Reading this mini-obituary of the great third baseman reminds me of a passage from Earl Weaver’s autobiography. It has a great passage . . . I don’t remember the exact words, but it’s relating some important game where Brooks Robinson was up at bat, and Earl told him, “Swing away, Brooksie.” Weaver always maintained that it was the players who won the game, not the manager, and that the manager’s job was to get the best out of each of his players. This is a good general principle, I think, and similar to Deming’s views on management.
Also it was good that when Weaver came back in 1985, his team did really badly. I mean, no, it wasn’t good, it was bad, it made me sad. But from a statistical perspective it was good to be reminded that even though Weaver was famous for being a genius, that wasn’t enough. Just cos you’re a genius, it doesn’t mean your teams will automatically win.
Speaking of team management, did anyone watch the Vuelta a Espana this year? Jumbo-Visma finished 1, 2, 3 on the GC podium, taking all of the Grand Tour wins this year (Giro, Tour, Vuelta), and it was American Sepp Kuss, normally in the domestique role, who took the Vuelta win. I imagine it was some pretty delicate managing, given that Roglic and Vingegaard (winners of the Giro and Tour respectively) had to settle for 3rd and 2nd to their main helper.
Sometimes it’s not just getting the best out of each player but convincing each player to do the best job for the team. Depending on the sport, those can be two very different things.
Jd:
Yes, Weaver also discusses contributing to the team. He relates a conversation he had with a player who didn’t want to platoon (bat only against lefties or against righties): the player said then he’d never get the salary he’d get if he were to start every game. Weaver replied that the player might be right, but by platooning he would help the team win, and his (Weaver’s) job was to help the team win, not to help the player maximize his salary. In his book, Weaver writes (or, I guess I should say, says to his ghostwriter) that he always tried to be honest with the players, and I guess that part of this was to be honest with himself, to admit that his role was not the same as the player’s role. One funny thing is that he mentions all these examples where he pulled the starter, even when the starter was in the lead, if he thought the inning was getting out of control—but I didn’t see him mention those games where Jim Palmer wanted to come out and Weaver kept him in because even if there was a chance that Palmer gave up some runs, he was still better than the middle relief that was the alternative. (That said, in the first Orioles game I ever went to. Palmer gave up a few runs in the first inning, was quickly pulled, and the Orioles ended up losing 9-1. A year later, we saw him win a 1-0 one-hitter.)
Anyway, to return to the main topic, Weaver has a bunch of other good management advice in his book. Perhaps had it been published 10 or 20 or 30 years later, it would’ve been called, “The Management Advice of Earl Weaver” and would’ve had about 1/5 the content, repeated over and over and over again.
To another point: Weaver was born in 1930 and the strategy book was published in 2002 which would make him 72 years old at the time. Clearly too old to be advising about strategy, given that he was more likely to be suffering cognitive decline by then. (actually the 2nd edition was 2002 and I can’t find any reference to the publication date for the first edition, probably due to my own cognitive decline)
Dale:
Earl Weaver’s autobiography (coauthored with Berry Stainback) was published in 1982.
On the general topic you’re discussing, we had a post a couple years ago, What’s the best novel ever written by an 85-year-old?. I do think there’s more to be said on this topic.
> his (Weaver’s) job was to help the team win, not to help the player maximize his salary
I bet this is a difficult sell in the world of pro sports, sometimes. Especially when the team isn’t very good to begin with (but the player is). I’m sure it is a lot easier if the game at hand is a playoff game.
I watched it!* It was riveting. I agree about the delicate managing, but I think this was done on the fly — the biggest drama was at Stage 17, when Roglic attacked his own teammate and race leader (Kuss) to win the stage. Clearly no one had persuaded him before that this was not something he should do! Interestingly, I’m sure that a factor that went into the change of management was public perception, most people not liking the notion of attacking your own teammate while he’s leading a Grand Tour, especially when that teammate is considered one of the nicest people in the sport. I wonder if public perception would be as much of a concern in years past?
* Watched meaning (i) NBC’s daily extended highlights on YouTube, which are pretty good, and on most days (ii) Chris Horner’s recaps, which are amazingly entertaining and informative despite not having video (e.g. https://www.youtube.com/watch?v=wPWbqmIxmWE). His explanations of tactics (and failure to apply tactics) are great.
The Vuelta was a bit of a disappointment, just because once Evenepoel fell out of contention there was very limited opportunity for drama in the general classification…or at least limited opportunity for the kind of drama I’m interested in. Yes, there was the chance of a soap opera if the teammates started fighting each other (it’s happened before of course: LeMond and Hinault; Froome and Wiggins), but that always seemed unlikely and sure enough it didn’t really happen.
I wish Evenepoel had tried to go head-to-head against the Jumbo Visma juggernaut even after he was out of contention, but instead he went for stage wins on breakaway stages. I was really hoping he would stay with the Jumbo guys on Angliru and then attack if he had the legs.
(Here’s a bit of an apolobrag: in May I biked Angliru…not the entire Vuelta stage, that would have been impossible in my current condition, but just that one climb. It was, by far, the hardest single climb I have ever done. It was nowhere near my hardest day on the bike, just because I only did the one climb — only about 45 minutes of it was really hard at the pace I went — but it’s the only road climb I’ve ever done where I really questioned my ability to finish at all, never mind the time. On the hardest section I averaged 2.5 miles per hour, paperboying all the way up. As I started that section I could see a guy walking his bike up the hill ahead of me, and it took me forever to catch up to him and pass him. And we could even have a sort of breathless conversation, because I was not maxed out…which, oddly, makes it harder: normally if I am on a super super steep section of a climb, it’s short enough that I can power through it, just let my heart rate do what it may, but on Angliru there are long sections over 20% grade, you just have to stay below your max heart rate or you will burn out. Most of the ride up, I was right on the edge of going too hard to be able to continue, while also going barely fast enough to keep the bike upright. In the unlikely event anyone wants to read about my experience, you can check it out here: https://julietandphil.blogspot.com/2023/06/biking-near-bilbao-near-arriondas-and.html )
Anyway, to get to the topic at hand — team management — I do wonder what it was like to be on Jumbo Visma’s bus. Were Vingegaard and Roglic lobbying to be let off the leash, at least on the final climb of each day? It wouldn’t be hard to make the argument: doing so would not jeopardize the _team’s_ chance of winning the race, they had that well in hand, so what’s the argument against letting the best rider on the team win the race, whoever that might be? But Vino and Roglic never tried to attack, at least not early enough to really challenge Kuss’s lead. One thought I had is that Kuss was their most valuable lieutenant, so they might not want to upset him by taking the jersey from him because he might not work as hard for them in the future. But that idea is somewhat undermined by the fact that Roglic just confirmed that he is leaving JV at the end of the season…perhaps a decision that hadn’t been made at the time of the Vuelta but he must have at least been considering it.
Maybe someday, after these guys retire, they will each give a candid report of what they were thinking during the race. Or maybe not: supposedly Hinault still denies that he was doing anything other than support LeMond in their final Tour together.
Raghu – I agree that it was interesting, but I do share Phil’s disappointment that Evenepoel capitulated. I suspect much of that was mental. I cannot remember a time when a rider cracked so badly, yet had such good legs afterward. I’ve never experienced that myself either, and although I was never at that level, I was at a high enough level to have some of the same sensations and experiences. In addition, Thomas was nowhere near his Giro form, and Ayuso and Mas were found wanting. So it just turned into a bit of a Jumb-Visma beat down with most of the drama among the team.
I think public perception did have a play in it, because cycling is purely a sponsor based sport. Appearances are very important. In addition, since they were looking for a sponsor, it’s not a bad idea to have a popular North American win. It could attract money from outside of Europe.
Phil – Evenepoel lost too much time the day he cracked to have any other option but stages. I really think much of this was mental, but who knows. He’s a bit of a petulant rider.
Wow! Super cool! Quite an experience to climb Angliru yourself. Big congrats on making it to the top! What gears did you ride?
I seriously doubt that either Roglic or Vingegaard were happy. My personal opinion is that I doubt either of them had a problem with Sepp winning. The problem was that I would bet that Roglic wanted to race it out with Vingegaard. Roglic always was the main GC man at Jumbo, for the last 9 years. He has won every stage race and Grand Tour except the Tour de France. He was undefeated in stage races this year until he had to give up the Vuelta. Vingegaard, over the last 2 years, has arguably become the best climber in the world, and the only rider able to challenge and beat Pogacar, winning 2 Tours in the process (Tour’s that Roglic either crashed out of or did not start). I suspect that Roglic wants to go back to the Tour and try to win it. But he can’t do that with Vingegaard there, unless he could have beaten him soundly at the Vuelta and earned a spot back as the main man at the Tour de France. So I would bet there is more to the story than Sepp Kuss. And it is no surprise that Roglic is leaving.
Phil and jd: Fascinating (including the Angliru climb recap)! To placate Andrew and return this thread to Earl Weaver: I know nothing about him, except that a handful of times in ~9th grade I played the Earl Weaver Baseball video game, despite having zero interest in baseball. It involved a lot of picking players / managing. Perhaps someone has resurrected it, and Andrew can play. And perhaps someone has made a similar game (which I have no interest in playing) for professional bicycling.
Raghu:
I think you would enjoy Buy That Guy, the game of political corruption.
This seems like a thread where I can go off topic without stepping on any toes.
I thought folks here might be interested in Francesca Gino’s shiny new website defending herself against claims of fraud:
https://www.francesca-v-harvard.org/
Not sure what she thinks she is doing by insisting all the claims against her are “cherry-picked,” as if her accusers should have spent just as much time with the data that is not suspicious.
Her website went live one day before the “Many Coauthors Project” website, which is supposed to launch tomorrow:
https://manycoauthors.org/
Grab some popcorn.
This is very interesting and I am anxious to see what others find. Gino finally is providing detail in her defense. It is partially convincing from what I’ve seen thus far. But I’m only on the first of her claims – what she refers to as cherry-picking the observations that datacolada analyzed. I’ve looked at the original data (the 101 people in the first tax study), along with datacolada’s analysis and Gino’s defense. It appears that her references to row numbers is slightly different (by 1) than what I see in the data I downloaded from OSF. But to cite one point of contention: datacolada says that 6 participant ids are out of order and Gino says there are 8. If I graph the participant ids as a function of row number, there are clearly 6 out of order. Now, if you manually scroll through the data set, the “order” of participant ids is somewhat ambiguous – graphing it is not. It is entirely possible that datacolada looked at it graphically and Gino looked at it manually, thereby reaching different measures for what might seem like an objective fact. I might be inclined to give both the benefit of the doubt here.
The other things that struck me about the ID and row order is that there are three clear series of IDs and they relate to the 3 conditions (sign at top, sign at bottom, control) used in the experiment. I am supposing that the manual entry of the data was to enter the observations sorted by ID, but separately for each condition. I can see no other way the ordering would end up the way it looks. Gino has this to say:
“One of the biggest assumptions Data Colada made in their blog post was that the data were electronically sorted twice: First by treatment, and then by participant ID. This assumption is almost certainly incorrect. It is far more likely that the data were never electronically sorted. This is a critical point.
Instead, here is what almost certainly happened: The RAs conducting the study simply stacked the paper copies by condition, and then manually entered the data in the order in which the papers were stacked. The sequence of paper entries tended to be in ascending sequence; this makes sense, since it was probably the same sequence in which the participants performed the tasks. On the other hand, it is also plausible that some of the papers got mis-ordered. After all, no one had any particular reason to care about sort order; the findings and statistical analysis depended on the data points themselves, not their sequence in a file.
Of course, there will still be skeptics who ask – “if that’s the case, then why were most of the papers in sequence, and just a few out-of-sequence?” The honest answer is, I do not know. I only know that working with paper creates the potential for mis-ordering to occur.”
She also offers a picture of the out of order points (8 in her view) claiming that there is no pattern to the claimed expenses associated with these participants: she says that some “are high, some are low, and some are in the middle.” My graph of the 6 points (and even her graph) is less than convincing that there is no pattern. While not all the points are at the extremes, enough are to affect the analysis. If someone were intentionally changing their data, presumably they would want to do it less than perfectly so as to seem possibly random.
I don’t have a view yet concerning who is right (and potentially both are right about some of the issues). But I’m glad to see Gino provide a detailed analysis and I was glad that datacolada provided a detailed analysis. Given that this is a case in court, I expect many more analyses to be conducted. But it is starting to look like analyses of the 2020 election results. As objective as some of these questions may seem, there are ambiguities enough to raise questions about any purported conclusion. I’ll await much more evidence before I reach any conclusions myself.
Matt, Dale:
I took a look at Gino’s document and had some comments which I hope will be helpful to Gino, to the Data Colada team, and to others who are interested in this issue. See below.
—
1. Gino writes, “When it comes to duplicate IDs, Data Colada included rows 52 and 53 (ID 49), but they omitted rows 5 and 75, which also have duplicate IDs (ID 13).” That appears to be correct, but Data Colada referred not to duplicate IDs but to duplicated _observations_. The two rows with ID 13 have different data so I would not call them duplicate observations. I noticed this myself; later I also noticed this point made by Data Colada in footnote 2 of their later post (https://datacolada.org/114).
2. Gino writes, “They included two observations that are in fact not out-of-sequence—namely, [Row 69, ID 101] and [Row 70, ID 7].” She also writes, “They did not even mention row 73 (ID 5), which is also out-of-sequence.”
Regarding Row 69; Data Colada mention this in their discussion of the Excel file, concluding that, “Row 69 used to be right before Row 34, and so it was in Row 33. If we look at what is row 33 now, it is the only observation in the control condition that is out of sequence. This suggests that the observations may have been swapped.” So, although it remains in order, they do give a reason for being suspicious of it.
Regarding rows 70 and 73: Looking at the file, it seems that these two observations are out of sequence relative to each other, but if they are switched, they would both be in sequence. Thus, it could be that row 70, ID 7 is out of sequence and row 75, ID 5 is in sequence, or vice-versa. In their discussion of the Excel file, the Data Colada team write, “row 70, now in Condition 2, used to be surrounded by rows that were in Condition 0.” If that is the case, then row 73 is not out of order.
3. Gino writes, “They mentioned Row 33 (ID 64) as being out-of-sequence, but they omitted it from their analysis.” I’m not sure about that. In their dataset, they have a column, “flag,” which lists their flagged observations. There are 9 observations, not 8, with flag==1, and one of these is row 33 (ID 64).
4. Gino criticizes Data Colada for only presenting the data for signing at the beginning and signing at the end, but not presenting the data for the control condition. Given that the title of the original article was, “Signing at the beginning makes ethics salient and decreases dishonest self-reports in comparison to signing at the end,” it makes sense to me that they focused on that comparison. Just based on the information I’ve seen, it doesn’t seem like “cherry picking” for them to focus on the headline comparison.
5. In trying to understand the Data Colada report and appendix 2 in Gino’s note, I went back to the summary of study 1 in the original article, which is, “We measured the extent to which participants overstated their income from the math puzzles task and the amount of deductions they claimed.” The outcomes reported in the article were “The percentage of participants who cheated by overclaiming income for math puzzles they purportedly solved,” “the average magnitude of cheating by condition,” and “claims of travel expenses.” So I think it makes sense to look at all three of these outcomes.
Here is a quick summary. For each, I computed the avg under condition 1, the avg under condition 2, the difference, the s.e. of the difference, and the z-score (i.e., diff divided by s.e.). You can compute the p-value if you want, but for reasons I’ve discussed many times elsewhere (for example here: https://www.stat.columbia.edu/~gelman/research/published/asa_pvalues.pdf) I really don’t like p-values and threshold-based decision making, so I’m not going to do that here.
For each of the outcomes ((1) proportion who cheated, (2) average amount of cheating, and (3) claims of travel expenses), I’ll give the results under 3 analyses:
(a) All the data in the file
(b) Removing Data Colada’s flagged observations entirely
(c) Trying to reconstruct the original data following Data Colada’s inferences from the Excel file, also counting duplicate observations once rather than entirely removing them.
(a) All the data in the file:
outcome 1: 0.371 0.788 0.416 0.110 3.788
outcome 2: 0.771 3.939 3.168 0.749 4.231
outcome 3: 5.27 9.62 4.35 1.31 3.31
(b) Removing Data Colada’s flagged observations entirely:
outcome 1: 0.433 0.767 0.333 0.121 2.755
outcome 2: 0.900 3.833 2.933 0.816 3.596
outcome 3: 5.96 8.39 2.42 1.21 2.00
(c) Trying to reconstruct the original data following Data Colada’s inferences:
outcome 1: 0.438 0.750 0.312 0.118 2.642
outcome 2: 0.969 3.812 2.844 0.787 3.615
outcome 3: 5.68 8.52 2.84 1.23 2.30
Each of these is “avg under condition 1,” “avg under condition 2,” “difference of the averages,” “standard error of the difference,” “z-score = diff/s.e.”
A quick look at the z-scores reveals that moving or removing the observations has a big effect. Again, I don’t think it makes much sense to look at p-values. That said, it is the claims of travel expenses (outcome 3) where results are on the border of statistical significance (z-scores of around 2), so I can see why they are all are focusing on that outcome. Also I did these calculations quickly so I could have made mistakes, and, beyond this, these are all just initial efforts, and it doesn’t seem like any of these three options represents whatever data were actually in the study.
6. Once you open the Pandora’s box that the data have not been recorded correctly, anything is possible. For example, a later Data Colada post (https://datacolada.org/114) writes that the forensic firm hired by Harvard “found evidence suggesting that many values of the dependent variables had been altered.” If values had been altered, whether on purpose or by accident, that’s not covered in my summaries above, which merely remove or change the conditions of the observations without changing the recorded measurements.
7. Setting aside all these data issues, even if there had been no data problems at all, I would’ve said that I think the published estimates in that article are likely to be way too high and quite possibly in the wrong direction, for reasons discussed in general terms in these articles: https://www.stat.columbia.edu/~gelman/research/published/retropower_final.pdf and https://www.stat.columbia.edu/~gelman/research/published/failure_of.pdf
Thanks for posting your work on this! I was hoping for a third-party review of the statistical claims. And sorry about the thread hijack.
I can’t tie Gino to Earl Weaver, but it does seem that she is “racehorsing” her defense in the hopes of creating the fog of war. Claiming that one (anonymous) scammer screwed everything up in one of the papers doesn’t really tilt the battle in her favor IMO. That’s arguing that the data integrity is so low that it is impossible to separate one problem from another. It also tacitly endorses the retraction.
Regarding number 3: it appears that observation is in the control condition so it was excluded from their analysis. https://x.com/JohnHBillings/status/1708187971889897531
The Gino story is in both today’s New York Times and the current New Yorker.
Here are links (both paywalled)
https://www.nytimes.com/2023/09/30/business/the-harvard-professor-and-the-bloggers.html
https://www.newyorker.com/magazine/2023/10/09/they-studied-dishonesty-was-their-work-a-lie
Hey, wait! This was supposed to be a thread about Earl Weaver.