I think one 209 00:15:35,360 --> 00:15:41,440 of the problems you can run into is these are all averages, and like 210 00:15:41,480 --> 00:15:46,759 I've tested, the averages are predictive even at the individual level. But the 211 00:15:46,960 --> 00:15:50,200 thing I say all the time is don't fire scouts, because if you're trying 212 00:15:50,240 --> 00:15:54,200 to apply macro level predictions to micro level individuals, like there is a lot 213 00:15:54,240 --> 00:15:58,799 of context there that is missing, and we do have to be aware of 214 00:15:58,840 --> 00:16:00,679 that. Even though, like I said, average, I'm not sold that 215 00:16:00,720 --> 00:16:06,840 ice time would be a huge help, it may matter in specific circumstances. 216 00:16:07,679 --> 00:16:10,840 You're basically assuming, by using clients per game, the coaches know what they're 217 00:16:10,879 --> 00:16:14,360 doing. I think that's a good assumption. On average, but that doesn't 218 00:16:14,399 --> 00:16:18,519 mean, for example, every coach knows what they're doing, so there could 219 00:16:18,600 --> 00:16:23,960 be a lot of basically niche circumstances at the individual level that can make some 220 00:16:25,120 --> 00:16:30,679 of these inferences difficult. When guys do transfer over, would be the primary 221 00:16:30,720 --> 00:16:36,759 thing I would look out for, if that makes sense. Are we factoring 222 00:16:36,799 --> 00:16:40,759 agent of this because you can say that the league is better or worse, 223 00:16:40,840 --> 00:16:45,000 But obviously you drop Patrick Kane into the HL, his translation is going to 224 00:16:45,000 --> 00:16:48,720 be a little bit different than some guy who's coming up and maybe playing the 225 00:16:48,759 --> 00:16:53,480 same minutes. You can't account for the fact that he's only nineteen coming up 226 00:16:53,480 --> 00:16:56,279 there or something like that. So do you factor that into the model? 227 00:16:56,759 --> 00:17:00,080 Yeah, I forgot to mention that at the beginning. That was another thing 228 00:17:00,120 --> 00:17:03,720 that NHL models won't be accounting for that I do have in the model. 229 00:17:03,759 --> 00:17:07,559 So I have an aging curve built into the model that tracks that allows for 230 00:17:07,000 --> 00:17:11,359 younger players to be expected to get better, because that's another problem when you're 231 00:17:11,359 --> 00:17:15,519 tracking scoring of OHL players who go to the NHL. But if you're twenty 232 00:17:15,599 --> 00:17:18,920 turning twenty one and then you transfer, you will tend to get better just 233 00:17:18,960 --> 00:17:25,079 because you're twenty turning twenty one, like the natural age progression of certain leagues 234 00:17:25,240 --> 00:17:29,160 will help boy their NHL E translations, and a lot of those junior leagues, 235 00:17:29,160 --> 00:17:30,960 like I said, because they're expected to get better when they leave. 236 00:17:32,000 --> 00:17:34,559 Anyways, part of the reason may be something to do with the league, 237 00:17:34,599 --> 00:17:38,119 but part of the reason will be the age. And I have age explicitly 238 00:17:38,119 --> 00:17:41,880 controlled for, which is something that won't be done in anything NHL E based 239 00:17:41,920 --> 00:17:47,319 as far as I'm aware. So I know we're going to get I keep 240 00:17:47,359 --> 00:17:49,599 teasing it, but we're going to get into some specific leagues a little bit 241 00:17:49,640 --> 00:17:56,920 later on whether they are improving, whether they're declining. But did do you 242 00:17:56,960 --> 00:18:04,960 have any major finings otherwise leagues that we thought were better but under your new 243 00:18:06,000 --> 00:18:11,920 model that appears to have been different leagues maybe that moved up in estimation or 244 00:18:11,920 --> 00:18:15,160 moved down or in estimation dramatically based on your study. Yeah, the big 245 00:18:15,200 --> 00:18:21,920 one in terms of moving up is the AHL. I have the AHL estimated 246 00:18:21,960 --> 00:18:26,400 is the second best league in the world now over the entire sample going back 247 00:18:26,440 --> 00:18:30,480 to two thousand. I don't believe the AHL has necessarily been the second best 248 00:18:30,519 --> 00:18:33,319 league in the world, but on the year to year level of the AHL, 249 00:18:33,359 --> 00:18:37,279 I have estimated as the second best league in the world and have for 250 00:18:37,400 --> 00:18:41,920 half a decade now, which I think is a pretty big flipping of people's 251 00:18:41,960 --> 00:18:48,559 perceptions. But as a lot of teams are starting to play their best prospects 252 00:18:48,559 --> 00:18:51,279 in the AHL, Buffalo is doing it a lot. You can see it 253 00:18:51,640 --> 00:18:53,799 in some of the organizations as they move their best players to the AHL, 254 00:18:53,960 --> 00:18:59,920 as guys who are middling NHL players fighting for a roster spot aren't as well 255 00:19:00,240 --> 00:19:04,839 to go overseas, specifically to the KHL for obvious reasons of not wanting to 256 00:19:04,839 --> 00:19:07,200 go to Russia. Right now, I think it makes a lot of sense 257 00:19:07,200 --> 00:19:11,799 that the AHL has been on the rise, and then the primary league from 258 00:19:11,839 --> 00:19:18,079 a professional level that's been falling off has been the KHL. And then at 259 00:19:18,079 --> 00:19:22,640 a junior level, there's a general rise in the American hockey pipeline, so 260 00:19:22,680 --> 00:19:26,400 the NCAA, the USHL, the NDTPI is part of the USHL, but 261 00:19:26,680 --> 00:19:33,519 there's been a general rise there relative to those CHL leagues. Yeah, for 262 00:19:33,599 --> 00:19:38,000 sure. Yeah, And I'm just trying to understand the other thing. You 263 00:19:38,039 --> 00:19:41,640 have to tell me whether this is already baked into the bread or whether this 264 00:19:41,759 --> 00:19:48,039 is another piece of it is different leagues play by slightly different formats, slightly 265 00:19:48,079 --> 00:19:52,599 different rules. European rinks are larger, penalties may be called different leagues or 266 00:19:52,599 --> 00:20:00,000 maybe slightly different rules. Is that and so the scoring environment doesn't necessarily translate 267 00:20:00,160 --> 00:20:03,720 to the talent I'm trying to I'm still trying to get through my thick skull 268 00:20:04,599 --> 00:20:10,799 whether we are measuring how we can measure that the players are good relative to 269 00:20:10,880 --> 00:20:15,640 a league if the league rules are slightly different. Is that what NHL E 270 00:20:15,880 --> 00:20:19,400 does basically, or is that an additional factor that you have to account for? 271 00:20:21,400 --> 00:20:25,359 So I guess we can go two ways to that it will be mostly 272 00:20:25,400 --> 00:20:30,559 baked into the bread. Because say, whatever rule in the SHL is really 273 00:20:30,599 --> 00:20:33,440 propping up these players scoring rights, then if they leave that environment, which 274 00:20:33,480 --> 00:20:37,119 is what we're tracking anyways, and that rule changes, that rule's gone, 275 00:20:37,240 --> 00:20:41,119 they're going to perform a lot worse, and the model's going to pick up 276 00:20:41,119 --> 00:20:42,720 on the fact that, Hey, these guys who leave the SHL are always 277 00:20:42,759 --> 00:20:45,920 not doing as well. The model won't tell you that it's the rule, 278 00:20:47,079 --> 00:20:49,720 but if there is some rule propping up scoring or like the rink dimensions or 279 00:20:49,759 --> 00:20:53,240 whatever, the model will be able to pick up on things like that. 280 00:20:53,599 --> 00:21:00,720 But my answer to what could go wrong with these models bring up something that 281 00:21:00,759 --> 00:21:03,480 I think is a better answer and something you should always keep in mind. 282 00:21:03,720 --> 00:21:07,720 And this is true in my model and all sorts of models. Pretty much 283 00:21:07,720 --> 00:21:12,519 everything will be regression based if you're aiming for interpretability, which will tell you 284 00:21:12,640 --> 00:21:18,920 the average effects of a change in some variable. This is great in almost 285 00:21:18,960 --> 00:21:22,799 all circumstances, even like medical journals and stuff are going to be using average 286 00:21:22,799 --> 00:21:29,599 effects. But there are key interactions that may or may not exist as well. 287 00:21:29,920 --> 00:21:33,640 So, for example, something I'm a little worried about that I don't 288 00:21:33,720 --> 00:21:37,279 have an answer for is if you look at like Byron Bader's model, for 289 00:21:37,359 --> 00:21:44,839 example, he measures NHL and he doesn't say this publicly, but he doesn't 290 00:21:44,839 --> 00:21:49,519 explain how. But I've basically figured this out that his NHL estimates very closely 291 00:21:49,599 --> 00:21:53,680 tracked what happens when you look when players jump from each league to the NHL. 292 00:21:55,240 --> 00:21:59,680 So he has way higher estimates of the OHL than everyone else publicly. 293 00:22:00,160 --> 00:22:03,160 Hey, and if you look at what happens when players go straight from the 294 00:22:03,200 --> 00:22:07,359 OHL to the NHL, you happen to get an estimate that is basically the 295 00:22:07,400 --> 00:22:11,599 exact same as OHL estimate, which tells me I'm fairly certain that's what's doing. 296 00:22:12,079 --> 00:22:18,119 Whereas the other public models like Patrick Bacon's, Max's, Chateel's, all 297 00:22:18,160 --> 00:22:22,119 of these will measure the web of interactions, so not just the OHL to 298 00:22:22,160 --> 00:22:26,880 the NHL, but the OHL to the AHL, all of that and bake 299 00:22:26,960 --> 00:22:30,680 that into the math. And what happens is if you look at estimates of 300 00:22:30,720 --> 00:22:34,039 a lot of these, say the Russian leagues in particular as an example, 301 00:22:34,079 --> 00:22:40,559 because they we can keep going on this theme is Bader's estimates are a lot 302 00:22:40,640 --> 00:22:48,359 lower relatively on the Russian league's than the other public models. And I think 303 00:22:48,480 --> 00:22:56,799 there's a potential that the Russian leagues feed into each other better than they feed 304 00:22:56,839 --> 00:23:00,160 to the North American game because of the different ices and stuff like that. 305 00:23:00,240 --> 00:23:03,839 So a model that is measuring the averages will be able to say that, 306 00:23:03,920 --> 00:23:07,680 like, on average, these things are happening, but maybe when you're switching 307 00:23:07,799 --> 00:23:14,119 from different ice surfaces you're scoring couldn't translate five percent worse or something like that 308 00:23:14,640 --> 00:23:18,079 could be at play here, so that maybe you know, the KHL is 309 00:23:18,400 --> 00:23:22,160 whatever best league in the world, but when the KHL is translating to the 310 00:23:22,279 --> 00:23:26,079 NHL, it's actually five percent worse than it is on average overall, just 311 00:23:26,160 --> 00:23:30,000 because of the way the ice surface translates or something like that. So I 312 00:23:30,039 --> 00:23:37,079 think these kind of specific interactions are something that you would need specific scouting on 313 00:23:37,160 --> 00:23:40,960 to know and could mess with any model that based around averages. This is 314 00:23:41,000 --> 00:23:45,359 so interesting, and I just I want to reiterate that because I want to 315 00:23:45,359 --> 00:23:48,279 make sure people understood what you just said, because there's such a huge difference 316 00:23:48,359 --> 00:23:52,960 between guys like I don't know, say, Athan McKinnon, John Tavera, 317 00:23:53,119 --> 00:24:00,240 Sidney Crosby jumping straight from their junior leagues to the OHL, and guys who 318 00:24:00,279 --> 00:24:03,400 play two, three, four years in the OHL and then make the jump. 319 00:24:03,480 --> 00:24:07,920 Those are obviously going to be different jumps. And I don't know that 320 00:24:08,000 --> 00:24:15,640 it's fair to expect a similar jump and equivalency from even high end prospects that 321 00:24:15,839 --> 00:24:19,559 might have played two three years in junior compared to those generational talents. So 322 00:24:19,920 --> 00:24:26,160 it's not fair to assign the same probability, and especially guys with who play 323 00:24:26,160 --> 00:24:30,920 in Russia who won't necessarily make an immediate jump because they have a contract so 324 00:24:30,960 --> 00:24:34,680 they can't go next year to the NHL, so that might create some noise 325 00:24:34,759 --> 00:24:38,880 or some effects that aren't necessarily correct or predictable. I think that's kind of 326 00:24:38,960 --> 00:24:41,680 part of what you're saying too, right, Yeah, yeah, exactly, 327 00:24:41,799 --> 00:24:45,359 because, like I said, there will be specific interactions even if you look 328 00:24:45,440 --> 00:24:52,519 like compare first round picks to second round picks with similar NHL E values when 329 00:24:52,559 --> 00:24:55,920 they do translate, the second round picks tend to perform worse. Why because 330 00:24:55,960 --> 00:24:59,920 the scouts have correctly picked up on all sorts of these more subtle factors that 331 00:25:00,759 --> 00:25:03,960 do matter but aren't measured in the model, right, which would be why 332 00:25:04,000 --> 00:25:11,240 your Crosby's translate better than the Guard's going to translate better than Stankovin or whoever. 333 00:25:11,640 --> 00:25:15,720 Like that kind of thing. Yeah, exactly. And I think there's 334 00:25:15,720 --> 00:25:18,319 another thing you said earlier about the HL that I think is worth repeating or 335 00:25:18,319 --> 00:25:22,920 digging into a little bit, and that is there's a lot of interesting things 336 00:25:22,960 --> 00:25:26,720 when a player like Yuri Koulik is taken, right. I think he's a 337 00:25:26,720 --> 00:25:30,200 perfect example because in the range that he's taken, and you compare him to 338 00:25:30,640 --> 00:25:36,640 other players per se that have been that same we're taking in that same range, 339 00:25:37,160 --> 00:25:40,799 but we're maybe taking out of the CHL and you know that they're going 340 00:25:40,880 --> 00:25:45,200 to have to play, you know, a couple more seasons in the CHL 341 00:25:45,359 --> 00:25:48,119 and aren't eligible to move over. So there's going to be a big difference 342 00:25:48,160 --> 00:25:52,160 there. I think there's a lot that you can say about that, But 343 00:25:52,240 --> 00:25:57,359 also the fact that you have the control of that player right you can, 344 00:25:57,680 --> 00:26:03,400 like in Buffalo's example, you have basically the farm team right there close eye 345 00:26:03,440 --> 00:26:08,279 you can help direct their development, as opposed to being in Russia for one 346 00:26:08,319 --> 00:26:12,000 example, or even in the COCHL, where you don't have necessarily as much 347 00:26:12,039 --> 00:26:15,839 control over what happens with that player. So just looking at that draft, 348 00:26:15,960 --> 00:26:21,759 Cooliquent twenty eighth overall, and they are a bunch of players from the CHL 349 00:26:21,799 --> 00:26:26,160 who went around there. In terms of other forwards and USNVP types Reach Schaefer, 350 00:26:26,240 --> 00:26:30,240 Isaac Howard, Owen Beck, Jaggafercus. So you're not going to be 351 00:26:30,279 --> 00:26:34,559 able to put those players in the AHL and give them a more difficult competition 352 00:26:34,720 --> 00:26:40,039 level and also be able to have more say over their development run the same 353 00:26:40,119 --> 00:26:44,440 system. So all of those are factors that are hard to measure right in 354 00:26:44,519 --> 00:26:47,440 terms of having a large enough data set to be able to compare apples to 355 00:26:47,480 --> 00:26:52,039 apples, but I think it's certainly important that is a big difference. I 356 00:26:52,079 --> 00:26:56,319 think that that may be playing into some of these equivalencies as well. Then 357 00:26:56,359 --> 00:27:00,039 you think all one hundred percent, and like I said, I think I 358 00:27:00,039 --> 00:27:03,839 think back to the AHL being a lot better. Like I don't think it's 359 00:27:03,839 --> 00:27:07,279 an accident that Buffalo got Coolich overseas as quickly as they possibly could, but 360 00:27:07,319 --> 00:27:12,519 that will definitely end up playing a role. The tough thing from a developmental 361 00:27:12,519 --> 00:27:18,440 perspective, why I'm not sold it'll ever be easily measurable is there's so much 362 00:27:18,480 --> 00:27:22,039 selection bias too. You can probably measure it right now and say, hey, 363 00:27:22,079 --> 00:27:26,559 guys that do make that jump to the AHL will perform better than guys 364 00:27:26,559 --> 00:27:29,640 who are left in the CHL in their draft plus two year or whatever, 365 00:27:29,680 --> 00:27:33,720 but part of the reason will also be, like, there is presumably developmental 366 00:27:33,720 --> 00:27:37,519 effects there, absolutely, But if you ever go to measure statistically, the 367 00:27:37,519 --> 00:27:40,759 big problem you're going to run into is that teams are going to bring the 368 00:27:40,839 --> 00:27:44,920 better players over as best they can to play at these higher levels, So 369 00:27:45,440 --> 00:27:51,039 your statistical models aren't going to be necessarily as valid as they should be because 370 00:27:51,039 --> 00:27:53,720 there's just such a selection biases. Teams aren't bringing over random sample of players 371 00:27:53,759 --> 00:28:00,119 like Buffalo is bringing over Coolich but not the next European player selected most likely 372 00:28:00,279 --> 00:28:03,799 yet and the reason why it is just because cool itge is better. Right, 373 00:28:03,880 --> 00:28:07,599 So it definitely does matter, but it's going to be very difficult to 374 00:28:07,599 --> 00:28:14,160 ever find to measure that statistically. Yeah, so if I understand what's going 375 00:28:14,200 --> 00:28:18,839 back and forth here, the guys who are held back in the CHL for 376 00:28:18,839 --> 00:28:22,680 a certain amount of time tend to be if they're the ones who jump straight 377 00:28:22,720 --> 00:28:26,240 to the NHL. A lot of times it's because they're ready more quickly, 378 00:28:27,000 --> 00:28:30,240 and so the AHL guys who need to simmer for more years tend to not 379 00:28:30,319 --> 00:28:34,160 be as big of an impact when they come up. So that's a problem, 380 00:28:34,200 --> 00:28:38,279 and that's really a problem with NHL E if we are just looking at 381 00:28:38,440 --> 00:28:44,039 one step to the next as the way to translate it. Okay, Okay, 382 00:28:44,240 --> 00:28:47,920 I want to be fair to your model, because the greatest thing in 383 00:28:47,960 --> 00:28:51,839 the world is to have somebody on and say, really cool system you've got 384 00:28:51,839 --> 00:28:55,000 here. Now here's all the problems I've got with it, and it's broken, 385 00:28:55,079 --> 00:28:56,920 and why are you been talking to me, and I don't want to 386 00:28:56,960 --> 00:29:02,400 go there because to me, it seems like it's not that a model is 387 00:29:02,960 --> 00:29:10,839 perfect or horrible, it's that your iteration can be more accurate than the NHL 388 00:29:10,880 --> 00:29:14,720 lead. Everything steps up more and more in what you are accounting for is 389 00:29:15,359 --> 00:29:18,960 presumably better and more predictive. But let me get down to a level and 390 00:29:19,000 --> 00:29:23,880 you tell me if it's too granular to account for. So you distinguish between 391 00:29:23,880 --> 00:29:29,680 Ford and d but there's so many other things that impact whether a player's game 392 00:29:29,759 --> 00:29:33,319 is going to translate. You can be a forward going from the OHL, 393 00:29:33,519 --> 00:29:38,000 the CHL to the NHL. Maybe you're in the queue and you can say 394 00:29:38,039 --> 00:29:41,599 this is the scoring change. Some guys who are in the queue might be 395 00:29:41,759 --> 00:29:47,480 thriving on just having open ice for days and being able to do things that 396 00:29:47,519 --> 00:29:49,759 they can't do at a higher level, whereas other guys might be making that 397 00:29:49,839 --> 00:29:56,440 scoring by the way that their game plays. That you don't see the difference 398 00:29:56,440 --> 00:29:59,200 in the scoring at one level. You see it in the next level because 399 00:29:59,240 --> 00:30:02,799 the way you they play, and that's going to end up coming out in 400 00:30:02,799 --> 00:30:06,000 the sample, because some guys are going to perform higher, presumably and at 401 00:30:06,039 --> 00:30:10,720 the low end of the model because of factors, but very difficult factors to 402 00:30:10,720 --> 00:30:15,160 get data on. Is that a problem that there's any way to account for, 403 00:30:15,480 --> 00:30:21,559 or something you think that somehow NHLZ is able to smooth out. So 404 00:30:21,759 --> 00:30:26,240 I think it'll help smooth it out relative to an NHL model that just looks 405 00:30:26,279 --> 00:30:29,519 at guys who jump to the NHL, because if you just look at the 406 00:30:29,559 --> 00:30:32,759 guys to jump to the NHL, only the better of those two players you 407 00:30:32,839 --> 00:30:37,119 named will probably make that jump, so it will measure both of those guys 408 00:30:38,400 --> 00:30:44,160 the theoretical exact word I'm will leave, and if the league is propping up 409 00:30:44,160 --> 00:30:48,480 players because of this open eyes, the ones who do leave and are propped 410 00:30:48,559 --> 00:30:51,279 up by that will struggle, and the model will learn that and start to 411 00:30:51,839 --> 00:30:56,000 pull its estimate of that league quality down because these guys will start to perform 412 00:30:56,279 --> 00:31:00,160 worse than you would otherwise have expected when they do make that jump. But 413 00:31:00,480 --> 00:31:04,359 again back to applying the predictions to an individual level, that is absolutely something 414 00:31:04,440 --> 00:31:08,839 that can be an issue and one of the things I would love to have, 415 00:31:10,599 --> 00:31:14,240 not necessarily time on ice. We talked about why that could be a 416 00:31:14,279 --> 00:31:17,960 flaw, but what I think the easiest way to account for some of these 417 00:31:18,799 --> 00:31:22,839 league merchants would be just getting power play versus even strength time. I think 418 00:31:22,880 --> 00:31:29,000 a significant portion of these guys who have explosions could probably be explained by, 419 00:31:29,000 --> 00:31:30,319 Oh, yeah, he's playing four and a half minutes on the power play 420 00:31:30,359 --> 00:31:34,000 in the year before he was playing like twelve seconds or whatever. I think 421 00:31:34,680 --> 00:31:41,240 that would probably be the most helpful contextual variable at which point you could measure 422 00:31:41,359 --> 00:31:47,960 points relative to the distribution of their ice time rather than just the ice time. 423 00:31:48,000 --> 00:31:51,799 Would be the primary way in which I think time on ice could help, 424 00:31:52,279 --> 00:31:53,920 beyond just like the raw number, but actually knowing, hey, this 425 00:31:55,000 --> 00:31:57,759 guy played four minutes on the power play, this guy played two minutes shorthand 426 00:31:57,799 --> 00:32:04,319 instead, that's why the scoring differences might exist. Would be the primary way 427 00:32:04,359 --> 00:32:07,480 which I would think you would want context there. Things like the power play 428 00:32:07,559 --> 00:32:12,960 usage are huge if you can't get on the power play at your low level, 429 00:32:12,960 --> 00:32:16,839 and especially at a man's league quote unquote, your translation is going to 430 00:32:16,920 --> 00:32:21,680 happen very differently. But like you said, the survivor bias, the people 431 00:32:21,720 --> 00:32:24,400 who actually make it from one level to another will take some of that out. 432 00:32:24,680 --> 00:32:29,119 All right, let's take a quick part of it too. I think 433 00:32:29,119 --> 00:32:31,799 that problem is getting worse actually, which isn't a good thing. As I 434 00:32:31,839 --> 00:32:36,000 sit here with this model specifically for defenseman, which is why I want to 435 00:32:36,039 --> 00:32:40,200 separate them, because as the world basically has followed the NHL into the four 436 00:32:40,279 --> 00:32:45,000 forward one defenseman power play, it gets really hard to earn power play ice 437 00:32:45,039 --> 00:32:49,680 time for defense Only one defenseman is earning meaningful power play ice time, which 438 00:32:49,720 --> 00:32:53,400 is going to exaggerate that that'd be something we have to watch out for totally. 439 00:32:53,839 --> 00:32:57,079 All right, there's so much more to talk about. Let's take a 440 00:32:57,119 --> 00:33:00,160 quick break and come back some of the addition we talk we've got to do 441 00:33:00,200 --> 00:33:24,039 here. We're back with Chase McCullum and one of the cool things here. 442 00:33:24,400 --> 00:33:29,200 If some of you are glazing at this point, and even though it Chase 443 00:33:29,240 --> 00:33:31,079 has done a great job of explaining it, and Victor and I have done 444 00:33:31,079 --> 00:33:36,319 our best job to pull it out, maybe some of you are listening to 445 00:33:36,359 --> 00:33:37,960 like, Okay, that's all fine, but what the heck am I ever 446 00:33:38,000 --> 00:33:43,119 going to do with this? But you have written some articles on ep ringside 447 00:33:43,200 --> 00:33:49,519 that really shows how to apply this very specifically to some contexts and really actionable 448 00:33:49,599 --> 00:33:53,799 stuff. And one of them that caught Victor's eye. In my eye is 449 00:33:53,839 --> 00:33:59,279 the article about Russian hockey in decline, And you've talked about that already a 450 00:33:59,279 --> 00:34:00,759 little bit on the show, But can you expound on that. What were 451 00:34:00,799 --> 00:34:07,239 you able to find about Russian hockey specifically in the data and the a plausible 452 00:34:07,239 --> 00:34:13,320 explanation for it? Absolutely, So we talked a lot about the model's flaws, 453 00:34:13,320 --> 00:34:15,639 which I think is really important to drill into, Like it's worth understanding 454 00:34:16,119 --> 00:34:20,920 the classic saying that all models are rocks are useful, right, Like it's 455 00:34:21,039 --> 00:34:24,280 very important to understand that I don't have everything I wish I had, and 456 00:34:24,320 --> 00:34:30,800 there will always be imperfections, But I still think what has been measured here 457 00:34:30,920 --> 00:34:35,360 is worth caring about. And the reason why is I took these predictions, 458 00:34:35,679 --> 00:34:38,199 or I took our league quality estimates and found two very important things, and 459 00:34:38,199 --> 00:34:43,360 then we can show how to apply them specifically. So, my league quality 460 00:34:43,480 --> 00:34:47,360 estimates as measured in this model, if you take them and attach them to 461 00:34:47,440 --> 00:34:52,719 your prospects, and you include my league quality estimates in your projection of how 462 00:34:53,000 --> 00:34:57,599 likely a prospect is, to, say, make the NHL, you will 463 00:34:57,639 --> 00:35:01,719 find that players who play in more difficult leagues as measured by my model have 464 00:35:01,840 --> 00:35:06,960 been more likely to make the NHL, all else being equal, accounting for 465 00:35:07,000 --> 00:35:09,239 things like their age, they're scoring, and whatnot that we're already accounted for. 466 00:35:09,800 --> 00:35:15,920 And second, I mentioned time trends, so that leagues may be declining 467 00:35:15,079 --> 00:35:21,039 or improving over time. I split out the effects just to prove a point 468 00:35:21,079 --> 00:35:28,360 basically, and found that when the model recognizes leagues as in decline, players 469 00:35:28,480 --> 00:35:31,840 who are in a declined version of the same league are less likely to make 470 00:35:31,840 --> 00:35:37,480 the NHL, again holding scoring constant than players who are in the better version 471 00:35:37,519 --> 00:35:42,119 of that league. So the example we're going to dive into is the KHL. 472 00:35:42,519 --> 00:35:47,960 My model has KHL as getting significantly worse recently, which then if you 473 00:35:49,000 --> 00:35:51,679 apply to predictions, it will say that, hey, a guy who scores 474 00:35:51,719 --> 00:35:55,280 twenty points in the KHL is less likely as a first, say an eighteen 475 00:35:55,360 --> 00:36:00,159 year old guy's eighteen and a half years old and he scores twenty two points 476 00:36:00,159 --> 00:36:04,159 and thirty KHL games. If that guy did it ten years ago, you 477 00:36:04,159 --> 00:36:07,480 would expect the odds of him to make the NHL are higher. And I've 478 00:36:07,480 --> 00:36:10,760 tested that and was able to prove that. So that basically comes down to 479 00:36:10,800 --> 00:36:14,480 the so what of the model. It's got all these flaws and yet it 480 00:36:14,559 --> 00:36:19,400 ends up being useful anyways. So the big takeaway from this, I think 481 00:36:19,480 --> 00:36:22,719 the most important thing is you're seeing a lot of Russian records fall. Mitchikov's 482 00:36:22,719 --> 00:36:27,440 breaking them. Forget who one of the former NHL guys I used to break 483 00:36:27,440 --> 00:36:30,480 the Russian scoring record. All sorts of these crazy things are happening. And 484 00:36:30,679 --> 00:36:36,239 if you look at the changes in league quality estimates, some of the biggest 485 00:36:36,280 --> 00:36:40,400 declining leagues are the NHL, the KHL, and the VHL. So it's 486 00:36:40,440 --> 00:36:45,400 systemic to all of the major Russian leagues we would care about as it relates 487 00:36:45,400 --> 00:36:50,679 to prospect or NHL level analysis have been declining. And then I compare it 488 00:36:50,760 --> 00:36:54,000 to a sample of similar leagues. So if you take the KHL compared to 489 00:36:54,280 --> 00:37:02,800 other pro leagues in recent times, the KHL has declined by these estimates significantly 490 00:37:02,920 --> 00:37:07,840 more than the average of the other pro leagues, suggesting that we need to 491 00:37:08,000 --> 00:37:13,920 change the way we think about KHL scoring. Not that it isn't good, 492 00:37:14,199 --> 00:37:17,280 which is tempting to do when you see things like the KHL is not as 493 00:37:17,280 --> 00:37:21,440 good as it used to be, but just that it is worse than it 494 00:37:21,599 --> 00:37:23,480 was. So even once you account for this, somebody like Mitchkov, who 495 00:37:23,519 --> 00:37:28,960 has I think the second best NHL E of the millennium, he's still a 496 00:37:28,960 --> 00:37:31,480 good prospect. He's still first overall level scorer and all of those things, 497 00:37:31,880 --> 00:37:37,119 but he's not the second best score of the millennium once you account for this 498 00:37:37,159 --> 00:37:44,400 stuff, so you can track relative changes in prospect probabilities based on what the 499 00:37:44,440 --> 00:37:46,880 model is saying. And I think the most important change in those probabilities is 500 00:37:46,920 --> 00:37:52,159 that all else being equal to these Russian guys need to score more to be 501 00:37:52,159 --> 00:37:54,639 as good as they were. So like you're seeing Mitchkov outscore Ovi, that 502 00:37:54,639 --> 00:38:00,880 doesn't necessarily mean Mitchkov's a better prospect than Ovechkin was, because the league Ovechkin 503 00:38:00,920 --> 00:38:05,719 played in was significantly better, even though it is the same league. If 504 00:38:05,719 --> 00:38:09,199 that makes yeah, But it's Danilli You're Off as a guy I think about, 505 00:38:09,360 --> 00:38:13,920 and Victor's hate for Danila You're Off in his draft here was epic. 506 00:38:14,159 --> 00:38:16,960 He gave me the hardest time for liking Danilli You're Off. And Danili You're 507 00:38:17,000 --> 00:38:22,199 Off played twenty one KHL games. He had zero points in those games. 508 00:38:22,599 --> 00:38:28,840 He averaged under two minutes a game. Russia just screws around with these prospects. 509 00:38:29,079 --> 00:38:32,039 So it makes it so difficult. In Mitchkoff, we get some kind 510 00:38:32,079 --> 00:38:36,480 of a sample on him because at least they're using him and they're allowing to 511 00:38:36,480 --> 00:38:39,000 do things. They're setting him up there and so he's running up video game 512 00:38:39,079 --> 00:38:45,840 scores. But I just don't know how you can have the world where Mitchkoff 513 00:38:46,000 --> 00:38:52,079 and you're Off are on the same wavelength, although you're off now he's got 514 00:38:52,159 --> 00:38:54,159 twenty seven points in thirty four games this year. He's actually being used. 515 00:38:54,320 --> 00:38:58,760 We can give some kind of a credit there two years past this draft year. 516 00:38:59,320 --> 00:39:04,760 But yeah, that's what makes it so difficult to me to comprehend how 517 00:39:04,840 --> 00:39:08,760 to do this kind of equivalency because of especially in Russia, it seems to 518 00:39:08,800 --> 00:39:15,119 me the way that they screw around with these guys' deployment. Yeah, and 519 00:39:15,159 --> 00:39:20,119 there's it's not just Russia too. Russia seems worse like the worst of them 520 00:39:20,119 --> 00:39:23,719 all. I'm definitely with you there. And another problem with anything equivalency on 521 00:39:23,760 --> 00:39:29,239 the pro guys is from a statistical perspective, you have to be less certain 522 00:39:29,800 --> 00:39:36,199 about the pro guys because there's so many fewer events that matter at a pro 523 00:39:36,320 --> 00:39:40,079 level. So Connor Bdard and mattve Mitchikov say their NHL ease are similar. 524 00:39:42,159 --> 00:39:45,599 Connor Bdard had to do one hundred and twenty things to drive that NHL E 525 00:39:45,719 --> 00:39:51,239 goals and assists over seventy games. Mitchkov only had to do twenty things over 526 00:39:51,320 --> 00:39:54,880 twenty seven games to drive his NHL E. So there's a significant level of 527 00:39:55,000 --> 00:40:01,280 uncertainty based on there as well that we have to account for. Even you 528 00:40:01,320 --> 00:40:06,280 can see on my measure, I can't necessarily account for that, But the 529 00:40:06,599 --> 00:40:10,159 pro leagues have wider air of bars because there's just more uncertainty underlying pro league 530 00:40:10,199 --> 00:40:15,519 scoring because it's so much more rare too, which also plays into that and 531 00:40:15,559 --> 00:40:20,199 makes it very difficult, which is something that needs to be accounted for and 532 00:40:20,239 --> 00:40:22,159 set. And I think this is part of the reason why if you look 533 00:40:22,159 --> 00:40:25,679 at the big NHL, you guys, the ones who fail almost all come 534 00:40:25,719 --> 00:40:30,400 from pro leagues, because even though it's harder to score in a pro league, 535 00:40:30,480 --> 00:40:35,119 it's a lot easier to luck into twenty points than it is to luck 536 00:40:35,239 --> 00:40:37,840 into one hundred and twenty points like you would have to in the CHL To 537 00:40:37,920 --> 00:40:42,679 be one of those really high scoring guys, if that makes sense, totally 538 00:40:42,760 --> 00:40:45,719 sample size issues. We talk about these things all the time. There's going 539 00:40:45,800 --> 00:40:51,719 to be way more variability in every direction when you have twenty games, twenty 540 00:40:52,000 --> 00:40:55,159 examples, twenty incidents, and a lot of them just are noise and don't 541 00:40:55,159 --> 00:40:59,119 show up as much, and so we just discount that as these guys aren't 542 00:40:59,159 --> 00:41:01,360 very good and maybe there's some gems in there that we're not picking up on. 543 00:41:01,440 --> 00:41:06,320 And then there's the guys who maybe lucked into a bunch of secondary assists 544 00:41:06,840 --> 00:41:09,199 and all of a sudden they look like they're amazing, and I think this 545 00:41:09,239 --> 00:41:13,320 is really good. Like trying to recalibrate some of our ideas and expectations. 546 00:41:13,320 --> 00:41:16,280 I want to draw some attention specifically to last year's draft, because you also 547 00:41:16,360 --> 00:41:22,840 looked at most recently this decline in KHL and there were there's obviously geopolitical concerns, 548 00:41:22,880 --> 00:41:29,880 but we also saw some really interesting Russian players drafted in that. And 549 00:41:29,920 --> 00:41:31,519 of course there was Mitchikov, which I know you wrote a whole thing about 550 00:41:31,559 --> 00:41:34,840 and definitely people should go read that. But I also wanted to focus on 551 00:41:34,840 --> 00:41:37,960 some of the other guys like Dmitri Simyshev and Dlo Boot. This was really 552 00:41:38,000 --> 00:41:44,360 interesting because we had one team take to what some would consider big swings on 553 00:41:44,440 --> 00:41:47,559 these two players. And one of them's defensemen obviously, so that's different. 554 00:41:47,679 --> 00:41:52,719 But they're both big players and they're both in the KHL, and Samyshev didn't 555 00:41:52,800 --> 00:41:55,960 have as much going I actually neither one of them did. But I'm just 556 00:41:55,960 --> 00:42:00,960 wondering what your thoughts are on those in general, because there were people in 557 00:42:00,000 --> 00:42:02,679 every direction, some people saying, Wow, this is great, I love 558 00:42:02,719 --> 00:42:06,119 that they did this, and other people panning it, just saying, what 559 00:42:06,159 --> 00:42:08,920 are you thinking? There's such better options, even passing on Mitchkov, although 560 00:42:08,920 --> 00:42:13,679 there may have been other reasons why Arizona did that. But what are your 561 00:42:13,679 --> 00:42:21,440 thoughts on specifically those two? Yeah? I thought those were the really bold 562 00:42:21,639 --> 00:42:29,800 choices. I didn't like, necessarily hate them to the degree that most people 563 00:42:29,840 --> 00:42:35,400 who make draft models did like. Our model was relatively high on Simashev and 564 00:42:35,599 --> 00:42:43,840 relatively high on Daniel as far as I could tell, But I would have 565 00:42:44,039 --> 00:42:47,440 definitely went a different direction even I would have based on our model and whatnot, 566 00:42:47,599 --> 00:42:52,280 and based on the EP grades and everything I've heard from public scouting and 567 00:42:52,280 --> 00:42:57,719 stuff. I've maybe Mitchcov didn't want to play in Arizona or something, but 568 00:42:57,760 --> 00:43:01,320 if you look Mitchkov and Benson being the two guys that were drafted immediately after 569 00:43:01,360 --> 00:43:07,199 their picks is a tough look. Now, with that being said, I 570 00:43:07,239 --> 00:43:14,159 do respect the swing, especially on Simashev. I wouldn't have made them made 571 00:43:14,159 --> 00:43:19,840 these swings. But if you think you have identified something and you have Simyshev, 572 00:43:19,920 --> 00:43:22,480 this six' four physical defenseman who can skate like the win like he 573 00:43:22,599 --> 00:43:30,920 has every physical tool in the book. As much as people publicly criticize picks 574 00:43:30,039 --> 00:43:35,679 like Simashev for not swinging on upside because they weren't scoring and scoring is usually 575 00:43:35,719 --> 00:43:42,440 highly related to upside at the NHL level the specific pick, I respect the 576 00:43:42,639 --> 00:43:46,840 thought process of swinging for the player you've identified as potentially a complete game chain 577 00:43:46,920 --> 00:43:52,679 like Simyshev. If he hits a ceiling, there won't be other players like 578 00:43:52,800 --> 00:43:57,840 him in the league, which I respect the swing for. But I would 579 00:43:57,840 --> 00:44:01,880 be really worried that pretty much everyone I saw and as best as I could 580 00:44:01,880 --> 00:44:07,039 tell, other teams were not nearly as high on those players, even if 581 00:44:07,119 --> 00:44:13,119 again they are good prospects, they maybe were just overdrafted, at which point 582 00:44:13,199 --> 00:44:19,159 you start running into the scenario that you need to be right and everybody else 583 00:44:19,280 --> 00:44:23,000 needs to be wrong, which is a very bad situation to put yourself in 584 00:44:23,039 --> 00:44:30,039 when you have thirty one other teams all and the public scouts and the public 585 00:44:30,159 --> 00:44:32,880 models. There is a lot of opposition to those picks. And any time 586 00:44:32,920 --> 00:44:37,679 you put yourself on the opposite side of something that many smart people believe, 587 00:44:38,000 --> 00:44:42,440 I think you should be a lot of red flags should be going off. 588 00:44:42,719 --> 00:44:47,599 If that makes sense. Absolutely, you're digging your heels in positioning yourself opposite 589 00:44:47,599 --> 00:44:52,639 everyone else. That's yeah, you better be right, because otherwise there's plenty 590 00:44:52,639 --> 00:44:55,760 of people to point out, yeah, oh you were the outlier. There's 591 00:44:55,760 --> 00:44:59,760 some other interesting things here. There's so many interesting things we could talk about. 592 00:45:00,000 --> 00:45:06,079 I wanted to shift to another potential. What I view is sometimes a 593 00:45:06,119 --> 00:45:10,760 somewhat similar situation. So in Europe there's other leagues. I always find it 594 00:45:10,840 --> 00:45:16,760 interesting and sometimes give a little feather in the cap of players who are prospects 595 00:45:16,760 --> 00:45:21,159 who are playing professionally at a young age, So that sometimes happens in Russia. 596 00:45:21,239 --> 00:45:27,039 Obviously, there's Finnish examples or sweetiest examples. More recently the Czech league 597 00:45:27,079 --> 00:45:30,679 has propped up a little bit. It is difficult, and I think especially 598 00:45:30,719 --> 00:45:32,400 the Finnish League, which is known to be a very low scoring league, 599 00:45:32,400 --> 00:45:37,239 and you already mentioned how that can make it difficult to assess because you're talking 600 00:45:37,280 --> 00:45:40,559 about more rare events. So I wonder if you could just speak on those 601 00:45:40,559 --> 00:45:45,400 a little bit. Is it still you think worthwhile to give special props to 602 00:45:45,440 --> 00:45:51,920 players like that who are playing significant games in the top level in Sweden, 603 00:45:52,159 --> 00:45:54,440 Finland you can include check if you want there, maybe even the NL, 604 00:45:54,559 --> 00:45:59,360 there's some examples there. But I do find it's difficult, especially when the 605 00:45:59,400 --> 00:46:04,280 player is and we see this a lot are playing fifteen to twenty twenty five 606 00:46:04,280 --> 00:46:07,800 games in the SHL, but they have a couple of apples, and then 607 00:46:07,880 --> 00:46:09,480 it's up great that they're playing, but they're not getting many points, and 608 00:46:09,519 --> 00:46:14,719 then it's how reasonable is that with twelve minutes of time highce right, yeah, 609 00:46:14,840 --> 00:46:17,559 exactly, I definitely agree. I think there is a significant amount of 610 00:46:17,599 --> 00:46:22,440 value in just showing up in pro games. So I had mentioned our model 611 00:46:22,519 --> 00:46:25,079 was a lot higher on the on SIMMISHEV than the other public models, and 612 00:46:25,079 --> 00:46:29,320 that was the reason why, because our model will specifically account for the fact 613 00:46:29,360 --> 00:46:32,440 that, hey, this guy played thirty KHL games or whatever it was as 614 00:46:32,480 --> 00:46:37,480 an eighteen year old, even if you're not scoring. People who play at 615 00:46:37,480 --> 00:46:42,239 these high levels, this means a professional coach has decided that you, at 616 00:46:42,239 --> 00:46:45,440 eighteen years old, are worth playing because they give you a better chance to 617 00:46:45,440 --> 00:46:49,880 win every night. That's a massive signal, even if you're not necessarily scoring 618 00:46:49,920 --> 00:46:53,320 points. That's really important. And then it also builds into the fact, 619 00:46:53,400 --> 00:46:57,599 so I think you have to look at them as two separate effects. When 620 00:46:57,639 --> 00:47:00,519 they're in a pro league, Yes there is signal in and of itself, 621 00:47:01,000 --> 00:47:05,519 and then it goes two different directions. So when you're in a pro league 622 00:47:05,559 --> 00:47:09,599 and not scoring, you have to respect the fact that it's easier to be 623 00:47:09,679 --> 00:47:15,880 good and not score in a pro league because those events are so much more 624 00:47:15,960 --> 00:47:19,320 rare, so that the variance being against. You can't tank your NHL e 625 00:47:19,440 --> 00:47:22,159 just by hitting two crossbars or something like that. And then yes, there 626 00:47:22,239 --> 00:47:25,360 is value in and of itself even if you are scoring like a mitgejob, 627 00:47:25,760 --> 00:47:30,119 but you will have to respect the fact that the variance goes the other way. 628 00:47:30,119 --> 00:47:35,400 With these high scores where it's probably more likely that they can be inflated 629 00:47:35,440 --> 00:47:38,039 by NHL e So it's a nonlinearity I think, where yes, there is 630 00:47:38,159 --> 00:47:42,639 value in and of itself, but you need to be careful when talking about 631 00:47:42,679 --> 00:47:47,920 the scoring of these prospects because it can affect the low scores and the high 632 00:47:47,920 --> 00:47:52,480 scores differently, I think from an NHL perspect or, like from an equivalent 633 00:47:52,519 --> 00:48:00,079 seed perspective, awesome. I want to go in as slightly different direction a 634 00:48:00,320 --> 00:48:02,719 similar topic, and that is looking at some of the North American leagues, 635 00:48:02,840 --> 00:48:07,079 and I think you mentioned and I would definitely say that the NC DOUBLEA is 636 00:48:07,119 --> 00:48:13,400 different from not including the AHL, but like the other non professional leagues, 637 00:48:13,480 --> 00:48:19,000 right, NCUBLEA, and you're talking USHL and CHL leagues WHL, OHL, 638 00:48:19,760 --> 00:48:23,239 MHL, and you could maybe lump in the AHL, BCHL, but those 639 00:48:23,239 --> 00:48:27,920 are so much lower I think, and I want to get your take on 640 00:48:27,960 --> 00:48:31,880 this. There's a pretty big difference between those few players who play their draft 641 00:48:31,960 --> 00:48:37,320 season in the NC DOUBLEA. You're talking like last year Fantilly there was Jack 642 00:48:37,400 --> 00:48:42,559 Eigel, Paul Korea. You're talking this year maclen Celabrini. That is a 643 00:48:42,760 --> 00:48:45,199 very different level and to me, should boost them up a lot. But 644 00:48:45,320 --> 00:48:51,320 also there's the other rarity effect that you're talking about. These are not you 645 00:48:51,360 --> 00:48:53,599 know, it's not the same as probably the generational talents that are like exceptional 646 00:48:53,920 --> 00:49:00,199 status in the OHL, but some of them are as high. But they're 647 00:49:00,280 --> 00:49:05,639 so you're talking about how do you compare these guys to someone who's a few 648 00:49:05,679 --> 00:49:08,800 months younger. And that's sometimes the challenge with playing in college. But it's 649 00:49:08,840 --> 00:49:15,320 not necessarily available to everybody, but it certainly can mean a lot if it 650 00:49:15,360 --> 00:49:17,559 goes really well. And then the other flip side of that is sometimes it 651 00:49:17,599 --> 00:49:22,880 goes okay, and then they just look like not the best prospect when somebody's 652 00:49:22,960 --> 00:49:25,880 killing it in the Q or in the USHL. But it's such a lower 653 00:49:25,960 --> 00:49:30,199 equivalent sye. So how do you marry it those thoughts in those discrepancies. 654 00:49:30,400 --> 00:49:36,159 Yeah, I have a theory that I haven't tested yet, but I think 655 00:49:36,320 --> 00:49:39,239 may help marry it. And I know I believe Jeremy Davis. I think 656 00:49:39,280 --> 00:49:44,599 it was he had the job that I currently have an EP before I did. 657 00:49:45,159 --> 00:49:50,519 I remember way back when he had different age adjustments for different leagues, 658 00:49:51,239 --> 00:49:55,320 and if I remember correctly, it was that as an example, the pro 659 00:49:55,480 --> 00:50:00,039 league guys got bigger age adjustments than the junior guy, so then applied to 660 00:50:00,079 --> 00:50:07,639 a more the different junior leagues. I think there's an effect there where the 661 00:50:07,800 --> 00:50:13,159 younger you are, there's probably larger effects the more difficult the league. So 662 00:50:13,280 --> 00:50:20,000 like these guys who are in college performing all right, it could be more 663 00:50:20,119 --> 00:50:24,119 impressive just because they're in A The NC DOUBLEA is significantly better than any of 664 00:50:24,159 --> 00:50:29,519 the other junior leagues of USHL and the OHL and whatnot, So I think 665 00:50:30,119 --> 00:50:37,039 they may be punished for being relatively young in the league more so than guys 666 00:50:37,079 --> 00:50:40,800 who are playing in these weaker leagues. Is something that I think probably exists, 667 00:50:40,800 --> 00:50:45,280 but I have not adjusted for yet. One of the interesting things about 668 00:50:45,280 --> 00:50:53,199 the differences between these leagues NCAA and HL, I would think there's relative stability. 669 00:50:53,800 --> 00:50:59,400 You tend not to get jerked between leagues. In those things, those 670 00:50:59,440 --> 00:51:01,960 players will pretty much sit there first season. It could be traded. But 671 00:51:02,760 --> 00:51:08,360 overseas leagues SAHL, there does seem to be some loney going back down KHL. 672 00:51:08,400 --> 00:51:12,800 Man they could just apparently they could just do whatever. Between the VHL, 673 00:51:12,960 --> 00:51:16,559 MHL and KHL. People are getting jerked around all the time, and 674 00:51:17,280 --> 00:51:22,199 it's really got to change how you can translate between those leagues because those things 675 00:51:22,239 --> 00:51:28,000 are varying in quality even throughout the year. It is that all does that 676 00:51:28,039 --> 00:51:31,119 all come out in the wash as well, just because you could say all 677 00:51:31,159 --> 00:51:34,719 that, but at the end of the day, when those guys come over 678 00:51:34,800 --> 00:51:37,760 to the NHL, they all were in the same boat at the time. 679 00:51:39,000 --> 00:51:43,880 Yeah, that's a tough one to account for from purely a statistical respect. 680 00:51:43,880 --> 00:51:49,079 Then, like a presumably does matter. Evan Demodov has played MHL, VHL 681 00:51:49,199 --> 00:51:52,840 and KHL games this year. Like nobody in the OHL is ever going to 682 00:51:52,920 --> 00:51:57,840 have played in the I don't even Canada doesn't really have equivalents. But nobody 683 00:51:57,880 --> 00:52:00,199 in North America would ever play in three leagues in the same year, unless 684 00:52:00,199 --> 00:52:04,840 you're talking to like World Juniors and stuff. But that's different. So it 685 00:52:05,679 --> 00:52:10,320 presumably does matter just at a personal level for these kids, but adjusting for 686 00:52:10,440 --> 00:52:13,960 it in a model is going to be really difficult. I think that's the 687 00:52:14,000 --> 00:52:16,679 point where you have to lean on your scouts, because some individuals may thrive 688 00:52:16,760 --> 00:52:21,360 in that environment, some individuals may really struggle with that right. Some people 689 00:52:22,280 --> 00:52:24,760 might take longer to adapt as they're changing back and forth. I don't know 690 00:52:25,000 --> 00:52:30,880 how much it matters on average, but there are presumably small edges with the 691 00:52:30,920 --> 00:52:36,519 individuals that you may be able to find there. Okay, yeah, Chase, 692 00:52:36,800 --> 00:52:39,679 We've we've poked, we've prodded, we've been all around this thing. 693 00:52:40,119 --> 00:52:46,199 Do you think there's anything else that people out there can learn that you want 694 00:52:46,199 --> 00:52:50,519 to explain to people about your model? And I'm sure that the next question 695 00:52:50,559 --> 00:52:52,880 everybody you have is can you just give me a spreadsheet with all the equivalencies 696 00:52:52,920 --> 00:52:59,000 for these guys? Please? I do have a spreadsheet of all The One 697 00:52:59,639 --> 00:53:04,280 really ian thing to go through with the model is that there's been a systemic 698 00:53:04,440 --> 00:53:12,360 change recently in the non NHL league's quality relative to the NHL. And what 699 00:53:12,440 --> 00:53:19,159 I have been seeing is that the NHL as a whole has been improving relative 700 00:53:19,440 --> 00:53:24,920 to these other leagues. So on average, most leagues are worse relative to 701 00:53:24,960 --> 00:53:30,159 the NHL than they currently are, which I think explains why you're seeing like 702 00:53:30,199 --> 00:53:36,000 all sorts of guys are scoring at these near record paces from an NHL equivalency 703 00:53:36,039 --> 00:53:39,840 perspective, And I think if these leagues are farther away from the NHL than 704 00:53:39,880 --> 00:53:43,639 they used to be. I think that would explain a lot of it. 705 00:53:43,719 --> 00:53:49,119 Also, even somebody like Jack Hughes who buy NHL equivalency, was like a 706 00:53:49,159 --> 00:53:53,480 borderline generational prospect, but Dards the first rookie we've seen true rookie, not 707 00:53:53,880 --> 00:53:58,519 you know, twenty year olds and whatnot coming over a lot of the top 708 00:53:58,599 --> 00:54:02,920 picks right out of the draft. Everybody's rookie season has sucked since Matthews basically, 709 00:54:05,079 --> 00:54:07,639 and the Matthews in line A, there's been a lot of very poor 710 00:54:07,679 --> 00:54:10,920 rookie seasons, which lines up basically with my model. In twenty eighteen, 711 00:54:12,880 --> 00:54:16,920 rest of the world as a whole started to get significantly worse compared to the 712 00:54:17,000 --> 00:54:21,840 NHL. So I think there's a distancing of these leagues which could be driven 713 00:54:21,880 --> 00:54:25,519 by a lot of things. The NHL is getting smarter, the NHL's expanding, 714 00:54:25,599 --> 00:54:29,920 so the best players from a lot of these other leagues are leaving, 715 00:54:30,039 --> 00:54:34,800 and there's forty extra NHL jobs or whatever it is on any given night compared 716 00:54:34,800 --> 00:54:37,199 to what it used to be. So I think that's really important to recognize 717 00:54:37,199 --> 00:54:44,320 that on average, a one hundred point OHL player or KHL player that don't 718 00:54:44,480 --> 00:54:47,920 exist. But like you get the idea or BHL player or whatever is probably 719 00:54:49,119 --> 00:54:52,119 farther away from the NHL than the same guy would have been if he was 720 00:54:52,159 --> 00:54:58,159 doing that in twenty ten compared to somebody doing it right now. Wow. 721 00:54:58,320 --> 00:55:00,840 Yeah, and that will jack up your sample, But yeah, you definitely 722 00:55:01,119 --> 00:55:06,760 have that drift accounted for in some of those error bars. Yeah, and 723 00:55:06,800 --> 00:55:14,000 that's why I posted a chart ware of the thirty best NHL seasons when you 724 00:55:14,000 --> 00:55:15,880 don't account for these time trends, like nine of them have come from this 725 00:55:16,000 --> 00:55:21,960 decade, despite the fact that we're only three years into this decade. This 726 00:55:22,000 --> 00:55:25,719 decade had a pandemic where a lot of guys missed. Twenty twenty one, 727 00:55:25,760 --> 00:55:30,280 specifically, a lot of guys didn't even get a shot at setting NHL ree 728 00:55:30,480 --> 00:55:34,320 records, and twenty twenty two was universally recognized as one of the worst draft 729 00:55:34,360 --> 00:55:38,480 classes you've ever seen. And yet we're still seeing elite seasons out twice the 730 00:55:38,559 --> 00:55:44,400 rate this decade as we've seen previous decades, and mixing in the fact that 731 00:55:44,480 --> 00:55:47,079 rookies have actually been worse despite this fact. So I think it explains a 732 00:55:47,079 --> 00:55:51,239 lot of kind of things that are happening in the hockey world at the macro 733 00:55:51,360 --> 00:55:52,800 level right now, this has been so great. I just wanted to follow 734 00:55:52,840 --> 00:55:55,840 up with one last thing. I know that you've been brought into the EP 735 00:55:57,000 --> 00:55:59,840 team and it's super exciting to have your insights there, and I know that 736 00:56:00,119 --> 00:56:02,599 I'm pretty sure you were there for the first draft meeting, and yeah, 737 00:56:02,639 --> 00:56:07,199 I'm just wondering, obviously your role is different and a lot of the other 738 00:56:07,239 --> 00:56:10,480 scouts are really obviously relying on the eye test, in person, video scouting 739 00:56:10,960 --> 00:56:16,280 and other things. What do you bring to that discussion. I'm wondering what 740 00:56:16,360 --> 00:56:20,239 kind of specific insights you point out and things like that. What has been 741 00:56:20,239 --> 00:56:22,760 your role there. Yeah, so I'm easing into it. I think I'll 742 00:56:22,840 --> 00:56:25,519 have a better answer for you in six months than I do right now because 743 00:56:25,519 --> 00:56:30,320 I've just been there. But for example, my favorite thing to do is 744 00:56:30,440 --> 00:56:36,639 test things like I in grad school, my best class was econometrics, which 745 00:56:36,679 --> 00:56:40,559 is just causal inference, So what's the average effect of X on Y? 746 00:56:42,119 --> 00:56:46,000 So just basically testing scientific questions like that, and I think that's one of 747 00:56:46,079 --> 00:56:49,719 the primary things where I can add value. And as an example, in 748 00:56:49,760 --> 00:56:53,239 the last scouting meeting, we were talking about one of the prospects and the 749 00:56:53,239 --> 00:56:59,599 scouts have brought up how this guy was dramatically outscoring his closest teammates, which 750 00:56:59,679 --> 00:57:01,119 sounds like a good thing. I assume that was a good thing. I 751 00:57:01,159 --> 00:57:06,440 made a variable for my draft model to account for guys who were doing that, 752 00:57:06,639 --> 00:57:09,000 and I assume to because I assumed it was a good thing. And 753 00:57:09,039 --> 00:57:13,000 what I actually found when I fed into the model was the model would actually 754 00:57:13,800 --> 00:57:19,880 learn to punish players who are way out scoring their teammates. The model will 755 00:57:19,880 --> 00:57:23,320 actually give more credit to guys who score on good teams than guys who score 756 00:57:23,360 --> 00:57:29,239 on bad teams. The logic, I think being that points are a scarce 757 00:57:29,320 --> 00:57:32,599 resource. There's only so many puck touches to go around, So if you're 758 00:57:32,599 --> 00:57:37,719 beating out all stars for puck touches, you're probably that's probably more impressive than 759 00:57:37,760 --> 00:57:42,239 beating out guys on the worst team in the league. So just being able 760 00:57:42,280 --> 00:57:47,159 to as E gauge test things like test conventional wisdom, because a lot of 761 00:57:47,199 --> 00:57:51,440 the time conventional wisdom is right, but it can also be wrong in all 762 00:57:51,440 --> 00:57:57,239 sorts of different circumstances. And just having somebody with the ability to test whatever 763 00:57:57,320 --> 00:58:00,960 hypotheses we think we have in the sport, I think and be really valuable, 764 00:58:00,960 --> 00:58:02,960 and that's one of the things that I'm doing right now as well as 765 00:58:04,039 --> 00:58:08,360 just like I wrote an article about statistically age adjusting their draft grades and things 766 00:58:08,400 --> 00:58:12,360 like that, because mentally, it's going to be very difficult to say, 767 00:58:12,360 --> 00:58:15,719 hey, how much more impressive does an eighteen year old have to be to 768 00:58:15,880 --> 00:58:19,360 make up for the fact that he's eighteen relative to a seventeen and a half 769 00:58:19,400 --> 00:58:22,480 year old. So I think there's just finding pockets like that, whereas statistical 770 00:58:22,480 --> 00:58:27,559 analysis can really help you because it's hard to do mentally, And I hope 771 00:58:27,559 --> 00:58:30,559 I'm able to help them out with all sorts of these things and probably find 772 00:58:30,639 --> 00:58:36,199 more ways to contribute. As I stick around, you will find more ways 773 00:58:36,199 --> 00:58:38,320 to contribute, and we will find more ways to consume Chase, because you've 774 00:58:38,320 --> 00:58:43,079 got some great stuff going out there. I hope everybody will check it out, 775 00:58:43,320 --> 00:58:46,880 just once more for posterity. Where should they go, Chase to see 776 00:58:46,920 --> 00:58:51,639 all of your information? Yeah, so they can find if you just search 777 00:58:51,760 --> 00:58:54,320 my name at hon Ei Prospects, anything that I've written should come up on 778 00:58:54,360 --> 00:58:59,920 ep ring side and then I post everything to Twitter at CM Hockey sixty six 779 00:59:00,039 --> 00:59:01,920 as well. As just random thoughts and whatnot. That's probably the best place 780 00:59:01,960 --> 00:59:06,440 to find everything I'm doing. I also just tease all sorts of stuff I'm 781 00:59:06,440 --> 00:59:12,320 working on whatnot there, So that's probably the optimal place to find outstanding everybody, 782 00:59:12,320 --> 00:59:14,920 go out and do that. Thank you so much for coming on today, 783 00:59:15,000 --> 00:59:19,519 Chase and broadening our minds as to new ways that we can find the 784 00:59:19,599 --> 00:59:31,840 prospects who are going to be most successful in our line ay. A couple 785 00:59:31,920 --> 00:59:35,679 things to mention before we get out of here. Our show is brought to 786 00:59:35,679 --> 00:59:38,760 you by fan Tracks. Did you know that you should? Because their sponsor 787 00:59:38,920 --> 00:59:42,519 of the show, you should move your leagues over to fan Tracks. Ask 788 00:59:42,559 --> 00:59:45,519 them. They'll help you. You can start new ones, the most options 789 00:59:45,519 --> 00:59:50,320 for scoring salaries, contracts, customizing your rookie eligibility, start up your leagues 790 00:59:50,360 --> 00:59:52,280 the day after the season ends, anything you can think of. They even 791 00:59:52,320 --> 00:59:57,360 have a pretty good chat feature while other platforms are pulling that one back. 792 00:59:57,920 --> 01:00:00,760 You can look at the fan Tracks hqu side. I write there sometimes. 793 01:00:00,800 --> 01:00:05,119 Actually I wrote most recently on baseball, but don't tell anybody, but there 794 01:00:05,119 --> 01:00:07,760 are plenty of articles on fantasy hockey. That group has geared up like nobody's 795 01:00:07,800 --> 01:00:13,679 business this year. We'd like to thank our entire FHL crew doing the work 796 01:00:13,719 --> 01:00:16,960 over here. We got a posse. We've got a family who are doing 797 01:00:17,039 --> 01:00:23,280 lots of work for us and participating heavily, including content curator Kevin Adams who's 798 01:00:23,280 --> 01:00:28,840 been helping with show prep. Ryan Downey, the commissioner of the Tidy Legs, 799 01:00:28,840 --> 01:00:32,159 affectionately known as the Tidy Admiral Net Guy. We keep him busy. 800 01:00:32,599 --> 01:00:37,000 Brandon, website guru, doing a fantastic job over there. Jeremy Vee you've 801 01:00:37,000 --> 01:00:42,199 heard him on the show. He's our lead scout and he keeps those scouting 802 01:00:42,360 --> 01:00:46,679 reports coming. We got people really digging into the video. We've got kind 803 01:00:46,719 --> 01:00:51,519 of the professional service that Victor has been able to get us into where you 804 01:00:51,559 --> 01:00:54,159 can view a lot of video on these prospects, and Jeremy is coordinating a 805 01:00:54,199 --> 01:00:58,559 lot of the scouting reports we pull out of that and we share with you. 806 01:00:58,679 --> 01:01:01,519 Jason is helping with our aspect ranks victors, making those better and better 807 01:01:01,559 --> 01:01:06,639 every day. Paul is assisting with some workflow and processes because there's too many 808 01:01:06,639 --> 01:01:09,440 things going on. If you've got some skills you'd like to lend the show 809 01:01:09,840 --> 01:01:14,719 to let us continue to build and build. Hit Victor up in the Discord, 810 01:01:14,760 --> 01:01:19,000 email or Twitter. We are brought to you by Dabber Hockey and Daber 811 01:01:19,039 --> 01:01:22,199 Prospects Victors and Editor over there. You can follow his work there, as 812 01:01:22,199 --> 01:01:28,480 well as his other podcast, Daber Prospect Report, he does with Peter Harlane. 813 01:01:28,480 --> 01:01:31,000 If you like this, you're probably gonna like that because they're talking a 814 01:01:31,000 --> 01:01:37,920 lot about the prospects and the different leaks where the prospects play. I do 815 01:01:37,079 --> 01:01:42,519 a solo show, Dynasty Sports Life. I talk four different Dynasty sports sometimes 816 01:01:43,079 --> 01:01:47,280 at the same time. Sometimes we even cover some fantasy hockey, but usually 817 01:01:47,440 --> 01:01:51,599 on that show, I don't know specific fantasy hockey episodes, but I talk 818 01:01:51,639 --> 01:01:55,039 about dynasty playing theory. I think we might even have an upcoming show on 819 01:01:55,159 --> 01:02:00,320 Dynasty ethics. We'll see about that. But this week it's going to be 820 01:02:00,360 --> 01:02:05,880 about quarterbacks coming out in next year's NFL Draft with a great guest who knows 821 01:02:05,920 --> 01:02:09,119 all about that stuff. Follow us on x at fan Hockey Life at Victor 822 01:02:09,199 --> 01:02:15,079 Newno. Twelve, Rate and review us on Apple Podcasts, Spotify, wherever 823 01:02:15,159 --> 01:02:19,840 else you like to get pods. There's so many good places that you can 824 01:02:19,880 --> 01:02:23,199 get this stuff, So check those things out, listen to us, We 825 01:02:23,239 --> 01:02:35,079 appreciate you, and until next time, keep living that fantasy hockey life.