Projpred results interpretation

We’re working on a project that examines whether a set of variables used in the clinic plus some metabolites is capable of tracking changes in another clinical variable over an intervention. We have created our model and run projpred to try & identify a sparser full model. The results are shown below:

I have interpreted this as “there is no sparser model that is useful for predicting changes in our clinical variable”. This is not a surprise to us - we expect the outcome to be difficult. But I’m second guessing myself now and wondering if this result could also be interpreted as “any single variable is as good as the full model”.

Any advice on interpretation would be appreciated.

Or you could also say “adding variables doesn’t improve predictive performance compared to the null model”

You can also compare the RMSE to the standard deviation of the data to see whether the full reference model is predicting any better than guessing with the mean of the data (ie null model)

Thanks Aki, perfect and what we’d broadly predict from the physiology.

1 Like