In order that they might rest their arguments on facts: The vital role of government-collected data

Our founding fathers recognized the value of data. James Madison observed that information on agricultural, commercial, and manufacturing interests would enable any legislature to represent the interests of its citizens more effectively. Madison pointed out that this kind of information had never been obtained in any country, and that it would enable the United States to grow (Hutchinson and Rachal 1963).

If gentlemen have any doubts with respect to [this information’s] utility, I cannot satisfy them in a better manner, than by referring them to the debates which took place upon the bills, intended, collaterally, to benefit the agricultural, commercial, and manufacturing parts of the community. Did they not wish then to know the relative proportion of each, and the exact number of every division, in order that they might rest their arguments on facts, instead of assertions and conjectures? — James Madison speaking on February 2, 1790, quoted in Gales [1834, 1145–46]

The federal government has therefore always played a vital role in developing the data that Americans rely on to make well-informed decisions. As mandated in the Constitution, the U.S. Census Bureau enumerates the U.S. population, periodically adjusting congressional and state legislative districts to reflect the changing population distribution. In this respect and in many others, data are often key to the functioning of our republic. Data are required for the proper exercise of citizenship and for holding public officials accountable. Without accurate information regarding the state of the economy and the effects of public policy, citizens would be unable to make fully informed choices about elected officials and to demand that the government reflects their priorities.

Today the decennial census is just one of the useful statistical products made freely available to the public. In areas ranging from the rural economy to energy markets to the labor market, among many others, federal statistical agencies make valuable contributions to the public good of increased knowledge and understanding. In fact, the Bureau of Labor Statistics alone received 36.8 million web page views monthly in 2013 (BLS 2013).

Importantly, the public benefits of statistical reporting come at a relatively small cost. The principal statistical agencies spend only about 0.18 percent of the federal budget (Figure A). This is slightly lower than its 2004 level (OMB 2004; 2016). Given the increasing importance of data to the modern economy, the additional opportunities to collect and analyze new information, and the need for informed policy making, it is striking that this aggregate cost has remained such a small share of the federal budget.

Over the years, the federal statistical agencies have refined a set of practices that ensure the quality and impartiality of its statistical reporting. Because the reports are of such value to the private sector and the public at large, financial markets carefully scrutinize them, reacting quickly to many of the releases. This makes it especially vital that the statistical reports are generated and published in an impartial fashion that puts every user of the data on an equal footing.

To this end, government data are closely scrutinized by nonpolitical, career employees on a “need-to-know” basis, with individual employees often focusing on an isolated part of the process and unaware of the overall statistical result (Malone 2016; OMB 2000, Circular No. A-130). Government agencies also provide source data for outside experts to review and comment on their methodology (Bureau of Economic Analysis [BEA] 2016a). In the case of the BEA, political appointees have very limited access to the data until after journalists receive them—one hour before the estimates are made public—further ensuring the impartiality of the process (BEA 2016b).

Objective, impartial data collection by federal statistical agencies is vital to informing decisions made by businesses, policy makers, and families. These measurements make it possible to have a productive discussion about the advantages and disadvantages of particular policies, and about the state of the economy. This document demonstrates a portion of the breadth and importance of government statistics to public policy and the economy.

Many of our data sources are principally designed to yield national estimates of important variables: the unemployment rate, GDP, and so forth. These statistics are familiar, and their accuracy is vital. But much of what is valuable to private business is necessarily local. A business that produces a specific product requires a specific type of worker, and a retailer selling a product or service must understand and address a particular market.

Figure 1 highlights the ability of the American Community Survey to yield local estimates of specific types of workers—in this case, engineers—that a business may want to hire. This estimate is potentially useful for firms deciding among different locations for new branches; a firm that locates in a region without the necessary worker skills is unlikely to prosper. The usefulness of this analysis is not limited to businesses seeking engineers; firms looking for other types of skilled or unskilled labor can make similar calculations using ACS data.

Engineers are plentiful in the northeast corridor from Washington, DC to Boston, as well as Denver and the California Bay Area, to take a few examples. The ACS sample size even allows for estimation of more-detailed subpopulations, like the number of graduates with chemical engineering majors. Businesses and workers can observe how much the typical worker is paid in a given location and occupation, further improving their analyses and informing their decisions.

Private businesses make extensive use of the American Community Survey in determining when and where to open or expand stores, distribution centers, and warehouses (Kleinhenz 2015). Target and Kroger, in particular, have reported that they use ACS data to tailor their product mixes and advertising appropriately across locations (ESA 2015). Other businesses analyze and interpret ACS data for analyses they sell to firms: Acxiom, an Arkansas-based marketing firm with more than $1 billion in revenue, creates market indices and other products that depend heavily on data from the ACS (NRC 2013, 110–111).

The ACS is useful for all these purposes because it provides detailed coverage of local areas. In figure 2 the capacity of the ACS to yield such information is demonstrated with a particular example: the share of the elderly in each local population. As with many other groups, the elderly are not evenly distributed across the United States, but cluster in certain areas. Florida, Arizona, and parts of the eastern seaboard all have relatively high elderly representation in their populations.

Beneficiaries of this information include firms that offer services to the elderly, as well as local and state policy makers who must respond to the age-related needs of their constituents. The ACS provides even more detailed information than shown here: a firm catering to Hispanic women aged 65 to 74, for example, could learn where its customers are located and adjust its approach accordingly.

The H-1B program allows employers to temporarily sponsor foreign workers in the United States on a nonimmigrant basis when those foreign workers have skills that are difficult or impossible to obtain in the domestic labor market. Current law limits the number of foreign workers who qualify, making the visas a scarce, valuable commodity.

Firms, researchers, and policy makers alike are interested in the jobs associated with H1-B visas. In order to inform discussions about the program, the Department of Labor’s Office of Foreign Labor Certification publishes calculations from its administrative data regarding visa recipients and the occupations in which they work. As figure 3 shows, H-1B visas are particularly common in the computer and information technology fields. Administrative data also show that in fiscal year 2016, California accounted for 19 percent of all H-1B visas and Deloitte Consulting was the top employer of H-1B workers, employing 10 percent of such workers (Employment and Training Administration 2016).

Because of its large size and distance from other developed economies, the United States is a relatively closed economy, with international trade amounting to a fairly small portion of its economic output. However, the role of trade has increased over recent decades, with imports persistently exceeding exports. Figure 4 depicts these developments using data from the Bureau of Economic Analysis.

To assemble its trade data, the BEA draws on export and import data collected by the Census Bureau and U.S. Customs and Border Protection, respectively. BEA classifies and adjusts these data to align with the definitions used for calculation of GDP and other national measures (BEA 2014). Notably, GDP calculations likely understate output growth to the extent that they fail to capture all improvement in hard-to-measure information technology. This mismeasurement is the subject of ongoing research (Byrne, Fernald, and Reinsdorf 2016).

Information on international trade, and particularly changes in trade, is important to understanding the health of the U.S. economy and its likely future direction. In addition, detailed export and import data may be useful to firms that seek to understand the international business environment in specific sectors of the economy.

Energy economics took on special significance in the second half of the 20th century as energy demand rose dramatically, and as energy supply at times experienced significant volatility. Though U.S. energy exports once equaled imports, due to reliance on large domestic deposits of coal and oil, supply soon proved unable to keep up with rising demand. Since the 1960s, the U.S. economy has relied on large net imports of energy, and particularly imports of oil.

Over the past decade, domestic production of energy shot up as new technologies facilitated the extraction of previously unobtainable deposits. Estimates from the Energy Information Administration (EIA) provide crucial context for this development. Figure 5 shows exports rising sharply beginning in the 2000s, but still lagging energy imports. However, according to EIA projections based on these data, exports are forecast to exceed imports by 2026, marking a startling shift from the historical U.S. position as a large net importer of energy. Of course, projections are always subject to considerable uncertainty, and may miss the mark if unexpected economic or technological changes occur.

Businesses that participate in the energy market benefit from precise information about the scale and timing of market fluctuations. These firms are intensely interested in the likely path of U.S. production, which will have important effects on prices and availability of energy. For example, Charles Schwab cites EIA data on shale output to discuss the market outlook for energy investments (Sorensen 2017); EIA market forecasts also underpin discussions of future natural gas production and U.S. sales of combined cycle gas turbines for power plants (Ray 2015).

Though the agricultural sector now employs a small fraction of U.S. workers, agriculture is highly efficient and production continues to rise. Using data from the Census Bureau, the U.S. Department of Agriculture’s (USDA’s) Economic Research Service tracks the evolution of agricultural output over time by type of production. In recent decades the U.S. agricultural economy has diversified, with revenues from grain, oilseed, and livestock constituting roughly equivalent shares.

Figure 6 shows that agricultural exports ranging from grains to livestock have grown since 2000. This growth is fueled by steadily increasing demand in foreign economies (projected to exceed domestic growth in coming decades) as well as falling energy costs for agricultural producers in recent years (USDA 2016). USDA projects that growth in agricultural exports will continue, rising to $176 billion in 2025. Though accurate projections are hard to make even with great data, they are nearly impossible to make with poor or nonexistent data.

Understanding how the agricultural economy has evolved and will likely change in the future is important for many firms that either produce or rely on these products. Moreover, policy makers negotiating international trade agreements will benefit from a clear understanding of the economic impact of agricultural trade.

There are variety of measures of labor market health collected by the Bureau of Labor Statistics—including the payroll jobs count, the unemployment rate, the labor force participation rate, and more. The unemployment rate, defined as the number of people without jobs who are currently seeking work divided by the number of people in the labor force, is the labor force’s most cited summary statistic. The unemployment rate rises during recessions and falls during recoveries, is generally higher for disadvantaged groups, and tends to be lower in thriving local economies.

The BLS, in conjunction with the Census Bureau, has created a consistent measure of unemployment that dates to 1948. Today, this measure is generated by a monthly nationally representative survey of about 60,000 housing units called the Current Population Survey, and is published only weeks after data are measured (BLS 2014).

The human cost of unemployment, and especially long-term unemployment, means that changes in the unemployment rate are carefully watched by economists and policy makers. Unfortunately, the unemployment rate is frequently misunderstood, leading to confusion about what precisely it is and isn’t measuring. This statistic only includes those who are in the labor force, and therefore does not account for those who have become discouraged after a lengthy job search and stopped actively searching, for instance.

The BLS publishes a number of alternative measures that add information to provide a more comprehensive picture of the labor market (BLS 2017). One of these measures, called U-6, includes discouraged workers and others who are outside the labor force but want work, as well as those who have part-time work but would prefer full-time work. As shown in figure 7a, this measure is substantially higher than the traditional unemployment rate.

The CPS also generates the labor force participation rate, an estimate of the fraction of the adult population that is either working or seeking work. The decline in labor force participation, especially among prime-age men, is one of the most discussed problems in the labor market today. The share of men between the ages of 25 and 54 that did not participate in the labor force has increased from less than 9 percent in January 2007 to just over 11 percent in January 2017 (BLS 2017). This information draws attention to the importance of policies aimed at getting more of these men off the sidelines of the economy.

State governments, researchers, and businesses rely heavily on the CPS, which provides timely information about local labor markets in addition to the national unemployment rate.

For example, the stock market reacts to the monthly release of new employment figures on the first Friday of each month. The samples in other surveys, e.g., Gallup polls, are benchmarked to CPS data (ESA 2014). The CPS has been instrumental in documenting a number of trends, including the economic benefits of education (Council of Economic Advisers 2014), the well-being of military veterans (Olsen and O’Leary 2011), and the high rate of occupational licensure (BLS 2016a). The CPS also informs the BLS’s Occupational Outlook Handbook, which provides valuable career information for prospective college students and others in the workforce (ESA 2014).

One of the most important policy changes over the past quarter century was the shift to a work-based social safety net. These reforms particularly affected unmarried, low-skilled women with children who were both pushed into the labor market through the time limitations put on cash welfare benefits and pulled into it through the expansion of the Earned Income Tax Credit, which supplemented labor market earnings. To estimate the impact of these policy changes, analysts have examined employment trends among women aged 19 to 44, comparing unmarried mothers to single childless women and married mothers. As shown in figure 8, in the time period immediately following the pro-work policy reforms, unmarried mothers’ employment rates climbed by approximately 10 percentage points, while employment rates of the comparison groups were stagnant or falling. Since the early 2000s employment rates have been falling among all groups—a trend that analysts are still trying to understand.

The data for figure 8 come from the Current Population Survey, which helps answer a variety of important questions. Some of these questions are addressed in annual supplements, like the “Current Population Survey Annual Social and Economic Supplement,” conducted every March. The “March Supplement” contains information concerning family characteristics, health insurance coverage, income, work experience, receipt of government benefits, geographic mobility, and poverty.

This is particularly important to the study of the impacts of government programs that aim to address poverty. This survey has important limitations: research suggests that the “March Supplement” significantly understates the receipt of government transfers from programs like the Supplemental Nutrition Assistance Program (SNAP, or food stamps) and cash welfare, and therefore understates the benefits of these programs (Meyer, Wok, and Sullivan 2009). Administrative data, obtained directly from agencies that oversee the transfers, can help correct survey estimates. However, it would be a mistake to rely exclusively on administrative data: because of the breadth of the survey, the CPS affords insights that cannot be gleaned from administrative data alone. Using survey and administrative data in conjunction often yields better results than using either type of data alone.

Family structure shapes children’s outcomes. Children who grow up in a family with married parents do better on a variety of dimensions, from education and behavior during childhood to employment outcomes (American Enterprise Institute/Brookings 2015). Statistically, growing up with a single parent is associated with higher rates of poverty during childhood and lower rates of upward mobility when children reach adulthood. High marriage rates appear to have positive spillovers, as well. A recent study by Raj Chetty and Nathaniel Hendren found that the share of adults who are married in a geographic area positively predicts upward economic mobility of children who live in the area—regardless of the child’s own family status (Chetty and Hendren 2016).

The share of children living in married families has declined steadily over the past half century, with around 65 percent of children overall now living with married parents. However, the fraction of children living in married families—and the rate of decline in marriage—varies dramatically across socioeconomic status. Drawing on data from the Current Population Survey, figure 9 shows that children whose mothers hold four-year degrees are nearly as likely to live with married parents today as they were in 1979. Strikingly, their counterparts with mothers who have only a high school diploma or some college have seen the share of children living with married parents decline sharply over the past 40 years.

That the statistical agencies have been in place for decades, collecting much of the same information over that span of time, is a great advantage. Users of the data can confidently make historical comparisons and, just as importantly, can rely on the expectation that necessary data will continue to be collected and generated.

However, it is also important for data to adapt to the changing circumstances of an evolving economy, which sometimes presents new opportunities. In 2015 the Bureau of Labor Statistics introduced three new questions regarding an institution that had gradually become much more relevant to the labor market: occupational licensing. The new data revealed that roughly one quarter of employed workers currently hold a license issued by a local, state, or federal government. Typically issued at the state level, licenses are legally required in order to practice in a given profession.

Figure 10 shows that workers who are licensed are different from otherwise comparable certified workers—i.e., workers with a job credential that is not legally required to practice—in their propensity to move. Though licensed workers move within state at about the same rate as comparable certified workers, licensed workers are much less likely to move across states than are certified workers. This is what one would expect in a state-based system with very limited licensure reciprocity: once a worker has obtained a license, the cost of re-licensure in a new state is often prohibitive.

The BLS data will permit new research that helps state and federal policy makers rationalize their excessively burdensome occupational licensing regulations. Without information about how many are licensed, what occupations they work in, and how often they move, among other questions, it would be more difficult to subject this labor market institution to careful economic scrutiny.

While manufacturing employment has fallen steadily over time as a share of employment—from 28 percent of nonfarm workers in 1960 to 8 percent today—it nonetheless is an important gauge of local labor market health and the subject of policy interest (BLS 2016b). Manufacturing tends to be a cyclical industry that prospers during strong economic times and languishes in bad, and the Great Recession that started in 2007 was especially destructive to the sector. However, even with the prolonged economic recovery of the 2010s, manufacturing employment levels continue to fall short of their prerecession levels.

Using a large survey of business establishments conducted by the Bureau of Labor Statistics, businesses and researchers can track these trends. Every state in the United States experienced a drop in manufacturing employment during the Great Recession, and only Michigan has regained its November 2007 manufacturing employment level. In other states, declines are often as large as 10 to 15 percent. However, these declines occurred while real manufacturing output increased slightly over the same time period (BLS 2007–2016). As technological progress renders individual manufacturing workers more productive, manufacturing output can increase even while employment declines.

The BLS survey that produces these data—the Current Employment Statistics—uses a very large sample, benchmarked annually with even more comprehensive unemployment insurance records (BLS 2016b). Consequently, it is more carefully monitored on a month-to-month basis than are alternative measures. Financial markets in particular respond immediately and sharply to unexpected changes in payroll employment, because these provide valuable information about the current health of the economy and the future course of Fed policy (Huang 2015).

A long-run story of America, and of industrial civilization, is one of continuous migration away from the agricultural economy and into the manufacturing and service sectors. As agricultural technology improved, fewer workers were needed to produce the same amount of output. These workers flooded into cities and suburbs that were better integrated into the emerging manufacturing and service sectors.

In recent years, this migration pattern has largely persisted, with rural populations continuing to decline in most of the country. The trend is particularly pronounced in the eastern half of the United States, as shown by calculations from the USDA’s Economic Research Service.

Government-collected data show migration trends that inform state and local policy.

The federal government’s social insurance programs have changed considerably over time. Accurate information about their coverage and performance is essential to having a productive discussion about the possibilities for reform.

Social Security Disability Insurance is a good example. Benefits provided by this program are intended to help eligible former workers who have become disabled, and are unable to work, due to a medical condition lasting at least one year. Though SSDI recipients are not the only disabled individuals receiving government assistance—such benefits are also provided through veterans’ disability compensation and Supplemental Security Income, for example—about two thirds of those who are disabled and receive benefits are on SSDI, and it has received the most research and policy attention.

There have been large increases in the number of beneficiaries relative to the number of workers over the past 40 years. In 1970, for every 100 people in the working-age population, 1.3 received SSDI benefits. By 2015 that number had increased to 4.5. While some of the increase is an artifact of the aging of the baby boom generation, another portion has been driven by screening rules that have been loosened over time.

The medical conditions that lead to SSDI receipt changed markedly during the large increase in disability rolls starting in the mid-1980s, as shown in figure 13a. Data from the Social Security Office of the Actuary allowed Autor and Duggan (2006) to explore the role of loosened screening rules for certain disabilities, namely mental and musculoskeletal disorders, in determining the number of SSDI recipients and total cost of the program. The authors find that liberalized screening rules have contributed to increasing use of the program.

The U.S. economy is extremely large and diversified across many different types of activities. In recent years the role of the United States as an energy producer has increased in striking fashion. The advent of cost-effective hydraulic fracturing, among other technologies, has generated booms in places like North Dakota that have recoverable oil and natural gas.

These booms have altered local labor markets in important ways. Government data like the Longitudinal Employer–Household Dynamics shed light on trends in employment, hiring, job creation and loss, and earnings. Through this Census Bureau program, those interested in local economies can see these trends with unprecedented detail going back several decades (Census Bureau 2017). In addition, the Census Bureau provides these data through free online tools that create detailed maps and figures, showing details such as commuting flow and the impact of disaster events like tropical storms on local economies (Census Bureau 2017).

Restricting focus to workers with a high school diploma or less, much of the uptick in hiring during the energy boom has come in the form of workers who were already employed, rather than those who were persistently nonemployed prior to hiring. This type of detail is useful for understanding the labor market impacts of technological shocks.

It is well known that attainment of a four-year college degree confers sizable earnings benefits for the average worker. After 10 years in the labor market, the typical four-year college graduate can expect to earn $27,000 more per year than the typical high school graduate. Given the high and increasing expense of a college education, this is no small part of the motivation for most students’ enrollment.

What is less well-understood is the variation in earnings that can be expected across different groups of graduates, for example across different college majors. As shown in figure 15, those with a degree in computer engineering earn a median salary of $89,200 at 10 years after graduation; those with a degree in drama and theater arts earn less than half this amount with a median salary of $41,800. To be sure, majors may also differ in terms of the nonpecuniary benefits they confer and in the types of students who enter them. Nonetheless, clear information about likely earnings trajectories is highly useful to students—and their parents—as they make large investments in their skills.

The Census Bureau’s American Community Survey makes analysis like this possible, because it is large and covers a range of economic and demographic topics. A large sample size means that there are enough individuals to draw statistically valid conclusions not only about the most popular majors, but also about smaller ones. Furthermore, having college major and earnings in the same dataset as geographic location and demographic characteristics allows users to analyze differences across important dimensions.

In an economy that is continuously remade by technological progress, workers are increasingly forced to adapt to changes in the demand for skills and occupations. As businesses shift their needs toward some occupations and away from others, it is important that individual workers—young and old alike—have the benefit of good information about the earnings trajectories associated with different careers.

Within a major field of study, graduates go on to pursue a range of occupations that vary widely in terms of annual earnings. For example, among 35- to 44-year-old college graduates who majored in finance, about 25 percent are employed in the top three occupations: 10 percent are accountants and auditors, 9 percent are financial managers, and 6 percent are other managers. Average earnings vary widely across these occupations, with financial managers outearning accountants and auditors by 28 percent for males and 46 percent for females.

Data on earnings by major field and occupation are useful factors for students to consider when making consequential decisions about their educational investment and career aspirations. Policy groups like The Hamilton Project and others take the raw data collected by the Census Bureau to produce user-friendly data tools that help provide this needed information.

The modern U.S. economy is characterized by a tremendous number of constantly changing prices. Making sense of these is no easy task, yet doing so is an absolute necessity for tracking economic progress over time, adjusting wage contracts for changes in cost of living, properly interpreting financial market prices, and conducting monetary policy, just to take a few examples.

Dating all the way back to 1913, the BLS has calculated the standard measure of changes in prices, known as the Consumer Price Index (CPI). Though the BLS has improved its methodology over the years, in part due to input from experts on the 1996 Boskin Commission, the fundamental essence of the measurement is the same: observe the basket of goods and services purchased by consumers, then track changes in those prices over time, with periodic updates made to the composition of the basket (Gordon 2006). Of course, different consumers make different purchasing decisions, so the index will necessarily reflect a typical consumer experience; people who buy other baskets may experience lower or higher inflation rates.

The modern procedure for calculating the CPI is the product of decades of research and experience. It is not sufficient simply to observe posted prices for goods and services; one important complication is that the quality of many products is always improving. For example, a $1,000 computer in 2000 was much less powerful and useful than a $1,000 computer in 2017, indicating that the effective price of computing services has fallen even if consumers are paying the same dollar amount for a computer. The BLS has implemented procedures for estimating this sort of change in quality. Prices of characteristics or components of the good (e.g., hard drives inside personal computers) are used to calculate the change in price of the good (e.g., the computer) that would be observed if quality were held constant (BLS 2010). Other technical challenges to inflation measurement rise because consumers differentially substitute across goods when relative prices change, when new products are introduced and must be incorporated into the market basket, and when consumers change where they shop—e.g., to warehouse or online stores. The BLS is continuously studying these issues to get the most accurate, unbiased measurement of inflation possible.

Inflation calculations are important in part because payment increases under Social Security and other compensation programs are indexed to them. If inflation is over-stated, the total cost of the benefits will rise, causing a ripple effect through the budget. On the other hand, if it is under-stated, seniors who depend on the benefits will have a harder time making ends meet. In addition, families often receive cost-of-living increases in their pay that are calculated from government inflation statistics. Some wage contracts are written in a way that explicitly accounts for changes in the CPI, providing for automatic cost-of-living raises (Miller 2015).