JMP Software comes complete with a large number of sample data tables.
The other day I found one containing body measurement data:
A quick look at the data in the Multivariate platform shows that there are a number of interesting relationships:
It’s also clear that these potential predictor variables are strongly correlated with each other which makes the modelling process potentially challenging. But in the end my final model selection was remarkably simple and reasonable.
Rather than discuss methods of dealing with collinearity I’m interested in how to present the final result. JMP has made great strides in allowing content to be delivered in an web browser, but I can’t recall seeing such output – not even on the official JMP blog.
So here is my final model as a web-enabled profiler:
[swf:http://www.pega-analytics.co.uk/blog/wp-content/uploads/2015/11/Body-Measurements1.swf 700 300]
Just as in JMP you can resize the frame sizes. It would be nice if I could script that automatically so that it isn’t necessary to scroll – if anyone knows how please leave a comment below!
Here is an alternative view – based on a framework I’ve developed for automatically generating a web page based on a column formula:
Go get your tape measure and see how well does the model performs. Measurements are in kg and cm. Would you have chosen the same model for this data?