Chairman Coelho, Chairman Blackwell, and members of the Census Monitoring Board:

With only 360 days until we mail Census 2000 forms, we are at the point in the Bureau's planning cycle when all major design decisions have to be made. Fortunately, issues that had delayed final operational decisions are now resolved.

One was the legal uncertainty regarding the use of statistical adjustment in determining apportionment counts. On January 25, the Supreme Court ruled against using such procedures for the apportionment counts. Thus, the Census Bureau will follow up on 100% of the housing units for which a form is not returned.

The other factor on which final operational decisions rested was the Census Bureau's evaluation of procedures tested in its three Dress Rehearsal sites. The Dress Rehearsal occurred later in the decennial planning cycle than we would have wished, and thus postponed until the last possible moment the opportunity to make modifications based on Dress Rehearsal experience. The evaluation is now sufficiently completed that the Bureau can, with confidence, determine which procedures worked as anticipated, which require modification, and which should be eliminated. An example of the first is the variable wage rate to minimize turnover in the enumerator labor force; an example of the second is the construction of the master address file, which required serious re-engineering (now, of course, underway); an example of the third is a second mailing of census forms.

Timing

I want to emphasize that there simply is not time to consider major additional operations, which would be untested and not carefully integrated into the current operational plan. However well-intentioned, efforts at this late stage to introduce new procedures to the design will place Census 2000 at risk.

The census has unique features, making it one of the most complicated operations conducted by the U.S. government. Think of it as a three-dimensional task -- a count, an address list, a date. Count every resident of the United States -- estimated to be 275 million in 2000. Identify every residential address in the United States -- estimated to be 120 million in 2000. Assign 275 million people to 120 million addresses on a fixed, single date, April 1, 2000. Each of these operations by themselves is enormous -- people are on the move, addresses come and go, and this movement and transformation does not conveniently pause just because Census Day is April 1.

It is because the task is huge and complex that the Census Bureau is very careful in how it proceeds. To the extent possible, and especially for procedures not used in prior censuses, we test everything, and weigh what works and what does not. Hours of deliberation, even argument, precede a decision to build any given procedure into the census. Census staff take turns challenging each other to prove the merits of a given operation. Nothing is haphazard, nothing is casual. Every step is carefully, deliberately considered. In this lengthy process, which started for the Census 2000 12 years ago (just as in 2000 we will test procedures that might be incorporated in 2010), we select and discard based on one overriding criteria -- will this lead to a more accurate and complete count. Selection among alternative procedures is based not on what is more or less difficult, but what is more or less productive. To suggest that the Census Bureau excludes a particular procedure because it "would be too much trouble" reflects a serious misrepresentation of the dedication and commitment of the Census Bureau career professionals.

When the pieces are all put together, when the whole is assembled -- testing starts all over again. For now we must determine how well the integrated system will work, not just the individual pieces. This too is a painstaking task. For example, twenty-five major software systems have had to be designed, and linked together. These software systems allow us to track 175 million forms, pay hundreds of thousands of workers, monitor tens of thousands of partnership programs, produce 12 million maps -- and all of these things, and more, on a very tight schedule. Every step, every operation, every procedure is at huge scale and is interdependent with every other step, operation, procedure.

Our Operational Plan, which we presented to the Congress in January and which was refined in an Update based on Dress Rehearsal experiences and made public in February, is a plan to conduct Census 2000.

This census plan is now being documented in excruciating detail in what the Census Bureau terms its Master Activity Schedule. That Master Activity Schedule is 4,000 lines of code. But it is more than that. It is a software program that shows how each step connects to every other step. Every procedure links to predecessor operations, horizontally to dozens of other operations, and then to successor operations.

We believe it critical to lock-in procedures now. I have asked the Congress not to impose on us the risk of going into the census with "just-in-time" programming, with untested procedures, with additions whose consequences for other operations will not be discovered until they happen. The operational machinery that constitutes a census is not something to be taken lightly.

Accuracy

My second observation is about expected accuracy levels. Following the 1990 census, President Bush and the Congress urged the Bureau to design a census for 2000 that would reverse the decline in measured accuracy from the 1980 to the 1990 census. The initial design recommended by the Bureau to meet this difficult challenge was set aside by the Supreme Court.

The Bureau has turned its full attention to a full enumeration design -- one that is operationally more robust and innovative than what was implemented in 1990, and that includes extensive partnership efforts, paid advertising, locally-based recruitment of enumerators, user friendly census forms in six languages, state-of-the-art data capture technologies, and much more. Our first and most important effort is to put a census form into the hands of every single household in America, as well as enumerating those who live in group quarters (such as college dormitories) or who have no usual home. This robust design notwithstanding, we do not anticipate that 2000 will be more accurate than 1990.

There is a simple reason. The factors that led to a higher undercount in 1990 than in 1980 are even more present in American society today. More Americans are living in irregular housing or are transient; more Americans are linguistically isolated or eager to avoid any attention from government; more Americans are cynical or uncooperative when it comes to what they view as government intrusion. The census forms have to fight through more junk mail to even be noticed. Civic participation rates -- note voter turnout -- continue to decline.

In short, the Bureau -- using traditional counting methods -- must run harder to stay in place. It will run harder; it hopes to stay in place.

Moreover, the Bureau knows that the rate at which persons are missed is not evenly spread across all population groups. This pattern was again confirmed in the Dress Rehearsal sites. The characteristics that make it difficult to locate and secure the cooperation of residents cluster disproportionately in Hispanic, African-American, American Indian, and Asian population groups. Although Census 2000 includes major operations specifically focused on the differential undercount, it is not likely that these will eliminate the differential undercount.

The Census Bureau strongly hopes to be proven wrong in its assessment that neither the overall accuracy levels nor the differential undercount rates will in 2000 show improvement over 1990, but it has a professional responsibility to share with the country its best current judgment.

Finally, we consider a question much discussed since the Supreme Court decision -- will there be adjusted numbers for purposes other than apportionment?

The task of a statistical agency is to provide the most accurate numbers that it can, not to determine how those numbers will be used. The latter is a decision for the political/legal system. Consequently, the Census Bureau does not itself have an answer to whether adjusted numbers will be used for purposes other than reapportionment. It does, however, as I announced on February 18, intend to include as part of Census 2000 a final quality check, as it has done for a half-century. This 300,000 household survey is called Accuracy and Coverage Evaluation, or ACE. This quality audit tells the country how many persons were missed in the census, as well as their characteristics and the areas where they live. It is the Census Bureau's report card on its own performance, for the Bureau takes as much professional pride in carefully reporting how inaccurate it is as it does in working to be accurate in the first place.

The Bureau has informed the Secretary of Commerce that it is feasible to generate these more accurate counts on a schedule and at a level of geographic detail that makes them available and suitable for redistricting purposes and for the formulas that allocate federal funds, as well as for statistical and program purposes.

Specific New Proposals for Taking the Census

At a hearing before the Subcommittee on the Census on March 2, I was asked to address Chairman Miller's 10 proposals that come under the heading "America Counts Today (ACT)."

In my remarks to the Subcommittee, I strongly emphasized the concerns I have already expressed to you about timing. I also emphasized that ACT does not constitute a census plan. It is a series of isolated initiatives. For example, they have little connection to the huge operation described in the Census 2000 Operational Plan as the Master Address File. Except indirectly, and not helpfully, they have nothing to do with the enormous optical scanning operation planned for Census 2000. They do not help us with the difficult issue of unduplication, or with the operations needed to validate that housing units are vacant. And on and on. I stated that on seven of the items we welcome the approach taken by the Chairman. They are consistent with what the Census Bureau has learned about how to strengthen the census. We readily embrace a more extensive advertising campaign, a census-in-the-schools program that will involve 100% of the nation's schools, greater resources for the partnership program, additional enumerators, a partnership with the National Service Corporation, and the waiver initiative.

Let me just mention three things we are doing that parallel with Chairman Miller's proposals. In our operational plan for a census using traditional census-taking methods, which we submitted to the Congress on January 14, 1999, we noted that we plan to expand our partnership program and expand and enhance the paid advertising and promotion program. Both of these programs are geared toward greater public awareness of Census 2000, which in turn should lead to greater public response and cooperation.

o We plan to expand the partnership program to increase Bureau partnership staffing and assistance. This expansion will allow the Bureau to form additional partnerships with both nongovernmental organizations that represent historically hard-to-enumerate groups and with governmental entities, including tribal governments, that have not yet taken the opportunity to be included in the partnership program. Already, more than 10,000 partnership agreements have been signed. The expanded program includes "in-kind" funding to support partners by providing services, such as printing locally designed promotional materials for Census 2000.

o We plan to expand and enhance paid advertising and promotion by developing and implementing additional advertising messages. One additional message, which we will use before Census Day, will target information about community benefits to areas with historically low participation in the census. Another message will seek the public's cooperation with enumerators during the Nonresponse Followup operation. This is increasingly important now that we must make followup visits to 45 million housing units, instead of 30 million.

We also plan to expand the "Census in the Schools" program to allow all schools to participate instead of only those in selected areas. Nontraditional advertising methods also would be pursued. Fact sheets and promotional materials will be available on a larger scale with the expanded program. And finally, we plan to conduct special publicity events that would bring the Census 2000 message to communities across the Nation.

Now I will discuss why we cannot support three items -- the language initiative, the second mailing, and local government review of mailing addresses -- because the Census Bureau believes it has already presented more efficient programs than the suggestions advanced by Chairman Miller. Indeed, as I told the Subcommittee, if some of these initiatives were legislated in the manner now before the Subcommittee, they would disrupt and even place at risk Census 2000. Let me now describe my concerns about these three proposals.

First, how to reach the linguistically isolated in our population. We welcome interest in this most difficult area, and can assure you that we intend to be as linguistically friendly as we possibly can. We do, however, believe that the program set forth in the Operational Plan reaches Chairman Miller's goal more efficiently than printing census forms in 33 languages. We are printing forms in the six languages that account for 99% of all households in the United States.

Does this mean that we are indifferent to the other 1% of the households, which speak, by the way, not just another 27 languages but perhaps as many as 130 or more languages? The Census Bureau gave a lot of attention to how to reach those population groups, but it of course wanted to do so in a manner that did not place at risk other census procedures -- such as how many pages of the form can be optically scanned.

We subjected this issue to what we call a business analysis that contains 28 pages of detailed analysis listing all the pros and cons of not just one, but of four major alternatives. In the end, we designed a careful operation to reach those linguistically isolated households. We did consider Chairman Miller's idea -- and rejected it not because it was "too hard" but because it would not do the job. Instead we have set forth an integrated language program that involves 15,000 paid temporary staff positions in the Questionnaire Assistant Centers, drawn from a wide range of language communities and the preparation of 15 million assistance guides in several dozen languages. We have also included a language focus in our partnership agreements with community organizations. All of this to reach that 1% of the population that does not speak one, or more, of the six languages already covered in the census operation.

Were Chairman Miller's bill to pass, the following would have to happen. We would have to renegotiate our largest contracts -- including nearly 20 printing contracts, the contracts for the Telephone Questionnaire Assistance program; Data Capture; and Data Capture Service Centers. The entire workflow for receipt, image capture, transcription, and/or key-from-paper would have to be modified.

Let me offer one simple example--the pre-census letter. It will go to 120 million households. The letter wording has been carefully designed to minimize confusion and maximize cooperation. There was extensive staff discussion on how to announce the availability of questionnaires in five languages other than English. You would now be asking that we figure out how to announce another 27 languages, each of which is of concern to a fraction of a million households, without confusing the remaining 119 million households. And this is just one simple procedure that could not at this late date be tested.

Next, the second mailing. Again, there is research, there is analysis, there is deliberation, there is judgment, there is Dress Rehearsal experience -- all of which indicates that its value is outweighed, substantially so, by the risks that it introduces in other census operations, not least of which is the deterioration in data quality in Nonresponse Followup. The Bureau has decided not to implement either a targeted or blanket second mailing. We considered a targeted second mailing, but printing vendors told us they would require at least a month to send a second mailing targeted only to nonresponding housing units. A targeted second mailing would, thus, have significantly delayed the start of the Nonresponse Followup operation, and our experience and research indicate that to ensure high quality and accuracy we need to begin this critical operation as soon as practicable after Census Day. Next, we considered a blanket second mailing, that is, mailing a second questionnaire to every housing unit, whether or not we had received a response to the original mailing. But a National Academy of Sciences panel advised us that there is considerable risk that a blanket second mailing could reduce the accuracy of the census. It could increase costs in dealing with duplicate forms, lead to massive duplication, and create adverse public reaction. Our Dress Rehearsal evaluations showed that a blanket second mailing did increase the overall mail response rate. However, the evaluations also indicate that about 40 percent of the households that mailed back a second questionnaire had also mailed back the initial questionnaire. Thus, the Dress Rehearsal processing had to be extended three weeks to handle the complexity introduced by the large volume of duplicate forms. For Census 2000, a workload of this magnitude would significantly delay data processing operations and potentially introduce significant errors into the data.

A final point I would make is that while a second mailing can help to increase mail response, there is no evidence that it reduces the undercount. In fact, if persons who mail back the first form are the easiest to count, those who would mail back a second form are the next easiest to count. They would almost entirely be included in the Nonresponse Followup operation and we do not expect great difficulty in enumerating them.

Third, Post Census Local Review. The Post Census Local Review program provided some limited benefit within the context of the 1990 census. But since early in this decade, the Census Bureau has been working to find a better way to improve the address list and to correct the deficiencies of the 1990 local review process. One such improvement is providing state, local, and tribal governments an earlier and more effective opportunity to correct the Census 2000 address list for their areas. We believe we have done that through the Local Update of Census Addresses, or LUCA, program made possible by the Census Address List Improvement Act that the Congress passed in 1994, which for the first time allows the Census Bureau to share its address list with its partners. And, I am pleased to report that we are looking at ways to modify our ongoing procedures for the Census 2000 address review process to respond to concerns raised by our governmental partners about newly constructed housing units.

The 1990 Post Census Local Review process cost too much on a per-case basis and took too much time relative to the extremely small number of people it added to the census. It came too late in the process and gained participation of only one-fourth of the nation's 39,000 governments. The 1990 Post Census Local Review added only one-twentieth of one percent to the overall population count, or about 125,000 persons. If we were to conduct such a program again, it would cost more and add even fewer people to the count than in 1990 because it would be redundant to the precensus address list review activities that are already underway.

The address list review process the Census Bureau has designed for Census 2000, with congressional and participant encouragement and support, is innovative, responsive to our partners, and efficient. The LUCA program thus far has been a success. We have encountered problems, but because we started early, we have had time to make refinements and correct the problems. Through our extensive outreach efforts, twice as many local and tribal governments are participating in LUCA as participated in the 1990 Post Census Local Review; those governments cover about 85 percent of all addresses. So, thus far, response to the program has been great.

We have heard increasing concerns from our governmental partners that housing units newly constructed in the months before Census Day may not be adequately accounted for under our existing plan. Any benefits from a Post Census Local Review would derive particularly from local governments' information on newly constructed units in their jurisdiction. Such information can and should be used in the census, but much earlier in the process than would occur under Post Census Local Review.

To address these concerns, we have developed a plan that we are refining in consultation with members of the Secretary of Commerce's 2000 Census Advisory Committee. Although

these plans are still being refined, we have developed a program that will allow local and tribal governments an additional opportunity to review the address list for their area starting in mid-January 2000 and ending on Census Day, April 1, 2000. The purpose of this review is to bring to our attention any newly constructed housing units that are not on the Census 2000 mailing list. This opportunity will be offered to all governments in which the Census Bureau will have the U.S. Postal Service deliver Census 2000 questionnaires -- approximately 94 million addresses. Local and tribal governments will have this opportunity whether they participated in LUCA or not.

Any units that are identified during this procedure will be verified during our coverage improvement field activities, and enumerated if they, in fact, were occupied on April 1. This opportunity for local and tribal governments to bring new construction to our attention will complement our planned check by the U.S. Postal Service that is scheduled to be conducted mid-January to mid-February 2000 and that also will result in the addition of new addresses to our list. In the more rural areas of the country, our plan already requires census enumerators to look for newly constructed units at the time they deliver the questionnaires or conduct the enumeration.

In addition, we have established a program to validate the boundaries for every local and tribal government in the United States. As we do before every decennial census, in late 1999 we will provide maps to jurisdictions showing their latest reported boundaries and ask that they provide any corrections needed to make them accurate as of January 1, 2000, the official boundary date for Census 2000. Then we will provide a set of maps showing each government that we incorporated their changes. This is unlike 1990, when governments did not see the results of their efforts until after the census. These boundary validation maps will be sent in the late spring of 2000 and we will seek any residual corrections by early summer of 2000.

That brings me to what I believe is the critical weakness of H.R. 472, Chairman Miller's proposal, "The Local Census Quality Check Act." It would mandate a program that Census Bureau professionals believe will not effectively and efficiently contribute to the overall accuracy of the census. The bill proposes a Post Census Local Review process that is very similar to the 1990 census process and that would face the same problems that the 1990 process faced and that I have described above. The bill does make two changes to the 1990 Post Census Local Review; it gives local officials more time to conduct the review and provides an opportunity for feedback. But because of the extensive address list review program that will take place before the census, including the new program to add newly constructed units that will carry up to Census Day, there will be nothing left for local officials to contribute in terms of adding housing units to the census. They will receive essentially the same address list they received in January. And this unproductive review, under the bill, would be conducted late in the process, extending even later than the 1990 program because it would give local officials three times as much review time as they had in 1990. The bill does nothing to improve upon the 1990 process. We have looked at ways to improve that process and concluded that an earlier review and input from local governments process would be most efficient. That is what we have designed.

In closing, I want to emphasize that the census clock is ticking. Given the lateness of the hour, we must acknowledge the hard reality that we cannot take a chance on untested operations or late additions. The most we can do is to modify already existing operations to enhance quality where that is possible, and we continue to look at ways to do so. The largest peacetime mobilization in U.S. history must go forward relying on the considered professional judgment of the career scientific and operational experts at the Census Bureau. I look forward to working with the Monitoring Board to achieve that goal.