Table 1.
Test | Year | Disattenuated Correlationa | NIbCR | NIcMC | Sample size | Accuracy by Unidimensional (%) | Accuracy by Bi-factor (%) | Accuracy improvement |
---|---|---|---|---|---|---|---|---|
English Literature | 2010 | 0.778 | 3 | 55 | 20,000 | 60.90 | 65.56 | 4.65% |
English Literature | 2009 | 0.77 | 3 | 55 | 20,000 | 63.04 | 67.17 | 4.13% |
English Language | 2009 | 0.81 | 3 | 55 | 20,000 | 63.62 | 67.56 | 3.94% |
English Language | 2010 | 0.807 | 3 | 54 | 20,000 | 63.37 | 67.15 | 3.78% |
European History | 2009 | 0.92 | 7 | 80 | 20,000 | 68.77 | 70.33 | 1.56% |
World History | 2009 | 0.89 | 3 | 70 | 20,000 | 69.70 | 70.75 | 1.05% |
US History | 2010 | 0.908 | 3 | 80 | 6936d | 69.32 | 70.24 | 0.92% |
European History | 2008 | 0.89 | 7 | 80 | 20,000 | 68.14 | 69.06 | 0.92% |
World History | 2008 | 0.91 | 3 | 70 | 20,000 | 70.92 | 70.07 | 0.85% |
Physics B | 2008 | 0.96 | 7 | 70 | 11,941 | 76.73 | 77.06 | 0.33% |
The disattenuated correlation denotes the estimated true score correlation between the subtest scores for the MC and CR items; they were provided by College Board.
The number of CR items included in the tests.
The number of MC items included in the tests. To maximize the comparability of results, we used 3 CR items and 55 MC items for all 10 tests.
US History 2010 data set had a sample size of 20,000. However, only 6936 students chose the same three CR items (i.e., #1, #3, and #4 items).