Julia M. Rohrer: — Subium • Bluesky web client

Julia M. Rohrer @dingdingpeng.the100.ci · 16 days

@pwgtennant.bsky.social lol I just wanted to tag you

I love it! We need to get this message out far and wide. And figure out answers to people who defend and post hoc justify their favourite ratio.

Pretty sure mass/surface has some contexts in which its absolutely the only correct measure...

tire pressure, building statics, less convincingly human epi...

There are some genuinely valid ratios (E.g. True rates). Other ratios may appear valid, but that doesn't mean they are valid for analysis by correlation and regression. A key concern is whether the denominator varies between units of analysis.

Ratio variables can give you spurious results if there is measurement error in the denominator.

Oh no, marginality extremists! 😜 onlinelibrary.wiley.com/doi/10.1002/...

The marginality principle revisited: Should “higher‐order” terms always be accompanied by “lower‐order” terms in regression analyses?onlinelibrary.wiley.com The Biometrical Journal publishes papers on statistical methods and their applications to life sciences, encompassing medicine, environmental sciences & agriculture.

uh, I didn't even think of ratios as predictors yet!

Yeah, I guess I'm a marginality extremist. Except that 'higher' and 'lower order' terms is a syntactic distinction that doesn't really matter. Not losing degrees of freedom down the back of the couch is the main principle, parameterization be damned.

If you want to be a marginality extremist you have to not understand that it’s just a syntactic distinction, Will. I don’t make the rules. You’re a true marginality believer if you disagree with the BPM example and think that modelling beats and minutes separately is better.

I have to not understand that? That's unfortunate for a budding extremist, as I do :-)

A quick question though: Where do the "Model B as a count with log link and offset(log(M)) on the RHS" people land in the marginality bestiary. Are they modeling M or not?

Who can I thank for this ingenious depiction of my feelings towards 98% of indicators variables out there?

happy to be of service for the greater good

Oh no... this is going to wake up the propensity scores people.

What do you think about event history analysis and other analyses of life course data, where the event history data is severely processed starting from data which is often measured in many different ways (eg. retrospective calendar vs prospective panel)? I for one am perfectly happy with distances

I don't know what exactly I have in mind, so my preliminary stance is "if you are perfectly happy with distances in those scenarios, so am I" 😂

I do see that in the social sciences we do compute indices like hell (not so much today as in the 90ies) totally ignoring errors in the variables left and right

It's probably different for you guys, but whenever I see an index being introduced I think to myself "here's some folk who would like to measure a thing but would rather not derive a measurement model for it because that would lead to intrusive thoughts about measurement validity, DIF, and all that"

I think there is a more innocent reason for index building: It is easy and most did not get trained in measurement models, may not have even heard of them. It is not meant as an excuse because one should read up on stuff, but I think this is often behind indices (in political science)

I don't think it's malign at all. I think the intrusive thoughts are sort of diffuse, nagging worries that what they're doing sounds quite close to the part of that thing they read a while back, didn't quite get, and hoped wouldn't be relevant to their work.

Are there any good examples?

Psychologists like to calculate all sorts of different profile correlations to index similarity between self- and other-reports of personality. I think the process was "we want one number to summarize similarity",>

Off the top of my head: fractional citation counts in innovation research, most of the literature on political polarization, and more or less all the conflict event data literature.

BTW I just saw this published last week! bsky.app/profile/pwgt...

"Ratios in regression analyses with causal questions" - interesting new paper in @amjepi.bsky.social warning about the casual use of ratio variables in regression analyses when interested in causal effects. #EpiSky academic.oup.com/aje/advance-...

academic.oup.com

I think design-wise there is still some room for improvement. I recommend the style of Ottomar Zimmermann, staunch fighter for "pi is ratiobal!" linap.de/wp-content/u...

linap.de

Woaaah, ancient memes! That's really stylish, how do you know this?

I studied physics in Hamburg, and had a number of mathematics lectures/seminars/tutorial in the Geomatikum building, where Mr. Zimmermann held his rational-pi vigil and handed out his elaborate hand bills. I probably have a few original copies somewhere at home.

Wait, was he serious? The flyer looked like a wonderful piece of art, but taking it seriously does change the vibes

I am convinced that he was utterly and completely serious, and apparently spent a large part of his life patiently designing and distributing those flyers.

Bonus leaflet linap.de/wp-content/u...

linap.de

Oh wow I guess I can respect the dedication. And the graphical design skills.

Post