D. L. Kostanich and L. S. Cahoon (1994) "Effect of Design Differences Between the Parallel Survey and the New CPS," CPS Bridge Team Technical Report 3.
The official monthly labor force estimates from January 1994 onward will be based on data from a comprehensively redesigned Current Population Survey (CPS). The redesign incorporates changes in the basic questionnaire and collection methods. In addition, these estimates will be constructed using the 1990-census-based population controls rather than 1980-census-based population controls.
In order to gauge the effects of the CPS redesign on published estimates, a Parallel Survey (PS), intended to provide annual average labor force estimates, was conducted from July 1992 to December 1993. The PS used a separate monthly sample of households approximately one-fifth the size of the regular CPS sample.
Annual average estimates from the PS were compared to the CPS for 1993. These comparisons provide estimates of the overall effect of switching to the new questionnaire and collection methods. This study showed an expected .45 (or rounded to .5) percentage point increase in the unemployment rate. (For a general discussion of these results, see the February 1994 issue of Employment and Earnings; for a more detailed discussion, see Polivka 1994). These results do not provide an estimate of the effect of switching from the old CPS to the redesigned CPS, because not all design differences were accounted for. We have known for some time that at least one design difference had a measurable effect on the estimated unemployment rate. That is, the change in the population controls resulted in an additional .10 percentage point increase in the unemployment rate; implying a total effect of .55 (or rounded to .6) on the unemployment rate. (See table D, The Employment Situation: January 1994.) With the completion of the PS, it became possible to estimate the effects of the design differences.
To better understand the effect of design differences between PS in 1993 and CPS in 1994, the CPS Bridge Team has obtained estimates of several design differences. Three design differences, present in January 1994, which we have examined are: 1) The redesigned CPS uses 1990-census-based population controls rather than the 1980-census-based controls; 2) The PS used fewer post-stratification controls than the CPS due to smaller sample sizes and different sample designs; and 3) About 18 percent of the PS sample was interviewed at centralized telephone facilities, as compared to 12.5 percent in the redesigned CPS.
Accounting for these design differences, the expected increase in the January 1994 unemployment rate is marginally greater—.48 percentage point. This estimate, as well as each of the individual effects given below, are subject to sampling variance; thus, the actual effects will differ from the estimated effects.
The specific estimated effects are:
- An increase of .10 percentage point on 1993 CPS due to the change to 1990-census-based population controls. (The actual increase in the PS was .12. This is a less accurate measure due to the smaller PS sample size.)
- An increase of .01 percentage point due to fewer post-stratification controls.
- A decrease of .08 percentage point due to less centralized interviewing in the new CPS. Except for the .01 percentage point increase due to fewer post-stratification controls, the overall effects and the other individual effects are believed to be statistically significant; although the centralization effect was marginally significant.
After January 1994, two additional changes will occur. The composite estimator will be introduced in early 1994, and the proportion of centralized interviewing will equal the proportion present in the PS by June 1994. Increasing centralized interviewing removes the decrease of .08 percentage point present in January to give an expected increase of .56 percentage point on the unemployment rate. The effect of the composite estimator is much less certain. As the composite estimator is introduced, the expectation is that the unemployment rate may drop by as much as .06 percentage point, or it may increase by .02 percentage point.
Applying the 1993 CPS seasonal adjustment factors to the time series for the PS and the CPS showed no additional effects. This is primarily due to use of the same adjustment factors on both surveys. However, it will take some time before we know the seasonal pattern for the new CPS.
Additionally, the switch to using 1990-based population controls will increase the variance on many estimates. Overall the increases are expected to be small.
Our best estimate of the combined redesign effects on the annual average unemployment rate is an increase of .56 percentage point. This is before compositing or seasonal adjustment. It will be awhile before we can determine a more precise estimate of the effect of compositing or seasonal adjustment.
Last Modified Date: July 19, 2008