Wouldn't be possible without @colinsmithpsych.bsky.social and @peterkvam.bsky.social!

The thread above gives the general rundown but if your to-read folder isn't full enough:
Paper: doi.org/10.1521/soco...
Pre-print: osf.io/preprints/ps...

Extending this work in several directions, stay tuned!

100%! I'll send the materials your way, Max

Thanks so much for the shoutout😁

The tug-of-war model is inspired from popular theories of semantic/linguistic similarity and was first proposed by Kvam et al., 2024
link.springer.com/article/10.3...
Hope this tutorial makes it more accessible and I'd be happy to answer any questions! (13/13)

*Crucially* in being mathematical, the model posits a concrete understanding of what associations are, aiding better theory-building. Besides the TOW, we also provide code and tutorials on fitting the QUAD and basic DDM for anyone interesting in modeling IAT data acc. established models (12/13)

*Crucially* in being mathematical, the model posits a concrete understanding of what associations are, aiding better theory-building. Besides the TOW, we also provide code and tutorials on fitting the QUAD and basic DDM to IAT data to help researchers interested in modeling IAT data (12/13)

Model parameters were estimated w/Bayesian stats. Psych data are always nested (eg. associations are held by individuals but thresholds are individual x condition). Parameters in the model was drawn as shown below.
JAGS/R code to run our model and a video tutorial provided! osf.io/cv2a5/ (11/13)

Threshold: amount of info required to make a decision
Start bias: the extent to which one is already predisposed to one response
Drift rates: the avg speed at which each stimulus "emits" info
Drift variability: the variance in drift rates..(10/13)

What can the TOW model tell you? Like with all models, it's all in the parameters! They are: association, thresholds, start point bias, drift rate, non-decision time,& drift variability...(9/13)

Model parameters also do a great job predicting self-reported behavior towards target groups, showing the predictive validity of the model (8/13)

But there's more to the results! Here's a what they look like for the Race IAT. Check out the preprint for full discussion (7/13)

But there's more to the results! Here's a what they look like for the Race IAT. Check out the preprint for full discussion (7/13)

Results summary: Ps showed more positive associations for White people, Democrats, Cis people, Apples, and Dogs while showing corresponding negative associations for Black people, Republicans, Trans people, Snickers, and Cats - 87%, 60%, 64%, 73%, and 65% respectively (6/13)

Data come from 5 IAT studies with widely different targets: Apples/Snickers, Black/White People, Cats/Dogs, Democrats/Republicans, and Trans/Cis People. (5/13)

In an IAT, the evidence accumulated for one response ("categorize this stimulus as Bad/Black") builds up faster if those concepts are aligned & are "pulling" together. Since the model involves concepts pulling together/away from each other, we call it the tug-of-war model..(4/13)

..and (90-180°) angles respectively. This conceptualization is combined in the Decision Diffusion Model (DDM) framework of decision making where people sample information about 2 options until they reach a threshold of acceptable support for one option (right panel above) (3/13)

..the angle between axes representing cognitive concepts (eg., angle between valence & race axes). If a category is not associated with positivity/negativity, the angle should be 90° (orthogonal concepts). Similarly, positive and negative associations will create (0-90°)..(2/13)