Skip to content
 

Published in 2018

Enjoy. They’re listed in approximate reverse chronological order of publication date, so I guess some of the articles at the top of the list will be officially published in 2019.

9 Comments

  1. Max Griswold says:

    Young scientist here: How do you manage to be so productive?

    • Alf says:

      And to think that writing is probably only 25% of all the work he does. Amazing…

    • Phil says:

      I’m sure everyone is a bit disappointed Andrew hasn’t answered this question..but hey, I guess he’s too busy teaching and writing papers and updating textbooks and meeting with the Stan team and writing new blog entries, so maybe it’s understandable.

      As an occasional collaborator with Andrew I can say something about this, although not nearly as much as Andrew can. And of course, the fact that I worked at about one deci-Andrew in terms of productivity during my quasi-academic career shows that just because I know some of the things that make Andrew so productive doesn’t mean I could do them (or, in some cases, even wanted to do them).

      So, here are 7 ways in which Andrew is way above average:

      1. The hardest thing to emulate is: Andrew is really smart. DIfferent people are smart in different ways — Einstein vs Shakespeare and all that — but in ways that are relevant for statistical modeling and analysis he’s, I dunno, let’s say top 10,000 in the country (I just made that number up). Lots of people with his level of analytical ability go into other sciences, or engineering, or finance, or whatever, so among people who go into statistics he’s among the elite in terms of raw ability. That counts for a lot. I’d say that in general he uses this ability for breadth, not depth: he doesn’t do any single thing that I couldn’t do with my much much more modest analytical ability, but he can do what he does a lot faster so he can get a lot more done in a given amount of time: analytical work that would take me days to work out and another week to check carefully and really understand, Andrew can work out in an hour and really understand in a day. An example from early in my/our career — Andrew and I are contemporaries, indeed went to junior high and high school together — is a paper called “All maps of parameter estimates are misleading”. I had recognized a general problem with mapping parameter estimates, and discussed it with Andrew. He suggested we write a paper about it. We started writing, and I generated some examples using actual data that I was analyzing at the time. And then ‘suddenly’ Andrew wrote a section based on some mathematical/analytical generalizations of the principle we had come upon. Admittedly Andrew had, and still has, the significant advantage that he had taken graduate courses that taught how to do that sort of thing, and I had not…but still. It would have taken me a week to learn how to do it and another day to do it; it would have taken most of you statistics graduate students many hours to do it as part of a problem set, and most of you would have gotten it slightly wrong somehow; but Andrew knocked it off in a few hours and stuck it right in the paper.

      2. Andrew enjoys working, so he spends a lot of time doing it. Again, I’ll compare him to me. I spend a lot of ‘work’ time screwing around: sending emails, surfing the web, shooting the breeze, wasting time in other ways. Andrew does a lot less of this. Even when he’s “wasting time”, he’s often doing so in a way that is somewhat productive compared to what I do when I waste time. I don’t mean that he _never_ just plain wastes time — he and I both squander way too much time playing blitz chess online, for example. And he spends plenty of non-work time with his wife and kids — but for the most part, when he’s working he’s working. This is something many of us could emulate, but most of us don’t.

      3. Andrew is a good writer, and fast. In part this comes from reading a lot — if you read a lot, you learn what works and what doesn’t — and in part it comes from having a good ability to analyze one’s own state of understanding and figure out how to convey it and what are possible areas of confusion. I’m good at this too, I think. We are both admirers of the techinical writing of animal physiologist Knut Schmidt-Nielsen: we read some of his work almost thirty years ago and I know it influenced both of us to some degree. KSN’s writing was extremely clear and you never had to puzzle over what he was trying to say. Supposedly he was an excellent editor, too..and I don’t mean the kind of editor who sends papers out for review, I mean he would actually edit other scientists’ work. He would take a sentence like “Desert-dwelling mammals, though well-adapted to life in arid environments, nevertheless show signs of distress when deprived of hydration for long periods” and change it to “Even camels get thirsty.” None of us have any trouble understanding that first sentence, but the second version is a lot better. This particular example is a fake: after I read one of Knut’s books, I made up this example when describing it to Andrew. But for years, when Andrew and I found ourselves hip-deep in verbiage when trying to explain something in writing, one or the other of us would say “Even camels get thirsty” as a reminder of what we were trying to achieve. Often we found a way to do it. Like anything, this gets easier and faster with practice, and by now Andrew has had a lot of practice. Once he is clear on what he is trying to say, he can write it really quickly. (By the way, this is orthogonal to being ‘smart’ in the sense of number 1, above. The great Russian physicist Lev Landau, who was far above even Andrew’s level at analytical understanding and calculation, had a terrible time writing anything down. It has been said of the series of physics books by Landau and Lifshitz that ‘Not a word came from Landau. Not an idea came from Lifshitz.’ )

      4. Andrew often doesn’t think in terms of “I’m going to work on this problem until I understand it, and then I’ll write it up.” Instead he starts writing right from the start, or at least as soon as he knows he will have something to say. Most other researchers I’ve worked with have thought of writing as a separate task, after you have gotten your results pretty much worked out. Andrew taught me years ago that you don’t have to do it that way and that it’s better not to. It’s better to write as you go. Yes, sometimes you will have to go back and change something in a major way, but on the whole you come out ahead. For one thing, if you wait until you have worked everything out before you start to write, you will find, when you do start to write, that you have to re-derive or re-prove a bunch of stuff, or that you no longer understand the simulation code that you wrote two months ago.

      5. Andrew understands that just about any new result is publishable and should be published. It doesn’t need to be really hard or really complicated, if you learn something that other people don’t already know, as determined by asking colleagues and doing a few quick literature searches, then it’s worth writing up. Some of this will end up being little stuff that nobody will notice, and that either will never be needed or that people will have to rediscover in the future, but some of it will turn out to be useful to people. If you have something worth saying, even something small, go ahead and write it up. Short papers are OK. Papers in obscure journals are OK. Don’t waste anyone’s time writing garbage, but don’t let potentially useful results die with you, either.

      6. Seek out collaborators and ask them to do a lot of work. When Andrew and I have worked together we have generally done something not terribly far from a 50-50 split in effort…certainly with the range 30-70 to 70-30 on everything. But I know of cases in which Andrew provides some advice and does some editing of the final paper, and is a co-author for a modest time investment. One of the characteristics of statistical ‘consulting’ is that you can sometimes help people out a whole whole lot without putting a lot of time into it. Andrew is not taking advantage of these people, he’s helping them out a huge amount. I’m just saying it doesn’t always take a lot of effort. On the other hand, I know of other examples where Andrew puts in a huge amount of work and his ‘collaborators’ don’t do jack, yet somehow they still end up as co-authors (this is true of one of the papers Andrew and I wrote with collaborators, so I know this first-hand). Andrew somehow just shrugs and carries on: OK, this one didn’t work out, these people didn’t really pull their weight but they’re still on the paper..it’s a little frustrating but it doesn’t stop Andrew from quite freely seeking out collaborators.

      7. Andrew is good at finishing papers. Many of the rest of us have some work that has been sitting at the ‘90% finished’ stage for months or even years, and some of this work never ends up being published.. I’ve done this myself with a few papers: you need to generate publication-quality versions of the figures, you need to finish the references, you realize that you should have expressed things in different units, you realize that you don’t have permission to share the data on which you based the paper so you need to re-do all of the examples with different data…there are all kinds of things that don’t change the substance of the paper but that can take all the wind out of your sails. Andrew just pushes through those kinds of things, either by sucking it up and getting it done or by figuring out how to reframe things so you don’t need a problematic example after all, or you can submit to a different journal that doesn’t force you to use a specific format for your article, or whatever.

      Speaking for myself, #1 is the only major thing I just can’t do. But I’d say that alone is only responsible for maybe 20% of the mammoth gulf between Andrew’s productivity and mine.

      Andrew is at or above the 90th percentile in 2, 3, 4, 5, 6, and 7, too, whereas I’m only up there with #3 (arguably). #2 is a really big one.

      Good luck, everyone!

      • Max says:

        This is good advice, thank you Phil. You’ve inspired me to finish one of my “90% of the way there” papers and start writing sooner. Cultivating persistence and consistency seems key.

  2. Joe says:

    From the Gaydar paper:

    “Both the studies under discussion here measure the perception of sexual orientation
    in a context-free way. Decontextualization—bringing a phenomenon “into the
    lab” for careful study—is a characteristic step of scientific measurement, but it
    can cause problems in fields such as ecology and social science, where context is
    all. Reductionism—breaking a complex phenomenon into simpler parts to enable
    understanding—is a necessary part of the scientific enterprise, but bracketing the
    social for an inherently social phenomenon causes its evaporation, not its reduction.”

    It’s interesting how homosexuality advocates frequently claim that homosexuality is “not a choice”, suggesting that it’s biological in nature and can’t be changed through, for example, gay conversion therapy. (I’m inclined to agree.) But in this paper you claim “context is all”, suggesting that homosexuality has no biological component. What gives?

    “human bodies are not biometrifiable”

    Well this statement clearly proves too much. If I arrange to meet my friend in the coffee shop, how do I figure out which of many people in the coffee shop is my friend? Using biometrics such as their facial appearance, height, hair color, etc.

    “In these cases, however, too much has been stripped away from social reality
    to make any general conclusions about differences between gay and straight people.”

    With this sentence, you are kind of assuming what you are trying to show. The interesting thing to investigate here is if there are underlying biological factors that are upstream of both a person’s sexuality and the shape of their face. The question under investigation is whether there is anything more to homosexuality than “social reality”. If we completely strip away social reality and we’re still able to classify people as gay or straight, that suggests there is more to the picture than just social reality.

    Anyway, like your paper says, reductionism is an essential part of science. It feels like arguments of the sort found in this paper could be used to condemn a wide variety of useful research. No dataset is going to perfectly replicate the real world, and it’s not very interesting to point this out because it is always true. This paper disappointed me because it felt like you were making bad arguments for a position just because it was fashionable, and you are a scientist I highly respect.

    • Andrew says:

      Joe:

      Right near the part of our paper that you quoted, we wrote, “Given that no census or representative sample exists of images of gay people (or, for that matter, straight people), any statistical analysis will always have to deal with the extrapolation problem, and we recommend using some sort of multilevel model that explicitly allows for variation among and within different subgroups of each population.” The point of such modeling is to more effectively generalize from sample to population. It’s similar to what we do with polling. When we consider how a survey question can be improved or consider ways of poststratifying or consider ways in which a particular survey does no line up to its target of interest, this does no condemn survey research. Survey research can be great, but it doesn’t always line up to its promise; much depends on the details.

      Regarding the rest of your comment: I don’t quite understand what you’re saying, but I think you’re arguing with people other than us. We never suggested that “homosexuality has no biological component”; indeed I’m not even sure what such a phrase means in this context. I’m also not sure what you mean by “homosexuality advocates.” I think we can all agree that there are various differences between gay and straight people, on average, and that these differences themselves vary across populations. In addition, as often happens with people, many of the differences are intentional, with gay people working out codes (“gaydar”), straight people noticing these codes and reacting to them (not wanting to look gay), and so on.

      It’s similar to the field of economics, in which available data such as prices represent the result of complex processes involving many people. Also as we wrote in the paper, similar issues would arise when comparing any subpopulations.

      I guess, sure, when we say “context is all,” we don’t literally mean “context is all“; what we mean is that, when studying gaydar, context is the most interesting aspect, because gaydar is an intentional manipulation of appearance in order to communicate in a subtle way.

  3. Phil says:

    Hey, “R-squared for Bayesian Regression Models” should have mentioned page 929 of http://www.stat.columbia.edu/~gelman/research/published/price.pdf Nothing wrong with self-citation, especially if I’m on it too!

Leave a Reply