FEDERAL V/ 0 R K S AGENCY
Work Projects Administration
Howard 0. Hunter, Commissioner
SAMPLING PROCEDURES AND METHOD OF OPERATION OF THE
WPA MONTHLY REPORT OF UNEMPLOYMENT
Prepared by
Division ox Research
Labor Market Research Section
October 191*1
Howard B. Myers
Director of Research
A3167
SAMPLING PROCEDURES AND METHOD OF OPERATION OF THE
WPA MONTHLY REPORT OF UNEMPLOYMENT
In response to many requests for a statement as to the
method of operation and the sampling procedures under which this
survey is conducted, the present statement of method and scope has
been prepared.
Since inquiries vary from a request for a general and .
nontechnical description to requests for information on detailed pro¬
cedures, this statement is divided into two parts, the first of which
is general and nontechnical, and the second somewhat more detailed
and technical. For those who want only a general outline of the
survey operation, the first section should suffice; for those who want
greater detail, the second section provides a description of the sampl¬
ing and estimating procedures.
It should be noted that the statement presented here refers
to the method under which the survey is now operating. In the 21
months during which this survey has been in the field continued efforts
have been made toward improvement, with the result that the procedure
at the present time is different in some respects from the original
plan. At a later date it is planned to issue a fuller account of the
method and scope of this survey with a description of the original
plan, the alterations made during operations, and the reasons therefor.
It is hoped that the present statement will meet, in part at least,
the requests for information on the operation of the sample survey.
GENERAL STATEMENT OF METHOD AND SCOPE
In December 1939 the Division of Research of the Work
Projects Administration put into operation a regular monthly survey
of unemployment to provide accurate measurements of unemployment,
employment, and the size of the labor force on a systematic basis.
This was the first attempt by any public or private agency to pro¬
vide, through actual field counts, the much-needed answer to the
question, "How many workers are out of work or working today?" The
guiding principles of this national survey are accuracy and speed
at small cost.
General procedure. The Work Projects Administration survey
interviews a scientifically selected cross-section of the total popu¬
lation each month. The superiority of this method arises from the
fact that unemployment is measured directly rather than estimated
indirectly as is necessary in the method used for other periodic
unemployment estimates. No additions or subtractions, based upon
assumptions as to unemployment reported in the 1930 or I9I4O censuses,
net average increase in the labor force, or changing age structure,
are necessary when direct measurement is made; such assumptions are
essential to all other methods of estimating unemployment. The result
of the sampling and direct interview procedure is an extremely fast
A31*67
- 2 -
and sensitive report on the state of the labor market as of one week
each month. In some respects, the procedures followed are similar to
those of the public opinion polls, whose accuracy has been demonstrated
again and again by the Gallup polls and fortune magazine.
Coverage. The national sample consists of 61* counties
located in~Ii3T states. (A list of these counties is given on page 6.)
The counties were chosen on the basis of (l) location, (2) population*
and (3) economic characteristics.
Strictly rural, as well as highly industrial counties, are
included; small agricultural counties, as well as the great metropolitan
counties of New York City, are covered; each section of the country is
fully represented; the range in unemployment is from a few percent of
the working population in the most prosperous counties to one-fifth
or more of the working population in the most severely depressed areas.
By making use of every available piece of information and
applying scientific methods of sampling which have been developed and
tested over a period, of five years' experimentation, the national
sample is truly a cross-section of the country to an extent consider¬
ably beyond that considered safe by the public opinion polls.
County operation. Within each of the 61* counties a sanple
of households is selected for interview, by a method which insures
that every part of the county population—rural* village, and city—
is covered. The total number of households to be interviewed within
a county is apportioned among city, village, and open country in
accordance with the population figures for that county. The same
households are interviewed for four to six months in succession
and then the sample households are changed so as not to arouse
irritation through too prolonged inquiry.
Information obtained. The questions asked provide infor¬
mation on the size of the household, the number of persons under and
over li* years of ago, and, for all persons lit years and older, their
relationship to the labor market during the census week, that is,
the entire week immediately preceding the week of interview. Each
person is asked one or more simple questions to determine whether he
or she was employed, unemployed, or out of the labor market. In
addition to figures on the volume of unemployment and employment,
data are secured on the age, sex, duration of unemployment for the
unemployed, industry of the employed, and other characteristics of
workers and nonworkers in the population. The fact that all persons
±h years of age or older are questioned about their labor-market
activity insures a sensitive report of unemployment that takes full
account of changing conditions.
Speed of operation. Beginning with the first day of the
enumeration week, field schedules are tallied daily in each county.
A3167
- 3 -
At the close of the enumeration week, a summary of the Salient facts
is mailed directly to the Washington office for consolidation and
interpretation. These summary figures are released in memorandum
form before the close of each month; an average of two weeks elapses
between the first interview in the field and the completed memorandum.
Final tabulations. After completing the preliminary field
tallies, each coamty supervisor mails the completed schedules to a
central machine tabulating office where additional information is
taken from the schedules. These additional facts provide the basis
for a much more detailed analysis.
Possibilities of extension. By simple extensions the
possibilities of the sampling method of measuring unemployment directly
can be greatly increased. Other types of inquiries may and have
been conducted using the same principle and the existing field
staffs.
The speed, accuracy, and small cost of gathering important
information by interviewing scientifically selected samples of the
population, are just beginning to be appreciated within the Government.
At a time like the present, when timeliness is an essential element
in fact-gathoring, the resources of the sample survey method should
bo fully utilized.
A3U67
- k -
SURVEY COUNTIES AS OF OCTOBER I9J4I
STATE
COUNTY
STATE
COUNTY
Alabama
Jefferson
Nebraska
Dawson
Arizona
Cochise
Mew Jersey
Ocean
Arkansas
TJnion-x-
New Jersey
"Warren
California
Alaneda-x-
New Mexico
Chaves
California
Los Angeles
New York
Albany
Colorado
Weld
New York
Bronx (N.Y. City)
Connecticut
Tolland-x-x-
New York
Kings (N.Y. City)
Delaware
Sussex-x-
New York
New York (N.Y.City)
Florida
St. Johns**
New York
Queens (N.Y. City)
Georgia
Paul.ding
New York
Richmond (N.Y. City)
Georgia
Worth
North Carolina
Buncombe*
Idaho
Payette
North Dakota
Cass
Illinois
Cook (Chicago)
Ohio
Previa
Illinois
Scott
Ohio
Scioto
Indiana
Allen
Oklahoma
"Woodward •
Iowa
Scott-x-
Oregon
Clatsop
Kansas
Ellsworth
Penn sylvania
Allegheny*
Kansas
Wyandotte
Pennsylvania
Philadelphia
Kenfucky
Christian
Rhode Island
Providence**
Kentucky
McCracken
South Carolina
Georgetown**
Louisiana
Rapides
South Carolina
Greenville
Maine
York
South Dakota
McCook-x-x-
Maryland
Baltimore City*
Tennessee-
Washington
Maryland.
C-arrett
Texas
Archer**
Massachusetts
Franklin
Texas
Jefferson
Michigan
Calhoun*
Utah
Snnpcte**
Michigan
Wayne (De t r 0 i t)
Vermont
Orange-x-x-
Minnesota
Pope
Virginia
Appomattox
Minnesota
Stevens
Washington
Snohomish
Mississippi
Stone
Washington
'Whitman**
Missouri
Madison**
West Virginia
McDowell
Montana
Silver 3ow*
Wisconsin
Fond du Lac
-x-Added to survey operations in October I9I4O.
-x~x-Added to survey operations in October lplil.
A3467 - 5 -
DETAILED STATEMENT OF SAMPLING- AND ESTIMATING PROCEDURES!/
In view of limitations as to time, cost and centralized
technical control the method of stratified samjjling and sub sampling
was adopted for this survey. This involved the selection of a
relatively small number of primary sampling units (counties) and
a subsample of secondary sampling units (households) within each of
the larger units. To insure maximum statistical efficiency both the
primary and secondary sampling units \\rere selected on the basis of
stratified sample.
The nrinary sampling unit. The county wo.s chosen as the
primary sampling unit because of the following considerations: The
county was a small self-contained geographic unit for which it wan
possible to obtain statistical information that could be used for
stratification purposes. Thb variance of the unemployment rate
between counties was low. Excellent county naps were available that
made the county a very satisfactory unit from which to subsample.
Finally, administrative operations in the average size county could be
handled easily by a local supervisor with six or seven enumerators.
The secondary sampling unit. Both administrative and
statistical conditions were taken account of in the selection of the
secondary sampling unit. The household was chosen mainly becauso it
was the smallest unit that could be sampled readily within a county.
Such units were easily defined and located, whereas it would have been
practically impossible to select a random sample of individuals
without first having listed all persons within the county. The house¬
hold was desirable also because of certain statistical reasons. A
great deal of valuable household data such as size, number of
workers, and employment status, could be obtained only by enumerating
household units. Moreover, for a great many characteristics, the
household is a more efficient sampling unit than the individual
becauso of the heterogeneity of individuals within households.
The intercounty sample. All of the counties in the United
States were stratified by (l) size, (2) location, and (3) economic
situation. Because of the great difference in county characteristics
associated with population size, counties were first divided into
the following strata:
(1) The nine counties in which the five largest cities are
located (containing 14 percent of the U. S. population)
(2) The 447 counties with 1930 populations of 45,000 and
over (containing 50 percent of the U. S. population). These counties
are referred to herein as "urban" counties.
(3) The 2,641 counties with 1930 populations of less than
45,000 (containing 36 percent of the U. S. population). These
counties are referred to herein as "rural" counties.
l/ The sampling and estimating procedures outlined in this section
were developed by Lester R. Erankel and J. Stevens Stock.
A3467
— 6 —
Because of the great heterogeneity among the five largest
cities, no attempt was made to seloct a sample of these nine counties;
all were included in the sample. Samples of counties were selected
from each of the other two strata. The method of selecti6h was the
same for both the "urban" and "rural" counties and, therefore, the
procedure will be explained in terms of the urban counties only.
All urban counties were further classified by population
size into three substra.ta (terciles): small, medium, and large.
The same counties were then cross-stratified by Stages into three
broad geographic locations: Northeast, Southeast, and West, so that
there were about an equal number of counties within each of these
locations.
The percentage of persons unemployed in each of these
counties -according to the 1937 Census of Unemployment was used as the
third mode of stratification. At that time these data were the most
current and reliable which were available for all of the counties in
the United States. The counties were then divided into three strata
(terciles) according to economic situation.
The urban counties were stratified into 27 cells: three
population size groups, by three geographic locations, and by three
economic levels. One county was selected from each of 26 of these
cells, and two counties selected from the 27th cell because of the
relatively high proportion of total urban population in this cell.
Within the 27 cells, the selection of counties was random except
that a deliberate effort wa„s made to maximize State coverage.
Twenty-seven counties were selected from the rural counties
in a similar manner, using the sane nodes of stratification.
In all 64 counties were selected for the sample—9 metro¬
politan counties, 28 urban counties, and 27 rural counties.
The intracountv sample. The household sample xirithin
counties \ic.s stratified according to population distribution. In
each sample county the number of households selected within each
city, town, village, or other minor civil division was proportionate
to the population in those places.
The procedure for selecting the household sample was dependent
upon whether the unit fell in en urban place (city, town, or village)
or in the open country. In the former, a sample of city blocks was
chosen by numbering a map and selecting every nth block; all dwelling
units in these blocks were then listed and a subsanple of households
was selected from these block lists. The sampling ratio was fixed so
that tnere iiras an average of no more than 3 households selected per
block.
In open country areas the household sample wa,s allocated
among the townships or similar minor civil divisions in multiples of 5.
In each township one mile sections were chosen at random in the Washington
A3467
- 7 -
office fron naps of the sanple counties. Five households are inter¬
viewed in each of these sections. The order of selection of households
was determined arbitrarily by starting at the southwest corner of the
section and proceeding around the section in a counterclockwise
direction. Alternate sections were indicated for each township to be
used when the original sections did not yield five households. The
sections were numbered consecutively at random over the entire county
and the enumerator proceeds fron the lowest numbered section to the
next higher in a township, interviewing up to 5 households per section
until the quota for the township is obtained.
The intracounty sample is changed every four to six months.
This is a compromise between having a constant group of households
to study monthly changes, and having new households to keep the
sample dynamic and to overcome interviewee resistance.
The intracounty sample excludes persons in institutions and
the armed forces.
The making of estimates. National estimates of labor force,
employment, and unemployment are made by expressing these characteris¬
tics in each sample county as a pei-centage of the total population
interviewed, and then proceeding to a weighted average of these per¬
centages for the country as a whole. The weights are proportionate
to the populations of the cell or segment of the universe each county
is chosen to represent. Thus, the weight of each of the nine metro¬
politan counties is proportionate to the total population of each of
those counties. The weight applied to percentages derived from a
county in any urban or rural cell is proportionate to the total
population of all counties in that particular coll. 2/
Thus, national estimates are obtained directly in terms of
percentages of total population. The total population in this case
is defined as all persons not in the armed forces or in institutions.
In presenting labor market data from this survey, the
results in percentage form are converted from total population to
the population 14 years of age and over. Estimates of national totals
are presented in monthly memoranda in absolute as well as percentage
form.
2/ By the modes and levels of stratification outlined above, it is
possible to obtain national estimates with as few as 27 counties. This
may be done by selecting the 9 metropolitan counties, 9 urban counties,
and 9 rural counties. Both the urban and rural counties must be selected
in each case in the form of a Latin square. From information obtained
from the 9 urban counties, for example, it is possible to provide
estimates for each of the 27 urban cells. Estimates for the urban and
rural cells, and metropolitan counties would then be averaged together
in the manner described above.