blocks in an urban area, or a geographic segment well defined by
roads, streams, or other landmarks in a rural area.
In some
locations, addresses were selected at the central survey office
from listings previously compiled by interviewers. In other
locations, interviewers listed addresses and selected specific
households during the same visit to the sample location, following
explicit instructions for both the listing and the selection
processes. Once households were selected, no replacements or
substitutions were permitted.
Oversampling the Poor.
Locations
in the lowest socioeconomic quartile (identified either on the
basis of percent of families below the poverty level in 1969, or
average value of housing units in the area) were selected with
higher probability than were other locations. The purpose of the
oversampling was to include in the sample a larger number of poor
households than would have been included in a proportionate sample
of the total household population. In the analysis of survey data,
weights were used to compensate for differences in probability of
selection.
Response Rates
Altogether,
2,224 households (occupied housing units) were included in the
survey sample. Interviewers normally made up to four visits to
sample households, as necessary, to complete the interview.
Interviews were completed for 1,455 households, or 65 percent of
those assigned. Interview completion rates were highest in
nonmetropolitan areas and lowest in central cities of metropolitan
areas.
Weighting Procedures
Weights
were calculated for each sample household to (1) compensate for
differences in probabilities of selection by socioeconomic
quartile, (2) adjust for differences in interview completion
experience in different sampling locations, and (3) expand data for
sample households to estimates for the total universe (all
households in the United States as of that time).
Survey
Reliability
All survey
data are, of course, subject to random sampling variability.
Sampling error figures are estimates of the differences that might
be expected between the survey results and results of a complete
enumeration using the same survey procedures. Sampling error may
have a substantial effect on survey figures, particularly when
results are reported for small subgroups of the total sample.
Results are
also subject to response and nonreporting errors. Weighting
procedures were used to adjust for variations in interview
completion experience in different interviewing locations, but no
additional corrections were introduced for nonresponse patterns.
Estimates based on the sample, however,