Step 9
Having excluded from the algorithm all cases flagged in Steps 4 and 7, prepare to re-
order all cases flagged in step 5 as having been wrongly bunched together in the same
pregnancy, or wrongly ordered chronologically.
The re-ordering not only has to move each outcome date to a new slot, but also every
one of the 29 variables associated with that particular baby (birth weight, outcome,
etc), as well as the 10 variables associated with that particular pregnancy (number of
babies in pregnancy, whether smoked, whether used contraception, etc). In particular,
the corresponding pregnum is set to a value of 1, indicating a single, not multiple
birth. As documented above, once a bogus ‘multiple’ pregnancy is separated out into
two or more single-baby pregnancies it is unfortunately not possible to tell which of
these pregnancies these latter variables (smoking/contraception, etc) relate to. A
decision was made to copy the same pregnancy data into each of the new pregnancy
slots; users should therefore be careful in the way these data are interpreted. The
cases affected are all those listed in Appendix 1.
A dummy set of variables in vector format of length 40 is set up so that all the 29
‘baby’ variables can be copied and indexed by the slot in which they erroneously
appeared.
Having been first copied to the dummy vectors, the original 29 variables are set to the
system-missing value in each of the 40 slots.
Similarly, dummy vectors of length 8 are set up for the ten ‘pregnancy’ variables; the
data are copied from the original variables, which are then set to system-missing.
Step 10
Complete the re-ordering process. Copy the data for the most recent baby into the
‘slot 1’ versions of all 29 variables, applying the ‘KEY1’ variable from Step 6 as the
index for the vector; then copy the data for the next most recent into the ‘slot 6’
versions by applying ‘KEY2’, etc., then ‘slot 11’ by applying KEY3’, etc. until all
babies carried have been accounted for.
We know that all such pregnancies are single outcomes, as the multiples and dubious
cases have been excluded from the algorithm. So the pregnancy variables in slots 2-5,
6-10, 12-15, ... etc are bound to remain system-missing. It should be noted that for
these cases the re-ordering addresses both problems outlined in (A)-(C) above (i.e. the
‘bunching’ as well as the ‘out-of-order’ problem).
Similarly, copy the data for eight of the ten ‘pregnancy’ variables into the correct
positions. Of the other two, pregnum is set to 1 (as multiple pregnancies have been
excluded) and morepreg is set to 1 or 2 depending on a machine-check as to whether
other earlier pregnancies exist.