Missing Data - Reasons for screening data
In case of any missing data, the researcher needs to conduct tests to ascertain that the pattern of these missing cases is random.
Create dichotomous variable - non-missing vs missing for a specific variable. Run a simple independent samples t-test on a different variable in the collected sample to see if there are any significant differences.
Handling missing values:
1. Delete missing data (good idea if there are only a few missing cases)
2. Delete variables containing missing values (good idea if most of the missing values are concentrated to only a couple of variables. Still problematic if they are important to the ultimate goal of the research)
3. Estimate missing values
4. Prior knowledge
5. Replace missing values with the mean (main concern: lowers the calculated variance as compared to the unknown actual variance)
One variation involves using group means for missing values for cases involving group comparison analysis
6. Regression approach: use several IVs to explain the DV (that includes several missing values). Predict missing values using IV values.
7. Concerns include finding proper IVs that explain DV, estimates obtained from prediction more consistent with the scores used to predict them compared to the real values.
8. When we use any of the techniques described above, as a researcher we have to ascertain that our solution hasn't changed the results of the analysis (run the tests, with and without the treatment).