The nonpartisan Urban Institute publishes studies, reports, and books on timely topics worthy of public consideration. The views expressed are those of the authors and should not be attributed to the Urban Institute, its trustees, or its funders.
This report is available in its entirety in the Portable Document Format (PDF).
Chapter 1
Introduction and Background
This report completes the methodological discussion of 1997 NSAF estimation that was begun in
report no. 3 in this series. Here our goal is to describe how we brought the 1997 NSAF up to
census undercount-adjusted control totals. In report no. 3, a description was given of the first
estimation approaches used for the 1997 survey. (Report no. 3 also explains how we brought the
survey up to census-level controls.)
The use of census-undercount controls allows researchers to more readily compare NSAF results
with those of most other large national surveys, like the Current Population Survey (CPS) or the
National Health Interview Survey (NHIS). As will be seen in chapter 2, this re-estimation effort
also allowed us to refine some other steps taken earlier.
Chapter 2 explains the new weighting procedures used to revise the weights described in report
no. 3 and how we produced new weights that could be used to make undercount-adjusted
estimates from the 1997 NSAF data. The most significant revision was a change in the population
control totals used in the last step of raking.
The revised weights developed here have controls that reflect the undercount of the population
from the 1990 census. The undercoverage represents the population not enumerated in the 1990
census. This population is carried forward to 1997 through special adjustments on specific agerace
groups. About four million persons residing in the United States on census day were not
enumerated in 1990. As already noted, the result of the undercount adjustment is to make NSAF
estimates more consistent with estimates from other large national surveys.
As described in chapter 2, two other adjustments in the weighting procedures were implemented.
One of these was to better account for households that did not have anyone who spoke either
English or Spanish. There was also an attempt to improve national estimates of American Indians
and Asian/Pacific Islanders. These two ?fixes? were implemented prior to the third and final fix,
which was the use of the new control totals. We believe that for the low-income population,
especially the minority population, the new 1997 NSAF results will be somewhat improved by the
above three refinements.
In the last section of chapter 2, we describe several additional sets of weights that were developed to support special analyses. As was done in Report no. 3, we simplify the presentation by showing the results for one study area, Michigan.
An important concern in any survey is the extent to which all the members of the target
population have an opportunity to be interviewed. This is called survey coverage. Any omission
of units from the sampling process is called undercoverage. Chapter 3 considers coverage in the
1997 NSAF.
In chapter 3, coverage ratios are defined and examined. First, overall NSAF coverage ratios for
different population groups are reviewed. The chapter then goes on to examine the household and within household-components of coverage. Comparisons of NSAF coverage to the coverage
in other household surveys are also made.
National coverage ratios in the NSAF are 94 percent for children and 86 percent for adults.
Coverage in the CPS is about the same for children and somewhat better for adults. The poorer
adult coverage in the NSAF may be due partly to incorrectly classifying some childless households
as having only persons 65 and over. Possible weaknesses in the nonresponse adjustment may also
be a factor.
Coverage ratios for blacks are noticeably lower (70 percent for children and 67 percent for adults) than for Hispanics and for non-Hispanic non-blacks. Black coverage is also lower than in the CPS. Again, more examination of the NSAF nonresponse adjustments may be a place to look for
possible causes.
Examining the components of undercoverage, it appears that the undercoverage in the NSAF is
primarily whole-household undercoverage rather than within-household undercoverage.
However, no firm conclusion about the magnitude of the two components is possible.
When the original Snapshots publication was under preparation, there was concern about what differences using undercount-adjusted results would make. The belief was widely held, however,
that for NSAF statistics expressed as percentages (as they were in Snapshots), the change in
conventions would make little difference. But we needed to check this and chapter 4 does just
that.
To this end, in chapter 4 each of the January 1999 Snapshots tables has been reproduced ? both as originally published and after adjusting the survey as described in chapter 2. Graphical and tabular comparisons are made. These confirm, it might be added, that the conjecture of only
minor differences was basically correct.
We also took the opportunity in chapter 4 to look at the effect of editing differences on the
Snapshots results. Over the year since the Snapshots were prepared the 1997 survey files were being changed as more was learned. We wanted to let researchers see what impacts this had on the original results. The data changes, also as we conjectured, have proved to be mostly of minor importance. This is not to say that individual researchers do not have to beware of further data problems. Some undoubtedly exist on the public use files being released. Still, we are relatively comfortable with the work done on the 1997 NSAF. It should be a sound baseline for comparison with the 1999 survey results coming out this year.
One final comparison in chapter 4 looks at what would have happened if we had not done a
special sample of non-telephone households and had relied only on a random-digit-dialing (RDD)
design. Large differences exist between this RDD approach and the original Snapshots estimates,
especially for households under 200 percent of poverty. An RDD-only approach would not have
worked at all well for this most important group.
The basic results about NSAF coverage have already been summarized in report no. 1 in the 1997
methodology series. Report no. 3 in this series has already been mentioned. The papers on
estimation published separately in report no. 16 may be worth further examination.
This report is available in its entirety in the Portable Document Format (PDF).
The nonpartisan Urban Institute publishes studies, reports, and books on timely topics worthy of public consideration. The views expressed are those of the authors and should not be attributed to the Urban Institute, its trustees, or its funders.
Usage, posting and reprint of materials on the UI web site:
Most publications may be downloaded free of charge from the web site in PDF format. This information may be used and copies made for research, academic, policy or other non-commercial purposes. Proper attribution is required.
Copyright of the written materials contained within the Urban Institute website is owned or controlled by the Urban Institute. Posting UI research papers on other websites is permitted subject to prior approval from the Urban Institute—contact paffairs@urban.org.
If you are unable to access or print the PDF document please contact us or call the Publications Office at (202) 261-5687.
Latest Research
Browse by Topic:
PDF
Printer-friendly summary
