Statistical Design of the 1995-1996 Status of the Armed Forces Surveys.
Final rept. 29 Sep 94-31 Oct 97,
RESEARCH TRIANGLE INST RESEARCH TRIANGLE PARK NC
Pagination or Media Count:
This report evaluates the original sampling designs and missing data compensation procedures for the 1995 Sexual Harassment Survey SHS and the 1996 Equal Opportunity Survey EOS in light of actual response rates experienced in each of the surveys. The average design effect, a measure of the efficiency of the design, computed over the 124 reporting domains used in the SHS design was 1.80. For the EOS, the average design effect computed over a total of 333 domains was 1.97. These results indicate both designs are reasonably efficient. Weighting class adjustments were used to compensate for non response in the SHS and inverse response propensity weighting was used to compensate for non response in the EOS. A comparison of the two shows that similar results were obtained. The modeling approach did produce slightly higher variances, but may have been more effective in reducing biases because of the additional variables used for the adjustments. Other means of reducing missing data biases are also discussed along with their advantages and disadvantages. These include adding additional mailings to the data collection schedule, decreasing the size and complexity of the questionnaire, and employing multiphase sampling designs.
- Personnel Management and Labor Relations
- Military Forces and Organizations