iiinova — where your digital self keeps evolving

Aaron Caldwell @arcstats.bsky.social · 08/02/2025

Read it and had to do a double take to understand. Looking at this, I’ve never felt so American 🤣

Dr. Benjamin Runkle @drbenrunkle.bsky.social · 25/11/2024

Reposted by Aaron Caldwell

We're hiring at the U. of Arkansas! Asst. Prof. with the aim of a research program focus on circular bioeconomy systems, teaching, service, and mentoring graduate students and research staff. Tenure will accrue in Dept. Biol. & Agric. Engr. Apply: uasys.wd5.myworkdayjobs.com/en-US/UASYS/...

01212

Aaron Caldwell @arcstats.bsky.social · 20/12/2024

⚠️ Job Alert ⚠️ We're looking to add a Assistant/Associate Professor of Biostatistics to our crew at UAMS in Northwest Arkansas! Please share widely! Awesome team, great work-life balance, and you get to help make real impact in community health. #stats uasys.wd5.myworkdayjobs.com/en-US/UAMS_A...

uasys.wd5.myworkdayjobs.com

Assistant/Associate Professor (Fayetteville)

Current University of Arkansas System employees, including student employees and graduate assistants, need to log in to Workday via MyApps.Microsoft.com, then access Find Jobs from the Workday search ...

044

Aaron Caldwell @arcstats.bsky.social · 22/11/2024

Mind sharing the DOI? I’m curious about this one.

020

Aaron Caldwell @arcstats.bsky.social · 13/11/2024

Hey folks, I intermittently get on social media. If you ever need me, send me a message through my contact form on my website aaroncaldwell.us

Mark Rubin @markrubin.bsky.social · 03/04/2024

Reposted by Aaron Caldwell

New article from me: “Inconsistent multiple testing corrections: The fallacy of using family-based error rates to make inferences about individual hypotheses” Open access: doi.org/10.1016/j.me... #Stats #Methodology

34417

Aaron Caldwell @arcstats.bsky.social · 26/01/2024

You can make these mean difference plots easily in SimplyAgree aaroncaldwell.us/SimplyAgree/

aaroncaldwell.us

SimplyAgree R package

An R package for agreement and reliability estimation

000

Reposted by Aaron Caldwell

Communications in Kinesiology (CiK) @cik.bsky.social · 17/01/2024

New article: "Effects of preferred versus nonpreferred music on bench press performance". By Jasmin Hutchinson, @jennymurphy2.bsky.social, et al. doi.org/10.51224/cik...

001

Reposted by Aaron Caldwell

Jenny Murphy @jennymurphy2.bsky.social · 18/01/2024

Check out our replication study as part of the larger replication project by the ssreplicationcentre.com Thanks so much to Jasmin and team for their brilliant work on this

ssreplicationcentre.com

Sports Science Replication Centre

0106

Reposted by Aaron Caldwell

andy™ @andylevy.net · 14/12/2023

gonna tell my grandkids this was musk’s twitter, because it was

tweet from “gentile news network”:

⚠️ HERE ARE THE RULES ⚠️

🚨Post a picture of a Jew.

🚨 Say "_____ is a Jew." (1st line)

🚨Include what they are known for/who they are (2nd line)

🚨 #NameThem. (3rd line)
Please post only clean images of the person i.e. no stars of David etc.

Let's get this trending. ❤️

63786142

Aaron Caldwell @arcstats.bsky.social · 08/12/2023

Ah, I had forgotten about this post. I like the simplicity of his MLM approach!

020

Yup, not necessarily wrong approach either. IMHO, I’d prefer to report Glass delta (pre-intervention SD) peerj.com/articles/103...

Yeah, it’s feature of the design (I’m guessing this is SMD of the change scores). Lack of concurrent control provides a conveniently large effect size

020

Yeah, that’s pretty much how I’d do it (at a glance)

So the mixed model you had but with t1 on the “right” and t2 and t3 on the “left”

130

Then I’d only include t1 as a covariate

When is the experimental treatment exposure? Before or after t1?

Communications in Kinesiology (CiK) @cik.bsky.social · 06/12/2023

Reposted by Aaron Caldwell

New article: "Model specification in mixed-effects models: A focus on random effects". By Keith Lohse et al. doi.org/10.51224/cik...

064

Aaron Caldwell @arcstats.bsky.social · 06/12/2023

No, that sounds entirely unreasonable. Are these limits for limits of agreement or an equivalence test.

100

If you don't think it would be any better your conclusion/claim would be better stated as "Effect sizes are function of the experiment design and analysis approach. So describing an effect outside of the context of the experiment just is not meaningful."

Claim 1: "Effect size is a function of the experiment design and analysis approach as much as, if not more than, the underlying effect. So describing an effect’s Cohen’s d outside of the context of the experiment just isn’t meaningful." Is misleading the reader then, no?

100

Not sure I agree, all depends on what information/inference you are trying to make with the comparison. Let me ask you this though, what makes you think an *unstandardized* mean difference would be any better for comparing between experiments?

100

And heterogeneity does not mean the effect sizes can’t be compared. Random effects models in ma exist for a reason

210

To put it another way: if I had two different studies that used different measures with wildly different reliabilities (within subject variation) I wouldn’t be surprised by differences in the SMD.

There are many reasons to distrust an SMD (Senn’s paper on the matter is one of my favorites). But, you aren’t even just changing the design *your changing the data generating process* which modifies the SMD. doi.org/10.1198/sbr....

220

Okay, but then your whole objective is wrong, the underlying variance structure changing would change the SMD. Regardless, the first claim is misleading.

200

Use a multilevel model and use the variance components. I'd probably choose the sum of variance components but there could be argument for only using the residual variance.

100

You used one specific SMD calc. One that would be inappropriate for the given data which I have mentioned in your previous posts on the topic

100

The first claim is not supported by your simulation. If you use an inappropriate standardization then *of course* it will be an ineffective measure outside the context of an individual experiment. I dislike SMDs too but this is really unfair and unsupported.

100

100%. This is social media and more of a poking fun post. But eponyms are mostly annoying to me in practice.

030

SMD directly tells the reader what the value represents: a *standardized* mean difference. The methods section should then show how the standardization is accomplished.

130

"Just give the formula" - agreed. But, the use of "Cohen's d" or "Hedges's g" since it 1) isn't abundantly clear what those values aren't meant to represent and 2) the use of possessive nouns would imply a *specific* value/calculation (which is almost never the case).

110

Can we all just agree to stop calling standardized mean differences "Cohen's d" in our scientific manuscripts? It's misleading, and IMHO leads to lazy reporting. If you want shorthand I'd recommend SMD (i.e., what Cochrane uses), and *always* report how the SMD was calculated #stats

1213

Wait, wouldn’t d be equal to the t-stat if they mixed up the SE and SD?

000

I think it is also possible that they used the residual variance from the ANCOVA. That could also inflate the SMD estimate.

I wouldn’t say it is “ambiguous” but there is a great deal of flexibility and options. But that is the case for reporting an effect size for pretty much any experimental design.

110

Yeah, I’d just reported the mean difference in most scenarios from the marginal means output (I’d also model it with a random intercept and slope by subject). You could the report the SMD as mean difference over the sqrt of the sum of the variance components.

100

Oh like a cross over replicate design?

100