Connor McCabe

I have been minimally involved on this platform for some time now, but you can now find me on Bluesky (bsky.app/profile/connor…). Colleagues, please feel free to follow me there as this will be my last post here. Take care, everyone.

The longitudinal association of cumulative depression with cannabis use disorder among young adults (just accepted in @JSADjournal) @DrOesterle @UWPsychiatry @cshrb_uw @ASU_SIRC @ASUSocialWork @uwsocialwork jsad.com/doi/abs/10.152…

Had a great time providing a workshop this morning on GLMs at #aps2024sf! I have posted all my materials at this OSF link: osf.io/ymbn6/. See the Rmd/html files for some helpful code in interpreting GLMs using marginal effects.

I'm very excited to announce our workshops at #aps2024sf! Including amazing instructors like @letsnotbehayest @JkayFlake @emilynordmann @andrewang91 @connorjmccabe @spicy_thai @jacasiegel @jgeller_phd @AndrewDevendorf psychologicalscience.org/conventions/20…

Longitudinal examination of alcohol use motives, item-level protective behavioral strategies, and alcohol-related consequences doi.org/10.1111/acer.1…

👏 plot 👏 the 👏 conditional 👏expectation. 👏 The 👏 prediction 👏 is 👏the 👏 estimate

@tcarpenter216 I’ve encountered this across 2 different machines in R. I believe the issue was that package versions were different (updated on one environment but not the other). Long shot, but maybe this?

Two code environments They are identical Code is identical Data are identical Model consistently performs 10% better in one of them Something tells me that I'm wrong somewhere....

@KMKing_Psych @IsabellaGhement @SolomonKurz @camjpatrick @dingding_peng @rubenarslan Our paper in MBR speaks to this as well. We (and I think William Berry's work) called it "model-inherent nonlinearity" that induces interaction independent of the product term on nonlinear scales: …line-com.offcampus.lib.washington.edu/doi/epdf/10.10…

Excited to announce that @EmorieBeck and I are co-editing a special issue of JRP on Modeling & Interpretation of Personality/ID constructs. We're looking for papers heavy on theory-model intersection. Deadline to submit *proposals* is Nov 15 (but tell us if you need more time)!

Beyond thrilled to share that my Pathway to Independence Award was funded this week by NIAAA! Will be working with the All of Us Research Program dataset to address intersectional differences and mechanisms of AUD using novel machine learning methodologies. Time to get to work!

@seriousstats @jonathanleehelm Awesome, thanks Jon and Thom for the refs!

Need help from the quant Twitterverse for a reference: who coined the phrase “polynomials are variables interacting with themselves”? Love the phrase and use it often, but I can’t remember where I first heard it. Thanks!

Excited to share my first #rstats package, InteractionPoweR! github.com/dbaranger/Inte… InteractionPoweR uses simulations to estimate power for interactions between continuous and/or binary variables in cross-sectional data sets.

@richarddmorey @KMKing_Psych I think we used CSHS, which may already be available to public via Max’s work. I can look into this in the next couple days

**NEW PREPRINT ALERT** "Best Practices for Addressing Missing Data through Multiple Imputation" Preprint includes a worked example using the ECLS-K: 2011 and three software packages (Stata, R mice, and Blimp) Comments and suggestions welcome! psyarxiv.com/uaezh

Aguinis (2005) seems to be the go-to citation when talking about small effect sizes in interaction/moderation research (psycnet.apa.org/doi/10.1037/00…). Does anyone know of more recent work along the same lines?

@KMKing_Psych True, and it was very helpful. It was probably the start of my discrete data interests

Note this is under review at the Journal of Open Source Statistical Software, so please consider this in-progress. Posting to the Twitterverse for any and all feedback so it can best accommodate your needs!

The big addition is plotting functionality for marginal effects. These plots can now be generated using the margplot function, which produces plots such as this one. Confidence regions on these are generated using parametric boostrapping a la King, Tomz, Wittenberg (2000)

Just released a package version of our modglm function corresponding with our MBR paper on interaction effects in GLMs of probabilities and counts. More info at my GitHub and in thread: github.com/connorjmccabe/…