Multiple Imputation of Right-Censored Data: an Application to Wage Data and Understanding the Changing Wage Gap by Gender in Germany

• Activity Number: 379
• Type: Contributed
• Date/Time: Wednesday, August 9, 2006 : 8:30 AM to 10:20 AM
• Sponsor: Business and Economics Statistics Section
• Author(s): Hermann Gartner*+
• Companies: Institute for Employment Research
• Keywords: multiple imputation ; missing data ; gender wage inequality ; censored wage data

Abstract

In many large data sets of economic interest, some variables, as wages, are top-coded or right-censored. For example the US Census's CPS, the Austrian social security records and the IAB-employment register (IABS). We treat this problem as one of missing data, where the missingness mechanism is not missing at random but rather missing by design (Little & Rubin,2002). Our approach is based on the multiple imputation (Rubin, 1978, 2004) of the right censored variables by drawing values from appropriate truncated distributions. This approach makes an important difference to realistic answers. We illustrate the approach with the German IABS, and analyze the gender gap by a Juhn-Murphy-Pierce decomposition. One important conclusion is that the main source for the narrowing gender wage gap between 1991 and 2001 in Germany is an improvement of women's position within the wage distribution.