I calculated the percent of (weighted) respondents with margtax=0 for various age categories, from the March CPS, 1992-2014. There is a significant spike in this percent in 2004. Any idea why this is in the data?
Prior to the 2004 ASEC, the Census Bureau randomly assigned mean amounts for certain tax variables (see below), based on IRS aggregate tables. In 2004 and 2005, these tax variables were instead imputed by matching to the IRS data of similar households. Due to a three-year lag between the ASEC and the most recent IRS file, the Census Bureau had to make adjustments that they have acknowledged (p. 2) were “erratic in range, distribution, and aggregate amounts” and especially problematic for 2004. Indeed, the Census Bureau changed methods again in 2006.
However, I unfortunately cannot confirm the new tax model in 2004 is responsible for the spike in zero marginal tax respondents. You might consider contacting the Census Bureau directly for additional information on this issue.
Here is a list of imputed tax variables for 2004/2005:
Self-employed health insurance deduction
Self-employed savings deduction (SEP, SIMPLE and qualified plans)
Child and dependent care expenses