The Bureau of Labor Statistics (BLS) calculates employment data with biases. If the errors were random, revisions would swing back and forth. Since revisions are almost always downward, this must be by design. The government knows that the media and markets pay more attention to preliminary figures than to revisions.