JackEdTaylor/selker

Shiny app demonstrating the threshold estimation method outlined by Selker et al. (2019)

selker

This Shiny app demonstrates Selker et al.'s (2019) method of describing an arbitrary number of threshold locations with just two parameters, showing how well the method can describe thresholds of known locations.

Running the App

You can run the app locally or remotely:

• Run it locally in R, with shiny::runGitHub(repo="selker", username="JackEdTaylor")
• Go to https://jackt.shinyapps.io/selker/
• Go to https://rstudio-connect.psy.gla.ac.uk/selker/

More on the Method

Models estimating threshold locations on latent distributions are very useful for analyses of ordinal ratings. A common way to account for participant or item variablility in response biases is to estimate thresholds separately for each participant or item, but for large rating scales this requires a large number of parameters to be estimated. Selker et al. suggest a method for estimating the locations of any number of thresholds which only requires the estimation of two parameters: $a$ (describing scale) and $b$ (descrbing shift). Applying this to estimate per-participant and per-item thresholds can be more efficient.

Threshold $\lambda_c$ in position $c$, when there are $C$ ordered regions in the latent distribution, has location:

$$\gamma_c = log ( \frac{c/C}{1-c/C} )$$

$$\lambda_c = a \gamma_c + b$$