"The economy would need to be growing at breakneck speed for unemployment to drop to 7.8% from 8.3% in the course of two months..."
"Let's get real. The unemployment data reported each month are gathered over a one-week period by census workers, by phone in 70% of the cases, and the rest through home visits. In sum, they try to contact 60,000 households, asking a list of questions and recording the responses.
Some questions allow for unambiguous answers, but others less so. For instance, the range for part-time work falls between one hour and 34 hours a week. So, if an out-of-work accountant tells a census worker, "I got one baby-sitting job this week just to cover my kid's bus fare, but I haven't been able to find anything else," that could be recorded as being employed part-time.
The possibility of subjectivity creeping into the process is so pervasive that the BLS's own "Handbook of Methods" has a full page explaining the limitations of its data, including how non-sampling errors get made, from "misinterpretation of the questions" to "errors made in the estimations of missing data."