Valuing Federal Statistics
THE VALUE OF FEDERAL STATISTICS rests on their unique character.20 Federal statistics meet the definition of a “public good,” similar to the judiciary, clean air, and national defense (see Box I.1), and are valuable for many important public purposes.
That federal statistics are a valuable public good justifies expenditure of public funds to ensure that relevant, accurate, timely, and credible statistics are collected and made available to all current and potential users. In fact, spending on statistical programs is a tiny fraction of overall federal spending: in fiscal 2016, the $6.5 billion budget request for all statistical programs identified by the U.S. Office of Management and Budget (2017:Table 1), excluding the decennial census, amounted to less than 0.2 percent of the budget authority of about $3.6 trillion for the federal government. The $6.5 billion is equal to about $20 annually for every U.S. resident (325 million as of December 1, 2016).21
Whether current spending on federal statistics, in total or as divided among individual statistical programs, is optimal in terms of the value of the statistics produced is not easily determined. Yet there are compelling justifications in terms of the uses of the data, which include political representation, economic decision making in the public and private sectors, administration of programs, scientific research, and contributions to public opinion and debate. After brieﬂy reviewing these uses, the section concludes
20 This section draws heavily on Citro (2016) and National Research Council (2013b:101–102); see also U.S. Office of Management and Budget (2017:3–4); see Prewitt (2010) for an overview of the federal statistical system and its data users and uses.
21 The decennial census adds about $40 per resident over a 10-year period, or an additional $4 annually. Of the total of $6.5 billion, the 13 principal statistical agencies accounted for 38 percent.
with a brief consideration of the limits of a formal cost-benefit approach to assessing value.
The bedrock justification for federal statistics on the population rests in the U.S. Constitution, which mandated a decennial census of population every 10 years (the first census was taken in 1790) for determining the allocation of seats in the U.S. House of Representatives among the states. At first, Congress added seats to the House each decade so that the decennial apportionment would not reduce the number of seats for any state even when its percentage share decreased. In 1929, Congress fixed the size of the House at 435 seats, so that reapportionment means that slower growing states lose numbers of seats in addition to percentage shares.
Reapportionment in turn triggers the redistricting process by which states, using census data, redraw the boundaries of congressional districts to accommodate changes in the number of seats and in the distribution of population geographically (e.g., movement from urban to suburban areas) and by race and ethnicity. Typically, data on voting patterns by party are also used in redistricting as political parties attempt to maximize the number of seats they can expect to win in elections. States and many local governments also use census and other data to reapportion and redistrict their legislative bodies. The redistricting process for congressional and local seats can become highly contentious, generating legal cases that, in some instances, reach the U.S. Supreme Court (see McMillen, 2012; National Research Council, 1995:App. C).
Legislation that affects voting participation and hence political representation and power in at least one instance hinges explicitly on federal statistics. Thus, section 203 of the Voting Rights Act mandates provision of language assistance to voters in areas that meet certain population criteria as determined by the U.S. Census Bureau. The current method uses the American Community Survey (ACS) to estimate single language minority groups with limited English speaking proficiency and educational attainment in counties and tribal areas.22
ECONOMIC DECISION MAKING
In the early days of the republic, James Madison expressed the view that expanding census statistics beyond the headcount and obtaining
22 See, e.g., “Voting Rights Act Amendments of 2006, Determinations Under Section 203,” 81 Federal Register 87532 (December 5, 2016); available: https://www.federalregister.gov/d/2016-28969 [April 2017]. The ACS asks questions formerly included in the decennial census “long-form sample.”
detailed information on the economy as well as the population would be of great benefit for legislative deliberations (Eberstadt et al., 2017:1–2). From these beginnings, the scope of federal economic statistics, broadly defined, expanded enormously as described in the “Brief History of the U.S. Federal Statistical System” section above.
For some economic statistics, it is possible to assess their value by comparing the dollars spent on them to the dollars they drive in the economy and society. The federal government currently labels 38 statistics—e.g., gross domestic product (GDP), unemployment, monthly retail trade, weekly natural gas storage, grain production, money stock, and others—as “principal federal economic indicators.” Statistical Policy Directive No. 3 (U.S. Office of Management and Budget, 1985), issued in the 1970s in response to attempted manipulation of Bureau of Labor Statistics monthly unemployment statistics (see Norwood, 2016) and strengthened in 1985, requires these indicators to be published by the cognizant statistical agency on specified release dates under procedures designed to protect the integrity and credibility of the estimates (see Practice 2).
An example of the consequential effects of these key statistics is provided by the CPI. It determines annual cost-of-living adjustments to Social Security monthly benefits, which in April 2017 amounted to $76.8 billion provided to 61.4 million people.23 Consequently, a 1 percentage point difference in the CPI would amount to $768 million in additional or reduced Social Security benefits on a monthly basis, or about $9 billion annually. Annual changes in the CPI also affect commercial and residential rents, public- and private-sector wages, and components of the federal income tax code. Monthly changes in the CPI are a major input to Federal Reserve Board decisions on short-term interest rates and financial decision making throughout the public and private sectors. By comparison, the BLS program to produce the CPI and other price series had an estimated budget authority of $216 million in fiscal 2016.24
Federal statistics also feed directly into allocations of billions of dollars of federal funds to states and local governments by use of legislated formulas. For example, formulas for allocating federal funds for education of disadvantaged children to the nation’s more than 13,000 school districts use statistical estimates of school-age children in poverty produced by the U.S. Census Bureau from a model that incorporates data from the ACS and administrative records (see National Research Council, 2000). Reamer (2010) traced the allocations of $416 billion in federal funds in fiscal 2008 that depended on estimates from the ACS, postcensal population estimates
23 See http://www.ssa.gov/policy/docs/quickfacts/stat_snapshot [April 2017].
24 See https://www.dol.gov/sites/default/files/documents/general/budget/2016/CBJ-2016-V3-01.pdf [April 2017].
developed by updating census data with administrative records, per capita income estimates from the Bureau of Economic Analysis, and other federal statistics (see also National Research Council, 2003).
Less obvious, perhaps, but just as important as the use of federal statistics in fund allocation formulas is their use to inform legislative proposals to modify existing programs or add new ones in a wide range of areas. Federal statistics are used extensively, for example, in microsimulation and other kinds of policy models to estimate the likely benefits and costs of proposed changes to tax policy, social welfare programs such as food assistance, and many other policy proposals. Such models are used by executive branch agencies and nongovernmental organizations to make a case for or against particular proposals. The Congressional Budget Office and the Joint Committee on Taxation have formal roles in costing out policy proposals before they can be acted on by Congress, and they make extensive use of survey and other data from federal statistical agencies and other sources (see, e.g., National Research Council, 1991, 1997a, 2010b).
The Economics and Statistics Administration (2014) discusses additional uses of government economic data for public- and private-sector decision making. This study also estimated the size of the “government-data-intensive-sector,” including news media, market researchers, investment advisors, pollsters, and firms that repackage government data to add value for their customers.
There may not be and often is not a dollars-and-cents calculus that establishes value by relating the data from a statistical program to specific consequential decisions as can be done, say, for the CPI. Nonetheless, qualitative assessments that trace through the uses of statistical programs can often provide evidence of substantial value. An example is an effort to comprehensively identify nonfederal uses of data from the American Community Survey by state and local governments, nongovernmental organizations, the media, and others (National Research Council, 2013a; see also National Research Council, 2007b:Ch. 3). This effort identified consequential nonfederal uses in such areas as disaster preparedness, economic development and workforce planning, public health surveillance, and regional transportation planning.
More generally, statistical programs have important uses in taking the pulse of social and economic trends, designing and evaluating government programs, and supporting foundational social science research:
- Monitoring the social and economic health of the nation, states, and localities—Regularly published social and economic indicators
from statistical agencies are widely cited in the media and consulted by the public to identify trends and, when estimates are available for states and local areas, to compare across areas. Some of these series, published annually or biannually, include America’s Children: Key National Indicators of Well-Being from the Interagency Forum on Child and Family Statistics; The Condition of Education from the National Center for Education Statistics; Income and Poverty in the United States from the U.S. Census Bureau; and Science and Engineering Indicators from the National Science Board and National Center for Science and Engineering Statistics.25
- Providing empirical evidence for developing and evaluating federal, state, local, and private-sector programs—For example, the American Housing Survey, sponsored by the Office of Policy Development and Research in the U.S. Department of Housing and Urban Development and conducted by the Census Bureau, provides valuable data on housing condition and housing finance to inform housing policy (see National Research Council, 2008). The Commercial Buildings Survey and the Residential Energy Consumption Survey, sponsored by the Energy Information Administration in the U.S. Department of Energy, provide valuable data for public- and private-sector policy making on end uses of various types of energy for heating, cooling, information technology, and other uses (see National Research Council, 2012a:Ch. 1).
- Providing input to important social science research that, in turn, informs the public and policy makers—Many policy-relevant insights have resulted from analysis of long-running federally funded surveys, including longitudinal surveys that follow individuals over time (see, e.g., National Research Council, 2005). A few of these surveys are the Education Longitudinal Study of 2002 sponsored by the National Center for Education Statistics; the Health and Retirement Study sponsored by the National Institute on Aging and the Social Security Administration; and the National Longitudinal Surveys of Labor Market Behavior sponsored by BLS.26
25 Websites [as of April 2017] for the cited series are: https://www.childstats.gov/americaschildren/; https://nces.ed.gov/programs/coe/; https://www.census.gov/library/publications/2016/demo/p60-256.html; and https://www.nsf.gov/statistics/2016/nsb20161/.
26 Websites [as of April 2017] for the referenced surveys are: https://nces.ed.gov/surveys/els2002/; http://hrsonline.isr.umich.edu/; and https://www.bls.gov/nls/.
Some attempts have been made to estimate costs and benefits of statistical programs or improvements in programs as input for making decisions about program priorities. However, formal cost-benefit analysis applied to federal statistical programs arguably is too dependent on questionable assumptions to be useful for this purpose.
A line of research begun in the 1970s has focused on the costs and benefits of marginal improvements in the accuracy of statistics used in congressional reapportionment, legislative redistricting, and allocation of federal funds to states and localities. As reviewed by the National Research Council (2015:51–70) (see also National Research Council, 1995:40–43), studies have found only modest effects on any of these from correcting for known net census population undercount. Studies have found more marked effects on fund allocation of using alternative estimates of per capita income.
Spencer (1997) has attempted to measure the impact of spending for improved data quality on specific policy decisions. As he acknowledges, a daunting problem is to determine an appropriate counterfactual—what the decision would have been with better (or worse) data—given that decision making rarely moves linearly from reviewing the evidence to framing appropriate policy alternatives to deciding among them.27
Moreover, a focus on specific policy decisions is too narrow for valuing investment in federal statistics. As one example, there is no immediately pending policy decision about the nation’s rising income and wealth inequality and declining social mobility (see, e.g., Chetty et al., 2014, which used federal income tax data), but estimates of those changes have brought them to the forefront of public and policy attention. This, in turn, argues for improving federal statistics on income and wealth—increasingly problematic in surveys (see, e.g., Czajka, 2009)—to ensure as accurate a picture as possible of trends in inequality and mobility in the United States over time on which to base future policy making.
In reviewing data needs on natural gas for the Energy Information Administration, the National Research Council (1985:Ch. 3 and App. 3A) raised valid concerns about formal cost-benefit analysis of federal statistical programs. The report concluded that such analysis is largely infeasible because of the difficulty of: (1) uncovering who has used an agency’s statistics—many uses are indirect without the user being aware of the data source, a problem magnified by the Internet; and (2) putting a quantitative value on each known use. The report recommended that statistical agencies emphasize regular in-depth contact with users through advisory committees
27 For a discussion of how social science evidence is filtered in various ways in policy formation, see National Research Council (2012b).
and other mechanisms to set priorities among their programs, in preference to formal cost-benefit analyses.
The fundamental characteristic of federal statistics as a public good and the demonstrated policy, planning, research, and informational value of today’s portfolio of statistical programs justify adequate budgets for federal statistics. Such funding needs to provide for research and development for continuous improvement in relevance, accuracy, timeliness, and accessibility (see Practice 10). In turn, it is incumbent on federal statistical agencies to communicate the value of their programs to policy makers and others and to analyze the costs of their programs on a continuing basis so that they can ensure the best return possible on the tax dollars invested in them.