At the Intersection of Official Statistics and Public Policy:

At the Intersection of Official Statistics and Public Policy:

Confronting the Challenges

Katherine K. Wallman, Chief Statistician of the United States

On January 19, 1999, the President of the United States opened his State of the Union Address as follows:

- America has ... nearly 18 million new jobs,

- wages rising at more than twice the rate of inflation,

- the highest home ownership rate in history,

- the smallest welfare rolls in 30 years,

- and the lowest peacetime unemployment since 1957.

This exceptionally high profile attention to the products of our Nation's statistical system bears witness to the extent to which our policy makers and our public depend on data to measure and understand conditions and trends in our economy and society.

Our democracy and economy demand that public and private leaders have unbiased, relevant, accurate, and timely information on which to base their decisions.Data on real Gross Domestic Product, the Consumer Price Index, and the trade deficit, for example, are critical inputs to monetary, fiscal, trade, and regulatory policy.They also have a major impact on government spending, budget projections, and the allocation of public funds.Economic data, such as measures of price change, have as well a significant influence on interest rates and cost-of-living adjustments that affect every American who runs a business, saves for retirement, or mortgages a home.Taken together, official statistics on demographic, economic, and social conditions and trends are essential to inform decisions that are made by virtually every organization and household.

Among the most influential statistics produced by our Federal government is the Consumer Price Index, or CPI, which measures the average change in prices over time for a fixed market basket of goods and services.As an economic indicator, the CPI is used by the Executive Branch, the Congress, and the Federal Reserve Board to determine and evaluate government economic policy.As a consequence of statutory requirements, the CPI directly affects the incomes of millions of our citizens through Federal programs that deliver benefits to individuals.These include more than 48 million Social Security beneficiaries, close to 20 million food stamp recipients, and almost 4 million military and Federal Civil Service retirees and survivors.Changes in the CPI also affect more than 26 million children through adjustments to the School Lunch program, and more than 2 million workers whose wages are tied to the CPI under collective bargaining agreements.According to a recent estimate, a one percent reduction in the CPI's annual growth rate would reduce Federal outlays by approximately 80 billion dollars over a five year period.

Similarly, information from our Decennial Census of Population and Housing affects Americans every day.Data on the number and characteristics of the population are used by State and local governments to plan schools and highways, by the Federal government to distribute close to $200 billion annually for health care and other programs, and by businesses in making their economic plans.Census data are used to reapportion Congressional seats among the States and to draw legislative districts within the States.

More specific examples of the uses of official statistics in informing policy discussions and fostering policy changes affecting the Nation's population abound.

- Statistics from the National Health Interview Survey in the late 1950's, and a more sharply focused Social Security Administration Survey of the Aged in 1963, showed that only 50 percent of the population over 65 years of age had any form of comprehensive health insurance coverage, and that much of that insurance coverage was inadequate, as measured by the proportion of the hospital bill that was paid.Testimony and legislative reports leading to amendments to the Social Security Act in 1965 reflected these facts, with the result that nearly everyone in this age group now has Medicare or Medicaid coverage for at least a major part of the costs of hospital care.Similarly, more recent information is informing proposals such as those that would allow Americans in the 62 to 65 year age cohort to purchase Medicare coverage and would offer a Medicare "buy-in" to workers between the ages of 55 and 62 who have lost company-sponsored health care coverage.As we look to the future, the Medicare population is expected to double over the next 30 to 35 years.But will the 85 year old of tomorrow look like the 65 year old of today in terms of health care needs?And what will be the financial status of this population?To keep the program relevant, relevant data will be crucial.

- In the early 1980's, statistics from the National Health and Nutrition Examination Survey (NHANES) provided, for the first time, national data on levels of lead in the blood.These data indicated that 78 percent of Americans had blood lead levels that were considered too high.Over the course of the survey, a clear relationship was shown between the lower amounts of lead in gasoline sold and the lower blood lead levels being observed in study results.The results from an analysis of these data by an expert committee of statisticians were presented as primary evidence at Congressional and Environmental Protection Agency (EPA) hearings, which in turn led to a decision by the Office of Management and Budget to reverse its proposal to relax requirements for reducing lead in gasoline and to instead support the EPA proposal to set a firm timetable for the complete elimination of sales of leaded gasoline.Additional programs were implemented to remove all lead from paint and other commercial products.By 1991, the next NHANES survey showed that the percentage of Americans with high blood lead levels had dropped sharply to just over 4 percent.The same survey also identified groups still at risk, and formed the basis for restructuring prevention efforts in the 1990s.

- During the 1980s and early 1990s, statistics garnered from the Nation's vital records system together with data from hospitals highlighted high and rising rates of cesarean deliveries in the United States.When compared with data from other industrialized countries, the analyses indicated that U. S. cesarean delivery rates generally were higher, and that the reliance on this procedure had not translated into better health for the mother or the infant.The data provided by the National Center for Health Statistics prompted a review of obstetrical practice and revision of standards for performance of cesarean deliveries.Using the new guidelines and new procedures in hospitals, cesarean rates leveled off in the late 1980s and have declined during the 1990s.

- In 1983, the publication "A Nation at Risk" called into question the quality and effectiveness of the entire educational enterprise in the United States.A major reason that this publication gained attention from both policy makers and the public stemmed from the data presented -- the fact that our students scored worse on international assessments of mathematics and science than those in most other countries participating in the study.The statistical information initiated what is now more than a decade and a half of reform movements to improve educational performance.

In recent years, international assessments have indicated that while our 4^th graders outperform their peers in other countries in mathematics and science, our 8^th graders are near the average, and our 12^th graders rank near the bottom.These findings have triggered changes in our laws to support State and local standards-based reform efforts and speed the use of technology in education.Early results from these policy shifts in the mid-1990s indicate that all but one of the States have now developed content standards at least in the areas of reading and mathematics.

Similarly, statistical analyses have been at the crux of education policy changes affecting our youngest Americans.Data have indicated that preschool programs for disadvantaged children improve the participants' subsequent performance in regular school, reduce their rates of criminal activity, reduce their chances of premature parenting, and improve their likelihood of attending college.Proposals to achieve full funding that would permit every eligible child to participate in Headstart are firmly grounded in statistics concerning the effectiveness of the program.

As these examples -- and many others we could cite -- indicate, the terms of reference for interaction between statisticians, policy makers, and the public have changed markedly since the days when members of the statistics profession were largely concerned with "coding, tabulation, and recognition and correction of obvious errors in data collection."Whether we as statisticians like it or not, we are faced with what economist James Bonnen has described as "a growing, intimate embrace between statistics and public policy decision making" that "has greatly increased the significance and decision value of the statistics we produce."As sociologists William Alonso and Paul Starr observed several years ago,

"So well institutionalized are statistics such as the unemployment rate, the money supply, and various price indices that the date and even the hour of their release are regular events in the political and economic calendar, setting off debates on the performance of administration policy and influencing both stock markets and elections."

The public policy uses of statistical data have increased exponentially over the last half century.Today, we use statistical data to redistribute income and to fine tune the economy.We mandate the use of official statistics in many of our country's laws, and we use government statistics to evaluate policy alternatives.Competing interpretations of statistics on income, inflation, and unemployment figure prominently in our election campaigns.We use statistics to evaluate risk in legislative policy on toxicity and the environment, and we use statistics in our governmental efforts to expand the coverage and contain the costs of health care.

With each session of Congress, the list of uses gets longer; our legislators have for some time been in the process of essentially shifting the burden of detailed decision making to formulas based on statistical series.In many cases, the data are not designed for such tasks -- and are woefully inadequate for the legislated purposes -- but still are called upon to bear the burden.It is little wonder, then, that the decibel level continues to rise when questions are raised about the accuracy of the Consumer Price Index or the Decennial Census population counts.

At this intersection of statistics and politics, we find some ironies.Some concern desires (or a seeming lack thereof) for accuracy of data.Others have to do with interest (or an apparent lack thereof) in adequately funding statistical programs.One irony that has always struck me is the debate that took place a decade ago concerning the content and sample size for the 1990 decennial census.While that census ultimately captured virtually all of the information that data users have come to rely on, the concerns that were laid on the table in 1987 (about the burden and cost of data collection) continue to engage the attention of decision makers in the Administration and the Congress.Indeed, as we have prepared for Census 2000, a cardinal rule for justifying any question we will include has been that the data must be required by some Federal statute.

Such disputes about the content of the Federal government's statistical endeavors are not as new to the scene as some embroiled in the debates of recent decades may have thought.At the time of the first decennial census, according to an account provided by historian Margo Anderson, Congress discussed whether the census should collect any information beyond a mere head count necessary for apportionment.James Madison suggested a rather elaborate schedule to categorize the population.The Senate deleted the proposed census of occupations, because, Madison wrote to Thomas Jefferson, they considered it "a waste of trouble and supplying materials for idle people to make a book.

The debate a decade ago over the need for the decennial census long form data brings clearly into focus the disconnect between the substantial uses of official statistics that are called for in legislation and the support that is provided by the Congress for producing the data on which these legislatively based programs rely.

Somewhat hazier at present, but appearing on our horizon, is the statistical gap we must bridge in an era of devolution.National mandates are bringing new responsibilities to State and local governments to plan and evaluate programs.One is generally hard pressed, however, to find within these recent laws any provision of resources to support the statistical infrastructure required to efficiently and effectively administer programs - let alone any allocation of resources to ultimately evaluate these initiatives with nationally comparable statistical data.As programs devolve to the States, we have from the statistical point of view a "double whammy" - not only do we lose the data sources (administrative records) that previously were available, we also stand to lose comparability of data across States and local areas.

Yet another key issue that demands our attention concerns what I believe is the thorniest - and potentially most consequential - issue facing our national statistical system.This is the growing tension between the need to protect the confidentiality of individual responses, and the desire to provide the broadest possible access to information.Government statistical offices derive their mandate for data collection and dissemination from a citizenry that demands at once both quality information to drive public policy and protection of the individual respondent from privacy invasion and administrative harm.In response to these cross-pressures, statistical agencies are actively pursuing approaches such as encryption and licensing agreements that extend responsibility for confidentiality protection to the data user.Striking the proper balance between permitting access to accomplish compelling and legitimate research, and incurring the risk, however remote, of inadvertent revelation of individual information is, I believe, a fundamental concern and challenge for official statisticians.

Many researchers find access to government data increasingly desirable.The newer data bases are more comprehensive, of better quality, and - with improved data base management techniques - better structured.At the same time, the individuals and institutions that provide the data residing on government data bases - as well as the agencies that sponsor the collection of such information - are well aware that the same technologies that extend analytical capabilities also furnish the tools that threaten the confidentiality of data records.This awareness has the potential to erode -- or at least to undermine-- our respondents' confidence in the privacy of the information they provide.

Electronic dissemination is truly a boon to national statistical offices anxious to make their data more accessible and useful -- and to user communities equipped to handle the wealth of available information.But this technology remains to a degree a bane, for while we have taken monumental strides in making our nation's statistics electronically available, attention to documentation in electronic media has lagged.And, I continue to argue, as I have for almost a decade, that the gap between our citizens' computer literacy and their "statistical literacy" remains significant.

As U. S. government statisticians seek to confront the challenges at the intersection of official statistics and public policy, we do so in an exceptionally charged partisan political environment.A confidant of the Speaker of the U. S. House of Representatives recently stated, for example,

The Speaker and virtually every GOP leader believe no single vote will have greater ramifications on the future of the Republican majority than the vote to block President Clinton from changing the way we conduct the Census.

I believe we have two critical assignments:

- We must ensure that our science is relevant to our society, and
- We must ensure that our society understands our science.

Personally, as I consider our challenge to ensure our relevance to society, I find myself an ardent subscriber to the thesis that our Census Bureau Director Kenneth Prewitt espoused more than a decade ago (at the conclusion of his article on APublic Statistics and Democratic Politics@ in Alonso and Starr's The Politics of Numbers):

The task involves analysis of the political role of numbers as well as a professional commitment to make the numbers perform according to the burden that a democracy places upon them.

If there is one pre-eminent credo for my role, and for other leaders in our national statistical system, it is that we must at all times be politically sensitive, i.e., aware of the political milieu in which we carry out our work, yet we must always be vigilant to ensure that our official statistics are not in fact - and are not perceived to be - overwhelmed, or even influenced by, political concerns.As Senator Grassley observed at a recent forum on statistics related to the aging population,

We need sound data for sound decisions.Sound data won't ensure sound decisions - but let us be sure we can blame the political process for unsound decisions.

I find myself struck by a paradox.On the one hand, our data are so relevant to policy that policy makers who use our statistics to allocate power and money - sometimes in applications the data were not designed to support - increasingly question our science and our product.Yet, these same policy makers criticize us for not having available from our statistical offices exactly the data they need, right then, for consideration in policy discussions and decisions.As we look to the future, we hardly need a crystal ball to give us the message - we will continue to be challenged to ensure that our science is relevant to our society.

Even more challenging as we look ahead - and perhaps here a crystal ball would indeed help - is the task of ensuring that our society understands our science.Almost 50 years ago, the American Statistical Association's president Aryness Joy Wickens challenged her colleagues as follows:

We owe the public ... a clear description of the general nature and limitations of statistics, with a simple measure of their accuracy, if it can be measured, and a warning if it cannot. We cannot, and should not try, to explain every technical detail, or to make statisticians of everyone ...

We still owe the public this debt.

Our citizens encounter statistics at every turn in their daily lives.Yet often they are unequipped with the statistical literacy required to evaluate the information presented to them.The "rationale for teaching statistics" set forth at the beginning of this decade in the American Statistical Association's Guidelines for the Teaching of Statistics K-12 Mathematics Curriculum, elaborated this encounter between citizen and quantitative information:

Raw data, graphs, charts, percentages, probabilities, averages, forecasts, and trend lines are an inescapable part of our everyday lives.They affect decisions on health, citizenship, parenthood, employment, financial concerns, sports, and many other matters. ... Informed citizens should understand the latest news polls, the consumer price index, and unemployment rates.

Back in 1992, when I was privileged to serve as the American Statistical Association's president, I defined "statistical literacy" as the ability to understand and critically evaluate statistical results that permeate our daily lives - coupled with the ability to appreciate the contributions that statistical thinking can make in public and private, professional and personal decisions.

Addressing the ASA membership on this topic at that time, I noted, among other things, that "policy makers and the public have raised issues that focus not so much on the scientific merit of our method but rather on matters (such as respondent burden and data confidentiality) that attend the collection and use of statistical information."Regrettably, as the policy implications of statistical data have become more widely appreciated - to wit, the applications of the Consumer Price Index and the Decennial Census population counts - even our methods have become grist for public debate, often leaving citizens in a quandary concerning the scientific versus the political aspects of arguments presented to them.As my colleague Edvard Outrata, President of the Czech Statistical Office, has said, "we face a newly skeptical public."

And little wonder the public is confused. In the current debate surrounding our Census 2000, we hear Citizens Against Government Waste explaining why the Census Bureau's plan is all a plot:

The Clinton-Gore Administration has announced that they think the last Census failed to count 10 percent of the population. ...To correct this problem, they developed a plan known as "sampling", whereby a poll, like a public opinion poll, would be conducted.This poll would use information from a group of people to make projections about the population as a whole. Based on the outcome of this "sampling", they would devise an estimate of how many Americans fall into various groupings, such as how many poor, how many of each minority, and so on. ... The net effect is that the Census outcome would be controlled by sampling, which can be manipulated to meet the political goals of the Administration.

Or, the public may have heard the recent assessment by Republican National Committee Chairman Jim Nicholson, "The Clinton Administration is implementing a radical new way of taking the next census that effectively will add nearly four and one-half million Democrats to the Nation's population."

What must the public think when they hear Representative Dan Miller, who now chairs the House Census Subcommittee, say, "I am a former statistics professor. ... I know all the statisticians say sampling is great.Statisticians would not have a job if we did not have sampling.That is what statistics is based on.Statisticians are biased to start with. " "Every governor, mayor, and county executive should be gravely concerned," Mr. Miller warned recently. "The Census Bureau is peddling snake oil and they're headed for your neighborhood."

We simply must extend our efforts to ensure that society understands our science.As former Office of Management and Budget Director Franklin Raines wrote in the July 12, 1998, issue of Science magazine,

"Not every American will become a scientist, and most will not be interested in the arcane details that so excite the scientific community.Yet scientists should make a difference where they can, most importantly by improving the science taught at the K-12 levels."