Interview with Kenneth Prewitt: The need to treat federal statistics as part of the nation’s scientific infrastructure
Kenneth Prewitt, former director of the U.S. Census and Carnegie Professor of Public Affairs at Columbia University, speaks with Stephanie Marudas about the important messages of the recent Annals volume he edited titled “The Federal Statistical System: Its Vulnerability Matters More than You Think.” Among the topics he addresses are the dangers of under-investing in the nation’s statistical systems and government’s need to keep the census a “nonpartisan, apolitical, scientific starting point for what becomes a quite political outcome.”
SM: You have edited a volume of The Annals about federal statistics that is out now, and as the former Census director for the 2000 census, you have firsthand knowledge about the role federal statistics play in American society. Can you walk us through how federal statistics, on a day-to-day basis, really shape our society?
KP: To begin with, you actually cannot talk about this society, or any other major complicated society, in the absence of some kind of empirical portrait of it, and that is what the statistical system gives us. Obviously, by the time we are into now, 2010, it is a very sophisticated system and we have important agencies that just worry about things like education or unemployment or crime or health. No major sector of American society is not measured by the statistical system. I think one of the original arguments that we have tried to put forward in this particular issue of The Annals is that we should treat federal statistics as a part of the nation’s scientific infrastructure. The production of statistics itself is a science. It is the science of survey research. It is the science of measurement. So not only is it heavily used by the scientific community, it is produced by the scientific community. And so I think that it is extremely important for this country to recognize that if you under-invest in the federal statistical system, it is like you are under-investing in any other part of science. All of our sciences have a really very practical part of what they do. They produce basic understanding of the physical characteristics of the Earth or of the human body or of the solar system, and from that we then take knowledge and use it. Well, it is no different from the statistical system. It produces basic knowledge about many, many characteristics of its population. Maybe I will put it this way, you cannot imagine a modern society that does not have maps. You could not function, so whether it is transportation maps or topographical maps. One of the very first things a society does when it becomes a modern nation is to map its territory. Well, the other thing that it does, right from the beginning, is it maps its population. And that is what a statistical system does. So we started with ours actually before we were a country, the colonies were already mapping us for their own purposes. But certainly with the establishment of the country in 1790 with the first Constitution, one of the very first things was to go out and take a census.
SM: And on that note, talk to us about what role the census plays, what its purpose is— considering your involvement.
KP: Well, the most important thing to say is that you could not have our form of representative government without a census. Our form of representative government allocates political power, distributes political power, proportionate to population size and that is what the founders wanted. And they also knew this country was going to grow and spread westward, which of course it did. And so one of the questions they had was when do you know it is time to have a new state. Well, you have got to have so many people before you can have a new state. So the census was mapping constantly the westward flow, until in fact we filled the entire continent and then even beyond the continent to Alaska and Hawaii, of course. But in that entire process, the only way you could create this constantly expanding system of representative democracy where power is allocated proportionate to population size was to do a census. So, really basic. But it is very important for anyone who wants to pay attention to these kinds of questions, which I think ought to be more people than do, to recognize that the census is a nonpartisan, apolitical, scientific starting point for what becomes a quite political outcome, which is competitive elections. The elections should be political. They are designed to be political. They are designed to resolve differences in this society, select different kinds of leadership depending upon what the mood of the society is— how well they think they are being governed. But the starting point for that is a correct and fair count of the population, because that is how we draw districts, decide how many House members go to a state and, as I say, in the very earliest years of the Republic even to decide whether there should be a state or not.
SM: You bring up the point that the census can become politicized, and in The Annals you write an article about this specifically. Can you talk to us about how this happens, or what agendas are being served when that does happen?
KP: We produce statistics because we expect them to be consequential, because we expect them to be used. We expect people to care about them. You cannot have something that is consequential in the political process, which as itself does not at some point attract political attention. So the census is very consequential in the way that I just said. It distributes political power and authority. It is also consequential in terms of allocating lots of dollars: federal grants for transportation; for building hospitals; for where food stamps are directed. A trillion dollars is now being distributed based on these kinds of counts. So it is consequential. So you would expect it to attract political attention. The argument that I make in that little essay on political interference is that you have got to protect the production of the statistics from political interference, but the use of the statistics can be political and should be political. And that is not the statistical system’s job. The job of the federal statistical agencies, the Bureau of Labor Statistics, the Census Bureau, criminal justice statistics, education statistics— their task is to produce the very best numbers that they can. That is a scientific task. If the politicians want to use these for fighting their political battles, that is what you would expect in a democracy. So I feel like we have to come up with practices and principles that protect the generation of the information from political interference, and it is not easy. And, as I say, finally it sets out the conditions for competitive elections. Well, that is very political, and so the parties fight endlessly about drawing the boundaries that create the congressional districts or the school districts or the state legislative districts and so forth. And those are very intensely political battles. But they start with a number, a distribution of the population by age, by gender, by race, ethnicity, depending upon the purpose. They start with what we hope is a completely accurate—never completely accurate because you cannot get total accuracy in something as complicated as counting the American public— but you start with something that has no political content to it, no inherent bias. It may be used in ways that are biased politically, but you would like to start from the point where the basic data set itself has been produced according to scientific principles and standards rather than by political choices and options.
SM: In your article, you do bring up that during the administration of George W. Bush, there were some instances of political interference. Can you describe what was happening then?
KP: Let me start by saying that I do not think that the temptation to get the numbers you want is a Republican or Democratic temptation, I think it is a temptation. And my particular experience of it was with a Republican administration, but certainly the Democratic Party was equally interested in getting numbers that would satisfy its political needs. And what happened in the lead-up to the 2000 census is that we tried a very complicated procedure; it is called dual-system estimation. I will not go through the technology of it, but it is a way to try to make sure that you end up with an equal proportion of whatever groups there are in a society, it is as if you had counted a total one hundred percent of the population. As it turned out, that procedure did not work very well and the Census Bureau decided not to use it. However, in the process of making that decision, the Bush administration had just come to town when the census was being finished in 2001; and one of the first things it did was it insisted that the Secretary of Commerce would examine the results of this particular statistical procedure before the data were to be released. And that is what we call “preclearance,” and the federal statistical system cannot actually tolerate preclearance. It is as if we took the unemployment numbers down to the Secretary of Labor and said, “Well, what do you think about this number?” And the Secretary of Labor says, “Well, I actually would like that number to be a little lower, and why don’t you do it this way instead of that way.” And it could be lower, all statistics are estimates. Why, we would have a political outcry. Everyone would like the numbers that make them look better and their opponents look not so good. But, if the statistical system is asked to do preclearance, which is to say to take major statistical products to their political bosses first for a clearance; then you have destroyed the integrity of a statistical system and this happened in 2001. It happened in a way that actually was not consequential, because the Census Bureau itself had already decided it probably would not use this. So it is just ironic, but certainly deeply unwelcome, that you would expect that a Secretary of Commerce could ask the [Census] Bureau to let him or her see a major part of a statistical analysis before its release to the public. And that I consider to be a very serious violation of principles of political independence and I tell that story in this chapter of The Annals piece.
SM: And you also say that former Census Bureau directors, like yourself and others who were in the field, were reluctant to speak up about it at the time.
KP: I said that many people, including the National Academies of Science and the former directors, in which number I certainly include myself, made less noise about this than they should have. Not because they were reluctant to do it out of any fear. They made less noise because it seemed not to matter since it was not overriding a decision of the Census Bureau. If the Census Bureau had wanted to use that particular procedure and get rid of the lumpiness in the numbers, and then had been overridden— then there would have been a huge outcry. But I, in retrospect, wish that I had made more noise and wish my colleagues had made more noise, because it was the principle that mattered, not in that case whether the numbers were or were not used. It also happened in 1990, so there is a precedent for this. But it happened in 1990 because it was a court order requiring it. There was no court order in 2000, which is why I think we should have made noise. We should have said, “This is pre-clearance. This is a violation of statistical independence,” and made enough noise that it would be less likely to happen again.
SM: When you were running the census in 2000, and looking now at the 2010 census, the budget has doubled for the 2010 census. Why this big increase? What is happening with the census today?
KP: You ask a very important question. The census costs doubled between 1990 and 2000, doubled again between 2000 and 2010 as you suggest— that is unsustainable. You cannot double the cost of something every ten years and sustain it in an indefinite future because sooner or later, in fact rather quickly, the Census Bureau will be taking a bigger and bigger bite out of the total federal budget. The reason for that increased cost is it is actually very hard to do a census. Your obligation is to try to count one hundred percent of the population. When you have the kind of conditions, you take 2010 with the foreclosure crisis, millions of Americans were not living in the home that you thought they were going to be living in and the families were distributed all over the place. Maybe some living with in-laws. Maybe one partner in the family had gone someplace to try to find a job, part-time job, living in some motel or whatever. The other partner in the marriage, the mother or the father, whichever, is still with the kids but she is living with her parents. So entire families got distributed, which had, prior to the foreclosure crisis, all lived together at an address where you could find them and so forth. So even finding those persons, if you will, those families, became very, very difficult. [Hurricane] Katrina still left a mark, of course, on that part of the country, which increased census challenges. But more generally, people are simply busy. They do not pay attention. It costs a lot of money to try to get their attention. So we had to run an unusually large advertising campaign; spent more than twice what we spent in 2000 to get the same level of awareness and the same level of cooperation. We had a huge partnership program in 2010. Technologically, you try to get more sophisticated; so you try to use certain kinds of technological breakthroughs, if you will, as we did with address canvassing. One part of the census is simply going out and finding every address, which you have to do. It is not like the country has a list of every address in the country; it has bits and pieces but it is not complete. Houses are being torn down and built constantly. Warehouses are being converted and old industrial spaces are being converted into condominiums. This is a process of renewal of the housing stock that is absolutely continuous; and yet you have to find, before you start the census, a complete listing of every possible residence in the country. It is very expensive to do those things, labor costs and so forth; and of course there are more people to count.
I should say one other important thing about the cost of the census, which is going to run between 14 and 15 billion dollars in 2010 when we get all the final numbers in. That includes not just what you see, that is these ten questions in ten minutes—the decennial census itself. That number also includes what is called theAmerican Community Survey, which is a very large, continuous survey of the entire population on a sample basis where we ask many more questions.
SM: It’s not sustainable to keep going at this level. How do you get Congress to buy into this? Because they are the ones who approve the census.
KP: Oh sure, they have to appropriate the money.
SM: So what do we do?
KP: When I say that doubling the cost every ten years is not sustainable, I really mean that it is a huge piece of the taxpayer dollars just to go out and count the population and sooner or later somebody is going to say, “Well, wait a minute.” And I see two or three things happening. All of them are really very consequential for the kind of federal statistical system we have. One of which is the government produces a ton of administrative data in the simple process of administering programs. If you actually could make use of all of the administrative data that the country already has as a by-product of administering programs, then you would know a lot about the country without having to ask questions.
The other thing that is growing at a huge rate, of course, is digital data. We all leave a digital footprint. So there are tons and tons and tons of digital data and also administrative data, and I am worried that U.S. Congress will say, “But wait a minute, we already know a lot of these things. Why do we have to go out and ask people at this very, very high cost?” And they will have a point. The problem is that the administrative data and the digital data are not themselves representative of the entire population; they get bits and pieces of the population. A second problem with those datasets is that we have not spent the kind of time and energy as a scientific community worrying about quality of those data as we have survey data. We have been doing very serious, sophisticated science on sampling theory, on measurement theory, on how to collect good data through sample surveys, since the 1930s where it all started. As a matter of fact, with the Census Bureau, sort of discovered sampling theory and that could be actually used and that produced a whole polling industry, market research industry and so forth. And so I worry a lot about the temptation to move to an alternative information system that has not yet been subjected to the kind of scrutiny that will be required sooner or later.
If you wanted to save a lot of money in a hurry on the census, you would go to a national registration system. There are countries in Europe which do not take a census. Germany does not take a census. Holland does not take a census. Instead, they have a national registration system. You are born, you got a birth certificate. You go to school, that gets recorded. You graduate from school and you marry, that gets recorded. You have children, that gets recorded. You have a job, that gets recorded. All of this is part of a statistical record, which is a product of, in effect, a basic national registration system that is connected to other kinds of administrative record systems. This country is close to having a national registration system. We give every baby born here now a Social Security number. But it certainly is not ready to imagine that you would link all of those datasets, the health records with the education records with the unemployment, with the employment records, with the housing ownership records or rent records— all the kind of records that exist about us. There is a real reluctance on grounds of privacy to link those. And so maybe our country will simply not do it. Well, if they want a census, then they are going to have to be prepared to pay this huge new cost every ten years.
I think it would be a much scientifically more sound statistical system if it could be based upon a national registration system. I am not in the majority on that position, but that is the way it is. But I do think we have the scientific capability to sort those things out and I make reference to this in The Annals piece. When I introduced this entire volume, I said that one of the things we have to prepare ourselves for is an information system that is not just based upon survey research but based upon these alternative sources of information; and we have to get it to a high level of scientific credibility and quality.
SM: So who is going to help us get there? Is it going to take certain members of Congress? Who is actually going to lobby for those changes? Or does that happen from the White House? Where can we see that?
KP: The important answer to that question is the scientific community itself is going to have to be the leading edge of this. You cannot do it without Congress, but I do not think you should wait for Congress. The three presidents of theNational Academies of Science and the President’s Science Advisor and the Chief Statistician of the United States and the head of O.M.B. [Office of Management and Budget], Peter Orszag— all of these people came to the symposium that produced this issue of The Annals that we are talking about. That was kind of the scientific community, not just the statistical community, but the scientific community recognizing the importance of these statistics. And that is where it will have to happen. People have got to realize, and it will partly come out of the commercial sector, which will say, “We need to know these kinds of things in order to make intelligent investments.” We need to know these kinds of things about our American population, its movement, its characteristics, its illnesses, its education level, all of these things in order to figure out how to run a complicated economy. So, I think it can come from more than one place. But I expect the scientific community to be the leading edge of this conversation.
SM: What got you interested in federal statistics to begin with? How did you get to where you are?
KP: I really did not understand what I was getting into certainly when I went to the Census Bureau. I had been engaged in the administrative roles in the social sciences, and that included a short term as the director of the National Opinion Research Center, which at least was a sample survey, an important one, associated with the University of Chicago. So I had some familiarity, of course, with the technologies of survey research and etc. But the census is at a scale by itself, so I was not prepared for the job. I did it, it worked out all right, but I take no credit so much as just that I had the wit to let the Census Bureau professionals do what they could do well. But it alerted me to the more complicated set of issues associated not just with taking a census, but with producing quality information for a society in the twenty-first century. So that just became one of the things that I then started to study and think about and write about and leading to things like thisAnnals article. I’m finishing a book on racial classification, and the fact that I think we do not ask the right questions to learn what we need to learn about our racial and ethnic groups in this society. Therefore, I am going to argue in this book that we need a completely different way to think about asking those questions that are more than two hundred years old, practically speaking. So it was an accident, it was a career accident. I would have done other kinds of things with my career but that came along and this became part of it. It is not all that I do in my career but it is just an important part of it. So I take no credit for sort of having the wisdom or the insight to sort of imagine how exciting and interesting it really is as a set of intellectual questions, but also as a set of questions that have real consequences for modern societies, modern democracies, certainly for this one.
SM: On that note, thank you for your time.
KP: Thank you.
Ken Prewitt on his Annals volume about the Federal Statistical System
Press play to hear recording.