In which I step out of the Ivory Tower

[philena]

I am an intellectual snob in many ways. I've noticed this about myself: when I read the newspaper in which the results of data-driven studies are reported, my first thought is not, "Oh, how interesting to observe this connection between X and Y," but rather, "Did they control for Z? Are they aware that there could be a selection bias? They know it'

Comments

[verrucaria]

I hope that the IRB rep you talked to was just subbing in for 15 mins. for someone else??? I mean I heard that IRBs are supposed to be composed of a wide variety of professions, but I wouldn't expect one of their least academic members to be the one person with providing advice to researchers requesting specific clarifications...

I wouldn't dare to do a statistical analysis without consulting a friendly and extremely patient biostatistician about what tests are most appropriate. Ideally, the statistician or someone else would also read through my experimental design to make sure that I didn't forget to gather data on some glaringly relevant factor. The scary thing is that sometimes your peer reviewer may not be fully capable of determining whether your choice of statistical analyses actually makes sense (not because the peer reviewer is a dumbass but because things can get really tricky really quick).

P.S. I think what impressed me the most about The Avengers was the actors' uncanny ability to deliver a one-liner after one-liner

[philena]

When it comes to actually doing statistical analysis of my own work (rather than criticizing someone else's), I'm also extremely cautious. I run everything by my advisor (who literally taught me everything I know about stats), and I still cringe to think about my qualifying paper, which I wrote when she blinked, and massively violated the independence assumption in my analysis section. She also has stories about writing papers in which the reviewers says something like, "Wait, why are you doing a Wilcoxon test? Why not just do a t-test?" because they aren't paying attention to the normality assumption. I think this might be less of a problem in biology than in linguistics, where data analysis and computational methods are new and shiny.

[verrucaria]

Using quantitative methods in linguistics must be really hard because how does one even organize the data in order analyze them (look at the nouns? look at the verbs? look at components of words? look some aspect of the syntax?*).

It seems that a linguist who uses stats has to have a pretty good grasp of them in order to be able to take a single step toward getting anything accomplished.

And in linguistics, "horizontal gene transfer" ("borrowing" of words, etc.) happens all the time. That happens in biology, esp. if you work with bacteria, but it's not as common in most eukaryotes. How does one control for borrowing?

I basically no longer remember which common statistical test does what. I just remember names for a few parametric and non-parametric tests. :)

*if so, what aspect; and how do you segregate/partition/order/whatever the presumably different syntaxes into groups or points on a continuum or whatever?

[philena]

In phylogenetic analyses of languages, the borrowing problem is a real hassle. Fortunately, though, I'm not a typologist or a historical linguist. I'm a psycholinguist, so I study language use. Here are some examples of how psycholinguists use statistical analysis

[dabroots]

While visiting with my big sister and other family a couple of weeks ago, my sister corrected me about something and after doing so, said that it felt good to do it because I've been correcting her "all the time."

I found you here after clicking to see who else at LJ likes Richard Russo. I'd like to add you partly for that reason, partly because I like the posts I've read so far, and of course there's also the tuxedo cat thing.