search across all the following databases:
Data and documentation
Common questions and answers.
Entire collection of data resources.
Users can match March CPS files across years. There are two reasons a user might wish to do so. The first is to create a larger file to enable more detailed analysis of small areas such as states or cities or to analyze small populations such as Asians or Mexicans. The second is to create a "longitudinal" file where respondents can be followed at two points in time.
One will almost always want to rectangularize the CPS data during the matching procedure as the "month in sample" variable is a household variable and the level of analysis is likely to be at the person level. See the Archive Notes on Extract for instructions on how to rectangularize data.
A respondent is in the CPS sample rotation for four months, then out for eight months, and then in for another four months. There is a household variable in the CPS called "month in sample." It ranges from 1 to 8 keeping track of the eight months the household is in the sample rotation.
I. Matching Files to Create a Larger CPS file
Select households which are in months 5-8 in year (x) and which are in months 1-4 in year (x+1). This results in a file with households that do not overlap in years (x) and (x+1).
II. Matching Files to Create a Longitudinal CPS file
Select households which are in months 1-4 in year (x) and which are in months 5-8 in year (x+1).
This results in a file where the same housing unit is in the sample in both year (x) and year (x+1). However, since it is possible that the residents of the housing unit have changed, it is necessary to perform additional matches to insure cross-year resident compatability.
One needs to work out what constitutes a match across time. Some changes across time are legitimate. For instance age of the respondent or household members can and usually will change. An older child may move out of or back into the household. Marital changes may cause the head of the household to be married and male in year (x) and divorced or widowed and female in year (x+1 ).
Details for Matching
The following is the information needed to uniquely identify each housing unit in each sample rotation. Because of changes in CPS procedures, the available information for matching housing units is not always identical.
|1968-1971||Variables: Random Cluster Code (F6-10) and Serial Number (F11-14)|
|1971-1972||Changes in CPS clustering procedures and the accompanying change of household identification numbers prevent matching 1971 and 1972 March CPS files.|
|1972-1973||The 1972 files uses 1960 random cluster codes while the 1973 file uses 1970 random cluster codes, thus precluding the matching of records.|
|1973-1975||Variables: Random Cluster Code (F7-11), Segment Number (F12-16), and Serial Number (F217-F218)|
|1975-1976||Variables: 1975: Random Cluster Code (F7-11), Segment Number (F12-16), and Serial Number (F217-F218)|
|Variables: 1976: Random Cluster Code (H35-39), Segment Number (H40-43), and Serial Number (H44-45)|
|1976-1977||Matching is not possible because variables required for matching are in a different format each year.|
|1977-1985||Variable: Household Identification Number (H18-29)|
|1985-1986||Matching is not possible because the 1986 file is based entirely on the 1980 census design sample.|
Variable: Household Identification Number (H320-331)
One can find this information in the technical documentation for the March CPS. The information for 1986 Forward is incorrect in the technical documentation. However, the above will not hold true forever. At some point the following statement will occur: "matching is not possible because the 199- file is based entirely on the 1990 census design sample." Thus, be sure to check the technical documentation for the most recent CPS files.