September 17, 1979
MEMO #49

Construction of new best measures of 1957 parental income.

    As described in COR #315-Addendum III, a number of runs were made to deter-
mine how to reconstruct the best measures of 1957 parental income (BMPINI1 and
BMPIN2).  These variables were originally created by using information from
PI5760 (Wisconsin tax data information) and YFML57 (1975 self-report on 1957
income).  When PI5760 was present, its value was used; when it was missing,
YFML57 was used.  Since the use of self-report data seemed to produce misleading
results, it was felt that some sort of transformation of YFML57 would be more
appropriate.  After trying linear and nonlinear transformations, it was decided
that a linear transformation worked about as well as a log-log function.  BMPIN1
and BMPIN2 were, therefore, reconstructed as follows:

(A)  When PI5760 was present, its value was used.

(B)  When PI5760 was missing and YFML57 was present, the value used was 
     equal to:

              26.56871 + .3987572 * YFML57.

    (This is based upon a regression of PI5760 upon YFML57 for 2851 male respon-
     dents in the 1975 survey who reported PI5760 and for whom YFML57 had been
     observed.)

(C)  BMPIN1 was truncated at $99,800, while BMPIN2 was truncated at $15,000.


    The correlation between PI5760 and YFML57 was .427, with an adjusted
    R square of .182.  The standard error of beta was .01582.

    While the best measures were truncated at $99,800 and $15,000, it is impor-
tant to note that, in almost all analyses using 1957 parental income, the cutoff
point has been roughly $15,000 in 1957-60 dollars (for example, Education,
Occupation, and Earnings by Sewell and Hauser).  In general, the user will want
to use $15,000 as the maximum parental income.