1972-1994 General Social Survey Cumulative File
In the original National Science Foundation grant, support was given for a
modified probability sample. Samples for the 1972 through 1974 surveys
followed this design. This modified probability design, described below,
introduces the quota element at the block level. The NSF renewal grant,
awarded for the 1975-1977 surveys, provided funds for a full probability
sample design, a design which is acknowledged to be superior.
Thus, having the wherewithal to shift to a full probability sample with
predesignated respondents, the 1975 and 1976 studies were conducted with a
transitional sample design, viz., one-half full probability and one-half
block quota. The sample was divided into two parts for several reasons:
1) to provide data for possibly interesting methodological comparisons; and
2) on the chance that there are some differences over time, that it would
be possible to assign these differences to either shifts in sample designs,
or changes in response patterns. For example, if the percentage of
respondents who indicated that they were 'very happy' increased by 10 percent
between 1974 and 1976, it would be possible to determine whether it was due
to changes in sample design, or an actual increase in happiness.
There is considerable controversy and ambiguity about the merits of these
two samples. Text book tests of significance assume full rather than
modified probability samples, and simple random rather than clustered random
samples. In general, the question of what to do with a mixture of samples
is no easier solved than the question of what to do with the "pure' types.
Investigators who have applied statistical tests to previous General Social
Survey data should continue to apply those tests. Investigators who have
refrained from applying such tests may now want to perform analyses on the
probability subsample. This would, of course, reduce the number of cases
by one-hall Whatever choice investigators make, it should be remembered that
the two subsamples represent the same universe.
Having allowed for the appearance of all items in the transitional sample
design, the General Social Survey then switched to a full probability sample
for the 1977, 1978, 1980, and 1982-1993 surveys. The variable SAMPLE
(Col. 3431) can be used to separate the block quota and full probability
samples on the 1975 and 1976 surveys.
A similar split sample transition design was used in the 1983 survey to
measure the effect of switching from the 1970 sample frame to the 1980
sample frame. Half of the sample was drawn from the 1970 frame and half
was drawn from the 1980 frame. The variable SAMPLE (Col. 3431) separates
cases from these two sample frames. Again in 1993, a split sample
transition design was employed on the 1993 survey to measure the effect
Of switching from the 1980 sample frame to the 1990 sample frame. Half of
the sample was drawn from each frame. More details on the 1970, 1980 and
1990 sample frames as well as the block quota samples appear below.
The adult, household population of the United States covered about 97.3%
of the resident population of the United States in 1985. Coverage varies
greatly by age group. For those 18-24, 9.4% of the population in 1980
lived Outside of households (mostly in college dorms and military quarters).
Among age groups from 25 to 64 the only 0.8-1.4% of the population lived
outside of households. For those 75 and older 11.4% were in group quarters,
mostly in nursing homes and long-term care facilities. For more details on
the non-household Population see Living Arrangements of Children and Adults,
Census of Population, PC80-2-413, May, 1985.
As defined for the GSS in 1983-1987, 98% of the adult, household population
is English speaking. The number of non-English speakers excluded is
indicated in Table A.3. Spanish speakers typically make up 60-65% Of the
language exclusions. About a dozen languages make up the remaining
exclusions.
BLOCK QUOTA
The sample is a multi-stage area probability sample to the block or segment
level. At the block level, however, quota sampling is used with quotas
based on sex, age, and employment status. The cost of the quota samples is
substantially less than the cost of a full probability sample of the same
size, but there is, of course, the chance of sample biases mainly due to
not-at-homes which are not controlled by the quotas. However, in order to
reduce this bias, the interviewers are given instructions to canvass and
interview only after 3:00 p.m. on weekdays or during the weekend or
holidays. This type of sample design is most appropriate when the past
experience and judgment of a project director suggest that sample biases
are likely to be small relative to the precision of the measuring instrument
and the decisions that are to be made.
Selection of PSUs
The Primary Sampling Units (PSUs) employed are Standard Metropolitan
Statistical Areas (SMSAs) or non-metropolitan counties selected in NORC's
Master Sample. These SMSAs and counties were stratified by region, age,
and race before selection.(1)
Selection of SamRIe within PSUs
The units of selection of the second stage were block groups (BGs) and
enumeration districts (EDs). These EDs and BGs were stratified according
to race and income before selection.(2) The third stage of selection was
that of blocks. The blocks were selected with probabilities proportional
to size. In places without block statistics, measures of size for the
blocks were obtained by field counting. The average cluster size is five
respondents per cluster. This provides a suitable balance of precision
and economy.
Interviewer Instructions
At the block or segment level, the interviewer begins a travel pattern at
the first DU (dwelling unit) from the northwest corner of the block and
proceeds in a specified direction until the quotas have been filled.
The quotas call for approximately equal numbers of men and women with the
exact proportion in each segment determined by the 1970 Census tract data.
For women, the additional requirement is imposed that there be the proper
proportion of employed and unemployed women in the location. Again, these
quotas are based on the 1970 Census tract data. For men, the added
requirement is that there be the proper proportion of men over and under 35
in the location.
These particular quotas have been established because past experience has
shown that employed women and young men under 35 are the most difficult to
find at home for interviewing.
Sampling Error
Although the mean squared error cannot be estimated directly from a quota
sample, one can make estimates of sampling variability using procedures
such as those outlined by Stephan and McCarthy.(3) Past experience would
suggest that, for most purposes, this sample of 1,500 could be considered
as having about the same efficiency as a simple random sample of 1,000 cases.
In making this judgment concerning the design effect, we are concerned with
the "average' effect upon a large set of different variables of the
clustering of households at the last stage of selection. Any statement of
sampling error assumes that the bias in quota sampling due to the lack of
control over respondent availability is slight for the study under
consideration.
For those persons interested in investigating the within-sample variability
of these data, we have included a "sampling error code" (see Q. 643).
Information about the use of this code is available from the GSS project
staff at NORC.
Probability, 1970 Frame
The NORC national probability sample is a stratified, multistage area
probability sample of clusters of households in the continental United
States.(4) The selection of geographic areas at successive stages is in
accordance with the method of probabilities proportional to size (p.p.s.).
Furthermore, the clusters of households are divided into replicated
subsamples in order to facilitate estimation of the variance of sample
estimators of population characteristics.
At the first stage of selection, Standard Metropolitan Statistical Areas
(SMSAs) and nonmetropolitan counties covering the total continental United
States were grouped according to size strata within the nine Census regions.
All population figures and other demographic information were obtained
from 1970 Census reports. Within each size stratum grouping based upon
geographic location, or racial characteristics (or both), was accomplished
before selection. The final frame was further separated into zones or
"paper strata" of equal population size in order to facilitate the selection
of replicated subsamples of primary sampling units (PSUs).(5) The selection
of PSUs was designed to produce four independent subsamples of equal size.
The four subsamples were randomly combined to form two larger subsamples
of 101 PSUs each.(6) The large subsamples are thus internally separable
into two replicated subsamples for variance estimation purposes.
NORC has selected one of the two large subsamples described above to serve
as its principal frame of households for the remainder of the decade. The
PSUs fall into 89 distinct SMSAs and nonmetropolitan counties. (New York,
a very large SMSA, represents five PSUs, whereas the smaller counties
represent only one PSU.)
The second-stage procedure involved the direct selection of Census block
groups or enumeration districts (E.D.s) within SMSAs or counties,
eliminating the traditional intermediate stage of clustering selections
within urban places or county division. The increase in geographic
dispersion within the primary areas has a negligible effect on field costs.
Before selection, the Census tracts, minor civil divisions, and Census
county divisions containing the block groups and E.D.s were carefully
stratified by geographic location, income, and race, in order to maximize
the precision of sample estimation within a PSU. Block groups and E.D.s
were then selected with probabilities proportional to size in numbers
sufficient to satisfy survey demands for households expected throughout
the decade. Lists of the separate households contained in the second stage
blocks or E.D.s were constructed by field personnel or obtained from
directories. Thus, the principal NORC national probability sample is, in
effect, an inventory of identifiable households, each with a known probability
of selection.
In a typical sample survey with equal probability of selection for individual
households (i.e., a self-weighting sample), households at which interviews
will take place are probabilistically selected from the available lists of
addresses for blocks and E.D.s. The method of probabilities proportional to
size results in the assignment of approximately equal numbers of interviews
in each final stage cluster, which in turn leads to increased precision in
the estimation of overall population characteristics.
The NORC national probability frame, with its broad geographic dispersion,
its reserves of additional SMSAs and counties, and its built-in replication,
provides sufficient flexibility for application to a wide range of survey
tasks. Its design is based on the consideration of sampling problems that
NORC and other organizations have encountered in past surveys, and we believe
that it substantially eliminates many of these difficulties.
Probability, 1980 Frame
1980 National Sampling Frame
The 1980 frame was designed, selected, and listed jointly by NORC and the
Survey Research Center. The new frame was selected, in most cases, in two
stages; about one-fifth of the second stage units were subsampled, producing
a third stage.
Eighty-four PSUs were selected at the first stage. The PSUs consist of
counties, SMSAs, independent cities and, in New England, parts of counties.
Prior to selection, the United States was divided into PSUs; the PSUs were
then grouped into 84 strata. The strata were formed by grouping metropolitan
and non-metropolitan PSUs within each of the four Census regions. Within
each region, additional variables were used to define strata. The stratifying
variables included within-region geography and size; size was measured by
the 1980 Census count of occupied housing units. One PSU was selected from
each stratum using a controlled selection procedure. This procedure ensured
proportionate representation along certain control dimensions (such as
percentage Hispanic in the West). The exact control variables (like the
stratification variables) differed somewhat from region to region. Sixteen
strata contained only one PSU, which was selected with certainty. The
remaining 68 PSUs were selected with probability proportional to size
(measured in housing units).
The unit for second stage selection was the block or enumeration district
(ED). The number of secondary selections within a PSU depended in part on
the stratum size. The number of second stage selections listed for NORC's
national frame in the 16 PSUs selected with certainty ranged from 24 to six
selections for PSU. In the remaining 68 sample PSUs, six second stage
selections were listed. (The same number of second stage selections were
listed for SRC's national frame; further, both organizations retained a
similar number of second stage selections as a reserve for future use.)
All total, the new frame includes 562 secondary selections.
Prior to selection, the second stage units within each sample PSU were
sorted by county, by minor civil division (in some areas), by Census Tract
or ED number, and by block number. Counties were ordered within PSUs
according to size and geography (e.g., in SMSAs, the county containing the
central city came first, then counties containing nearby suburbs, and so on).
In twenty states, information was available on the size and median family
income of minor civil divisions (MCDs), which are governmental units below
the county level (such as cities or towns). Where this information was
available, we sorted the block and EDs by MCD and ordered the MCDs by size
and income.
Next, we sorted all blocks and EDs by Census Tract number and then by block
or ED number; these sorts establish a geographic ordering. The secondary
selections were made using systematic zone selection; the probabilities of
selection were proportional to size (measured in housing units). Each
secondary selection included at least 50 housing units.
In enumeration districts and blocks with a large number of dwelling units,
a third stage of selection was carried out. The block or ED was subdivided
into pieces which were "field counted" by field staff from NORC or SRC.
In a field count, an area is scouted and a rough count of the number of
housing units is made. Based on the field count we selected a piece of the
sample block or ED with probability proportional to its size.(7)
Comparison of the 1970 and 1980 Frames
Table A-1 summarizes the main differences between the old and the new frames.
Aside from the obvious differences in the number of selections at each stage
and in the measures of size (people vs housing),
--------------------------------------------------------------------------------
Table A-1: Main Differences Between Frames
--------------------------------------------------------------------------------
1970 Frame 1980 Frame
First Stage
-----------
Number of Selections 100 84
Measure of size Persons-- Housing--
--at Least 10,000 --at Least 4,000 units
Methods of Selection Systematic Zone Selection Controlled selection
Stratification (Implicit) Census 84 explicit strata
division, urbanization plus control variables
% Black
Second Stage
------------
Number of Selections 1800 (18 per PSU) 562
Unit BG or ED Block or ED
Measure of Size Person Housing units--at least 50
Method of Selection Systematic Zone Selection Systematic Zone Selection
Stratification (Implicit) income, (Implicit) County: size,
% Black geography; MCD (20 states):
income, size; CT, ED block:
geography
Third Stage
-----------
Unit "Segment" Part of a block or ED
(in most areas, no 3rd
stage)
Measure of Size Housing units-- Housing units--
--at Least 100 --at Least 50
Method of Selection one selection with one selection with
probability proportional probability proportional
to size to size
--------------------------------------------------------------------------------
the two frames differ in three main respects. First, the stratification
variables differ. In particular, percentage Black, which was used as an
implicit stratifier at the first and second stages in the old frame is no
longer used in the new sample (except in rural areas of the Deep South).
Instead, the new frame uses a variety of geographic control variables at
both the first and second stages. Second, the new frame was, for the most
part, selected in two rather than three stages. A third stage was used
only to subsample within blocks and EDs too large to be listed in a single
trip to the field.
The most important change involves the method of selection at the first stage.
The new frame used a controlled selection procedure that assures proportionate
representation along the control dimensions as well as the dimensions used to
define strata. This procedure should yield additional statistical precision
(Leslie Kish estimates the gains at up to 20%). An added benefit of controlled
selection from explicit strata is that it allows us to use the Keyfitz
procedure to draw a new national sample. The Keyfitz procedure is a method
for updating national samples using new Census data; it maximizes the retention
of PSUs from an existing national sample. It could not be easily applied to
a sample drawn using systematic selection procedures (such as the old frame).
The controlled selection procedure used to select the PSUs has one drawback--
it will be more difficult to subsample PSUs. The zone selection procedure
used to select the old frame is easy to select a half-sample; in fact, the
old frame was selected in two replicates. Subsampling is not such an automatic
procedure with the new frame.
1983 GSS Sample. Fifty PSUs from the new frame were selected for the 1983 GSS.
All 16 certainty PSUs were selected for the GSS subsample. The remaining 68
PSUs were paired, according to size and region; one PSU was selected at random
from each pair. Within each of the 50 subsample PSUs, half of the segments
(a total of 141) were selected systematically for the 1983 GSS. Fifty PSUs
were also selected from the old frame with three segments per PSU (total of
150).
1982 Black Oversamples
In 1982 the National Science Foundation funded as a separate project an
oversample of Blacks on the GSS. Two different sample frames were used.
First, an additional sample of lines were drawn from the 1970 sample frame
(as the regular 1982 GSS cross section was). These households were screened
for race and eventually yielded 107 extra Black respondents. Second, a
special sample frame designed to sample Blacks was drawn from 1980 Census
data. In this sample frame localities were selected according to their
Black population rather than their total population. This procedure
increases the take-rate among selected lines yielding a more efficient
sampling of Blacks. This oversample proportionate to the Black population
resulted in 247 extra Black respondents. In all the Black samples added
354 extra Black respondents. When added to the 156 Blacks who were part
of the regular cross section, that made a total of 510 Black respondents
in 1982.
The three groups of Blacks can be added together to form a national
probability sample of Black Americans. The Blacks and Non-Blacks in the
regular 1982 cross-section can be used as a national sample of all races
without using the OVERSAMP (Cols. 3432-3436) weight. If one wants to use
all 1982 cases to form a national sample (the 1,506 from the cross section
and the black oversamples of 354) then the OVERSAMP weight must be used to
establish the proper racial balance. The OVERSAMP variable is designed to
match the racial distribution found in the regular cross section and
reproduce the total number of unweighted cases (i.e. 1,506 + 354 = 1,860).
For a report on the black oversamples including a comparison of the two
methods of oversampling blacks see Roger Tourangeau and A. Wade Smith,
"Finding Subgroups for Surveys," PUBLIC OPINION QUARTERLY, 49 (Fall, 1985),
351-365.
Information on response rates is given in Table A-3.
1987 Black Oversample
In 1987 the National Science Foundation funded as a separate project an
oversample of Blacks on the GSS. An additional sample of lines was drawn
from the 1980 sample frame (as the regular 1987 GSS cross section was).
These households were screened for race and eventually yielded 353 extra
Black respondents. When added to the 191 Blacks who were part of the
regular cross section that made a total of 544 Black respondents in 1987.
1990 National Sample
Like its predecessor, the 1990 National Sample was selected in two major
stages, with PSUs consisting of one or more counties selected at the first
stage and segments consisting of one or more blocks selected at the second.
In a few cases, segments were subsampled, a procedure that constituted a
third stage of sample selection.
The 1990 sample included 100 first stage selections. The PSUs consisted
of metropolitan areas or nonmetropolitan counties. The metropolitan PSUs
include metropolitan areas of all three types distinguished in the 1990
Census--Metropolitan Statistical Areas (or MSAs, which correspond to the
SMSAs used in 1980), Consolidated Metropolitan Statistical Areas (or CMSAs,
which join the metropolitan area of a very large city with the adjacent
metropolitan area of one of its suburbs), and New England County Metropolitan
Areas (or NECMAs, which are the whole county counterparts of the New England
MSAs). Prior to selection, the United States was divided into 2,489 PSUs;
the PSUs were then sorted into strata. The major strata again grouped
metropolitan and nonmetropolitan PSUs within each of the four Census regions.
The nonmetropolitan PSUs were further sorted by state; then, within state,
by percent minority, and, finally, within percent minority groupings, by per
capita income. Each PSU was classified according to the percent of its
population who are minority group members; this encompassed everyone but
non-Hispanic Whites. Percent minority groupings were formed by classifying
each PSU according to percent minority quartiles within its major strata.
The metropolitan PSUs were sorted b Census division, minority quartile, and
per capita income. The sample PSUs were selected using systematic selection,
with the selection probability for a PSU proportional to the number of
housing units. This selection procedure ensured proportionate representation
along each of the sort variables. Nineteen PSUs were so large that they had
to be included in the sample with certainty.
The second stage sampling unit in the 1990 National Sample was again the
segment, consisting of one or more adjoining blocks. The number of segments
selected within a PSU again depended on the whether the PSU was a certainty
selection. From three to 26 segments were selected in the 19 certainty PSUs;
in each of the remaining 81 sample PSUs, three segments were selected. All
told, the 1990 National Sample includes 384 second stage selections. Prior
to selection, the segments within each sample PSU were sorted successively
by a) whether they were within the central city of a metropolitan area or
outside of it (in metropolitan PSUs), b) state (in those PSUs that crossed
state lines), c) county, d) place, e) percent minority quartile within the
PSU, and f) census tract (CT) or block numbering area (BNA). The sample
segments were selected using systematic sampling with probability proportional
to size (in housing units). Undersized blocks were linked to adjacent ones
to assure that each segment included at least 50 housing units. Similarly,
census tracts with fewer than 50 housing units were linked with adjoining CTs.
In the smallest PSUs, it was possible for a segment to be selected more than
once. In such cases and when segments included unexpectedly large numbers
of dwellings, a third stage of sampling was carried out. The segment was
subdivided into pieces by a field count; based on the field count, one piece
of the segment was selected with probability proportional to its estimated
size.(8)
Comparison of the 1980 and 1990 Sample Frames
The two national samples are quite similar. Table A-2 summarizes the main
features of each sample design.
There are a number of differences between the two samples, most of them
minor. At the first stage of sampling, the 1990 National Sample included
more selections than the 1980 Sample (100 vs. 84), and it used a systematic
rather than a controlled procedure to make the selections. The two sampling
methods do not differ markedly; they have similar theoretical properties
(e.g., both allow additional control beyond that afforded by ordinary
stratified sampling) and yield similar results in practice. The systematic
procedure makes it somewhat easier to select subsamples from the 1990
National Sample.
At the second stage of selection, the two samples again differ in their
sample sizes (384 selections in 1990 vs. 562 in 1980) but the same method
of selection was used. There was a small change in how the second stage
units were defined. By 1990, the Census Bureau had divided the entire
nation into blocks and no longer used Enumeration Districts; segments in
the new sample could thus be defined exclusively in terms of blocks. At
both of the first two stages of sample selection, the 1990 National Sample
classified the sampling units by their minority population; the earlier
sample had not made such extensive use of this variable.
1993 GSS Sample
1980 NATIONAL SAMPLE. For the 1993 GSS, approximately 1,100 housing units
were selected from the 1980 National Sample. These dwelling units were
clustered within 141 segments in 50 of the sample PSUs. Within each of the
16 certainty PSUs, approximately one-fourth of the sample segments were
selected for the 1993 GSS, yielding a total of 39 segments. Within the
remaining 68 sample PSUs, a subsample of 34 PSUs was first selected; then,
within each of these 34 PSUs, three segments (out of the six available)
were included in the 1993 GSS. Overall, then, the 1993 GSS sample included
approximately one segment in four from the 1980 National Sample--one-fourth
within the 16 certainty PSUs plus one-half of the segments within a randomly
selected half of the 68 remaining sample PSUs. The subsampling of the
noncertainty PSUs was carried out by pairing PSUs from similar strata; one
PSU was selected randomly from each pair. The subsampling of segments was
done using a simple systematic selection procedure.
--------------------------------------------------------------------------------
Table A.2. Main Features of Two National Samples
--------------------------------------------------------------------------------
1980 National Sample 1990 National Sample
First Stage
-----------
Number of Selections 84 100
Measure of Size/Minimum Housing units/4000 Housing units/2000
Method of Selection Controlled selection Systematic selection
Stratification Region, metro status; Region, metro status,
additional control division/state, % minority,
variables per capita income
Second Stage
------------
Number of Selections 562 384
Unit Blocks or EDs Blocks
Measure of Size/Minimum Housing units/50 Housing units/50
Method of Selection Systematic Systematic
Stratification County (ordered by County (ordered by size &
size), MCD (ordered by state), place, % minority
size and income), census tract/BNA
census tract/ED number
Third Stage
-----------
Unit Part of block or ED Part of block or ED
Measure of Size/Minimum Housing units/50 Housing units/50
Method of Selection One selection per One selection per
segment with probability segment with probability
proportional to size proportional to size
--------------------------------------------------------------------------------
Note: In both samples, the third stage was generally not needed in most areas.
1990 NATIONAL SAMPLE. The new National Sample also contributed approximately
1,100 housing units to the 1993 GSS sample. These were drawn from 191 segments
in 68 of the Sample PSUs. Within each of the 19 certainty PSUs, approximately
one-half of the sample segments--a total of 71--were selected for the 1993 GSS.
Within the remaining 81 sample PSUs, a subsample of 40 PSUs was first selected;
within each of these 40 PSUs, all three available segments were sampled for the
1993 GSS. About half of the segments from the 1990 National Sample--one-half
of the segments within the 19 certainty PSUs and all of the segments within a
random half of the 81 other PSUs--were included in the 1993 GSS sample.
The subsampling of the noncertainty PSUs was carried out by pairing PSUs that
were nearest neighbors in the final sorted file from which the first stage
selections were made; one PSU was selected at random from each pair. The last
PSU of the 81 noncertainty PSUs could not be paired with another sample PSU;
a random procedure was used to determine whether it would be included in the
1993 GSS sample. Segments in the certainty PSUs were subsampled using a simple
systematic selection procedure.
References
(1) For selection procedures, see Benjamin Ying and Carol Richards,
"The 1972 NORC National Probability Sample." Chicago: NORC, August 1972.
(2) Ibid.
(3) Frederick Stephan and Philip McCarthy, SAMPLING OPINIONS.
(New York: John Wiley and Sons, 1958, Chapter 10.)
(4) Alaska and Hawaii are not included in this sample.
(5) The selection methods used are similar to those described in
standard textbooks, e.g., W. E. Deming, SAMPLE DESIGN IN BUSINESS RESEARCH
(New York: Wiley & Sons, 1960), and L. Kish, SURVEY SAMPLING (New York:
Wiley & Sons, 1965).
(6) In the actual implementation of the selection method, subsamples 1
and 4 resulted in 51 PSUs, whereas subsamples 2 and 3 produced only 50 PSUs.
The result was not unexpected and is due to a technical reason, details of
which will be provided on request. The inequality of subsample sizes does
not affect the equal probability characteristics of the sample.
(7) Steven G. Heeringa and Judith H. Connor, The 1980 SRC/NORC National
Sample Design and Development. Ann Arbor: Institute for Social Research, 1984.
(8) For further details on the 1990 National Sample, see Roger Tourangeau,
Robert A. Johnson, Jiahe Qian, and Hee-Choon Shin, SELECTION OF NORC's 1990
NATIONAL SAMPLE. Chicago: NORC, 1993.
(9) C. Bruce Stephenson, "Weighting the General Social Surveys for Bias
Related to Household Size," February 1978.
(10) Tom W. Smith and Bruce L. Peterson, "Problems in Form Randomization
on the General Social Surveys," July 1986.
(11) See Tom W. Smith, "Attrition and Bias on the International Social
Survey Program Supplement," GSS Methodological Report No. 642 February 1986.
(12) C. Bruce Stephenson, "Probability with Quotas: An Experiment,"
GSS Methodological Report No. 3, April 1979; Tom W. Smith, "Response Rates on
the 1975-1978 General Social Surveys with Comparisons to the Omnibus Surveys
of the Survey Research Center, 1972-1976," GSS Methodological Report No. 5,
June l968; Tom W. Smith, "Sex and the GSS: Nonresponse Differences," GSS
Methodological Report No. 9, August 1979; Tom W. Smith, "The Hidden 25%: An
Analysis of Nonresponse on the 1980 General Social Survey," GSS Methodological
Report No. 16, May 1981; Tom W. Smith, "Using Temporary Refusers to Estimate
Nonresponse Bias," GSS Methodological Report No. 21, February 1983; Tom W.
Smith, "Discrepancies in Past Presidential Vote," GSS Methodological Report
No. 25, July 1982; and Tom W. Smith, "Notes on John Brehm, THE PHANTOM
RESPONDENT: OPINION SURVEYS AND POLITICAL REPRESENTATION. GSS Methodological
Report No. 79, 1993.
(13) Tom W. Smith, "Rotation Designs of the GSS," February 1988.
1972-1994 General Social Survey Cumulative File