search

uchicago-computation-workshop / 2018_spring_conference

0 stars 1 forks source link

How do the results speak to each other? #26

Open tamos opened 6 years ago

tamos commented 6 years ago

Chelsea,

How do you results speak to each other? Can you elaborate on how we should interpret your results with respect to each other? #7

cernhofer commented 6 years ago

The most direct comparison between the regression vs. trees is in terms of variable importance and RMSE.

For variable importance, we can compare the jumps in R-squared with the decrease in node impurity to see how each method models the data/if they agree upon which variables are most important. I believe they do.

Using RMSE is purely to determine which method is better at predicting teen birth rates. Linear regression had the highest RMSE which suggests that there is a least some level of non-linearity present in the relationship. Because of this, regression trees would be a better tool for prediction.

Thanks for the question!!