diffPSI | R Documentation |
Will either return the difference Will estimate how p's rating of outcomes of each action deviate within situation s deviated from their average rating of outcomes of different actions in that situation
diffPSI(devPSI, locX = c(4:ncol(devPSI)), select_i = "max", dbl = T)
devPSI |
deviated PSI-structured dataframe (should have 'p','s','i' as their first three features) |
locX |
location of expected outcome ratings (default generously assumes the prescribed 'c(p,s,i,...,Likelihood)' data structure) |
select_i |
Return the maximum or minimum of i? |
dbl |
Double the deviation scores (default is T) |
Note that when nI = 2, the dpsi scores will be reflections of one another. In which case, keeping only the minimum or maximum i for each scenario can be useful to help avoid artificial inflation of sample size (although other strategies, like multilevel modeling in lavaan, may be preferable to address this issue).
When nI is greater than 2, a more sophisticated script should be used, which perhaps subtracts the focal action (e.g., A = 1) from the average of the contrasting actions (e.g., A = 0)
expected effects (difference) of focal action(s) from contrasting action(s)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.