From these data one may reasonably assume that retirement
for many rural Southerners will continue to be largely one of
relative economic deprivation, however improved their future
outlook may seem as a result of recent Federal social legislation.

 FACTORS AFFECTING RETIREMENT PROJECTIONS
 Various analyses of the S-56 data revealed how the attributes
affecting incomes of white and Negro families tended to cluster
differentially and to yield drastically unlike retirement expecta-
tions.9 For example, costs of home upkeep, medical care, life
insurance, and leisure activities, as well as net equities owned
by householders, differed quite generally from state to state and
according to family income levels, but more so for white than
for Negro families (2).
 Lack of homogeneity in the spending and saving habits of
families was apparent in some aspects of living, yet a consider-
able degree of similarity prevailed. To illustrate, the average
cost of home upkeep deviated according to state of residence and
family income for both whites and Negroes, but the occupations
of the Negro male heads were also determining factors. For
white families, costs of medical care varied by state of residence,
the health of the husband, and family structure; for Negro
families only in relation to family incomes. In total assets
accumulated, significant differences were noted for both white
and Negro families according to state of residence and family
income. Among whites, but not among Negroes, amounts of
total assets owned also differed according to farm or nonfarm
residence, home tenure, and occupation of the husband.
 Predictive attributes were difficult to isolate, yet six were
 identified as being of significance for white families, and three
 for Negro families, only two of which were common to both
 whites and Negroes.10 In general, more favorable retirement

 'Least-squares analyses, which contained five covariates, were run at
 the University of Georgia Computing Center. The covariates were (1)
 number in family, (2) formal education of husband, (3) formal education
 of wife, (4) age of husband, and (5) age of wife. It was assumed that,
 for a given race, each covariate had an identical effect on the dependent
 variables in all states. In addition, sums and averages were calculated
 and linear multiple regression analyses completed at the University of
 Florida Computing Center.
 'OPredictive variables were identified by linear multiple regression
 analyses. Regression equations were estimated in a step-wise manner. The
 criterion used for the inclusion of a term in the model developed was based
 on the reduction in the residual mean square. No term was added to it
 unless the reduction was significant at the 90 per cent (90%) level of