For graph-based command-of-evidence questions involving overall and subgroup results, first identify every component of the claim. Compare the groups within each category, then examine whether the categories are represented in different proportions. Select the choice that reports all data needed to connect the subgroup pattern to the overall result.
Hints
- Hint 1
Separate overall and category results
Do not assume that the system with the higher overall accuracy also performed better on each kind of page.
- Hint 2
Examine the page mixtures
Determine whether one system was tested on a larger proportion of the kind of page that both systems processed more accurately.
- Hint 3
Cover the complete argument
Look for a choice that addresses performance within both page categories as well as the composition of the two test sets.
Step-by-step
- Step 1
Identify the required evidence
The statistician's argument has two components: Lumen must not be more accurate within either page category, and the two systems must have been tested on substantially different mixtures of those categories.
- Step 2
Compare accuracy within each category
Scribe had 98% accuracy on printed pages, compared with Lumen's 96%, and 74% accuracy on handwritten pages, compared with Lumen's 70%. Thus, Scribe—not Lumen—was more accurate in both categories.
- Step 3
Examine the test-set mixtures
Printed pages were easier for both systems, but they constituted 80% of Lumen's test set and only 30% of Scribe's. Lumen therefore received a much larger proportion of the easier kind of page.
- Step 4
Select the complete evidence
The fourth choice reports both Scribe's advantage within each category and the unequal proportions of printed pages. It therefore most completely supports the statistician's explanation of the overall results.