improve R2 an

NourNashed

hello all,

I am struggling with the low R2 and R2 adjusted i am getting from the model generated by JMP. every reading has many replicates but the R2 is still low (0.17). i did Box-Cox transformation based on the suggested λ but still R2 very low. any help on how to improve the R2? and what does low R2 mean??

I am really stuck and cant trust my statistical results on which factor is significantly impacting the outcome. 

Ben_BarrIngh

Hi @NourNashed,

 

It depends on the quality of the data, because ultimately a low RSq is telling you one of three things: a) the data is very poor quality and too noisy b) there is no trend in the relationship or c) the model type you are applying does not suit your data. Supplying more details on your data (and the dataset) would help, but here's some general guides.

 

• You mention replicates - how much does the response vary with each replicate? If it's very wide (high standard deviation) then this will make it difficult to determine the difference between each replicate from a statistical standpoint.
• What do your residuals look like? Are there patterns in the row-to-row values? Are they randomly distributed?
• What terms have you included in your model? We can look at the lone effects of a parameter (i.e. the amount of stirring), but also interactions (the relationship between stirring and temperature) and curvature (does stirring cause the response to change in a quadratic or cubic form). You can take a 'catch all' approach by trying to apply a full set of all of these terms then filtering them out with something like stepwise selection or generalised regression.

 

If you haven't I would consider reviewing some of the modules in the STIPS course, particularly the module 'Correlation and Regression'.

Thanks,
Ben