Thread by @EikoFried

1/7 Here's your response, @RawsthorneMat!

From where I'm standing, this insight isn't new. @bringmann_laura, @aaronjfisher Ellen Hamaker, @fionnekebos, @Oisin_Ryan_, @jonashaslbeck, Marieke Wichers, et al have been grappling with this for 1/2 decade.

https://twitter.com/aidangcw/status/1444271775165165571

2/7 In our own work, we've written about this in detail, too. CF our 2017 challenges paper with @angecramer in which we have a dedicated section on the topic; our 2017 review paper that lists this challenge; and my 2021 Psych Inquiry paper features inference gap as core topic.

3/7 I also briefly went back to the very first workshop I taught 2016, which (like all future ones) had a dedicated section on this problem. So from where I'm standing, the field isn't "finally recognizing" this issue; it's well known, & folks have struggled & grappled with it.

4/7 I can only speak for myself, but I believe the lit has been grappling with this issue because it's really important to avoid the inference gap between theory and model/data, and because it's really difficult to do so.

5/7 Working with temporal data, it quickly becomes obvious that this doesn't solve most issues of dynamic inferences. E.g. folks use temporally invariant models, but dynamic processes are like variable over time.

+ issues of measurement, modeling, time frames, power, etc.

6/7 Temporal data don't buy you causal inference: causal assumptions + data do. From this perspective, I've really enjoyed the historical causal inference literature. Many within-person processes, e.g. smoking -> lung cancer, were initially observed in cross-sectional data ofc.

7/7 Conclusion: I've become a bit of a pluralist over the last few years, inspired by e.g. P Feyerabend & S Mitchell.

Using appropriate statistical models/data to help you bring your data to bear on your theories is really important, but really hard.

eiko-fried.com/on-theory/