Felix Thoemmes
@felixthoemmes.bsky.social
Played around with Causion by @isager.bsky.social
It's an impressive teaching tool for causal inference... and it also really pretty. Amazing color scheme and slick interface.
Shown below a demonstration of an extended front-door DAG in Causion.
It's an impressive teaching tool for causal inference... and it also really pretty. Amazing color scheme and slick interface.
Shown below a demonstration of an extended front-door DAG in Causion.
January 29, 2026 at 2:28 AM
Played around with Causion by @isager.bsky.social
It's an impressive teaching tool for causal inference... and it also really pretty. Amazing color scheme and slick interface.
Shown below a demonstration of an extended front-door DAG in Causion.
It's an impressive teaching tool for causal inference... and it also really pretty. Amazing color scheme and slick interface.
Shown below a demonstration of an extended front-door DAG in Causion.
Due to popular request ;)
December 2, 2025 at 3:41 PM
Due to popular request ;)
Rearranging variables can make this clearer. In plot below, it's relatively easy to see that total effect needs no adjustment, and that direct effect from A to Y needs to block the indirect path that traverses D and C. D is a collider, and C the child of the collider, hence adjustment on B is needed
December 2, 2025 at 1:59 PM
Rearranging variables can make this clearer. In plot below, it's relatively easy to see that total effect needs no adjustment, and that direct effect from A to Y needs to block the indirect path that traverses D and C. D is a collider, and C the child of the collider, hence adjustment on B is needed
2) One can also derive the shape of all possible path coefficients for a given model using Wright's tracing rules. Here is a picture of those for the example of the mediation model (X->M->Y).
November 10, 2024 at 1:58 AM
2) One can also derive the shape of all possible path coefficients for a given model using Wright's tracing rules. Here is a picture of those for the example of the mediation model (X->M->Y).
I also enjoy elliptopes - a while ago, animated this one (not quite as pretty as the ones in your blog post).
November 10, 2024 at 1:58 AM
I also enjoy elliptopes - a while ago, animated this one (not quite as pretty as the ones in your blog post).
I see your Ann Arbor fall walk and raise you an Ithaca fall bike commute to campus ;)
October 23, 2024 at 4:49 PM
I see your Ann Arbor fall walk and raise you an Ithaca fall bike commute to campus ;)
I'd say bite the bullet and explain potential outcomes and why in RCTs baseline bias and effect bias have expectation 0. Left is from Steyer's Causal Effects Xplorer (www.causal-effects.de/tools.php). Right from a slide that I use, adapted from Cunninghams mixtape (mixtape.scunning.com)
June 3, 2024 at 3:11 PM
I'd say bite the bullet and explain potential outcomes and why in RCTs baseline bias and effect bias have expectation 0. Left is from Steyer's Causal Effects Xplorer (www.causal-effects.de/tools.php). Right from a slide that I use, adapted from Cunninghams mixtape (mixtape.scunning.com)
One key take-away is the importance of the "full mediation" assumption of the front-door model. As soon as it is satisfied the front-door model is often superior to a regression estimator. If violated, unadjusted analysis is often preferable.
October 13, 2023 at 2:04 AM
One key take-away is the importance of the "full mediation" assumption of the front-door model. As soon as it is satisfied the front-door model is often superior to a regression estimator. If violated, unadjusted analysis is often preferable.
New co-authored paper with Yongnam Kim.
Bias and Sensitivity Analyses for Linear Front-Door Models, where we derive bias formulas for the linear case of the front-door model and compare the front-door estimator with an unadjusted regression estimator.
meth.psychopen.eu/index.php/me...
Bias and Sensitivity Analyses for Linear Front-Door Models, where we derive bias formulas for the linear case of the front-door model and compare the front-door estimator with an unadjusted regression estimator.
meth.psychopen.eu/index.php/me...
October 13, 2023 at 2:02 AM
New co-authored paper with Yongnam Kim.
Bias and Sensitivity Analyses for Linear Front-Door Models, where we derive bias formulas for the linear case of the front-door model and compare the front-door estimator with an unadjusted regression estimator.
meth.psychopen.eu/index.php/me...
Bias and Sensitivity Analyses for Linear Front-Door Models, where we derive bias formulas for the linear case of the front-door model and compare the front-door estimator with an unadjusted regression estimator.
meth.psychopen.eu/index.php/me...
About to send a letter back home, and always amazed how beautiful the #USPS global forever stamps are.
February 28, 2021 at 2:19 PM
About to send a letter back home, and always amazed how beautiful the #USPS global forever stamps are.
Yes, I did! Thanks @tcplny ❤️
December 19, 2020 at 5:19 PM
Yes, I did! Thanks @tcplny ❤️
Open science advocates are occasionally accused of having a somewhat religious zeal... nevertheless I love this cross-stitch gift that I got.
July 10, 2020 at 10:35 PM
Open science advocates are occasionally accused of having a somewhat religious zeal... nevertheless I love this cross-stitch gift that I got.
Cornell announced their pay cuts. The endowed side will see their retirement contributions cut from 10% to 3%.
July 3, 2020 at 7:36 PM
Cornell announced their pay cuts. The endowed side will see their retirement contributions cut from 10% to 3%.
Was running out of ideas to entertain kids at home. Started playing Monty Hall with my 9yr old. This "paradox" never ceases to amaze. His mind was blown that switching doors actually helps.
June 23, 2020 at 9:54 PM
Was running out of ideas to entertain kids at home. Started playing Monty Hall with my 9yr old. This "paradox" never ceases to amaze. His mind was blown that switching doors actually helps.
Future historians will regard these Notices of Non-discrimination as Rosetta Stones. I am ASTOUNDED by the number of languages that are represented. It even has Swiss German (Wann du Deitsch schwetzt...)
June 19, 2020 at 1:37 PM
Future historians will regard these Notices of Non-discrimination as Rosetta Stones. I am ASTOUNDED by the number of languages that are represented. It even has Swiss German (Wann du Deitsch schwetzt...)
I missed you too, Tompkins County Public Library. @tcplny #curbsidepickup
June 17, 2020 at 10:11 PM
I missed you too, Tompkins County Public Library. @tcplny #curbsidepickup
Was hoping to have a peek at @JeffRouder talk on ANOVA (and the lies it tells), but apparently WMG (Warner Music Group) blocked the video on YouTube. :(
https://www.sowi.uni-mannheim.de/en/erdfelder/research/one-world-cps/
https://www.sowi.uni-mannheim.de/en/erdfelder/research/one-world-cps/
May 6, 2020 at 1:31 PM
Was hoping to have a peek at @JeffRouder talk on ANOVA (and the lies it tells), but apparently WMG (Warner Music Group) blocked the video on YouTube. :(
https://www.sowi.uni-mannheim.de/en/erdfelder/research/one-world-cps/
https://www.sowi.uni-mannheim.de/en/erdfelder/research/one-world-cps/
New paper by Vanderweele et al. argues that factor analysis that shows uni-dimensional structure actually tells us nothing about the latent structure at all (if we assume other causally related factor). #measurement
https://arxiv.org/abs/2001.10352
https://arxiv.org/abs/2001.10352
February 24, 2020 at 9:09 PM
New paper by Vanderweele et al. argues that factor analysis that shows uni-dimensional structure actually tells us nothing about the latent structure at all (if we assume other causally related factor). #measurement
https://arxiv.org/abs/2001.10352
https://arxiv.org/abs/2001.10352
First day in Cardiff, Wales, and I realize that what I believed was a British urban legend is actually reality. ALL sinks in my house have two taps - one hot, one cold. And it makes no sense whatsoever.
August 4, 2019 at 7:41 PM
First day in Cardiff, Wales, and I realize that what I believed was a British urban legend is actually reality. ALL sinks in my house have two taps - one hot, one cold. And it makes no sense whatsoever.
Inspired by @minzlicht I shaved my head right before the start of my sabbatical. In a couple of days I will be joining @richarddmorey @chrisdc77 and @CandiceMorey at @cardiffuni
July 28, 2019 at 12:26 AM
Inspired by @minzlicht I shaved my head right before the start of my sabbatical. In a couple of days I will be joining @richarddmorey @chrisdc77 and @CandiceMorey at @cardiffuni
And here the standard errors of the treatment effect estimates.
Adjusting on nothing = large SEs (comparatively speaking), but adjusting on only the unbalanced covs = also large SEs! Adjustment on those doesn't buy you anything! Adjusting on all, or predictive ones =much better
Adjusting on nothing = large SEs (comparatively speaking), but adjusting on only the unbalanced covs = also large SEs! Adjustment on those doesn't buy you anything! Adjusting on all, or predictive ones =much better
April 13, 2019 at 1:09 AM
And here the standard errors of the treatment effect estimates.
Adjusting on nothing = large SEs (comparatively speaking), but adjusting on only the unbalanced covs = also large SEs! Adjustment on those doesn't buy you anything! Adjusting on all, or predictive ones =much better
Adjusting on nothing = large SEs (comparatively speaking), but adjusting on only the unbalanced covs = also large SEs! Adjustment on those doesn't buy you anything! Adjusting on all, or predictive ones =much better
These are the estimates of 5000 reps. All are unbiased - adjusting on nothing, and adjusting on only imbalanced covs looks pretty similar. So does adjusting on everything and adjusting on the predictive covs.
April 13, 2019 at 1:09 AM
These are the estimates of 5000 reps. All are unbiased - adjusting on nothing, and adjusting on only imbalanced covs looks pretty similar. So does adjusting on everything and adjusting on the predictive covs.
Wow, this funnel plot of PRE-REGISTERED replications of money priming is crazily syymetric https://research.tilburguniversity.edu/en/publications/a-comprehensive-meta-analysis-of-money-priming
March 31, 2019 at 1:42 PM
Wow, this funnel plot of PRE-REGISTERED replications of money priming is crazily syymetric https://research.tilburguniversity.edu/en/publications/a-comprehensive-meta-analysis-of-money-priming
Reading list for this semester's MSG (Methods Study Group).
Topic this semester: "The p-value".
Topic this semester: "The p-value".
January 15, 2019 at 7:46 PM
Reading list for this semester's MSG (Methods Study Group).
Topic this semester: "The p-value".
Topic this semester: "The p-value".
Interesting tidbit from Harvard's discrimination trial. Expert witness simulates data under H_0 and finds p=.02, then claims with (1-p)=99.8% confidence that H_1 (Harvard manipulated admissions process) is correct. Sounds like Prosecutor's fallacy.
January 3, 2019 at 4:04 PM
Interesting tidbit from Harvard's discrimination trial. Expert witness simulates data under H_0 and finds p=.02, then claims with (1-p)=99.8% confidence that H_1 (Harvard manipulated admissions process) is correct. Sounds like Prosecutor's fallacy.