How will the introduction of probability sampling affect the CES program?
The CES program is not only converting to NAICS, but has also been switching from a quota-based sample to a probability-based sample for preparing monthly employment estimates. Starting in March 2001 with the wholesale trade sector, this change in sampling technique was phased in over a two-year period. Currently, all industry sectors are estimated with probability samples, which offer several key statistical advantages:
Less biased estimates – Quota samples are now known to be at-risk for potentially significant biases, which may lead to an unrepresentative sample. Through random sample selection and improved estimation methodology, a probability-based sample more effectively ensures that estimates will properly represent the universe.
More representation for new firms – Quota samples lack a timely sample-based representation of employment from new business openings. Under probability sampling, a new approach called the net birth/death model has been developed for regular sample updates. This model better ensures representation of business births in the CES sample, and is expected to result in smaller adjustments.
Confidence intervals and sampling errors – In the past, the only accuracy measure for the CES survey’s quota sample was to measure the CES estimates against the annual benchmark (the Unemployment Insurance-covered universe of employment). The probability-based sample allows for the publication of confidence intervals and sampling errors.
Are estimates calculated the same way?
With quota-based samples, the estimator used to produce all employee estimates is an unweighted ratio known as the link relative, which uses a simple ratio of the over-the-month change in the sample and applies this same movement to the universe.
The new estimator under probability is defined as a weighted link relative, as a weight is applied to each sample unit in the estimation process. A sampled unit's weight is the inverse of its probability of selection. For example, a sample unit selected from an estimating cell where 1 in 10 units are selected will have a weight of 10, because it represents itself and 9 other units.