Phillies 2018 Season News/Notes/Discussion - Part Deux

[flyers19rw]

Perhaps the defense and hitting are connected? It’s rare to see a player’s entire skill set fall apart at one time. If he’s a head case, as some suggest, failure in one area could carry over to the other.

[Austinfan]

Analytical models of complex phenomenon are always wrong to some extent, and their real value is disciplining expert intuition.
Lawrence Klein would tweak the results of the Wharton Macroeconomic model, using his expert judgement.

And you want feedback to the model to refine it over time.

In a complex, dynamic system such as competitive sports, models can't be correct because there's an adjustment, if models make perfect predictions, the opponent will soon also make those predictions, and then take actions to partially invalidate those predictions (i.e. teach their hitters to adjust to how they're being defensed).

[andyb]

The preponderance of evidence is that the Phillies had below average fielders this season and were also below average in how they positioned the fielders they had. I am guessing they know this and know they need to get better. We'll see how they adjust. They'll get a bit of a pass for implementing new systems in the first year of a new regime. That pass will become a problem to address if it persists in year two.

[allentown]

In a complex, dynamic system such as competitive sports, models can't be correct because there's an adjustment, if models make perfect predictions, the opponent will soon also make those predictions, and then take actions to partially invalidate those predictions (i.e. teach their hitters to adjust to how they're being defensed).

What bothered me about the article was that there was no indication that there was any post-facto analysis of what the correct OF positioning was, e.g. what would have resulted in more outs. Fuld's judgement was viewed as correct and the accuracy of the model determined by how well it matched Fuld. Since the Phillies positioning or at least their overall OF efficiency stank, I think it fair to ask whether the model and its data might not have been correct, while Fuld, who made the final positioning decision might have been wrong. Other posters object to calling this positioning by Fuld's gut, since Fuld watched video, but gut judgements have a basis in observation and experience. Still it doesn't sound like Fuld quantified in video review enough for it actually to qualify as data.

I'll say the same about Klein. I'll call it the McNamara trick -- using a big pile of computer printout and reference to a model to give unwarranted weight to your personal opinion -- an opinion based upon his experience, but still opinion, on the same par with all the other economists making opinion/gut predictions based on their experience. When you tweak the results of a model to reflect your personal opinion, then you very simply are not working from an analytical mathematical model. That's not to say that Klein can't beat a decent mathematical model, just as chess masters have defeated many chess-playing models, but none of us would say IBM had a great, or was even using a chess-playing model if they had their own grandmaster tweaking the moves determined by the computer.

[Austinfan]

Except there's evidence in the forecasting literature that the combination of analytical modeling and expert judgment is more accurate than modeling alone.

[allentown]

That may well be if the models are sufficiently poor. Howver, that isn't the right question to ask. I'd ask these two questions: does the expert make a different prediction because of having access tot he computer model and does the combination of expert plus computer result in a better estimate that expert alone. The expert who uses the computer model will certainly say that yes of course it does, because that adds a scientific veneer to his estimate, but where the rubber meets the road is the testable question: does the expert who uses the model combined with his experience make a better prediction (consistently) than similarly qualified experts who don't use the computer model?

[Austinfan]

The reason the expert makes a better prediction with the model seems to be because the model is a "reality check" on his intuition, while his intuition often includes variables excluded from the model.
Of course, the expert has to actually be an "expert."

[BostonRoss]

As someone who develops models in a completely different field, I can say that in noon the Phillies should use the models as tools making their decisions but also use other tools, including human judgment. I don't know enough about the Phillies' analytical modes to comment on how they should be used, but it's hard for me to imagine that they are true predictive tools or have enough data and variables to be predictive. I don't know enough to tell whether whatever Fuld is adding is helping or hurting. The poor defensive results are not a good sign, but it's possible (not necessarily likely) that whatever Fuld is adding is a positive thing.

[allentown]

Also as a builder of models in a different field, I will say that in my experience analytical (or theoretical) models are constructed based upon data and then tested for correctness against other data, beyond the range of the independent variable(s) or time of the data used to create the model. A possibly valid model is able to predict the value of the dependent variable within this new, enlarged range of independent variable(s)/time. I have never seen analytical or even theoretical models tested against what some human thinks the answer ought to be. If an organization employs a human who knows with precision and high degree of certainty what the answer is, then that organization doesn't need a model, other than to have the computer crunch the numbers far faster than the human can form his opinion, but in that eventuality the human doesn't have time to 'correct' all the determinations the model is making. Now a human can certainly extend the range of independent models/time and see if the model outputs a clearly wrong/nonsense answer as some models of the future value of options did when the independent variables ran outside the 'plausible' range tested. The company I worked for lost $1 billion on that one, before going to court and successfully claiming naivete.

[BostonRoss]

In my experience, the predictive ability of a model depends on the ability to know the values of the independent variables. I don't think these can all be known in the case of baseball. A lot of this is because they are trying to predict human behavior. That's a big issue in my field, and I bet it is even more of an sports where each player's physical condition can be quite variable.

I agree that models are not tested against someone's a priori estimate of the actual answer. I don't know if that is what the Phillies are doing. I'd like to think that they use the model plus other using things they may have trouble quantifying through their experience rather than using either the model or their intuition alone.

[mdougherty12]

In a complex, dynamic system such as competitive sports, models can't be correct because there's an adjustment, if models make perfect predictions, the opponent will soon also make those predictions, and then take actions to partially invalidate those predictions (i.e. teach their hitters to adjust to how they're being defensed).

What bothered me about the article was that there was no indication that there was any post-facto analysis of what the correct OF positioning was, e.g. what would have resulted in more outs. Fuld's judgement was viewed as correct and the accuracy of the model determined by how well it matched Fuld. Since the Phillies positioning or at least their overall OF efficiency stank, I think it fair to ask whether the model and its data might not have been correct, while Fuld, who made the final positioning decision might have been wrong. Other posters object to calling this positioning by Fuld's gut, since Fuld watched video, but gut judgements have a basis in observation and experience. Still it doesn't sound like Fuld quantified in video review enough for it actually to qualify as data.

I'll say the same about Klein. I'll call it the McNamara trick -- using a big pile of computer printout and reference to a model to give unwarranted weight to your personal opinion -- an opinion based upon his experience, but still opinion, on the same par with all the other economists making opinion/gut predictions based on their experience. When you tweak the results of a model to reflect your personal opinion, then you very simply are not working from an analytical mathematical model. That's not to say that Klein can't beat a decent mathematical model, just as chess masters have defeated many chess-playing models, but none of us would say IBM had a great, or was even using a chess-playing model if they had their own grandmaster tweaking the moves determined by the computer.

From the article in question:

After the game, the computer model that told the players where to stand publishes a "postmortem" for Kapler and his coaches to review. Instead of a dead body, this postmortem inspects the defense. It is a daily audit, determining if the Phillies installed the best practices.
The report, compiled by a member of the R&D department, displays the result of each at-bat. It outlines where the fielder was positioned and compares it with where the computer suggested he play and to where the coaching staff told him to play.

There is nothing in the article that says that the analytical group is fitting the model to Fuld’s opinions. It merely says that as they have refined the model, Fuld has wanted to make smaller changes.

[allentown]

At least to me, the article strongly implies that the analytical team is adjusting the model to more closely match Fuld's opinion. He says he initially adjusted the computer answer by an average of 10 feet and now just adjusts by a few feet. No mention of ever testing the model against reality or changing the model based on adding additional variables or an increased data set. Clearly Fuld thinks his answer is the correct one. Fuld just may not understand what the analytical guys are actually doing..

[BostonRoss]

While there are a few places where the article uses language that to me would indicate possible misunderstandings about the model (including that it "told" the players where to stand, I suspect that a lot of the people involved, including the author and perhaps Fuld, are not articulating well how the model was used. If Fuld were simply every time moving the player 10 feet from where the model indicated was the best place to position him, then that could be a "gut" fudge factor that could be hard to justify. I don't think that was happening--if the average distance between the modeled and realized positions happened to be 10 feet, that in itself wouldn't be a problem, depending on what information was used to decide on the positioning.

The results indicate that the Phillies should probably change how they are positioning the defense. That could mean further revisions to the model or to the other information they're using, or perhaps bringing in additional information.

[allentown]

From the article in question:

After the game, the computer model that told the players where to stand publishes a "postmortem" for Kapler and his coaches to review. Instead of a dead body, this postmortem inspects the defense. It is a daily audit, determining if the Phillies installed the best practices.
The report, compiled by a member of the R&D department, displays the result of each at-bat. It outlines where the fielder was positioned and compares it with where the computer suggested he play and to where the coaching staff told him to play.

This part is good. It would be nice if the article mentioned which was closer to where the OF should have been, the model or Fuld's final decision. It would also be good to know which was closer to correct not just recently, but back at the time when Fuld was changing the model-prediction by an increased average number of feet.

[andyb]

Some nuggets in this Gelb article about future bullpen decisions and 1 thing he highlights is how the Phillies like to have multiple relievers with options. We will have 6 without options next year (Neshek, Hunter, Garcia, Neris, Ramos, Morgan) plus Senanthony which would give us limited flexibility with in-season bullpen management. So potentially we will trade 1 or 2 from that list.

One addition to the reliever list potentially is De Los Santos who they appear to be intent on converting into a late-inning bullpen arm where his stuff plays up a bit and his lack of a 3rd pitch (good fastball and change, not so good slider and curve) is hurting him. I did notice this a bit in repeat trips through the order for him in his earlier (admittedly small sample) starts.

Not sure who they might trade but Neris probably is building some value with his strong finish. With his heavy use of the splitter he is probably not a great long term bet to remain good. I also assume we will add a better left-handed reliever this winter, so Morgan is probably available too.

What using a second-inning pinch-hitter says about the...

[Austinfan]

Garcia and Morgan are my candidates, with Loup released.
Garcia has been too inconsistent with Ramos and Arano ready to replace him.

I think De Los Santos returns to Lehigh and starts next year, that gets him more innings to work on his stuff and keeps him stretched out for a spot start/long relief role. I think Drew Anderson is another bullpen candidate b/c I'm not sure he's strong enough to maintain his stuff for 100 pitches but should tick up in the pen.

If Irvin can't get his velo over 90+ on a regular basis, he's probably ticketed for a LOOGY role.

[allentown]

I don't think Morgan is tradable at this point. He just hasn't been good enough this season. Although his HR rate is down, his K-rate is down 1 K/9 while his walk rate is up 1 BB/9 compared to last season. A WHIP of 1.5 and 4 walks per 9 aren't going to get it done. He doesn't have a great ERA this year and it matches last season only because he's been a bit lucky, One thing that stands out about Morgan from last season to this is that he's pitched in about 55% more MLB games but thrown 20% fewer IP. Just another indication of how man more in-game warm-ups our relievers are doing this season. He's the sort of totally fungible LOOGY you can pick up from the AAAA ranks on the off-season, many with stronger L/R splits

[andyb]

The reason to trade a Neris is that his performance is probably not sustainable. If we think of trading the lesser guys like Garcia and Morgan we simply won't get much. Now it may be to our benefit to have a Yacksel Rios or De Los Santos over a guy like Garcia because of the options, but if Neris could bring back a prospect (i.e. a future controllable reliever with options) then it is better to have both Garcia and the pitchers with options.

The whole comment about De Los Santos is that the Phillies are making this switch. So he is probably not going to go down and learn to be a starter and stretch out. They think he is better suited for the bullpen and the way he has finished the season backs up that assessment.

[zambonir]

As much as relievers are fungible and youth is good, sometimes players without options are just better (and, obviously, more experienced) than players with options. But yeah, surely Morgan is gone and Garcia (who is no kid) is expendable. Considering Ramos and Neris both spent time in the minors this year it's hard to know what their value/future is. I think I'd rather have their potential, even Neris's hot finish and "closer" pedigree isn't going to get you much. Neither of them seem like sure-thing high-leverage relievers but both have stuff, and have both done it (and also failed).

If you think they are going to add a closer type or at least a lefty that's one more guy without options. Given their preference for churn it's entirely possible they won't do that (or anoint a closer again, I can't really see where it cost them any games this season. Bullpen wound up being the best piece of team).

Maybe they can find a somewhat overpaid LOOGY and trade a few guys they were gonna move on from anyway.

I would agree De Los Santos' career as a Phillies starter is over, especially if they don't trade or convert one of the current 3-5 starters (since one of them is unlikely be part of the Opening Day rotation).

[Philsmania]

It appears that Loup will be activated from the DL on Wednesday. At that point, the Phillies will have done it--had the entire available forty man roster active.

https://amp.mlb.com/294974234-phillies-set-to-activate-aaron-loup-from-dl.amp.html