Home Conference Resources Advocacy About APDU

A Lifelike Model of the Distribution of Wages: The Salamander

Tuesday, 3:45pm - 5:00pm

Speaker:

  • John Angle
    Statistician, Economic Research Service, U.S. Department of Agriculture
    bio

Session Organizer:

John Angle
bio

Abstract:

Being questioned about income makes many survey respondents uneasy. Some legislators have expressed unease with the Census Bureau's posing of income questions. To minimize this unease, the Bureau guards respondent identity from disclosure when survey data records are released in a public use sample by, among other methods, topcoding incomes bigger than the 99th+ percentile.That high a topcode leaves unprotected most of the largest incomes in conditional distributions, such as the nonmetro, with lower mean and more skew than the national distribution. Radical solutions to disclosure risk in public use samples, such as "artificial data" have been proposed. Lowering the minimum topcodeable income is a simpler way to guard against disclosure risk specifically associated with incomes, but a lower topcode truncates more of the right tail of the distribution. Yet, for the purpose of estimating statistics of the whole distribution of annual wage and salary income, the information in the right tail can be reproduced from redundant information in the rest of the distribution, even just the distribution below the third quartile. The Salamander, a model of the distribution of annual wage and salary income, demonstrates this fact. The Salamander is parsimonious, close fitting, and with parameter estimates from incomes below the topcode, can closely estimate statistics of the whole distribution of annual wage and salary income. Since Salamander dynamics mimic the dynamics of the empirical distribution, the Salamander forecasts the distribution well. The Salamander maintains the usefulness of annual wage and salary income in public use samples despite a lower topcode.

Back to agenda