Press Briefing -- June 21,2000
|DANNY SELNICK: Good morning, and welcome to the National Press Club.
I'm Danny Selnick, a member of the
Newsmakers Committee, and Vice President and Bureau Manager of PR
Newswire's Washington office. And I'll be today's host.
Before I introduce our guest I'd like to
take a moment to remind everyone of certain requirements. If you haven't
signed in already, please do so after the conference is over. We'll first
have an opening statement from Dr. Prewitt, and then we'll take questions
from you. When you ask questions, please identify your name and the
organization that you are affiliated with. If you're simply a member of
the Press Club, and not a working journalist, let us know that as well.
Transcripts, as well as archived video and
audio of events at the National Press Club are available on our Web site
Now to today's topic. Every 10 years the
United States Census Bureau attempts to conduct a complete accounting of
every resident in the United States, no matter where they live. The data
is used to help the federal government determine how communities
nationwide receive their fair share of annual federal funds for vital
programs and services in education, transportation, health care, housing
The 2000 Census, and important resulting data will help in the allocation of more than $185 billion in federal funds annually.
Joining us today is Dr. Kenneth Prewitt,
director of the U.S. Census Bureau, who will discuss the operations of the
2000 census and why the bureau calls this, quote, unquote, "a good
And now, Dr. Prewitt.
DR. KENNETH PREWITT (director, U.S. Census Bureau): Thank you, Dan.
As a matter of fact, obviously I've characterized Census 2000 as a
good census because that's the standard hyperbole expected of federal
agencies; and, therefore, it can get discounted. But I believe there is
something at stake that is greater than simply a federal agency trying to
What do we mean by a good Census?
Well, I remind you that starting just after
the 1990 census, continuing for a decade, the favored Congressional GAO,
and even media vocabulary for talking about Census 2000 employed such
phrases as "census at risk", or "another failed census in the making."
Indeed, February '97 GAO report, "there's a high risk to the nation of an
unsatisfactory census in 2000." A year later, March '98, "the nation runs
the risk of a very expensive and seriously flawed census in 2000." A year
later, January '99. "Major challenges and uncertainties have led us to
conclude that there's a high risk that the 2000 census will be less
accurate and more costly than previous ones." A year later, December '99,
"with less than four months remaining until Census Day, significant
operational uncertainties continue to surround the bureau's effort to
collect timely and accurate field data." Even as late as March of this
year, "the bureau still faces a high challenge in delivering data capture
Now, I know that it's the job of GAO to
point out risks and challenges and so I don't fault them, obviously, for
doing so. But it did create a climate around Census 2000 which called into
question, in effect, the competency of the Census Bureau.
Well, what is a good census? From one
perspective, it's as follows. Every major operation in Census 2000
performed better than was predicted by the GAO, by the congressional
subcommittee, by the Census Monitoring Board, and in a few important
instances, better even than the Census Bureau itself predicted. We have
continually been on or ahead of schedule. Also, this census will be a
black ink, not a red ink census. It is too early to know if we will be
simply at budget, marginally under budget, or substantially under budget.
But I do know that 10 years ago at this time the Congress had before it a
request for a supplemental appropriations because the 1990 census
operations had not unfolded as expected and budgeted for.
Later today, or tomorrow, Congress will be taking up the Commerce, Justice, and State appropriation and it need not worry that the Census Bureau will be requesting any supplemental funding for Census 2000. So from the perspective of a very large scale government operation, it has been a good census because it is being effectively and efficiently conducted. And in a period when there is so much cynicism about government competence and a corresponding rush to privatization that a federal agency can perform, a needed public function on schedule, on budget is itself non-trivial.
But there's a much more telling way to consider what has made Census
2000 a good census. It has successfully engaged the American people.
The most telling indicator is the initial response, the mailout operation.
As you know, we reported this number as 66 percent and widely described
this as halting a decades-long decline in civic engagement. This was a
conservative report. It is now clear, with late returns, that the number
is much closer to 68 percent, not only a halting of the decline but a
measurable improvement. Moreover, as those of you who have followed this
story closely know, the initial response rate is calculated on a
denominator which necessarily includes vacant housing units, seasonal
homes and the like. That is, households from which we could not have, in
fact, received a census form.
Not until we finish all of our field work
can we subtract these households from the denominator and calculate what
we'll then label the return rate. There is no doubt that this number will
be in the high 70 percent range, and it is now out of the question that it
will reach 80 percent. If so, the nation would have reached the very
ambitious goal of 1990 Plus Five, because the equivalent return rate in
1990 was 75 percent. This return rate that is calculated after the
denominator is cleaned up is, in fact, the best measure of civic
participation in Census 2000, and it is an extraordinary accomplishment by
the American people. We add to that the unexpectedly high level of
cooperation in the current operation when we go door-to-door to track down
the nonresponders. In that operation, as we all know, we started with 42
million households to visit, and today, well ahead of schedule, we have
fewer than 100,000 to complete.
This then is to talk about the good census
not as a set of government operations, but rather as an exercise in civic
responsibility. It happened across the country, mobilized by thousands
and thousands of local governments, businesses, churches, community
organizations, advocacy groups and so forth. The prevailing cynicism in
this town sometimes keeps us from seeing this bigger story, but the
evidence is in. It is a good census because the American people made it
When I was deciding whether to accept this
job, my wife encouraged me but did quip that I probably would lose my
innocence. Well, she was not wrong in this prediction, but surprising to
both of us, as I went around the country watching the civil mobilization
on behalf of the census, what I really lost was my cynicism about the
Let me shift this narrative a bit. A good
census is not a perfect census. There is a true count of American
residences as of the first of April, 2000. None of us know what that true
count is. Any statistical exercise is an estimate of that truth, and we
could describe Census 2000 as perfect only if its estimate is identical to
that true count. Perfection is not in reach; an approximation of the
We talk about the census as a vast
complicated undertaking. This is so. But there's actually a quite simple
way to understand it. A census is nothing more than a series of
operations designed to move the estimate as close to the truth as
possible. For example, had we stopped the census after the
mailout/mailback operation, we could have actually estimated the American
count by calculating the number of people per responding household and
extrapolating that ratio to the nonresponding households. This would have
been an estimate of the population, but not a very good one. So we did
another operation, non-response follow-up, which is now nearing
completion. The count estimate now is much closer to the truth but we have
time, resources and ideas about getting yet closer.
For example, we are now doing something called Coverage Edit Follow-Up which deals with population count discrepancies on census forms and related issues on approximately 2.5 million households. Also, we've just started a much bigger field operation called Coverage Improvement Follow-Up which will take us back to nearly 10 percent of the housing units in the country looking for new addresses and making doubly certain that a housing unit marked vacant, or demolished or seasonal, really is so. These operations and dozens of smaller ones that improve census quality keep moving the census estimate closer to the truth.
Finally, of course, is the Accuracy and Coverage Evaluation, or in the
beltway shorthand, the use of sampling in the census. Of course from the
perspective of the Census Bureau, this is no more, nor less, than another
operation to try to move the estimate that is the census closer to the
truth. But I need not tell you that this operation has been treated as if
nothing less than the future of the republic is at stake, twice reaching
the Supreme Court, and perhaps it will again.
Something as arcane as dual system
estimation has been made to do very heavy duty in the war of spin and
symbol manipulation much favored in this town, symbols such as on the one
hand, constitutional purity; and on the other, racial justice. The Census
Bureau was caught in the crossfire of this partisan battle but is simply
trying to get its estimate a bit closer to the truth by accounting for
some of the people who we know are traditionally missed in the census, and
others who are erroneously included - that is the undercounted, and the
We think that a good census can become a
better census. Not a perfect census, but a better census if we measure
the undercount and the overcount and correct for these miscounts. To
return to my earlier point, we believe that evaluating the accuracy in
coverage of prior census operations will move the estimate closer to the
And so, two final points. First, the
country, and especially the press discuss the census undercount in some
detail, especially by racial sub-group. You do so because, and only
because, the Census Bureau issues a report card on its own performance.
This report card is faithfully reported as fact in the press, in GAO
studies, in CBO reports, in scientific fora and by political leaders
especially in the Congress. Indeed, the 1990 undercount numbers were used
in Congress on both sides of the aisle to justify a 7.5 billion budget for
Census 2000. And the 1990 undercount corrections have been used to move
the estimate of labor force participation and inflation closer to the
truth over the last half dozen years. Indeed, every pension plan in
America is indexed to this improved estimate.
It is ironic that the Census Bureau's
report card has the confidence of the country for all kinds of purposes
but not, strangely enough, for the major purpose for which the report card
is prepared, to correct for the measured errors.
And to make my final point, permit me to be
slightly autobiographical. Why did I come to Washington? Not "Potomac
fever," which as a matter of fact, is difficult to catch out in Suitland.
I came because 20 years ago I wrote an essay on statistics and numbers in
which I argued that national statistics play a key role in democracy along
two critical dimensions. The intelligence with which the government
governs and the intelligence with which the public can hold those who
govern accountable for their mistakes and errors of judgment.
There is no democracy without
accountability. And when we judge our political leaders in terms of
whether education is improving, crime is down, the health system is
working, the inflation rate is in check, we are looking at social and
economic indicators that rest finally on a good census. The idea that the
Census Bureau could predesign a census to achieve a political outcome not
only lacks face validity, for reasons I'm happy to explain, but is
pernicious in the extreme. To casually erode public confidence in the
nation's numbers system is playing with matches in a flammable
environment. I came to Washington to assert, as strongly as I know how,
that national statistics have to be produced outside of and independent of
partisan battles. The stakes are high for the economy, for social
understanding and finally for democracy itself.
It is said that I'm a political appointee.
True. The director of the Census Bureau is a presidential appointee. So
is the head of the National Science Foundation, and the National
Institutes of Health, and of NASA. I don't believe, and doubt that you
do, that there is a Republican versus Democratic way in which NSF funds
nano-technology, or particle physics, or NIH wages its scientific war on
cancer and AIDS or NASA designs an exploration of Mars. At the Census
Bureau, we don't think that there is such a thing as a Republican versus a
Democratic way to conduct a census. There are ways to get the estimate
closer to the truth, and that is what we think we are about. Thank you.
SELNICK: I'd like to open it up for
questions. Again, please identify yourselves and the organization that
Q: D'Cohn, Washington Post. I'd like to ask
a question that many people have been unable to answer. I'm not sure you
want to take a stab at it this setting, but if you were trying to explain
sampling and how it works to someone who doesn't have statistical
background, what words would you use to do that in a way that wouldn't
take more than half an hour or so?
DR. PREWITT: Well, the idea of sampling
itself is a remarkably simple one, and we do take samples all of the time
in common life. You know, we taste the soup to see if it's too hot. We
don't taste the whole bowl, we taste one spoonful to see if it's too hot.
We take a blood sample, we don't draw out all of the blood to find out if
we're carrying any infectious diseases. People constantly, without
knowing about it, sample. They look out the window to see what the
weather is. They don't look out every window, they look out one window.
Those are all kinds of common sense ways which we sample.
Now, sampling the American population is
more complicated. There's something about sampling theory. The more
homogenous the thing you're sampling, the easier it is. And so taking a
blood sample is fairly straight forward because we presume all the blood
is homogenous. When you're sampling from something that's not homogenous,
like the American population, then it's a more difficult task.
Nevertheless, the idea itself, seems to me, fairly transparent and obvious
if we just think in a common sensical way about how we conduct our daily
life. And I know that's not quite what you mean, Dee, but I would start
with trying to get people just to appreciate why we have to sample in
order to function intelligently as individuals.
Now, what does it mean to sample the
American population for something as complicated as a census? What it
means is that if you can actually go out and take a sample of those people
who were missed in the census - by taking the sample you can find - you
could use all of your best resources, your best enumerators, a probing
interview - that is you're only talking to a few of them, as we've said,
and it translates to about one quarter of 1 percent of the American
population. You can actually do a better job of talking to one quarter of
1 percent of the population than you could talking to the entire
population. So when you're drawing a scientific sample you do it in such
a way as to allow you to give it your best shot, and that is all this
Accuracy and Coverage Evaluation is, our best shot. We then look at the
characteristics of the people and we say, my goodness, the characteristics
of the people that we missed are such-in-such. And that's really what the
Accuracy and Coverage Evaluation Survey does.
Once we have done that, then it's relatively easy to say
statistically, without going through the complicated thing, but it's
relatively easy to say, therefore, wherever those kinds of people live, we
know that we missed 3 percent of them, 6 percent of them, 1 percent of
them or what have you. So I guess I would back up and be redundant, but I
would first try to get people to appreciate the logic of sampling itself,
and then, hopefully, get them to appreciate why when you take that, even
to something as challenging and difficult.
The other thing, and obviously I made this
point in my opening comments - I think it would help the American society
to appreciate that any statistic is an estimate of the truth, and so all
we're doing is struggling to get that estimate closer and closer to the
truth. We do not think that any of us know what that truth is. I don't
know, 275, 311 million; whatever, lived in the United States on April
1st. None of us know that. The question is whether we have
designed operations that are reasonable ways of trying to get slightly,
slightly closer to the truth.
I'll stop there. That's not quite an
answer but we'll come back to this often over the next several months.
Q: Mark Wagner with Congress Daily. A week
or so ago the announcement was made that the decision of the sampling
would be transferred from the Commerce Secretary to you. Do you feel
comfortable with that authority, making that call, and what has the
reaction been on Capitol Hill?
DR. PREWITT: Let me first correct your
comment. That was not a delegation of authority to me, it was a
delegation of authority to whoever happens to be the Census Bureau
director in March of 2001, which is not likely to be me. I'm an
administrative appointment, I leave office on January 20th at
noon; so, therefore, this delegation is whoever happens to be sitting
there. And that's very important because it was somewhat characterized as
if it was political. But in effect, I don't know how it would be
political in a sense that we don't even know who's going to be sitting
there, what party they belong to, if they belong to any party and so forth
and so on.
So I guess my first observation - make sure
we all understand that it was a delegation of authority to the Census
Bureau director, and we will see who happens to be that director in
I certainly welcomed it. I think that in
1980 the then-president and the Secretary of Commerce did delegate a
decision to the then-Census Bureau Director Vince Barraba. And Vince told
me this when I was thinking about this job. He says the most important
thing you can do is to reassert the independence and authority of the
Census Bureau over the decision of what set of numbers to use, forgetting
the Accuracy and Coverage Evaluation, just period, what set of numbers to
use. How do you establish the scientific credibility and public
confidence in numbers if they don't appear to be the Census Bureau's
numbers? So I was just very delighted when the Secretary of Commerce this
year took the initiative to return to the 1980 pattern and ignore the 1990
pattern, which was the pattern in which the decision was made at the level
of the Secretary.
As to the third part of your question, the
response of this on the Hill, my guess is - I would characterize it as
reasonably or pleasantly quiet, which you would expect. I mean, this is
kind of a straight forward thing. I don't know if it wouldn't be treated
as this is a good thing, this is one more step in trying to make certain
that there's not even the appearance of anything partisan in the census.
I was personally gratified by Congressman Ryan's comment to that effect
that this established that the census was above partisan politics, and I
would hope that other members of Congress would repeat that theme.
Q: Ellen Ferguson, Gannet News Service.
This transfer of authority, was this something that you sought, did
Commerce Secretary Daley do this at your urging? How long had their been
discussions about this?
DR. PREWITT: Well, I certainly did not
disguise my own enthusiasm for this. I would say that quite honestly,
Ellen, I can't even recall how it first surfaced as an idea. It's sort of
around, it's an obvious idea. I do not think that the Secretary needed
any particular persuasion as to the merits to it. As soon as it was
discussed, sure, why not, that's what we ought to do. The Secretary saw
this as just an instance of good government. Just like we're trying to
talk about this as a good census, you've got good government trying to do
what seems to make sense.
I did, of course, as you know, in the
hearing just several weeks before the Secretary made that announcement, I
did rather strongly say to the Chairman of the Subcommittee who was
talking about the census at that time and whether it was or was not
political, I did rather strongly put on the record, indeed Congressman
Maloney is here, she will remember this as the ranking minority member,
that I said it would be my very strong hope that this delegation be made.
Q: I just wanted clarification going back
to any statistic as an estimate of the truth. For people who argue for
what they call the traditional head count, that that is the basis, that
should be the sole basis, for any sort of numbers. If the Census Bureau
stopped with - at the non-response follow-up and with the response rate
through the mail, would the number that you got from that, the numbers
that you got from that alone, would that be, quote, unquote, the truth or
would that be just another estimate?
DR. PREWITT: It would be an estimate. It
would have to be accepted as, quote, unquote, the truth, but we would know
it wasn't because we would know that we had missed a very large number of
people and double counted other people. So we would be giving to the
country an estimate which we felt we could improve upon. Nothing more
complicated than that. It does not mean we could not give the country
that number, but as a statistical agency, a scientific agency, insofar as
we have time, resources, and ideas we would like to continually try to
improve that count. So I would not like to stop with mail-out/mailback,
or non-response follow-up, or these new things I just mentioned, or
coverage and edit strategy or our coverage improvement follow-up effort,
very major effort going back to about 8 million households to make
certain. I wouldn't want to stop before that because I think by doing
that operation we'll get better; we'll find some cases, get rid of some
A.C.E. is just in that line. It's one more
way in which we will be able to improve the census. If we had more
resources we might think of other major ways to do this. At a certain
point, you have to get the data because memories begin to fade in terms of
who lived where on April 1st, and also you get a lot of movers
and it gets harder to reconstruct the population as of a given point in
time. So we wouldn't continue to do operations indefinitely, because we
would actually begin to get data deterioration. So we've designed the
Census to get as much as possible of the work done, as close as possible
to April 1st. But if we had stopped April 1st, we
would have had a much, much less effective or accurate estimate than the
one that we believe we will have.
Q: Amy Stringer, Bloomberg New Service. I
was just curious, when you do the Coverage Improvement Follow-Up, where do
you find those people? Where have you found the non-respondents? How have
you located people?
DR. PREWITT: Surely. What we do in this
particular procedure, what we call the Coverage Improvement Follow-Up; we
start, don't forget, the census with an address list of 120 million
addresses. We know that, at the edges, that address list has got some
problems; it's got some duplicate addresses in there, it doesn't get all
the new construction that's happened over the last three or four months -
that is, three or four months after we finish constructing the list. And
so when we go back - we've had people on the streets doing this
enumeration and they say, my goodness, this address isn't on my list but
it looks like it's a real address. They then write that down; that comes
back in to headquarters. So basically what it is is a way to go back and
make certain that our address list, our basic address file, is made as
accurate as we possibly can, so it picks up new units and takes out
I'll give you one other example. During
non-response follow-up, enumerators come in and they give us a form back
and the form says this house is vacant, or this is a seasonal home, no
resident there or this house has been demolished. We have a procedure in
which we do not take one person's statement about the status - if we think
it's an address and the enumerator comes back and says it's not, we send
somebody else out and double-check it, it's what's called a double kill.
We don't take anything off the address file until it's had two independent
investigations. And that's again, just being doubly certain that we don't
leave anyone out of the census. So what that process is primarily focused
upon are the addresses: are the addresses there, if we got them all, if
there's some that should not be there that we take out and it's that kind
of process. Hey, we go back to 10 percent of our address file, back -
well, not quite 10 percent. But somewhere between 8 and 10 million
households, we'll go back to. It's a very big field operation.
Just by itself, if you weren't doing a
census, a Non-Response Follow-Up, anytime you're doing something in this
country which involves 8 to 10 million addresses, it's a big deal. But
it'll hardly be noticed, because all the noise about the census is now
largely not behind us, but the noise is shifted to some other kinds of
But from our point of view, it's still a
big, serious challenge. One of the reasons why it's very difficult for me
to estimate on the budget front, is because we still have a big field
operation to go. And it's not until we've done that that we can really
come to terms. All I can say as of now is that it's black ink, not red
ink. But how much and whether it's just marginal, we simply don't know
Q: I wanted to follow up on an earlier
question. If you were to stop and not do the sampling, any idea how much
the numbers would be off by?
DR. PREWITT: Well, if we did not do the
sampling at all - see, the sampling process is the report card. When I
say we give the country a report card on our own performance, the only way
we know how to do that is to go out and do this Accuracy and Coverage
Evaluation study. So it is our own report card. If we don't do it, then
we would never know. The country would never know. How would the country
know? Indeed, I frequently say it is ironic that we have this intense
debate about the undercount and the overcount based upon a methodology
which is now being set aside for purposes of making the count better.
But the fact that we have the argument and have the discussion in the
country about the undercount and the overcount is a result of the fact
that in 1990 we went out and did it.
So the answer to you is if we did not
sample, we'd have no idea; the country would have no idea. There would be
no report card on Census 2000 except for a lot of anecdotes. You know, we
weren't counted. Somebody, you know, knocked on my door three, or four or
six times. I mean you'd have lots and lots of anecdotes, but nothing
If you think about it, it's an interesting
characteristic of the Census Bureau. It is one of the few big government
operations which the agency does, and it is the only agency which can
measure its effectiveness. No other agency in the government cannot,
somewhat, be measured by outside processes. But this is such a big,
complicated field- based thing. The only way you could measure our
performance would be maybe to go and contract it to some huge, huge
operation, I mean, in the private sector, a survey, research operation.
But nobody out there could actually do it.
It's like - well, I won't use that analogy
for fear of it being misused. Anyway, the point is that the Census Bureau
is one of the few organizations, federal agencies which has the capacity
and also the inclination. What about if we didn't do this? What if we
didn't go out and do this Accuracy and Coverage Evaluation, because we
didn't want anybody to know how many people we missed? We didn't do it;
didn't want to tell you. Or if it hadn't been funded. A long answer.
But let me just go back to why you ask it.
I think so we will be able to tell the country whether we did or did not;
how well we did or did not do; how many people were included and
erroneously included and missed. The debate is whether the numbers will
be used, not whether they should be collected. I am very pleased that, to
my knowledge, there is no one in Congress or elsewhere who seriously
wishes that we weren't conducting this Accuracy and Coverage Evaluation.
So that the argument is not whether we should know this. The argument is
simply how the numbers will be used when they're produced.
Q: Could you comment on the funding levels
for FY 2001 and the CJS budget? Are they adequate for completing the 2000
Census? And what about the other statistical agencies?
DR. PREWITT: Well, I think when you mean other statistical agencies, you mean other programs of the Census Bureau? Yes.
The funding for other programs in the
Census Bureau portfolio is less, seriously less, than what we would hope
and need. And I'm now talking about economic statistics, poverty
statistics. A whole array of things that the Census Bureau produces for
the country will be at jeopardy with the 2001 budget. I do not believe
that the particular 2001 budget that's now being debated will prevail, as
I believe the administration will fight hard for a budget that allows us
to do the other part of our work.
With respect to the decennial, based on
what I've seen on the 2001 budget, we do not think any major operation of
the decennial themselves are particularly jeopardized. And I haven't done
all the sort of close look at the 2001 budget. So I want to reserve the
right to slightly modify that, after I've done more looking at it.
But I think, all along, the United States
Congress has felt like the decennial should be well funded. And we've had
no major arguments with the Congress about the level of funding that's
been provided. We certainly will argue strongly for funding for other
important programs if we're going to have good economic statistics.
You know, as I said in the appropriation
hearing, there is, you know, no agency in the federal government studying
e-commerce except the Census Bureau. There's nobody else to go out and
get data on e-commerce and e-business. Well, with the current 2001
budget, it will be very difficult for us to do that work. And therefore,
this country - if it were to happen, this country will not have good data
about its basic economic processes in a very, very dynamic moment, of
course, for the economy.
Q: I received some calls and e-mail from
people saying that in the rush to complete the Census, to get 100 percent
of the follow-up, that there've been some problems. And I'm wondering if
any of these had come to your attention. In one case, someone in my
office has had, I think, nine different enumerators knocking on his door,
and he was told that one reason for that is some enumerators quit and run
off with their documents, so a new one had to come in and do it all over
again. In another case, I've been told that there's been talk of
questimating some totals at one local office.
Have you had any reports of - ?
DR. PREWITT: Let me just quip for a moment.
If somebody's being contacted by nine persons, that doesn't sound like a
rush to complete. That sounds like doing the job as thoroughly as
possible. And at the end of the Census operation, here's what happens.
This is true at the end of every operation. This was true at the end of
Remote Alaska, list/ enumerate. At the end of every operation, a large
number of enumerators have gotten so engaged with this task that they say,
"I want to keep doing it. I just think I can get that household. I just
think that person who said no to me three times won't say no the fourth
time." They don't want to stop. And left to their own devices, they
would stay in the field almost indefinitely. It's just in the spirit of
I get e-mail and letters all the time from
enumerators saying, you know, just let us keep working at this. That's
true after the 1990 Census. It's true after the 1880 Census. It's not a
characteristic of what we're experiencing right now.
So I think the desire on the part of large
numbers of enumerators and crew leaders, and so forth, is to sort of stick
with this task as long as possible. What they do not understand is we
have very good measures of how data begin to deteriorate the further you
move away from April 1st. And we always have very good data about the
enormous increasing unit cost at a certain stage. And so we have some
responsibility to the American taxpayers to run this Census in a prudent
fiscal way, and we also are concerned about data quality. So we have
procedures that say that on such and such a date we go into what we first
call final attempt, and then into closeout.
And so part of what you're hearing is kind
of tension around that set of activities. Fortunately, it's very, very
small-scale. I mean obviously we can all get our e-mail and our letters.
But when you sort of compare that against 500,000 people out there doing
this job, it's a pretty thin thing.
Now to the other part of your question, the
questimates, what we call curbstoning, people fabricating responses, and
so on, is a characteristic of any large operation in which we're dealing
with a temporary and marginally untrained work force. We have in place
quality control, upon quality control, upon quality control. And we're
out right now redoing interviews in a handful of cases, maybe as many as
four or five areas where we're actually redoing the work because we have
reason to think that the work wasn't done right. They either went to
closeout prematurely, or we have rates of what we call - we have something
called pop count. If we get a whole lot of houses that only have one
person in them, we have to go back and look at that work, because that's
somebody sort of rushing through and just putting one person down in order
to get the form in. And so we go back and look.
Sometimes it's perfectly explainable.
There are units, you know, let's say elderly housing units where only one
person does live in each of those apartments. But until we've been
convinced that this is the case, we go back out and take a look.
So at this stage of the Census, we're
looking at a lot of those. I don't mean to say they aren't there. They
are not large-scale. They're all highly particular to a certain
situation. We have procedures. We think we will catch most of it. At
the end of the day, will we catch every bit of it? No. It's not doable.
Do we think we're going to get a much higher percentage of those kinds of
cases, that is the questimate cases, than we got in 1990 and '80?
Absolutely. We have better procedures to go find it, check it and redo it
And that's why, again, as I say, we're
going back out in the field to upwards of 10 percent of the houses.
That's part of this operation, to make sure that we don't have - we get
rid of as much, you know, questionable stuff as we possibly can.
Q: When you say four or five areas -
DR. PREWITT: Yeah.
Q: - how many households might that be?
DR. PREWITT: Oh, redoing work? I would
guess certainly fewer than - it's somewhere in the neighborhood maybe -
this is a rough guess; it's an estimate; not the truth. But certainly
well under 100,000 households where we're going back and redoing work, you
know, based upon our own quality assurance checks and so forth.
Well, it's 20 percent of one instance. It's
40,000 cases. That's right. Yeah. Yeah. Okay. Half that. It's probably
not fewer than 50 and not more than 100.
Q: In the beginning you had given some
numbers about the initial response rate, the first 66 percent. And then
you had given us some other numbers. Are these better estimates, or -
DR. PREWITT: Well, fair enough, Mark. What
we first described to you as the response rate way back in the middle of
April at 66 percent, we're now fairly confident that response rate is
going to be very close to 68 percent. And the reason it's not exact yet,
is because not only do you add cases, you subtract cases. You know, you
got a lot of forms in that were blank, and that's part of what we also do
in this follow-up operation. We go back to those households that we got a
blank form [from] and try to figure out, you know, why did they mail it in
blank. Does it mean that it was just an accident? Somebody saw it and
stuck it in? Or was there somebody really living there, and so forth?
So that's part of our quality assurance.
So there are some adjustments both to the
denominator and the numerator on this initial response rate. But I'm
fairly confident it's going to be - by the time you round it, the number
will be 68 percent. Then when I say that will easily take us to the high
70 percent and maybe higher on the return rate, that is the really
effective participation rate. What I'm talking about is how many vacant,
seasonal homes, demolished housing units, duplicates would we be taking
out of that big denominator when we finally have done all this work. And
that could easily be as many as 10, 11 million households, undeliverables
as addressed, and so forth.
So that the number of housing units which
could effectively return a questionnaire because somebody lived there and
only lived there, that was their primary residence, will be a much lower
number than 120 million. If that gets down to 110 million, or 108
million, then the response rate necessarily jumps by as much as 10
percent. That's what I mean when I say the effective return rate, or
participation rate, what you think of as a civic participation rate, could
reach 80 percent. Now we will not know that until late August or early
September. That's after we've done all this work, you know, if we've gone
back out in the field and really taken a hard look at how many of the
housing units really are inhabited.
Q: When you refer to the 65 percent in
1990, is that a number that will -
DR. PREWITT: Yes.
Q: - Are you still working on that number to reach?
DR. PREWITT: We've tried to do apples to
apples is the point. The 1990 number that would be equivalent to the 68
percent, that is the final response rate, not the return rate, but the
response rate, was 66.4 percent. So that did creep up a bit after 1990.
And we set it publicly at 65 because that seemed more - that was closer to
what we were going to be measuring in the experience of the census itself.
So we tried to be as honest was we could. That's why I say it was
conservative in that sense.
But that does mean that it was 66.4 percent
in 1990. If it's 68 percent in 2000 for that initial response rate, it
will once again be a measurable improvement over 1990. And that's what I
feel. Whatever the final number will be, what I'm completely confident of
is that the initial response rate is going to be higher than it was in
1990, a measurable improvement, and that the return rate, which is this
other thing - sorry - which gets a cleaner denominator, if you will, but
also a cleaner numerator, we have every reason to believe that there's a
somewhat higher percentage of vacant, demolished housing units, seasonal
homes, and so forth, this year than 1990, which means that this is why I
say we could reach 80 percent. The equivalent number in '90 was 75
percent. Therefore, 80 really would be an improvement by 5 percent over
1990. It's a very big deal if that turns out to be the case. And we're
going to be close enough to that it is a big deal.
Q: This year for the Census -
DR. PREWITT: -
counted forms around the country. My recollection is the number that were
returned was about 650,000 cases. And most of those came in time that the
respondents did not get contacted in non-response follow-up. Some of them
came in. We had a deadline. Some of them came in after that deadline,
which means necessarily they had to get recontacted. And that's a fairly
small percentage of that.
So we only reinterviewed where we felt like
we had to. The other thing about the "Be Counted" form, however, is that
it has to be geocoded. It has to get back to an address. That's a big,
complicated clerical operation. That sailed along very smoothly, and a
large number of those - we could recode back to our basic Master Address
File. And in some instances, we went out and did field work to make sure
the housing unit was there. In a few instances, and I can't give you the
exact number, the "Be Counted" form was used to add people. The initial
form only allowed for six lines, as you know. And so some people used the
"Be Counted" form to appropriately add more people in that housing unit.
So it worked well. And the most important
thing, again, I want to say, it was not misused. The Census Bureau was
concerned about it. I mean 16 million forms out there that weren't
geocoded to an address could have been misused. That is, somebody could
have just sat there and, you know, filled them out all day long, and sent
them all in. And we did, of course, have processes to try to detect that.
We had certain thresholds that we looked at very closely as the "Be
Counted" forms were coming in. And we are quite certain that this was a
legitimate improvement of the census and it was not misused in any
particular jurisdictions or by any groups.
Q: The question may have been answered
before. But how, in terms of the questions that people thought were
invasive on the long form: how many people - what was the rate of people
not responding - ?
DR. PREWITT: Yeah.
Q: And how is that going to affect how federal funds are going to be allocated?
DR. PREWITT: Yeah. Surely. That's what we
call item non-response. That is, to what extent will the long-form
questionnaire actually be filled out? And we have not done that work yet.
We can't - you know, we can't report on that. We expect there will be a
number of long forms in which they only complete the short-form items.
But whether that's going to be higher than usual, we just haven't done the
Q: - entirely usual. How bad was that?
DR. PREWITT: It depends on how much higher
than usual it is, right? [Laughs.] As we said before, if you really got a
large number of long forms in which you got no information, you simply got
the short form, and if they were particularly geographically clustered -
see, it really has to do with how they're distributed. If you got, you
know - instead of getting one out of six, you got one out of seven of the
American households, but it was evenly spread across the entire country,
it wouldn't be as damaging. But if you got one of six in some places, but
one out of 23 in other places, then you've got a real problem making your
estimates in the areas where you got a really bad return rate.
The long-form sampling size is designed to
allow us to make estimates of population characteristics at the level of
the census tract. That's why we want as many as one out of six. And the
census tract is - the average population size of the census tract slips
out of my mind. Sorry.
MR. JOST: It ranges from one to eight thousand, with an average of six.
DR. PREWITT: Thank you very much, Steve.
So we were able to make estimates of population characteristics for
population groups in the neighborhood of five to six thousand people.
And, of course, what then happens is, in communities, you take all the
census tracts that belong to that community, and then you make character
statements about the population of Washington, D.C., or Alton, Ill., or
what have you. So the problem is, if you really get bad long-form data,
if it's not one out of six, but it turned out to be much lower than that,
or sampling ratio, you would then not be able to make estimates down to
that population size. You might have to make estimates down to that
population size. You might have to make estimates to no lower level than,
say, two or three census tracts combined, to population groups of 15,000,
or something of that sort. And that's what we would do to fix that
problem. We would simply say you can't bring your estimates down to as
low a population-size area as you would like to have done so. And that
will be a loss to the country. But we will figure out some way.
Certainly the Census Bureau is smart enough to do what it can do to
correct for that.
Q: Do you have any kind of assessment as to
how your efforts to do census in language worked out?
DR. PREWITT: Well, it's the same thing.
What we saw, as the system unfolded, gave us some confidence in that
program. That is, the numbers of language forms that came in was two, 2.5
million, about 2.5 million. That's kind of what we expected. The pattern
of them across the five languages was certainly what you would expect.
That is, you know, the vast majority of them were Hispanic forms, and then
a much smaller percentage of Vietnamese and Tagalog, and so forth. And so
the pattern looked right. They came in at the same time, a little bit
later as you would have expected, because they had to wait for it. And so
the number of calls we got on our Telephone Assistance Center line in
different languages, again, was a pattern that was predictable.
So based upon the experience, we think it
worked. Like I said, every census operation is nothing more complicated
than taking this estimate and trying to get it closer to the truth. The
fact that we were able to reach to some linguistically isolated households
simply meant that our final count is going to be closer to the truth than
it would have been in the absence of it. In this particular instance, it
will be, you know, a marginal improvement compared to big improvements
like non-response follow-up, or the vacant/delete program. But that
doesn't mean it's not critically important, because it got a part of the
population that otherwise would have had a very hard time participating.
But we've not yet encountered one of our
operations that did not, by and large, behave the way we hoped it would or
expected it to.
MR. JOST: We have time for one more question.
Q: I'll ask it. You said earlier you don't
expect to be Census director at the time that the sampling decision is
made. Is that because you're just reflecting political reality here, or
have you already made a decision that you do not intend to - ?
DR. PREWITT: Oh, no, I've made no personal
decision. I'm an appointee of this administration. When this
administration goes out of office, I go out of office. If any
administration were to ask me, then that's a separate issue. But as I
understand it, even if there were an administration that wanted me to stay
- it could be either administration - then I would have to be renominated
and reconfirmed. There's no way in which my appointment continues. And
so that's what I meant, not more complicated than that. It's not a
personal statement. It's a kind of a statement - as I've said before, I
think, in one press conference, to my knowledge, this is the only head of
a major statistical agency certainly in the industrial countries which
does not have a term appointment. And I feel, by the way, about that the
way I feel about the delegation. I think the Census Bureau director
should be a term appointment. I think the Census Bureau director should
be appointed either in a year ending in "7" or "3." And if you think
about that, that means that there's a couple of years - a "7" means you've
got three years to get ready for a decennial, and your term continues for
a couple of years beyond it. So you overlap a decennial. And the way it
is now is the accidents of administrations in census years means that
every 20 years, by definition, this happens. That is, you have a census
director whose term is ending just before you've really quite finished at
least the decennial.
But I do want to say on that point the
decennial is an important part of what the Census Bureau does. It's not
the only thing the Census Bureau does. And there're plenty of major tasks
for a Census Bureau director in the nondecennial phase of the Census
Bureau cycle. But nevertheless, if you don't get the decennial right, the
rest of the stuff is hard to get right, because that, again, provides the
controls and all the other kinds of survey work. But it's very, very
important work: economic statistics, the household statistics, the trade
statistics, the poverty statistics, unemployment statistics. All of those
things the Census Bureau is routinely producing for this society.
I might say, in conclusion, back to the
2001 budget, the Census Bureau - I'm obviously very proud to be part of
that organization. I said you could not get Potomac fever in Suitland.
You can actually get other illnesses in Suitland. We inhabit a building
which has infestation of pigeon dropping, which has asbestos, which has
bad water, which has broken pipes. Every day - not every day, but very
often - yesterday, I heard about a couple of more offices got flooded,
people have their work disrupted. And one of the things that we did
introduce in the 2001 budget was the importance of modernizing not only
the census and how it operates, but modernizing the space in which it
operates if this country's going to have good data over the next decade.
So I would hope that when the 2001 budget
finally is resolved, that there is an acknowledgment of the fact that
we've actually not only done a good census, but we've done it under really
quite challenging circumstances right in the headquarters. I had to
vacate my office three times because of major floods for a couple of days
at a time. And so, I urge you to follow the 2001 budget on many fronts,
but one of them is whether we end up having a building which is adequate
to the kind of tasks that the United States people expect of us.
SELNICK: Thank you, Dr. Prewitt. Again, if
you've come late to the conference, please sign in before you leave.
[END OF PRESS CONFERENCE.]