Hacker News Reader

Comment by mlthoughts2018 | Hacker News Reader

mlthoughts2018

Without more details about why homotopy type theory was mixed into that paragraph about causal inference, I call bullshit. I think this is bolstered by clicking on the link to the previous post about categorical data analysis, which offers nothing and even takes a pot shot at professional statistics (saying the field is “bad at specifying the domain and range of functions” which is a bizarre and incredibly inaccurate characterization of professional statistics).

I think someone is making a mountain out of a mole hill here. I’m sure there will be a flurry of grant money though.

View on HN

I think "bad at specifying the domain and range of functions" was somebody trying to describe type safety in more pedestrian terms.

I think the actually idea is while the theory of statistics (like all math) is well typed, professional statisticians tend to use shitty tools/programming languages without good type systems.

People scoff that type errors are easy, but when most of the work is encoding the domain knowledge, and the statistics isn't to bad and just the last step, you definitely want to make sure you model prior to drawling statistical conclusions is sound and rich.

tl;dr there's good reason to invest in better tools, and the homotopy type theory that's being funded is as much about "how to apply math" as "what math to apply".

No, it is not about static typing or the general use of typed data structures for describing statistical software. Statistical software has been among some of the best software around for a long time. In fact, I frequently revisit aspects of the C++ implementation of STAN and the interface design of pymc as inspiration for effective software design in other domains.

The idea that statistics is a field with poor software tools is just plain crazy.

> Stan interfaces with the most popular data analysis languages (R, Python, shell, MATLAB, Julia, Stata) and runs on all major platforms (Linux, Mac, Windows).

None of those languages have good type systems.

And like I said, this is about modelling more generally than just statistics.

Statistics has become politicized through questionable interpretations of polling and crime data specifically with respect to various demographic markers. Increasing skepticism towards the field is a product of ideological division. No science is immune from its scientists and most humans are irrational creatures with emotive drivers.

There certainly has been widespread misuse (and probably intentional abuse) of statistics, but is it fair to blame the field for that?

It's very often an outsider misusing statistics, while professional statisticians who point out the mistakes are ignored.

I think this is mostly unavoidable whenever your field of study bumps up against politics and money. Economics has the same problem but perhaps worse in some ways, as there is less rigor.

I don't know how much of the blame you should assign to the field, in that.

How is any of that related to this article? This is about applications of formal mathematics, and the criticism (not skepticism) used by the author is in no way connected to the point you’re making.

Reply on news.ycombinator.com