Using data from the Current Population Survey and a new matched survey
of employers and employees, this paper investigates error in the meas
urement of employer-provided health insurance. The often-used March CP
S gives lower coverage estimates than the April/May CPS, which focuses
on employer-provided coverage. In addition, individuals who are in bo
th the March CPS and April/May CPS often give inconsistent responses o
n their health insurance status, perhaps due to differences in the wor
ding of the health insurance questions. A new survey shows that worker
s tend to report higher rates of coverage than do firms and that many
individuals also disagree with their employers about their coverage. T
he differences in the firm and worker reports of coverage are uncorrel
ated with standard worker and firm characteristics, suggesting classic
al measurement error that does not bins the parameters of models expla
ining health coverage. When health insurance is used as an explanatory
variable, however, measurement error results in significant bias towa
rd zero.