Questions about statistics—“What is the population of London?” “How many people are diagnosed with cancer annually?”—have long been a staple at library reference desks. Most librarians are familiar with the print ready-reference sources that were traditionally kept within reach to answer them—the Statistical Abstract of the United States, The World Almanac and Book of Facts, and the CIA World Factbook, to name just a few—and, now, with the online equivalents of these publications. Not too many years ago, however, “data reference” was a specialized library service available primarily at research libraries. Few librarians outside of those institutions would ever encounter a question like “I need thirty years of time-series data on the production of beef in Texas at the county level,” and even fewer librarians would have been comfortable with the intricacies of working with data on reel-to-reel tapes and punchcards.
Two changes—the rise of the Internet, which has made disseminating data much less complicated, and the spread of statistical software packages into the undergraduate curriculum—have blurred the line between statistics reference and data reference. As more data and more user-friendly tools to work with data have become available, interest in finding and using quantitative data has grown. Yet data-related reference questions are often some of the most daunting questions for general reference librarians. My hope is that this guide makes those questions a little less daunting.
For Whom Is This Guide Intended?
Although I hope that all reference librarians find this guide helpful, its primary audience is librarians at public libraries, high school libraries, and the majority of academic libraries that do not employ staff dedicated to data reference. Librarians working with undergraduates at large research institutions may also find that the guide allows them to answer some basic data reference questions without having to refer students to a dedicated data services librarian.
Because the intended audience is primarily librarians at institutions that have not made a major financial commitment to data services, this guide focuses on freely available, online sources for data. In many subject areas, much of the most frequently used data can be found online for free if one knows where to look. On the occasions when a commonly requested type of data is not freely available from any online source, subscription databases or print series containing the data may be mentioned. Librarians whose institutions subscribe to many statistical databases may find Lynda M. Kellam’s guide Numeric Data Services and Sources for the General Reference Librarian (Chandos, 2011), which focuses more on subscription databases for data reference, to be helpful for their situation.
This guide focuses on data available through English-language interfaces, although it occasionally makes reference to data available in foreign-language interfaces. It does not include certain specialized scientific data sources, such as gene sequence databases or databases of chemical structures; nor does it include qualitative data sources. Instead, the focus is on quantitative social science data, broadly defined—that is, data that is primarily useful in the context of social science disciplines such as economics and sociology, as well as scientific data that frequently is or can be deployed in the context of government policy making, such as data on public health, climate change, or natural disasters.
In the following thematic chapters 2–26, the discussion is divided into three categories: major sources for U.S. data, major sources for international data, and minor sources. The sites listed as major sources are large databases, typically from the major international or federal government agencies with responsibility for a given area; they are the most likely resources for answering most common reference questions in their areas.
Minor sources were selectively chosen from the dozens of smaller data sources in each area because they fill a gap in the data that is available from the major sources or because they present a selection of data from the major sources in a more user-friendly interface.
Data Jargon
To communicate with patrons who need data, and to get the most out of the rest of this book, it helps to be aware of certain types of data jargon. As with any kind of reference question, before undertaking a search for data it is important to be sure you understand exactly what your patrons want. In the case of data that means understanding not only exactly the topic on which they need data but also the characteristics of the data they need: whether they need statistics or data, what their desired universe and unit of analysis are, whether they want time series or cross-sectional data, and more. The italicized terms, and other related concepts, are defined below.
Data versus Statistics
The terms data and statistics are often used as if they mean the same thing (sometimes even in this book), but in fact there is an important distinction between them. Data is raw input for some sort of statistical analysis. A list of all of the traffic accidents in New Jersey in 2010, with information about the drivers (e.g., age, blood alcohol content, whether they were using a cell phone at the time of the accident) and the accident (e.g., time of day, weather, number of cars involved) would be data. Unless you are interested in information about a specific accident (say, if you are a lawyer representing one of the drivers), this list is not likely to be terribly informative by itself. To be able to say anything about road safety in New Jersey generally, you would need statistics. Statistics, in this context, are the results of a statistical analysis of the data. Statistical analysis does not have to mean some sort of complicated multivariate regression. In many cases, it is simply an average, a percentage, or a frequency. For example, the percentage of accidents that occur during snowstorms, or the frequency of accidents involving teenage drivers, are examples of statistics that could be generated from this data.
Certain pieces of information can be treated as either statistics or data, depending on what the user wants to do with that information. Take, for example, the unemployment rate of the United States in November 1980: 7.5 percent, according to the Bureau of Labor Statistics.1 This number is a statistic—the product of statistical analysis of the data gathered from individuals in the labor force by the Current Population Survey—and, for a history student writing a paper about how economic conditions affected the 1980 presidential race, it might be all that he needs. However, an economics student who wants to test the hypothesis that changes in the price of oil affect the unemployment rate in the United States might treat that same number as a data point: one of many monthly unemployment rates that she will use to run regressions, thereby generating other statistics.
Subtypes of Data: Microdata and Aggregate Data
The term microdata is used to refer specifically to the kind of data that is, unequivocally, data rather than statistics: raw observations, survey responses, and the like that are not the product of any kind of statistical analysis or summary. Microdata often refers to data about individual people. A spreadsheet where each row contains a single person’s responses to the questions on a survey is an example of microdata. The traffic accident data mentioned above would also be considered microdata, as would information about individual stores collected as part of the Economic Census or daily rainfall totals for a specific location as reported by the National Weather Service.
The converse of microdata is aggregate data—data produced by some sort of statistical procedure, such as averaging or, in the most basic and perhaps the most common example, simply adding up the number of cases. Monthly unemployment rates are an example of what might be referred to as aggregate data, as are election results by precinct and data about retail establishments by county. If the description of a data table ends with “by state,” “by gender,” or something similar, you are almost certainly dealing with aggregate data.
Public-Use Data versus Restricted Data
In this book, I try to focus on freely available data, where “freely” means two different things: that no monetary payments are required to access the data, and that there are no onerous restrictions on who may have access to the data or the conditions under which they may use the data. Nevertheless, some of the resources mentioned in this book contain both public-use data (data with either no restrictions at all on access or with minimal registration requirements) and restricted data (data for which access is conditional and requires an approval process). Accessing restricted data may be as simple as filling out a short online form and waiting for approval, or it may be as complicated as completing an extensive certification process and traveling to a designated data enclave (a secure facility with equipment and policies designed to prevent the unauthorized sharing of confidential information) to use the data.
Why is some data public-use and other data restricted? The section “What Data Is Not Disseminated?” (p. 10) explains how concerns about privacy and confidentiality can lead to restrictions on data access.
Surveys, Censuses, and Administrative Data
Surveys are one of the most common methods of gathering data in the social sciences. In a survey, data is gathered from a sample (a small subset) of the population (sometimes called the universe), and that data is then used to make estimates about the entire population. For example, a typical public opinion survey might do telephone interviews with one thousand people (the sample) randomly selected from all adults over 18 years of age who reside in the United States (the universe). These thousand people’s responses would then be used to estimate how the entire adult population of the United States feels about, say, the president’s job performance. This contrasts with a census, which is often conducted like a survey in that people are asked to answer questions over the telephone or to fill out a form, but, instead of contacting a sample of the population, the goal with a census is to contact every single person (or, in the case of the Economic Census, the Census of Agriculture, or the Census of Jails, every single institution) in the population. It also contrasts with administrative data—in which official records (say, birth certificates, tax returns, or customs declaration forms) are used to gather data, rather than asking people or institutions directly to provide information about themselves.
Cross-Sectional, Longitudinal, and Time-Series Data
Many studies gather data at only a single point in time: a survey is written, people respond to it over a few days or weeks, the data is analyzed, and then the study is complete. This type of data is relatively cheap and easy to gather, but it is difficult or impossible to use it to examine changes over time. These types of studies are called cross-sectional studies or surveys. Because so many studies are cross-sectional, the types of data that are collected over time are more challenging to find. However, because they can be used to do more sophisticated types of analyses, they are particularly valuable to researchers.
The two main types of data that have been collected over time are longitudinal data (sometimes called panel data), which follows the same individuals (the “panel”) for months, years, or sometimes decades; and time-series data—any data collected at relatively regular intervals over an extended period of time. The major macroeconomic indicators—such as gross domestic product and the monthly unemployment rate—are time-series data, since they have been reported monthly or annually for many decades. Daily stock prices for a group of companies are also time-series data. Certain ongoing surveys and public opinion polls intentionally ask the same question in the same way over many years, which creates a time series of public opinion on certain topics. Longitudinal data is relatively rare, although several longitudinal data sets are mentioned in chapter 20. Time-series data on a variety of topics is readily available.
Unit of Analysis and Unit of Observation
In the context of data reference, the distinction between unit of analysis and unit of observation is subtle but important. The unit of analysis is the type or level of thing that the patron wishes to study. For example, in an education study, plausible units of analysis could be students, teachers, schools, school districts, states, or countries. The unit of observation, on the other hand, is the type or level of thing about which the original researchers gathered data. Different units of analysis often, but not always, require different units of observation in the data sets. For example, a researcher who wants to compare average test scores in different school districts (using school districts as the unit of analysis) would want different data than a researcher who wants to study how students’ socioeconomic backgrounds affect their test scores (using students as the unit of analysis), even though both might approach the reference desk asking for “data about students’ test scores.” The patron who wants to compare test scores across school districts may be able to use data with individual students as the unit of observation, as long as the data file contains a school district for each student; the patron could use that data to calculate average test scores by school district. The opposite is not true: the patron who wants to use students as the unit of analysis is not likely to be satisfied by data with school districts as the unit of observation.
North American Industry Classification System Codes
North American Industry Classification System (NAICS) codes (www.census.gov/eos/www/naics/) are used by the governments of the United States, Canada, and Mexico as well as some private data sources, to classify businesses and workers into industries for statistical purposes. These codes, which range in length from two digits (designating broad sectors) to six digits (designating specific industries) are arranged hierarchically, so that adding additional digits to the end of a broader code allows one to designate a more narrowly defined industry within the sector. For example, by adding digits to 62, “Health Care and Social Assistance,” one can move to 622, “Hospitals,” and then to 6222 “Psychiatric and Substance Abuse Hospitals.” For many industries, the most specific code is actually a five-digit code (or occasionally even a four-digit code), and the five-digit codes are the most specific codes that can reliably be used for comparisons with data disseminated by Mexico and Canada. In the cases where a five-digit code is the most specific, some data interfaces offer an identically labeled six-digit code with a “0” on the end as an option. The examples below illustrate the system, using the 2012 version of NAICS:
44–45 | Retail Trade |
445 | Food and Beverage Stores |
Specialty Food Stores |
|
44529 | Other Specialty Food Stores |
445291 | Baked Goods Stores |
51 | Information |
515 | Broadcasting (except Internet) |
5151 | Radio and Television Broadcasting |
51512 | Television Broadcasting |
515120 | Television Broadcasting |
Different sets of classification codes are used by the United Nations and other organizations, and also for international trade data. Although the codes differ, the basic concept of adding digits to numbers to designate more specific industries or products is incorporated into all of the systems. These systems include the Standard International Trade Classification (SITC), a product classification (e.g., “beverages,” “textile yarn”) that is managed by the United Nations; International Standard Industrial Classification (ISIC), an industry classification (e.g., “manufacture of beverages,” “spinning, weaving and finishing of textiles”) that is also managed by the United Nations; the Harmonized System (HS), a product classification that is managed by the World Customs Organization; and Schedule B, product classification managed by the U.S. Census Bureau that is based on the HS. These codes are primarily encountered in the sources listed in chapter 16.
Who Gathers and Disseminates Data?
Data collection and dissemination are expensive, time-consuming enterprises. Developing and testing a survey, hiring and training interviewers, paying postage or telephone charges, following up with people who do not respond, cleaning and analyzing the data: the costs add up quickly. Plus, in many areas it is not a profitable activity for the organizations that undertake it; much of the time, the decision to gather and share data is driven by academic or administrative motives rather than profit. These two factors of data gathering and data sharing—high costs and often low pecuniary rewards—mean that in many areas governments and intergovernmental organizations with large budgets and no need to show a profit are the best or only source of data. This is especially true for data about small geographic areas, such as provinces and counties. Thousands of times more respondents are needed to be able to calculate accurate estimates for each of the three thousand-plus counties in the United States than to calculate an accurate estimate for the entire United States, for example, which makes it substantially more expensive to conduct a survey capable of producing county-level data than one intended to produce only national data.
Additionally, to get comprehensive and complete data, it is helpful to have strong incentives to cooperate with the data collection process—another area in which governments have an advantage over private organizations. For example, in the United States business executives can be fined thousands of dollars for not responding to the Economic Census or for responding with false information.2 Private organizations that wish to survey U.S. businesses have a much harder time finding similarly effective incentives to convince businesses to cooperate with their efforts.
Both the U.S. federal government and the UN system have extensive mechanisms for gathering and disseminating data on a wide range of topics. A brief overview of both institutions’ data-gathering and data-disseminating agencies follows; for complete information about what data is disseminated by each of these agencies, please see the relevant listing in chapter 2.
United States
Although dozens of agencies, bureaus, and other divisions within the U.S. government produce some data resources, thirteen are considered the major statistical agencies:
Bureau of Economic Analysis
Bureau of Justice Statistics
Bureau of Labor Statistics
Bureau of Transportation Statistics
Census Bureau
Economic Research Service of the Department of Agriculture
Energy Information Administration
National Agricultural Statistics Service
National Center for Education Statistics
National Center for Health Statistics
National Center for Science and Engineering Statistics
Office of Research, Evaluation and Statistics of the Social Security Administration
Statistics of Income Division of the Internal Revenue Service
Approximately eighty-five other agencies, from the Consumer Product Safety Commission to the Fish and Wildlife Service, also collect some data on certain topics and release it to the public.3 These organizations cooperate and coordinate with each other in various ways, from jointly developing standards through the Federal Committee on Statistical Methodology to operating a portal, FedStats (www.fedstats.gov), with links to many of the federal government websites that contain data and statistics.
United Nations
As with the United States, in the UN system responsibility for gathering and disseminating data is spread across many separate agencies. Chief among these is the United Nations Statistics Division, which is responsible itself for gathering statistics in certain areas (including, among others, commodities, energy, and national accounts) and which also provides assistance and standards for the statistics-gathering activities of others. Other UN agencies that gather and disseminate data include the following:
Food and Agriculture Organization
International Labour Organization
International Telecommunications Union
Joint United Nations Programme on HIV/AIDS
United Nations Children’s Fund
United Nations Development Programme
United Nations Educational, Social, and Cultural Organization
United Nations Framework Convention on Climate Change
United Nations High Commissioner for Refugees
United Nations Industrial Development Organization
United Nations Office on Drugs and Crime
United Nations Population Division
World Health Organization
World Meteorological Organization
World Tourism Organization
Private Data Collection and Dissemination
Although government and intergovernmental agencies are often the best sources for data, on some subjects private organizations have a distinct advantage. For example, political constraints sometimes affect the data that government agencies can gather. In the United States, it is illegal for the Census Bureau to ask questions about religious affiliation. France forbids its government agencies from collecting data about people’s race or ethnicity. When restrictions such as these are in place, one must rely on nongovernmental sources for data. Also, government and intergovernmental agencies typically do not ask public opinion or polling type questions in their surveys, such as questions about whether people favor or oppose various pieces of pending legislation or how good of a job they think the president is doing. Again, for this type of data, private organizations are the best sources.
What Data Is Not Disseminated?
Not all data that is gathered is disseminated, often for reasons of cost. Calculating aggregate data and preparing microdata for public release cost money, so typically only the data that is likely to be of highest interest goes through the process that leads to public dissemination. When the microdata is released, however, patrons can often use it to calculate the aggregate data or statistics they want even if the original researchers did not calculate or disseminate those particular figures themselves. Luckily, many major surveys, including the General Social Survey (chapter 24) and the Current Population Survey (chapter 20), do disseminate much of their microdata to the public.
Even when microdata files are released to the public, as with the aforementioned surveys, data points that could compromise respondents’ privacy are usually not included in these public-use data files. Although these microdata files are composed of data about specific individuals, this microdata is carefully deidentified—information that could identify individuals, such as names and street addresses, is removed—and cleaned to be sure that it is either impossible or extremely difficult to identify the specific person or business who provided each data point. This process sometimes involves removing potentially useful data from the microdata files. In particular, very old ages and very high incomes are rare and make it relatively easy to figure out who a particular individual is in a microdata file, so these are often top-coded—that is, for ages or incomes over a certain level, the file does not give the specific age or income, just an indication that it is over that level. Sometimes, but not always, data with potentially identity-revealing data such as this is available as restricted-use data, which can be accessed only by well-qualified researchers under strict conditions of use.
The suppression of data to protect respondents’ privacy is a particularly important factor to consider when looking, not just for microdata, but also for data for small geographic areas. The Census Bureau and other government agencies do not release aggregate data about very small groups of people or businesses, since this data could allow someone to make accurate guesses about the characteristics of individual identifiable people or business establishments. For example, the Census Bureau conducts a survey called County Business Patterns that provides county-level aggregate data about the number of establishments, number of paid employees, and payroll by industry. When disseminating this data, the Census Bureau suppresses information such as the number of employees and annual payroll for industries with only a handful of establishments in that county. Otherwise, if you worked at Establishment A and knew for certain how many employees you had and what you were paying them, and if your friend worked for Establishment B and knew for certain how many employees it had and what it paid them, the two of you together could use the data from County Business Patterns to make a good guess as to how many employees Establishment C has and how much it is paying them—information that Establishment C does not want to be shared with its competitors.
Administrative data is more complicated; some is released only when it has been aggregated to the point that individual cases cannot be identified, but other types are intentionally disseminated as individually identifiable microdata. For example, by U.S. law political campaigns must release the names, addresses, occupations, and employers of people who donate more than $200 to any single federal campaign. People who have been arrested or convicted of a crime may have some of their personally identifying information revealed in conjunction with that legal action. In many states, the names, job titles, and salaries of public employees are public information that can be downloaded and used as data. These examples are, however, exceptional; for most subjects of interest, individually identifiable microdata is not available.
Beginning Your Data Search
For librarians accustomed to the relatively organized world of books and journal articles, trying to find data can be a frustrating experience. There is no single WorldCat-like portal for data; there is no data-specific set of standard subject headings; and the standards for citing data in publications are weak to nonexistent, making it challenging to track down a specific data set from a secondary source. (See appendix A for more information about the problems of data citations.) As a general rule, the following progression is useful. First, try the general sources listed in chapter 2. If none of those appear likely to contain the necessary data, move on to the subject-specific data resources listed in chapters 3–26. Then, if the data still cannot be found, try the ideas in chapter 27. As you become more familiar with many of the most common data sources and the dissemination of data more generally, you may begin to find that a different method of attack works better for your personal research style and the types of questions most commonly asked by your patrons.
1. “Data Series LNS14000000,” Bureau of Labor Statistics, http://data.bls.gov/timeseries/LNS14000000.
2. United States Census Bureau, “2012 Economic Census FAQs,” 2012 Economic Census Advance Information, http://bhs.econ.census.gov/bhs/ecad/SUR1_1.html.
3. The current list of federal agencies that spend more than $500,000 in a year on statistical activities can be found in the annual publication Statistical Programs of the United States Government (www.whitehouse.gov/omb/inforeg_statpolicy#sp/).