Thread by @jonaslindeloev

I've made a cheat sheet and a bunch of applets to give you an intuitive feel for various reaction time distributions. You can choose datasets, fiddle with parameters, and see working code examples: lindeloev.net/shiny/rt/ 1/n

You can choose among several pre-existing datasets, e.g., by @EJWagenmakers and @JeffRouder. You can also copy-paste in your own data and try fitting different distributions. 2/n

For example, take the Ratcliff diffusion model for a spin. You can upload your own data to see if it fits. Or you can plug in some parameters from a paper to get an intuition about their findings. 3/n

It is easy to infer these parameters using the *family* argument in brms::brm (by @paulbuerkner). The following code fits a mixed model using a shifted log-normal distribution:

fit = brms::brm(rt ~ condition + (1|id), data, family=shifted_lognormal())

4/n

Actually, my main motivation for doing this was to make it easy and compelling for everyone (including myself) to stop using the normal distribution for RTs. The alternatives have superior fits, are easy to do, and most of them are more interpretable. 5/n

I've come to love distributions where a change in just one "difficulty" parameter can fit long(er) and short(er) RTs and their associated larger/smaller variance (see ejwagenmakers.com/2007/Wagenmake… by @EJWagenmakers). This means you can regress on just one parameter. 6/n

I've dared to rank the distributions on fit and interpretability in a regression context. The *shifted log-normal* come out as the top descriptive dist. and *Ratcliff diffusion* as the best theoretical model. This is subjective, so make up your own mind. What do you think? 7/n

It would be great if RT-friendly distributional regression was someday available in popular statistics software like @Jamovi, @JASPStats, or (deep breath) @IBMSPSS. Until then, #brms (Bayesian) and the #gamlss package (frequentist) has you covered. 8/n

All code is available at github.com/lindeloev/shin…. Please chime in with improvements or ideas. Personally, this is my first #shiny app (@rstudio), but definitely not my last! 9/n

Thanks for great papers and materials, @HenrikSingmann @dora_matzke @twiecki @StatModeling @paulbuerkner @EJWagenmakers @JeffRouder.

Tagging @fusaroli @dataandme @sharon000 @Rbloggers @komaau @ProfAndyField @JPdeRuiter

@HenrikSingmann @dora_matzke @twiecki @StatModeling @paulbuerkner @EJWagenmakers @JeffRouder @fusaroli @dataandme @sharon000 @Rbloggers @komaau @ProfAndyField @JPdeRuiter OK, the shiny server has trouble keeping up with the interest. Here's the cheat sheet if it's down: github.com/lindeloev/shin…