Open mvuorre opened 1 year ago
We have quite a bit of available RT & correct/incorrect response that I have access to but is not topical in that way. I could ask some contacts (Amy Orben, or Ben Nelson). Also found this (https://journals.sagepub.com/doi/pdf/10.1177/00936502211038196) which had freely available data
The data you linked to looks very suitable to me. I have worked a bit with that data already (https://psyarxiv.com/dpuya) and so am somewhat familiar with it as well. Perhaps we can initially plan to use that unless something else comes along.
Would you RT & correct / incorrect data be suitable to model with a diffusion model (https://cran.r-project.org/web/packages/brms/vignettes/brms_families.html#response-time-models)? That might be pretty cool too! (But maybe not as the first example because of the additional complexity.)
You definitely could do a diffusion model, although we've done DSEM to answer slightly different questions.
One thing that occurs to me that could be a nice contribution following off a comment in McNeish & Hamaker is they say latent mean centering isn't available through MLM software. The implication being that normal MLMs cannot do this. But with this approach, hierarchical models could use latent centering (at least in theory). It might be really cool to show a standard kids-in-classrooms type example to show how the brms implementation is useful outside of DSEMs as well.
That would be a very nice thing to discuss and, if needed, to show an example of.
"Though person mean centering in multilevel software is the only option, this leaves estimates susceptible to both Nickell’s bias and Lüdtke’s bias. These disadvantages are overcome by DSEM in Mplus, as it uses latent mean centering by default to similarly partition within-person and between-person effects, while also guarding against these biases (Asparouhov et al., 2018)." (M & H, p. 631)
When I'm looking at something like equation 4(c) in McNeish and Hamaker, I'm not sure what DSEM means. It looks like hierarchical modelling to me: There's some regressions, and some parameters are modelled, and some might even be latent. What is achieved by this terminological multiplication?
Re terminology: $€£
I can dig up some data and show the centering for the hierarchical data
As an option, in the netherlands we have this share data sites dataverse, We can look for some data examples there. Or we could use an example data set from a book, like Growth-Modeling
Find a good dataset that is sufficiently simple yet detailed enough to be useful. Optimally, this would include something like two one outcome and an additional covariate, so that we can model the outcome on the lagged outcome and another (time varying) covariate. Some basic longitudinal study, maybe a short experience sampling study, would be cool.
For my interests, it would be A+ if this involved (any kind of) well-being outcomes and some technology related predictor(s).
We'll start with this (https://journals.sagepub.com/doi/pdf/10.1177/00936502211038196) but can swap if it turns out not suitable.