Draft Projections and Accelerated Rebuild – Salt City Hoops Podcast

May 30th, 2014 | by Andy Larsen
(Photo by Rich Barnes/Getty Images)

(Photo by Rich Barnes/Getty Images)

Welcome to the New Salt City Hoops Podcast! This podcast is the first of many to come recorded at the ESPN700 studios, as announced earlier this month.

In this episode, we focus on the #5 draft pick. Ben Dowsett explains collaborator Layne Vashro’s models to us, and reveals his choice for the #5 pick. We break Noah Vonleh down… can he become Chris Bosh? How good will he be right away? Then, we have SCH writer Denim Millward on the program, explaining his plan to have the Jazz be a winning team next year. Could the Jazz acquire Brook Lopez? How good could the Jazz be next season? Then, we have Jacob Frankel of Hickory High and ESPN Truehoop join us to talk about his draft projection model. What do the numbers say the Jazz should do at #5? Finally, we introduce our new weekly segment: the Crazy Trade Idea of the Week. Is our package good enough for Milwaukee to consider dropping down from the #2 spot? All that and more on this week’s podcast!

Thanks to David Glauser for producing.

Andy Larsen

Andy Larsen

Andy Larsen is the Managing Editor of Salt City Hoops, the ESPN TrueHoop affiliate for the Utah Jazz. He also hosts a radio show of the same name every Saturday on 1280 AM.
Andy Larsen

22 Comments

  1. Aged fan says:

    One minor quibble with the explanation of Vashro’s player comparisons: it sounded from the podcast like they should be interpreted as being capable of showing how VALUABLE a prospect is overall. But I’m pretty sure that the comps are as likely to show STYLE similarities as value similarities.

    Since standard deviations (which include a squaring operation, if I’m remembering right) are averaged and minimized (see Vashro’s explanation for comps: https://docs.google.com/spreadsheet/ccc?key=0Aipd_3kGfREgdHRtcThFd3ZtMlRPYldIRmdGb241RlE&usp=drive_web#gid=2) aspects of a players game that are statistically unusual in any way are magnified in this calculation (and thus the calculation finds as comps players with similar unusualness). For players whose style is not very unusual, the comp is more likely to turn up a player of similar value.

    So, if I’m understanding Vashro’s calculation correctly, while there is some element of value built in, the more unusual a player is the more likely the comp is to turn out a player with similarity in style than in value. (For example, Kyle Anderson’s closest comp is Luke Walton even though the value (EWP) comparison (14.2 vs. 3.0) is completely different (see also Marcus Smart’s comp). And Vashro seems to recognize this by including a “difference” calculation along with his comp to show whether the player is likely to be better or worse than the comp.

    In other words, it’s tricky to interpret the comps Vashro gives, and I’d be wary of tossing the names around without very carefully explaining what the calculation may or may not be doing.

    • Ben Dowsett says:

      Thanks for the comment – you’re mostly correct but just missing an element or two of context, and also I’ll absolutely agree with you that my on-air description of Layne’s models was relatively terrible, for which I apologize. Layne’s Comparable models certainly end up matching up plenty of players who compare stylistically but, while I don’t want to put words in his mouth, I believe if you asked him, that is more an inherent result of matching the players who most closely resembled each other across all variables in college rather than the other way around. That is, it’s natural to see plenty of stylistically similar players but not an express goal of the models. There are many examples, if you look through carefully, of players who match in Layne’s Comparables and weren’t similar whatsoever in style.

      One other thing to take heavily into account that you didn’t mention is the Standard Deviation column to the right of each comp – this should be used in conjunction with the Difference measurements. So for your example of Kyle Anderson, while Walton is indeed one of the closest comps, that particular comp showed a .51 SD across all variables, one of the very highest of any prospect, and all of his comps have similarly high standard deviations. As I’m sure you know, this means that while Walton was indeed one of the closest comps, the comparison itself isn’t particularly robust from the standpoint of the models. So it’s just about using all that info contextually within the process of evaluating a prospect.

      • Aged fan says:

        Yes, agree that the SD column should be used in conjunction with the difference column. In Kyle Anderson’s case, it adds to the conclusion that he’s truly one-of-a-kind.

        I’m not sure I’m totally following your first point. Are you saying that you think Vashro intended to create a value comp and also (as a side-effect) ended up with something of a style-comp (at least in certain circumstances). And that a value comp is really what this measure is? If so, I’m not sure I totally agree. Or at least by introducing a sum of squares element into the calculations (correct me if I’m wrong if that sum of squares element is not part of the calculation), I believe that he’s inadvertently creating a style comp at the partial expense of a value comp (of course I mean statistical style rather than visual style). If he was trying to produce a pure value comp, wouldn’t it make more sense not to involve squaring?

        In any sense, I think we agree that both style and value similarities are part of the output of this calculation, even if we may disagree on how they are balanced. But this fuzziness of interpretation is why I’m hesitant put much stock in the some of names that turn out as comps.

        PS — really enjoyed the podcast

        • Ben Dowsett says:

          Thanks – enjoyed recording it, the ESPN studio is excellent and we will be even better once we are more comfortable with it. As to my first point, that is fairly close to what I’m saying, if we define “value” in this case to mean only the statistical variables he uses to calculate the model. I can’t speak to the particulars of his calculations in the regard you’re speaking about, but I think we’re talking about pretty much the same thing anyway – isn’t “statistical style” really just “a player’s statistics”?

  2. Aged fan says:

    Let’s imagine a possible draft choice with the following statistical profile compared to an average college player (in units of standard deviations from average):

    points: +1
    rebounds: +1
    FG%: +1
    FT%: +1
    FT attempts: +1
    assists: +1
    steals: +1

    Then let’s imagine 2 possible comps:
    Player A:
    points: +2
    rebounds: +2
    FG%: +2
    FT%: +2
    FT attempts: +2
    assists: +2
    steals: +2

    Player B:
    points: +3
    rebounds: -1
    FG%: 0
    FT%: +2.5
    FT attempts: -.5
    assists: +1
    steals: +2

    Who is the better comp to the draft choice? Player B is by value (assuming all categories are weighted similarly), at 7 total standard deviations above average for both. But by style, player A is a better comp. That is, they have similar profiles of the categories in which they rate above or below average (though quite different total values).

    I’m assuming Vashro’s calculations produce the second (stylistic) kind of comp–producing a better match for player A, more than they do the first. But I’d have to have more detail about his method to be sure.

    • Ben Dowsett says:

      Yeah, this is certainly stuff that I wouldn’t want to comment on for fear of badly misrepresenting Layne’s work. I see what you’re getting at, though.

  3. casey says:

    I feel so vindicated that you also think Favs ideal position isn’t center. I always thought Favors was big enough to play center until this last year seeing him match up against guys like Robin Lopez and Dalembert. A lot of people just say that Favors wasn’t his best because of our defensive system, but I saw him get overpowered by big guys all the time. In fact, you should have Dan on your podcast and debate with him about Favs playing the 5 (argued w/ me on twitter about it). Ideally, because Favors can’t shoot either, we can get a three point shooting, shot blocking, 7′ center. Wait, no play like that exists?

  4. Alexander Buck says:

    Hmm, I don’t like the idea of Vonleh defending centers. Vonleh might have the size and bulk to defend centers in the future (though he’s smaller than Favors by a good bit now…), but putting him against centers designates him as the rim protector. With Favors guarding PFs, he’ll often be away from the rim since so many PFs today can shoot, leaving Vonleh to protect the paint. And Vonleh is not good at protecting the paint. I watched a decent bit of Indiana this year and he surrendered a ton of layups because it takes him so long to get into the air. With Vonleh defending centers and Favors being pulled from the rim, the Jazz would have a mediocre defense I think. I would prefer to keep Favors at the 5 and Vonleh at the 4.

    Favors does lack ideal size for center… But his skill level is far too poor for him to play PF next to true centers (which Vonleh is not I think).

    • Ben Dowsett says:

      Yeah, I cited these as potential red flags in my Vonleh piece last week and they’d certainly be huge factors to consider for the Jazz. That said, I think a lot of them are mental/timing related, and he at least does appear to possess SOME of the physical profile necessary. I’m still open to the possibility that Vonleh’s low IQ is as much a result of poor coaching to this point as anything else, and if this is indeed the case I think the Jazz can suss much of that out through the interview and workout process. If not, however, they’d certainly have to consider other options – I agree that pairing Favors with the type of player Vonleh is NOW will not be a viable long term option.

  5. LKA says:

    Like the trade idea except I would change one of the three, 23rd, 36th, or Neto.Instead throw in a couple of seconds in later years. Keeping the 23rd replace it with the 2017 GSW unprotected pick. This is a rich draft and you will still need a good backup point. Bucks could still get Vonleh, or Smart with the 5 pick..

    • Ben Dowsett says:

      I’d happily do that if the Jazz could get away with it, but my honest opinion is that the offer I proposed on the air would be slightly shorting the Bucks, given the historical value of trading picks that high as well as the perceived value of this year’s top 4.

  6. cw says:

    Coincidentally, I was just looking at kevin Pelton’s WARP projections compared to actuality for the past 4 or 5 years and his accuracy rate was terrible. Which is not a knock on him because it seems to me almost impossible to predict anything based on a year (or two) of college statistics. They players are new to the craft, are no where close to peak physical development, all play on different teams with different coaches and differing roles, all will go to different teams with different coaches, roles, all will face the possibility of career altering injury.

    Is there some place where the methodology is written down?

    • Ben Dowsett says:

      You just did a great job summarizing what makes the concept of draft evaluation so difficult in the first place – there’s literally no system known to man that does as reliable a job as we likely desire. You’re correct, it’s simply impossible to forecast this number of players, particularly when such a high percentage still have so much developing left to do.

      Do you have an ESPN Insider account? If so, I know Pelton’s detailed info is in there along with Ford’s and anyone else who does a model for them. I also advise checking out Layne Vashro’s model, which I mentioned in the piece, as well as Ian Levy’s Prospect Similarity Scores over at Hickory-High.com for another angle.

      • cw says:

        Lane Vashro’s model was what I was asking about. I’m curious what goes into it and also how accurate it has proven to be. It sounds like he has used historical data, so has he compared his systems historical draft projections with the players historical PER or whatever 4 or 5 years later?

        I have been curious about this lately (draft time) and am I am wondering if use comprehensive stats from games actually makes projection LESS accurate.

        • Ben Dowsett says:

          Ah, thought you meant Pelton’s stuff. Layne’s base EWP model is calculated using data stretching back to 1983 and spanning across all sorts of variables – from basic box score stats to things like college strength of schedule and margin of victory, even variables for coaching and prospects’ combine measurements (those who attend). Using over 30 years of such variables across hundreds of college seasons and their subsequent cumulative results in the NBA, Layne runs a mixed-effects linear regression to determine predictive values for each of the variables included. From there, he can then plug these predictive values in for current prospects to generate his predictive model. From this base model, he expands to others, such as the Humble model (combines EWP with popular scouting consensus, from DraftExpress and ESPN’s Chad Ford), Comparisons model (closest comparable college seasons since 1983 using only the statistical variables included), and others.

          If you want to go even further, I recommend the APBRmetrics boards, where Layne has done some major posting in the past. Here’s a thread he started in particular: http://apbr.org/metrics/viewtopic.php?f=2&t=8472

          I highly recommend his work, and have also worked with him separately on a project for Beacon Reader, an independent pay-to-read site that connects authors directly with their reader bases. If you’re so inclined, feel free to check out my project page there as well: http://www.beaconreader.com/ben-dowsett

          • cw says:

            Thanks, I’ll look at that stuff. But before I do can you tell me how accurate are the projections, if you know that? Like, are the draft projection numbers within %50 of the 4 or 5 year numbers. 40%? 30% Know what I mean?

        • Ben Dowsett says:

          Sorry had to respond to this message, won’t let me “Reply” to your most recent for some reason. I really can’t speak to those specific sets of numbers, though Layne does publicly post his data and if you’re familiar with R or another such computing software (Excel might even work for what you’re talking about), you could definitely do that sort of math yourself. One thing I can tell you (despite Layne’s absolute refusal to gloat about this or even really acknowledge it) is that when I first got in contact with him, I ran some quick correlations between his data and actual draft order since 83, with Layne’s own OBS (an overall player metric combining Win Shares and RAPM-Wins in the NBA) as the comparison point. Layne’s EWP and Humble models, over the course of the past 30 years, showed significantly higher predictive power to OBS than the actual draft order – in essence, though this isn’t really the appropriate way of stating such data, Layne’s models outperformed actual NBA GM’s over this time period.

          • cw says:

            Yeah, I was listening to a podcast where he said that. That is pretty good. Of course Cleveland and Minnisota totally skew the draft order towards dumbass, but still. He mentioned on twitter or somewhere that he had a retro model that had–I thought–the actual Win Production (or whatever it is) along with the EWP, but I can’t find that.

            Anyway, he’s got the two UCLA guys really high, contrary to a lot of other people/ What do you think about that?

        • Ben Dowsett says:

          The same link I posted yesterday should have a tab at the bottom titled “Retro”, which is his retrodictions. I believe the OBS column appears in those, showing a player’s NBA performance thus far.

          I’m super high on Anderson, and while I like Adams I’m not quite as high on him as Layne’s models due to my own personal opinion that he doesn’t really have any single elite skill.

  7. cw says:

    I get the spreadsheet things with the retro tab but no OBS column.

    ps. Thanks for your help.

    • Ben Dowsett says:

      Hm, he must have changed it. I know he was moving some things around in the last couple months, and is now travelling outside the country for most of the summer, so it’s possible he just didn’t reset everything. He also might be eschewing OBS altogether – it was only ever really used as a point of comparison, and Layne isn’t particularly worried about exactly how well his model stacks up to others or real draft order. And no worries, I can rap on this stuff all day.

Leave a Reply