2. The rise of probability sampling in official U.S. statistics

Constance F. Citro

Previous | Next

It is not an exaggeration to say that large-scale probability surveys were the 20th-century answer to the need for wider, deeper, quicker, better, cheaper, more relevant and less burdensome official statistics. Such surveys provided information with known precision in contrast to non-probability surveys; and they provided detailed information at greatly reduced cost and increased timeliness compared with censuses. Duncan and Shelton (1978) and Harris-Kojetin (2012) review the rise of probability sampling in U.S. official statistics.

It was not clear at the time when the theory and practice of modern probability sampling was being developed in the 1930s in the United States that probability surveys would gain such widespread acceptance. The arrival of Jerzy Neyman in the mid-1930s gave a boost to the work of W. Edwards Deming, Calvin Dedrick, Morris Hansen and colleagues at the Census Bureau who were developing the needed theory for sampling of finite populations. Small-scale sample surveys in the 1930s at universities and federal agencies on such topics as consumer purchases, unemployment, urban housing and health provided proofs of concept and practical tips.

Table 2.1
Selected ongoing U.S. statistical agency probability surveys, by year begun
Table summary
This table displays the results of Selected ongoing U.S. statistical agency probability surveys. The information is grouped by Decade and Year/Type of Survey (appearing as row headers), Repeated Cross-Sectional Household Survey, Repeated Cross-Sectional Business Establishment Survey and Panel Person Survey (appearing as column headers).
Decade and Year/Type of Survey Repeated Cross-Sectional Household Survey Repeated Cross-Sectional Business Establishment Survey Panel Person Survey
1940 1940 - Current Population Survey (CPS)

1947 - CPS Annual Social and Economic Supplement (CPS/ASEC)
1946 - Monthly Wholesale Trade Survey  
1950 1950 - Consumer Expenditure Survey (CE)

1955 - National Survey of Fishing, Hunting, and Wildlife-Associated Recreation

1957 - National Health Interview Survey (NHIS)
1953 - Advance Monthly Retail Sales Survey

1953 - Business R&D and Innovation Survey (BRDIS)

1959 - Building Permits Survey
 
1960 1960 - Decennial Census Long-Form Sample (became American Community Survey in 2005) 1965 - National Hospital Care Survey 1966-1990 - National Longitudinal Survey of Older Men
1970 1972 - National Crime Victimization Survey (NCVS)

1973 - American Housing Survey (AHS);

1973 - National Survey of College Graduates (NSCG)

1979 - Residential Energy Consumption Survey (RECS)
1975 - Farm Costs and Returns Survey and Cropping Practices and Chemical Use Surveys (combined in Agricultural Resource Management Survey in 1996)

1979 - Commercial Buildings Energy Consumption Survey (CBECS)
1972-1986 - National Longitudinal Survey of High School Class of 72

1973-present - Survey of Doctorate Recipients (SDR)

1979-present - National Longitudinal Survey of Youth (NLSY79)
1980 1983 - Survey of Consumer Finances (SCF) 1985 - Manufacturing Energy Consumption Survey (MECS) 1984-present - Survey of Income and Program Participation (SIPP)
1990 1991 - Medicare Current Beneficiary Survey (MCBS) 1996 - Agricultural Resource Management Survey (ARMS) 1997-present - National Longitudinal Survey of Youth (NLSY97)
2000 2005 - American Community Survey (ACS)   2001-2008 - Early Childhood Longitudinal Study (Birth Cohort)

The federal government’s young statistical Turks still had to surmount hurdles in the bureaucracy up to the White House before they could move sampling into the mainstream of federal statistics. Thus, “old timers” at the Census Bureau were skeptical about the possibility of using survey methods to get information on unemployment and politicians were divided about whether they wanted the estimates (Anderson 1988). In 1937, a major breakthrough occurred when a two percent sample of households on nonbusiness postal routes, designed by Dedrick, Hansen and others, estimated a much higher - and more credible - number of unemployed than a “complete” census of all residential addresses that was conducted on a voluntary basis. Picking up on that effort, from 1940-1942, the Works Progress Administration fielded the sample-based Monthly Report on the Labor Force, the forerunner to the Current Population Survey (CPS). The CPS continues to this day as the source of official monthly estimates of U.S. unemployment conducted by the Census Bureau and published by the Bureau of Labor Statistics (BLS).

Another breakthrough occurred when the Census Bureau, which struggled for decades to respond to demands for added questions on the decennial census without turning the instrument into a nightmare for respondents and interviewers, asked six questions on a five percent sample basis in the 1940 census. The success of sampling led to a decision to administer two-fifths of the questions in the 1950 census to a sample, and subsequent censuses followed suit. Table 2.1 lists selected ongoing U.S. household surveys, business surveys and panel surveys and when they began. The variety of subjects covered and the longevity of these surveys attest to the dominance and value of the sample survey paradigm in U.S. official statistics.

Previous | Next

Date modified: