PERC 2018 Abstract Detail Page
Previous Page | New Search | Browse All
| Abstract Title: | Comparing Methods for Addressing Missing Data for Concept Inventories |
|---|---|
| Abstract: | The most common method for addressing missing data in the PER literature is complete case analysis, where researchers only analyze matched samples. However, many statisticians recommend researchers use multiple-imputation (MI) to address missing data. We used simulated datasets to compare estimates of student learning using complete case analysis and MI. We based the simulated datasets on grades and concept inventories from 1,310 students in 3 physics courses and grade distributions from 192 STEM courses. We created missing data in the simulated datasets based on participation models from Jariwala et al. (PERC, 2017). Results showed that complete-case analysis tended to overestimate scores with a larger effect on the posttest but that MI only slightly overestimated scores. To improve the accuracy, precision, and utility of pre/post CI measurements, we recommend that researchers use MI and that researchers report descriptive statistics for both the participants and non-participants in their studies. |
| Abstract Type: | Juried Talk |
| Parallel Session: | Juried Talks II |
| Parallel Session: | Parallel Sessions Cluster III |
Author/Organizer Information | |
| Primary Contact: |
J. Nissen University of Maine |
| Co-Author(s) and Co-Presenter(s) |
R. Donatello, and B. Van Dusen |
