Abstract
Shail Butani, Lawrence Cahoon, Robert Fay, and Donna
Kostanich. (1994) "Measurement of Different
Design Effects," Proceedings of the Section on Survey
Research Methods, American Statistical Association, 778-783.
The Current Population Survey (CPS), the monthly labor
force survey that the U.S. Bureau of the Census conducts for
the U.S. Bureau of Labor Statistics (BLS), uses state
probability samples (state-based design) totaling about
60,000 households. In January 1994, CPS interviewers began
collecting data in an entirely computer-assisted environment,
using a new questionnaire. Before introducing the new data
collection method, the Census Bureau and BLS tested it on a
separate national probability sample of 12,000 households.
The results of this parallel survey (PS), which ran from July
1992 through December 1993, indicated that the CPS annual
average unemployment rate would have been 0.45 percentage
point higher in 1993 had the new data collection method been
used. For a general discussion of the results, see the
February 1994 Employment and Earnings (Bureau of Labor
Statistics, 1994, Vol. 41, No. 2); for a more detailed
discussion, see Polivka (1994).
The above difference of 0.45, however, provides an estimate
of only the overall effect on a 1993 annual average basis. In
this paper, we discuss the measurement of design differences
between the PS and the CPS in January 1994 and subsequent
months to better understand the transition effects that were
not accounted for. We concentrate our analysis on the effects
of: 1) differences in the proportion of households
interviewed from centralized facilities; 2) use of 1990-based
population controls, adjusted for census undercount, rather
than 1980-based controls formerly used; 3) differences in
ratio adjustment methods; 4) use of composite estimation to
reduce variance; 5) seasonal adjustment; and 6) 1990-based
population controls on variances.
In this paper, we first give a brief background on the
redesign of the CPS; discuss the design differences between
PS and CPS; present estimates of various design differences;
compare the expected estimates to actual estimates for the
early months of 1994, where possible; summarize our results;
and state a topic for immediate future research. Due to space
limitations, we give only estimates of effects for design
differences on the national unemployment rate. For estimates
of differences in design effects on several demographic
groups and on employment-to-population ratio and civilian
labor force estimates, see Kostanich and Cahoon (1994).
Last Modified Date: July 19, 2008