U.S.

Bureau of Labor Statistics

Current Population Survey (CPS)
Technical Documentation
November 2018

www.bls.gov/cps/documentation.htm#reliability

Calculating Approximate Standard Errors

and Confidence Intervals for Current
Population Survey Estimates

This document provides information about calculating approximate standard errors for estimates
from the Current Population Survey (CPS). It also includes examples of how confidence
intervals for estimates can be calculated. A November 2018 update of this document introduces
 and  parameters and a slightly different methodology for calculating standard errors.

The CPS sample size is designed to meet a specified reliability criteria for unemployment
estimates at the national and state level. The requirement is that, assuming a national
unemployment rate of 6.0 percent, an over-the month change of roughly 0.2 percentage points be
statistically significant at the 90-percent level of confidence.

A table showing changes in selected labor force indicators with statistical significance tests is
online at https://www.bls.gov/cps/documentation.htm#reliability. The table is updated each
month with the publication of the Employment Situation news release.

Approximate standard errors and confidence intervals for CPS estimates can be calculated using
instructions in this document and the parameter and factor tables (PF-1 through PF-16) available
online at https://www.bls.gov/cps/parameters-and-factors-for-calculating-standard-errors.xlsx.
(These tables mirror tables A-1 through A-16 of the monthly Employment Situation news
release.) The parameter and factor tables allow users to calculate approximate standard errors
for a wide range of estimated levels, rates, and percentages, and also changes over time. The
parameters and factors are used in formulas that are commonly called generalized variance
functions.

Reliability of Current Population Survey Estimates

An estimate based on a sample survey like the CPS has two types of error—sampling error and
nonsampling error. The estimated standard errors provided in this publication are
approximations of the true sampling errors. They incorporate the effect of some nonsampling
errors in response and enumeration, but do not account for any systematic biases in the data.
Nonsampling error
Nonsampling error refers to errors due to factors that are not related to sample selection. These
errors can be attributed to many sources, and many would affect results of a census as well as a
survey.

These sources include:

 The sampling frame may have imperfections. For example, the frame may exclude some
housing units. Other housing units may be included in error or appear more than once.
This can cause undercoverage or overcoverage.
 Household rosters may inadvertently exclude persons or include persons. Also, based on
scope definitions by age or other criteria, errors may cause a person to be inadvertently be
excluded or included. This may cause undercoverage or overcoverage that can vary,
particularly by demographic characteristics.
 Questions may not be designed properly, though extensive testing makes this a minor
issue for CPS.
 Responses to some items may be incorrect or may be missing entirely. For example,
questions may be misinterpreted, or a person giving information about another may not
have sufficient knowledge to provide correct information.
 Housing unit nonresponse may introduce bias. No response is obtained from about 10%
of the housing units determined to be eligible (in scope) for a number of reasons. For
example, some housing units refuse to cooperate, and no successful contact is made with
others. If responding households have different characteristics than nonresponding
households, the estimates may be biased.
 Statistical methods to fill in for missing data or to correct for housing unit nonresponse
are imperfect.
 Ratio estimation (used in second-stage weighting and composite weighting) is known to
introduce small biases.
 CPS composite estimation is affected by month-in-sample (MIS) bias. One trait of MIS
bias is that the estimated unemployment rate for a panel (or rotation group) is highest
when it is first included in the CPS.
 There can also be errors made by interviewers, even falsification. These errors are
minimized by thorough training, use of laptops for administering interviews, routine
quality control checks, and a reinterview program.
 Data transmission and processing errors are possible, but should be eliminated by CPS
quality control procedures.

The full extent of nonsampling error in the CPS is unknown. The effect is small on estimates of
relative change, such as month-to-month change. The effect is small on means and rates,
particularly unemployment rates. Estimates of monthly levels tend to be affected to a greater
degree. More information can be found in Current Population Survey Design and Methodology
(Technical paper 66, October 2006) online at https://www.census.gov/prod/2006pubs/tp-66.pdf.

2
Sampling error and confidence intervals
When a sample, rather than the entire population, is surveyed, estimates differ from the true
population values that they represent. The component of this difference that occurs because
samples differ by chance is known as sampling error, and its variability is measured by the
standard error of the estimate. A sample estimate and its estimated standard error can be used to
construct confidence intervals; when these estimates are unbiased, the statistical properties of
confidence interval “coverage” are known.

The following are examples of confidence intervals:

 A 90% confidence interval is the range from 1.645 standard errors below the estimate to
1.645 standard errors above the estimate. The true population value is unknown, but
there is an approximate 90% probability that the interval includes or “covers” the true
population value.
 A 95% confidence interval is the range from 1.96 standard errors below the estimate to
1.96 standard errors above the estimate. The true population value is unknown, but there
is an approximate 95% probability that the interval includes or “covers” the true
population value.

Confidence interval statements like these are approximately true for the CPS, and this is
especially so for rates and changes over time. For a more complete explanation of confidence
interval coverage refer to a standard survey methodology text, such as chapter 1.7 in the 3rd
edition of Sampling Techniques by William G Cochran (Wiley, 1977).

When examining differences between estimates, the question often asked is: are the two
estimates significantly different? The difference may be for the same characteristic at two
different time periods: is the change significantly different? The difference may be between
different subpopulations in the same time period, such as the unemployment rate for men versus
the unemployment rate for women. Given an unbiased estimate of difference/change and an
unbiased estimate of standard error for that difference/change, confidence intervals can be
constructed and statements such as the following can be made:
 If zero lies outside of the 90% confidence interval from 1.645 standard errors below the
estimate to 1.645 standard errors above the estimate, the difference/change is said to be
significantly different at the 90% level of confidence. If zero lies in the interval, the
difference/change is said to be not significantly different at the 90% level of confidence.
 If zero lies outside of the 95% confidence interval from 1.96 standard errors below the
estimate to 1.96 standard errors above the estimate, the difference/change is said to be
significantly different at the 95% level of confidence. If zero lies in the interval, the
difference/change is said to be not significantly different at the 95% level of confidence.

Approximate standard errors and confidence intervals for CPS estimates can be calculated using
the instructions below and the parameter and factor tables (PF-1 through PF-16) available online
at https://www.bls.gov/cps/parameters-and-factors-for-calculating-standard-errors.xlsx. (These
tables mirror tables A-1 through A-16 of the monthly Employment Situation news release.) The
parameter and factor tables allow users to calculate approximate standard errors for a wide range

3
of estimated levels, rates, and percentages, and also changes over time. The parameters and
factors are used in formulas that are commonly called generalized variance functions.

The approximate standard errors calculated using the parameter and factor tables (PF-1 through
PF-16) are based on the sample design and estimation procedures as of 2015, and reflect the
population levels and sample size as of that year. Guidance for calculating standard error
estimates for historical CPS data may be found in the “Reliability of the estimates” section of the
Household Data Technical Notes (Employment and Earnings, February 2006, pages 12-19 or
printed pages 193-200) online at https://www.bls.gov/cps/eetech_methods.pdf.

Information presented here may be of use to researchers working with official BLS estimates as
well as those computing estimates using the public use CPS microdata files, available from the
Census Bureau's FTP site (https://thedataweb.rm.census.gov/ftp/cps_ftp.html) or from their
DataFerrett tool (https://dataferrett.census.gov/). Note that estimates generated using the public
use microdata files, except for a few topside estimates, will not match official BLS estimates.1

Calculating approximate standard errors and confidence intervals

A table showing changes in selected labor force indicators with statistical significance tests is
online at https://www.bls.gov/cps/documentation.htm#reliability.

Using parameter and factor tables PF-1 through PF-16

These tables give  and  parameters that can be used with formulas to calculate approximate
monthly standard errors for a wide range of estimated levels, proportions, and rates. Factors are
provided to convert monthly measures into approximate standard errors of estimates for other
periods (quarterly and yearly averages) and approximate standard errors for changes over time
(consecutive monthly changes, changes in consecutive quarterly and yearly averages, and
changes in monthly estimates 1 year apart).

The standard errors for estimated changes in level estimates from one time period to the next (for
example, one month to the next or one year to the next) depend more on the monthly levels than
on the size of the changes. Likewise, the standard errors for changes in rates (or percentages)

1
Beginning with data for January 2011, the Census Bureau incorporated additional safeguards in the CPS public use
microdata files to ensure that respondent identifying information is not disclosed. In general, respondents' ages were
altered, or "perturbed," in the public use microdata files to further protect the confidentiality of survey respondents
and the data they supply. One result of the measures taken to enhance data confidentiality is that labor force and
other estimates from the public use microdata files will no longer exactly match most estimates published by BLS,
which are based on internal, nonpublic-use files. Although certain topside labor force estimates for all persons will
continue to match published data—such as the overall levels of employed, unemployed, and not in the labor force—
estimates below the topside level (such as employment status by age, sex, race, and Hispanic or Latino ethnicity) all
have the chance of differing slightly from the published data. In addition, estimates calculated using characteristics
such as industry, occupation, hours worked, duration of unemployment, along with all other characteristics not
expressly listed above, are subject to such differences. All such differences should fall well within the sampling
variability associated with CPS estimates.

4
depend more on the monthly rates (or percentages) than on the size of the changes. Accordingly,
the factors presented in tables PF-1 through PF-16 are applied to the monthly standard error
approximations for levels, percentages, or rates; the magnitudes of the changes do not come into
play. Factors are not given for estimated changes between nonconsecutive months (except for
changes of monthly estimates 1 year apart); however, the standard errors may be assumed to be
higher than the standard errors for consecutive monthly changes.

IMPORTANT: CPS data are published as levels in thousands, so 4 million unemployed

people would be displayed as 4,000 in a table. Similarly, an entry of 2,460 in a CPS table
represents 2,460,000 people. When calculating standard errors and confidence intervals,
you must use the actual level at full precision, not the data as published in thousands.

Standard errors of estimated levels

The approximate standard error se(x) of x, an estimated monthly level, can be obtained using the
formula below, where  and  are the parameters from parameter and factor tables PF-1 through
PF-16 associated with a particular characteristic, and N is the total civilian noninstitutional
population 16 years and over.

𝑥2
𝑠𝑒(𝑥; 𝑁) = √( + 𝑁) (𝑥 − )
𝑁

Note that x is the monthly level (not in thousands).

Illustration – single month level

Assume that, in a given a month, there are an estimated 4 million unemployed men, and the
civilian noninstitutional population is 250 million. Obtain the appropriate  and  parameters
from table PF-1 (Men, 16 years and over; Unemployed). Use the formula for se(x; N) to
compute an approximate standard error on the estimate of x = 4,000,000 where N = 250,000,000.

 = 1050.17
 = 0.00000883

𝑠𝑒(4,000,000; 250,000,000)
4,000,0002
= √(1050.17 + (0.00000883 ∗ 250,000,000)) ∗ (4,000,000 − ( )) = 113,235
250,000,000

𝑠𝑒(4,000,000; 250,000,000) ≈ 113,000

5
Standard errors of estimated levels for quarterly or annual averages or changes over time
Tables PF-1 through PF-16 provide factors that can be used to compute approximate standard
errors of levels for other time periods or for changes over time. For each characteristic, factors
(f) are given for:
Consecutive month-to-month changes
Changes in monthly estimates 1 year apart
Quarterly averages
Changes in consecutive quarterly averages
Yearly averages
Changes in consecutive yearly averages

For a given characteristic, the correct factor from tables PF-1 through PF-16 is used in the
following formula, which also uses the  and  parameters from the same line of the table. A
three-step procedure for using the formula is given. The f in the formula is frequently called an
adjustment factor, because it appears to adjust a monthly standard error se(x). However, the x
and N in the formula are not monthly levels, but averages of monthly levels (see examples listed
below).

𝑥2
𝑠𝑒(𝑥; N; 𝑓) = 𝑓 ∗ 𝑠𝑒(𝑥; 𝑁) = 𝑓 ∗ √(𝛼 + 𝛽𝑁)(𝑥 − )
𝑁

Note that x and N are averages of monthly levels over the designated period.

Step 1. Average monthly levels appropriately in order to obtain x. Levels for 3 months are
averaged for quarterly averages, and those for 12 months are averaged for yearly averages. For
changes in consecutive levels, average over the 2 months, 2 quarters, or 2 years involved. For
changes in monthly estimates 1 year apart, average the 2 months involved.

Step 2. Calculate an approximate standard error se(x), treating the average x from step 1 as if it
were an estimate of level for a single month. For a given characteristic, obtain parameters  and
 from the applicable PF table.

Step 3. Determine the standard error se (x; f) on the average level or on the change in level.
Multiply the result from step 2 by the appropriate factor f. The  and  parameters used in step 2
and the factor f used in this step come from the same line in the corresponding PF table.

Illustration – consecutive month change in level

Continuing the previous example, suppose that in the next month the estimated number of
unemployed men increases by 150,000, from 4,000,000 to 4,150,000. Also suppose the civilian
noninstitutional population 16 years and over increases by 200,000, from 250,000,000 to
250,200,000.

6
Step 1. The averages of the two monthly levels for the estimate and the total population are x =
4,075,000 and N = 250,100,000.

Step 2. Apply the  and  parameters from table PF-1 (Men, age 16 years and over;
Unemployed) to the average x and N, treating them like an estimate and a population value for a
single month.

α = 1050.17
β = 0.00000883

𝑠𝑒(4,075,000; 250,100,000)
4,075,0002
= √(1050.17 + (0.00000883 ∗ 250,100,000)) ∗ (4,075,000 − )
250,100,000
= 114,290

𝑠𝑒(4,075,000; 250,100,000) ≈ 114,000

Step 3. Obtain f = 1.10 from the same row of table PF-1 in the column “Consecutive month-to-
month change,” and multiply the factor by the result from step 2 to calculate the standard error
for the change between the 2 months.

𝑠𝑒(150,000) = 𝑓 ∗ 𝑠𝑒(4,075,000; 250,100,000) = 1.10 ∗ 114,290 = 125,719

𝑠𝑒(150,000) ≈ 126,000

For an approximate 90-percent confidence interval, compute 1.645 * 125,719 = 206,808 

207,000. Next, subtract the number from and add the number to 150,000 to obtain an interval of
-57,000 to 357,000. This is an approximate 90-percent confidence interval for the true change,
and since this interval includes zero, one cannot assert at this level of confidence that any real
change has occurred in the unemployment level. The result also can be expressed by saying that
the apparent change of 150,000 is not statistically significant at a 90-percent confidence level.

Illustration – quarterly average level

Suppose that an approximate standard error is desired for a quarterly average of the Black or
African American employment level. Suppose that the estimated employment levels for the 3
months making up the quarter are 14,900,000; 15,000,000; and 15,100,000, and the
corresponding levels of the civilian noninstitutional population 16 years and over are
249,800,000; 250,000,000; and 250,200,000.

Step 1. The averages of the three monthly levels are x = 15,000,000 and N = 250,000,000.

Step 2. Apply the  and  parameters from table PF-2 (Black or African American; Total;
Employed) to the average x and N, treating them like an estimate for a single month.

7
  = -592.49
 = 0.00000816

𝑠𝑒(15,000,000; 250,000,000)
15,000,0002
= √(−592.49 + (0.00000816 ∗ 250,000,000)) ∗ (15,000,000 − ) = 142,863
250,000,000

𝑠𝑒(15,000,000; 250,000,000) ≈ 143,000

Step 3. Obtain f = 0.86 from the same row of table PF-2 in the column “Quarterly averages,” and
multiply the factor by the result from step 2 to calculate the standard error of the quarterly
average.

𝑠𝑒(15,000,000) = 0.86 ∗ 142,863 = 122,862

𝑠𝑒(15,000,000) ≈ 123,000

Illustration – consecutive quarter change in level

Continuing the example, suppose that, in the next quarter, the estimated average employment
level for Blacks is 15,400,000, based on monthly levels of 15,300,000; 15,400,000; and
15,500,000. This is an estimated increase of 400,000 over the previous quarter. Also suppose the
average population level in the next quarter is 250,600,000, based on monthly levels of
250,400,000; 250,600,000; and 250,800,000.

Step 1. The average of the two quarterly levels (15,000,000 and 15,400,000) is x = 15,200,000.
The average of the two quarterly population levels (250,000,000 and 250,600,000) is N =
250,300,000.

Step 2. Apply the  and  parameters from table PF-2 (Black or African American; Total;
Employed) to the average x and N, treating them like an estimate for a single month.

a = -592.49
b = 0.00000816

𝑠𝑒(15,200,000; 250,300,000)
15,200,0002
= √(−592.49 + (0.00000816 ∗ 250,300,000)) ∗ (15,200,000 − ) = 143,878
250,300,000

𝑠𝑒(15,200,000; 250,300,000) ≈ 144,000

Step 3. Obtain f = 0.79 from the same row of table PF-2 in the column “Change in consecutive
quarterly averages,” and multiply the factor by the result from step 2 to calculate the standard
error of the change in quarterly averages.

8
 𝑠𝑒(400,000) = 0.79 ∗ 𝑠𝑒(15,200,000) = 0.79 ∗ 143,878 = 113,664

𝑠𝑒(400,000) ≈ 114,000

For an approximate 95-percent confidence interval, compute 1.96 * 113,664 = 222,781 

223,000. Subtract the number from and add the number to 400,000 to obtain an interval of
177,000 to 623,000. The interval excludes zero. Another way of stating this is to observe that
the estimated change of 400,000 clearly exceeds 1.96 standard errors, or 223,000. One can
conclude from these data that the change in quarterly averages is statistically significant at a 95-
percent confidence level.

Standard errors of estimated rates, ratios, and percentages

As shown in the formula below, the approximate standard error se(p;y) of an estimated rate or
percentage p depends, in part, upon the number of persons y in its base or denominator.
Generally, rates and percentages are not published unless the monthly base is greater than 75,000
persons, the quarterly average base is greater than 60,000 persons, or the yearly average base is
greater than 35,000 persons. The α and β parameters are obtained from the correct PF table.
When the base y and the numerator of p are from different categories within the table, use the α
and β parameters from the PF table relevant to the numerator of the rate or percentage.

(𝛼 + 𝛽𝑦)
𝑠𝑒(𝑝; 𝑦) = √ 𝑝(100 − 𝑝)
𝑦

Note that y (not in thousands) is the base of percent p, and se(p; y) is in percent.

Illustration – single month percent

For a given month, suppose y = 156,000,000 employed persons. Of this total, 27,000,000, or p =
17.3 percent, are classified as part-time workers. Obtain the α and β parameters from table PF-9
(Full- or part-time status; Part-time workers) that are relevant to the numerator of the percentage.
Apply the formula to obtain:

𝛼 = −1636.59
𝛽 = 0.00002042

(−1636.59 + (0.00002042 ∗ 156,000,000))

𝑠𝑒(17.3; 156,000,000) = √ ∗ 17.3 ∗ (100 − 17.3)
156,000,000
= 0.119 ≈ 0.1 percent

9
For an approximate 95-percent confidence interval, compute 1.96 * 0.119 = 0.233 percent  0.2
percent. Subtract this from and add this to the estimate of p = 17.3 percent to obtain an
approximate confidence interval of 17.1 percent to 17.5 percent.

Standard errors of percentages for quarterly or annual averages or changes over time
Factors from tables PF-1 through PF-16 can be used to compute approximate standard errors on
rates, ratios, and percentages for other periods or for changes over time. As with levels, there are
three steps in the procedure for using the formula.

(𝛼 + 𝛽𝑦)
𝑠𝑒(𝑝; 𝑦; 𝑓) = 𝑓 ∗ 𝑠𝑒(𝑝; 𝑦) = 𝑓 ∗ √ 𝑝(100 − 𝑝)
𝑦

Note that p and y are averages of monthly estimates over the designated period,
and se(p; y; f) is in percent.

Step 1. Appropriately average estimates of monthly rates or percentages to obtain p, and also
average estimates of monthly levels to obtain y. For changes in consecutive averages, average
over the 2 months, 2 quarters, or 2 years involved. For changes in monthly estimates 1 year
apart, average the 2 months involved.

Step 2. Calculate an approximate standard error se(p; y), treating the averages p and y from step
1 as if they were estimates for a single month. Obtain the α and β parameters from the PF table
that apply to the numerator of the rate or percentage.

Step 3. Determine the standard error se(p; y; f) on the average percentage or on the change in
percentage. Multiply the result from step 2 by the appropriate factor f. The α and β parameters
used in step 2 and the factor f used in this step come from the same line in the appropriate PF
table.

Illustration – consecutive month change in percentage

Continuing the previous example, suppose that, in the next month, there are 156,600,000
employed persons, and that 28,000,000, or 17.9 percent, are part-time workers.

Step 1. The month-to-month change is 0.6 percentage point—that is, the share of the employed
who worked part time changed from 17.3 percent to 17.9 percent over the month. The average
of the two monthly percentages of 17.3 percent and 17.9 percent is needed (p = 17.6 percent), as
is the average of the two bases of 156,000,000 and 156,600,000 (y = 156,300,000).

Step 2. Apply the α and β parameters from table PF-9 (Full- or part-time status; Part-time
workers) to the averaged p and y, treating the averages like estimates for a single month.

𝛼 = −1636.59
𝛽 = 0.00002042

10
 (−1636.59 + (0.00002042 ∗ 156,300,000))
𝑠𝑒(17.6; 156,300,000) = √ ∗ (17.6) ∗ (100 − 17.6)
156,300,000
= 0.120 ≈ 0.1 percent

Step 3. Obtain f = 0.99 from the same row of table PF-9 in the column “Consecutive month-to-
month change,” and multiply the factor by the result from step 2.

𝑠𝑒(0.6 percent) = 0.99 ∗ 0.120 = 0.119 percent ≈ 0.1 percent

For an approximate 95-percent confidence interval, compute 1.96 * 0.119 = 0.233  0.2 percent.
Subtract this from and add this to the 0.6-percentage point estimate of change to obtain an
interval of 0.4 percent to 0.8 percent. Because this interval does not include zero, it can be
concluded the change is statistically significant at a 95-percent confidence level.

Parameter and factor tables PF-1 through PF-16

An Excel file with the parameter and factor tables is available online at
https:www.bls.gov/cps/parameters-and-factors-for-calculating-standard-errors.xlsx.

These PF tables mirror tables A-1 through A-16 of the monthly Employment Situation news
release.

A table showing changes in selected labor force indicators with statistical significance tests is
updated monthly at https://www.bls.gov/cps/documentation.htm#reliability.

11