SPRITE stands for Sample Parameter Reconstruction via Iterative TEchniques. 

If you have a sample that's well-described (mean, SD, range, and sample size), it reconstructs a potential dataset for those parameters.

It does this by creating a proto-sample with the right mean (for instance, if you have mean=4.5 and n=100, it makes a starting distribution with sum=450 sorted into bins). After that, it shuffles the values around according to rules you control to make the right SD. If you change the shuffling rules and the starting distribution, you can get different outputs.

This means you can have one (or a hundred, or a thousand) potential pseudo-datasets for any sample. There are a lot of reasons you might want these.

Some additional fun points:

• SPRITE isn't 'published', it's just a preprint. I never saw the point of publishing it, because it works.

• SPRITE is available in R, Python AND Matlab. See the preprint.

• SPRITE can be calculated online directly through this RShiny app: https://shiny.ieis.tue.nl/sprite/ 

• SPRITE is also available in an R package! Type install.packages("rsprite2")from the prompt to install.

• SPRITE was a direct consequence of the work done on Brian Wansink's papers -- we would see samples with odd mean / SDs but had no way to find out what they might look like.

• SPRITE actually has a more sophisticated cousin, CORVIDS. It does more or less the same thing, but better / more completely (just slower). CORVIDS struggles with big samples.

Go back Home