Occupational Pay Relatives explanatory note
Technical Note
Pay relative controls and calculations
Pay relatives control for differences among areas in occupational composition as well as
establishment and occupational characteristics. Metropolitan areas often differ greatly in
the composition of establishments and occupations that are available to the local workforce.
For example, in Brownsville, Texas, the ratio of workers in the high-paying management,
business, and financial occupational group to the number of workers in all occupations is
under 6 percent, whereas nationally this ratio is over 8 percent.1 In addition to these factors,
the NCS collects compensation data for metropolitan areas at different times during the year.
Payroll reference dates differ between areas which makes direct comparisons between areas
difficult.
The pay relative approach controls for these differences to isolate the geographic effect
on wage determination. To illustrate the importance of controlling for these effects, consider
the following example. The average pay for professional and related workers in San Francisco
is $37.57 and the average pay for professional and related workers in the entire United States
is $29.76.2 A simple pay comparison can be calculated from the ratio of the two average pay
levels, multiplied by 100 to express the comparison as a percentage. The pay comparison in
the example is calculated as:
($37.57 / $29.76) x 100 = 126
This comparison does not control for differences between San Francisco and the nation in
the mix of occupations, industries, and other factors. A more accurate estimate of the
geographic effect of wages in San Francisco can be obtained by taking these differences into
account. Controlling for differences in occupational composition, establishment and
occupational characteristics, and the payroll reference date in San Francisco relative to the
nation as the whole, the pay relative for professional and related occupations in San Francisco
is equal to 117.
Sampling errors and statistical significance
Because the NCS is a sample survey, data are subject to sampling error. For the data
presented here, sampling error are differences that occur between the pay relatives estimated
from the sample and the true pay relatives derived from the population. It is important to
assess whether differences between each pay relative and the pay relative for the nation as a
whole is likely to be the result of sampling error or of true differences in pay levels. To
perform this assessment, a test of statistical significance is conducted.
The test constructs a 90-percent confidence interval that assumes the given area’s true
pay relative is equal to the national average. The confidence interval is constructed so that
there is a 90-percent probability the pay relative calculated from any one sample is contained
within the confidence interval. If from a single sample a calculated pay relative falls
within the confidence interval, then the pay relative is not statistically significant and
the hypothesis that the true pay relative is equal to the national average is accepted. However,
if the pay relative falls outside of the constructed confidence interval then the pay relative
is statistically significant at the 10-percent level. The hypothesis that the given area’s pay
relative is equal to the pay relative for the nation is rejected and one can conclude with
reasonable confidence that the true pay relative is different from the national average.
In addition to sampling error, pay relatives are subject to a variety of sources that can
adversely influence the estimates. The NCS may be unable to obtain information for some
establishments; there may be difficulties with survey definitions; respondents may be unable
to provide correct information, or mistakes in recording or coding the data may occur.
Non-sampling errors of these kinds were not specifically measured. However, they are expected
to be minimal due to the extensive training of the field economists who gathered the survey
data, computer edits of the data, and detailed data review.
Historical pay relative data are available for 1992-1996, 1998, 2002, 2004, and 2005.
There are several differences between the recent pay relatives and the pay relatives for
earlier years, including different industry and occupation classification systems, varying
methodology, and different survey designs. These differences limit comparability. The pay
relatives for 2004, 2005, and 2006 were calculated using the same industry and occupation
classification systems, methodology, and survey design. Nonetheless, comparisons between the
estimates for these years should be made only with a high degree of caution.
Pay relatives were estimated using a multivariate regression technique methodology to
control for interarea differences. This technique controls for the following ten characteristics:
- Occupational type
- Industry type
- Work level
- Full-time / part-time status
- Time / incentive status
- Union / nonunion status
- Ownership type
- Profit / non-profit status
- Establishment employment
- Payroll reference date
Even accounting for the characteristics used in the current regression analysis, there is
still significant wage variation across the areas. The variation is due to differences in
wage determinants that were not included in the model. Examples of these determinants include
price levels, environmental amenities such as a pleasant climate, and cultural amenities.
The pay relative regression methodology introduces another type of error. Regression
models are subject to specification error. The significance test does not specifically
measure specification error. However, care was taken to minimize this form of error by an
extensive search across specifications for the model that performs best in terms of predictive
accuracy.
For more details, see Maury B. Gittleman, "Pay Relatives for Metropolitan Areas in the U.S."
Monthly Labor Review, March 2005, pp. 46-53, and Parastou Karen Shahpoori, "Pay Relatives
for Major Metropolitan Areas," Compensation and Working Conditions, Spring 2003.
1 Data for this example are based on the May 2006 Occupational Employment and Wage
Estimates, http://www.bls.gov/oes/current/oessrcma.htm.
2 Average pay for professional workers in San Francisco and for the United States are
based on wage estimates published in the San Francisco-Oakland-San Jose, CA National Compensation
Survey, March 2006 and the National Compensation Survey: Occupational Wages in the United
States, June 2006, http://www.bls.gov/ncs/ocs/compub.htm.
Last Modified Date: October 11, 2007
Bureau of Labor Statistics
