On some job records on the 1996 cross-sectional files, the values in the TSJDATE#, TJBOCC#, and EJBIND# variables* may be incorrect. The affected records are those for any job (for clarity, we will call it "Job K") that: (1) was first reported in a wave in which the job holder had at least one other job ( "Job K+1" ); and (2) was held in at least one subsequent wave.

For such jobs, the values of TSJDATE#, TJBOCC#, and EJBIND# are correct on the cross-sectional file for the wave in which Job K was first reported, but may be incorrect on subsequent waves. (The initially-reported values of these variables for Job K should have been assigned to them in any subsequent wave in which Job K was held; for the affected records, however, the subsequent values were assigned from Job K+1 rather than from Job K.)

* These variables are defined as:

TSJDATE#: Starting date of Job # (When did ... start this job?)

TJBOCC# : Occupation classification code of Job #

EJBIND# : Industry code of Job #

The universe for the TPRFTB# variable (Net profit or loss: "What is your estimate of the net profit or loss, that is, the difference between gross receipts and expenses, during the reference period?") was incorrectly programmed. The correct definition is:

EINCPB# not equal 1 and EHPRTB# =1 ( where # represents 1, 2).

Values for this variable for observations not in this universe should be replaced with "0".

The data for the TSBDATE# variable (Date operation of business began: "When did ... begin operating this business?") in the files for Wave 1 and Wave 2 were corrupted by an error in the public-use file-creation process.

Do not use this variable for these two waves.

While continuing to analyze the SIPP 1996 cross-sectional files, we discovered that we incorrectly assigned zero-dollar amounts (that is "no earnings") to the TPM#SUM variables* on the cross-sectional files for some persons in the sample who had jobs, but whose information on earnings was not captured. The problem may have affected around 1.5 percent of the observations in the unweighted monthly earnings distributions; its impact upon various subsets of these distributions (for example, persons with earnings under $1000 in the month) is proportionately greater.

We are investigating the scope and likely impact of the problem.

*This variable is described in the data dictionary as: "Earnings from job received in month #:What was ...'s gross pay before deductions in month # (where # represents 1,2,3, or 4)?"