**TL; DR**: I describe some priorities of academic statistics and about how these priorities influence the actual practice of data analysis in science.

- what people think (or hope) academic statistics is about

one of my strongest beliefs is that statistics should be about service to the broader scientific enterprise, alas it is not

- what academic statistics is actually about
- how other disciplines have responded

non-probabilistic data analysis dumb shit

what about the people who do actually do care about the

*practice*of statistics? the people who are actually doing statistical servicewhy i think we’re in a stable equilibrium

the values of statistics (the academic field) and how these influence the actual practice of statistics

**TL; DR**: This post explores the role of the academic field of statistics within the academy. An insider’s summary of the academic field of statistics, with particularly attention

how do methods actually get diseminated

to be a statistician is not that bad. to be a scientist interested in methodology is to suffer.

To be a methodologist is to suffer.

as methodologists, you might imagine that statisticians care at the disciplinary level about the practice of data analysis in science at large.

my goal is not to make normative claims about how methods should be d

As an academic discipline, statistics occupies

near-universal adage that working with statisticians is a royal pain in the ass

A central tension

painful to care about the practice of data analysis

To be a methodologist is to suffer.

tension about whether or not statistics is a service discipline, by which i mean a discipline that primarily exists in support of the research pursuits of other disciplines.

social scientists largely feel abandoned by statistics as a discipline

- can see that an enormous, obvious statistic mistake has been made, and the interpretation of the results is simply incorrect

sub-fields within statistics: - theory - methodology - applied

what statisticians actually do: - publish in JASA and JRSS and AOS, the more theory the better

what statisticians sometimes think of themselves as doing: - construct and disseminating statistical methods

data analysis / “real world data” examples in statistics papers

put in other words: academic statistics is a deeply insular institution that on occasion wishes to exert cultural capital to change the practice of applied statistics. so far as i can tell, very few statisticians actually have much pull here, with the possible exception of Andrew Gelman, who roasted psych so hard he started the replication crisis (with help), a movement which has now been almost entirely co-opted by psychologists who are still making egregious statistical mistakes, just slightly different from the ones they were making before.

in a similar sense, i think that statisticians as a field know we need to embrace computation, but i think we have done something similar with respect to coding practices. it’s not like senior faculty all of a sudden went out and learned how to program and took programming classes. they just started doing it however they could make it work and teaching what they knew, which is mostly a cludge of things they taught themselves, rather than any sort of systematic or structured curriculum. i have long believed that the correct place to learn how to program is in a computer science department, and while there is genuinely a lot of quality computational work in statistics these days, i also think that a lot of it is pretty accidental, and the result of young faculty who straddled the statistics-computer science borderline during their training.

twitter drama a couple years when a prominent psychologist suggested that psychologists should start teaching statistics instead of statistics faculty. this was a person who is a respected a methodologist within psychology, but i have also seen make basic statistical mistakes! i was mad! i, also, like, get where they were coming from, and how frustrating it is to try to learn how to analyze data from statisticians.

since stat doesn’t actually really do service, what happens is that in a lot of applied departments, there are 1-2 methodologists. i have extremely mixed feelings about these methodologists. they teach the stat/data analysis classes in their departments. they often know more statistics than their colleagues, and they often fill the extremely critical translational niche, writing tutorials that turn statistics papers into content meaningful and worthwhile for their peers. and yet, a lot of the work that these folks do is excellent applied statistics. a lot of it is also mediocre, and a decent portion of it is straight up garbage. these folks are unironically in a horrible spot, doing the actual service work that their peers wish statisticians do, but they don’t get any credit from statisticians, and they also tend to have a hard time justifying their own methodological work within their actual field of study.

is there any resolution here? is there some social structure that

culturally, statisticians do not seem to be particularly charismatic relative to other fields, nor do they seem to mind all that much.

there are, of course, many

i think that many applied scientists would be shocked at how little data analysis many phd students in statistics do

the role of cognitive constraints and deep knowledge heirarchies

want to explain why statisticians are so often frustrated, why this isn’t likely to change anytime soon, and the

other fields to comment on: - psychology - econometrics - computer science - applied math & engineering

some deeply demoralizing facts about the function of statistics

not letting phd students in my department analyze my data – i would do the theory that they do if i could!