Recently, I met with the professor and the grad student in charge of the research project. I discovered that when reviewing the results of the project, that the models we had created were awful. God awful. To give you an idea, statisticians use a measure called R2 to determine how well a model fits that ranges from 0 to 1. A R2 of 1 means that the model perfectly predicts the data. A R2 of 0 means that the model has no relation to the data at all. A good R2 is around 0.8, a decent R2 is around 0.5, and a poor R2 is around 0.3. Our average R2 was around 0.1.
At this point, my professor decided that it would be better for us to reuse the method, but use a different method. While we had been using data at the provider level, she wanted us to aggregate providers to a State level. This would give us a coarser resolution, and hopefully some patterns would emerge. I'll keep you updated.
No comments:
Post a Comment