Everybody seems to agree that the official measure of the US unemployment rate is too low, but it is uncertain by how much. Indeed, the method of measurement is very poor, as it is based on a survey which is rife with errors. Follow-ups have uncovered frightening errors both on the side of the interviewers and interviewees. And even those reinterviews are not reliable, because they are too small, too infrequent and still full of errors. Thus they cannot be used as a standard for evaluating measurement error bias. But the survey is a short panel, meaning that interviewees are followed for a few months, which allows to improve the accuracy of the measures.
Shuaizhang Feng and Yingyao Hu do this by assuming that there are latent variables underlying the process, and these variables have some level of persistence and measurement error probabilities that varies across groups (race, gender, age). The resulting "true" unemployment rates are on average 35% higher than official ones, more during recessions. Right now, the unemployment rate would be close to 15%. Employment rates do not differ significantly, indicating that unemployed often claim not to be in the labor force, especially females.