Figure 3: Example of Decision-Tree Models Fit to Two Samples from NELS88 Dataset

This figure is from the article "The Pitfalls of Prediction," February 26, 2013. 

Panel A

Did the student have many discipline problems?

1. If no, was the student of low socioeconomic status?
   1. If no, the student had a 5.5 percent risk of dropping out.
   2. If yes, the student had a 26 percent risk of dropping out.
2. If yes, did the parents have low academic aspirations for the student?
   1. If no, the student had a 16 percent risk of dropping out.
   2. If yes, did the student rarely change schools?
      1. If yes, the student had a 33 percent risk of dropping out.
      2. If no, the student had a 65 percent risk of dropping out.

Panel B

Did the student have a low grade composite?

1. If no, the student had a 6 percent risk of dropping out.
2. If yes, was the student rarely subject to discipline?
   1. If yes, the student had an 18 percent risk of dropping out.
   2. If no, was the student often absent?
      1. If yes, the student had a 58 percent risk of dropping out.
      2. If no, were the student's parents married (or in a marriage-like relationship) and living together?
         1. If yes, the student had a 28 percent risk of dropping out.
         2. If no, the student had a 60 percent risk of dropping out.

Source: Greg Ridgeway, "The Pitfalls of Prediction," February 26, 2013, nij.ojp.gov: https://nij.ojp.gov/topics/articles/pitfalls-prediction