Aggregating Comparable Categorical Responses to the Unit of Observation in Employer Surveys

• Activity Number: 365
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Jeremy Pickreign*+
• Companies: NORC at the University of Chicago
• Keywords: Survey Analysis ; Categorical Variables ; Variable Aggregation

Abstract

The California Employer Health Benefits Survey collects data from employers on up to five plan types offered to workers. Data are frequently analyzed at the employer level (unit of observation). Responses, however, can vary by plan type. Aggregating continuous variables (e.g., annual premium cost) across plan types is straight forward, but aggregating categorical variables can be problematic. Also, enrollment into each offered plan type is not uniform adding to the problem. In this paper, five methods are contrasted for aggregating categorical plan level variables (e.g., self-insurance indicator) to the employer level. Methods include: (1) a simple average across plans; (2) a weighted average across plans by plan type; (3) a random proxy of a representative plan; (4) the proxy representative plan having the largest enrollment; and (5) stacking the data making the plan the unit of analysis. T-tests found no significantly different estimates between each method pairing. While each method provides statistically comparable estimates, method (5) may be best as it uses more information and minimizes the standard error.