Linking and the CPS
- CPS Rotation Pattern
- CPSID and CPSIDP
- ASEC vs March Basic Monthly Survey
- CPSIDP vs MARBASECIDP
- Weighting Linked Datasets
- Unlinkable Samples
- Further Reading
CPS Rotation Pattern
Households are in the CPS for a total of eight months. Respondents are included in the CPS for four consecutive months, then are not interviewed for the next eight months. Respondents are then included in the CPS again for the next four months.
This video tutorial explains the CPS panel rotation pattern and introduces the CPS Rotation Pattern Exploration Station (RoPES)
The CPS Rotation Pattern Exploration Station is a tool to help users visualize what samples have linkable records.
CPSID and CPSIDP
CPSID is a 14-digit number that uniquely identifies households across months. CPSID is made up of the year in which the household first enters the CPS,
the month in which the household first enters the CPS,
and the serial number that uniquely identifies the household within a given month.
CPSIDP is used to link persons across months. For each person in the household, CPSIDP appends a unique person identifier to the end of their household's CPSID value.
CPSIDP identifies people everytime they appear in the Current Population Survey. The variable MISH indicates the respondent's "month-in-sample" - that is where in the CPS rotation pattern a respondent falls in a given month. MISH has values of 1-8; persons have MISH values of 1-4 before the 8-month break and values of 5-8 after this break.
ASEC vs March Basic Monthly Survey
When constructing CPS panels, we recommend using either the March Basic Monthly Survey (BMS) or the ASEC from any given year. Because the ASEC includes all individuals in the March BMS from a given year as well as oversample individuals from different months of the CPS, including both the March BMS and the ASEC as part of a panel essentially results in two records from the same month for all individuals in both the BMS and the ASEC. Rather, users should choose either the March BMS or the ASEC to include as part of their panel. The ASEc includes nearly all information from the March BMS in addition to variables from the Annual Social and Economic supplement.
Users should note that when performing this merge, some individuals in the ASEC will not to match to the March BMS because of the ASEC oversamples. ASEC oversample records are indicated by the variable ASECOVERP.
While we recommend choosing either the March BMS or the ASEC as part of a CPS panel, there are certainly instances when a user might want information from both files to use in the panel. For example, you may want information about respondent HEALTH or POVERTY, which are only available in the ASEC. In that case, we recommend first merging the ASEC onto the March BMS before linking to other BMS months you wish to analyze.
CPSIDP vs MARBASECIDP
CPSIDP should be used for all linking requirements across BMS months and to link ASEC to BMS months. MARBASECID will also correctly link the ASEC to the March Basic Monthly. However, it is not necessary to first link the ASEC to the March BMS before linking the ASEC information to other months; CPDIDP will make all possible links on its own. For more information on MARBASECIDP, see detailed documentation here.
Weighting Linked Datasets
The Bureau of Labor Statistics delivers cross-sectional weights, but they provide only limited longitudinal weights (PANLWT for linking adjacent months for gross flows analysis). IPUMS-CPS has generated longitudinal weights for several different types of links. These weights are based on the cross-sectional weights and arrived at by iterative proportional fitting (ipf) or raking. For example, we create new weights for the set of people who link from time 1 to time 2 based on the population counts of the people who were eligible to link from time 1 to time 2. The eligible counts are based on the intersections at time 1 of
These are the same characteristics that the CPS raking is based on in the second stage for cross-sectional weights.
|Longitudinal Weight Variables|
|LNKFW1MWT||Longitudinal weight for two adjacent months (BMS only)|
|LNKFW1YWT||Longitudinal weight for two adjacent years|
|LNKFW8WT||Longitudinal weight for a full 8-month panel (BMS only)|
|LNKFWMIS14WT||Longitudinal weight for the first four months of the panel rotation (MISH 1-4) (BMS only)|
|LNKFWMIS45WT||Longitudinal weight for linking across the 8-month survey break (MISH 4 to MISH 5) (BMS only)|
|LNKFWMIS58WT||Longitudinal weight for the second four months of the panel rotation (MISH 5-8) (BMS only)|
|Create your own||Stata replication files for creating LNKFW1MWT for January 2012|
Users should note that availability of these variables in a given sample is contingent on the availability of the samples required to make the given type of link. For example, LNKFW1MWT will not be available for January of 2019 until February 2019 data is also available through IPUMS-CPS. Similarly, LNKFWMIS14WT will not be available for January of 2019 until February, March, and April 2019 basic monthly data are also available through IPUMS-CPS.
Due to changes in the system for numbering households in the CPS, the some samples cannot be linked using household identification numbers. The following samples are unlinkable:
- 1976 samples cannot be linked to 1977 samples containing supplements
- 1976 samples cannot be linked to 1978 samples
- 1977 samples samples containing supplements cannot be linked to 1977 samples without supplements
- 1977 samples cannot be linked to 1978 samples
- samples before May 1985 cannot be linked to samples after May 1985
- samples before April 1995 cannot be linked to samples after April 1995
Longitudinal weight variables are unavailable in samples for which the linking type is impossible.
Papers about linking the CPS
J.A. Drew, S. Flood, J.R. Warren, Making full use of the longitudinal design of the Current Population Survey: Methods for linking records across 16 months, Journal of Economic and Social Measurement 39 (2014).
S. Flood and J. Pacas, Using the Annual Social and Economic Supplement as Part of a Current Population Survey Panel, Minnesota Population Center Working Paper No. 2016-4, (September 2016).
B.C. Madrian and L.J. Lefgren, An approach to longitudinally matching Current Population survey (CPS) respondents, Journal of Economic and Social Measurement 26 (2000).
S. Feng, The Longitudinal Matching of Current Popu.ation Surveys: A Proposed Algorithm, Journal of Economic and Social Measurement 27 (2001).
S. Feng, Longitudinal Matching of Recent Current Population Surveys: Methods, Non-matches and Mismatches, Journal of Economic and Social Measurement 27 (2001).
A. Katz, K. Teuter and P. Sidel, Comparison of Alternative Ways of Deriving Panel Data from the Annual Demographic Files of the Current Population Survey, Review of Public Data Use 12(1)(March 1984).
A. Pitts, Matching Adjacent Years of the Current Population Survey, Unpublished Manuscript, Los Angeles, CA: Unicon Corporation, 1988.
Papers that use CPS Longitudinally
C.L. Shandra, Disability and social participation: The case of formal and informal volunteering, Social Science Research 68 (2017).
K.A. Couch and R. Fairlie, Last hired, first fired? black-white unemployment and the business cycle, Demography 47 (2010).
B. Ward, A. Myers, J. Wong, and C. Ravesloot, Disability Items From the Current Population Survey (2008-2015) and Permanent Versus Temporary Disability Status, AJPH 107(5) (May 2017).
B. Hardy, T. Smeeding, and J. Ziliak, The Changing Safety Net for Low-Income Parents and Their Children: Structural or cyclical Changes in Income Support Policy?, Demography 55 (2018).