Examples

• In this section I present some examples from the current form of the newpan file
• The criterion used to determine whether a seam change constituted a ``true'' employment status change in these examples is maximally permissive:
  • Broad employment status (employed; unemployed; inactive) must match
• In some ways this is clearly too permissive: many other rules could be implemented

• Seams occur between pairs of records where the spells overlap or abut exactly, where the first record is a wINDRESP

• Source encoding
  • Wave number $\times$ 10 plus 0 for wINDRESP or 1 for wJOBHIST, plus wJOBHIST record number divided by 100
  • Thus 10, 20, 30, etc. , for wINDRESP
  • 11.04, 11.03, 11.02, 11.01 for records from AJOBHIST in temporal order (earlier spell gets higher record number)

• Next page shows example data for three individuals

  
  

       pid source spcens episno epdate epend flag spdate  spend   empstat  soc   sic  seamcode 
  10127666     10      0      1   1083  1140    0   1083   1103 employed/  371  9320       n/a  
  10127666     20      0      1   1083  1140 1083   1104   1114 employed/  371  9320 tight ext  
  10127666     30      0      1   1083  1140 1083   1115   1126 employed/  371  9320 tight ext  
  10127666     40      0      1   1083  1140 1083   1127   1137 employed/  371  9320 tight ext  
  10127666  51.01      0      1   1083  1140 1083   1138   1140 employed/  371  9112 tight emp  
  10127666     50      0      2   1141  1184    0   1141   1151 employed/  371  9112       n/a  
  10127666     60      0      2   1141  1184 1083   1152   1161 employed/  371  9112 tight ext  
  10127666     70      0      2   1141  1184 1083   1162   1174 employed/  371  9611 tight emp  
  10127666  81.01      0      2   1141  1184 1129   1175   1184 employed/  371  9112 tight emp  
  10127666     80      0      3   1185  1198    0   1185   1186 employed/  199  9112       n/a  
  10127666     90      1      3   1185  1198 1185   1187   1198 employed/  232  9112 tight emp  
  
  10129901     10      0      1   1072  1112    0   1072   1103 employed/  363  9510       n/a  
  10129901  21.01      0      1   1072  1112 1072   1104   1112 employed/  420  9510 tight emp  
  10129901     20      0      2   1113  1130    0   1113   1114 employed/  420  9510       n/a  
  10129901     30      0      2   1113  1130 1114   1115   1126 employed/  363  9510 tight emp  
  10129901  41.01      0      2   1113  1130 1096   1127   1130 employed/  363    -9 tight emp  
  10129901     40      1      3   1131  1138    0   1131   1138 employed/  363  9510       n/a  
  
  10539743     20      0      1    981  1148    0    981   1113 f-t educa   -8    -8       n/a  
  10539743     30      0      1    981  1148  981   1114   1125 f-t educa  720  6410 tight emp  
  10539743     40      0      1    981  1148  981   1126   1137 f-t educa  720  6410 tight ext  
  10539743  51.01      0      1    981  1148 1137   1138   1148 f-t educa   -8    -8 tight emp  
  10539743     50      1      2   1149  1151    0   1149   1151 employed/  622  6620       n/a  
  10539743     60      0      3   1152  1174 1137   1152   1162 f-t educa   -8    -8 empstat m  
  10539743     70      1      3   1152  1174 1137   1163   1174 f-t educa   -8    -8 tight ext  
  10539743  91.04      0      4   1185  1186    0   1185   1186 employed/  990  9900 apparent   
  10539743  91.03      0      5   1187  1186    0   1187   1186 employed/  594   100       n/a  
  10539743  91.02      0      6   1187  1191    0   1187   1191 employed/  630  7700 tight emp  
  10539743  91.01      0      7   1192  1196    0   1192   1196 employed/  173  6670       n/a  
  10539743     90      1      8   1197  1199    0   1197   1199 unemp/job  430  7902       n/a
  

  • In this example we see a number of features
    • Source information showing wINDRESP and sometimes multiple wJOBHIST records, e.g., for PID 10539743 before wave 9
    • Seam-codes indicating most sort of matches: n/a, tight ``extensive'' matches where employment status and SOC/SIC match, matches where employment status matches closely but SIC or SOC do not, employment status mismatches (10539743 between waves 5 and 6) and gaps (10539743 missed wave 8)
    • We also see apparent negative durations: 10539743 in record 91.03 - end date is defined as the start of next spell minus one, so if two spells begin in the same month, the first had EPEND apparently before EPDATE
    • We can also see how ``splits'' are put together as episodes, with either wJOBHIST records reporting explicit ends or unacceptable state changes (e.g., employed to full-time education) marking the transitions
    • Note also how episode dates are drawn from spell dates; FLAG contains the reported start, which is ignored when it appears to refer to a spell whose start has already been reported

• The distribution of SEAMCODE:

            seam status |      Freq.     Percent        Cum.
  ----------------------+-----------------------------------
                    n/a |      46433       37.83       37.83
  tight extensive match |      42818       34.88       72.71
    tight empstat match |      23947       19.51       92.22
    loose empstat match |       2553        2.08       94.30
       empstat mismatch |       6537        5.33       99.63
           apparent gap |        454        0.37      100.00
  ----------------------+-----------------------------------
                  Total |     122742      100.00
  

• ``n/a'' means this split explicitly represents the start of a new spell, as the previous one was a wJOBHIST record explicitly describing the end of the previous spell
• Many of the ``tight extensive matches'' are non-employed spells, whose SIC and SOC are coded as -8
• Gaps represent spells whose start date is after the end date (or date of end of observation) of the previous spell

• In practice, we will often want to collapse newpan into a file with one record per episode:

  aggregate out="episode.sav"/break=pid,episno
    /epdate=first(spdate)
    /epend=end(spend)
    /empstat=first(empstat)
    /sic=first(sic)
    /soc=first(soc)
    /epcens=last(spcens).
  

• If we wish to work with employment status spells (e.g., ignore job-to-job changes) we can ignore episno:

  aggregate out="episode.sav"/break=pid,empstat/presorted
   . . .