PETERSON-MIDDLEBURY
I included a unit on the census following discussion of opinion polls and survey sampling. Material from the Quantitative Literacy Series booklet Exploring Information from Surveys and Samples proved quite helpful throughout this discussion. In particular, the section on capture-recapture was a good lead-in to the PES methodology.
Particularly helpful articles:
Bailar, Barbara. The miscounting of America Washington Post, March 6, 1988, pC3.
Hamilton, David P. Census adjustment battle heats up. (statistical adjustment of its decennial count of the American population), Science 248, (May 18 1990), 807-808.
Barringer, Felicity. Commerce Dept. declines to revise '90 census counts; 5 million people missed; critics say administration's ruling will hurt minority groups and big cities. The New York Times, July 16 1991, pA1(N), pA1 (L).
Passell, Peter. Can't count on numbers (the elaborate effort to adjust the census by using statistics adds up to ambiguity). The New York Times, August 6 1991, pA1 (L).
Freedman, David. A. Adjusting the 1990 census. Science 252, May 31 1991, 1233-1236.
(I did not do much with the homeless discussion. Moore's Concepts and Controversies text talks about this in the context of sampling. See also articles in the source list at the end of this profile.)
The QLS booklet does a good job discussing the assumptions of the basic capture-recapture model. For animal populations, the term ``trap-happy" and ``trap shy" are introduced. A trap-happy animal becomes easier to catch after being caught once; a ``trap-shy" animal becomes more elusive. Exercises had students map notions of ``trap-happyg" and ``trap-shy" animals, births and deaths into corresponding problems with PES.
Class Discussion Questions:
1. As we have seen, many people are suspicious of statistical polls and predictions. And no one is claiming that the adjusted numbers will be more accurate on a block by block basis. Do you think the government can convince the public that an adjustment, which will have demonstrable flaws, is still an improvement?
2. Opponents of adjustment decry any tampering with the ``virgin data." (Mosbacher said he was ``deeply concerned that adjustment would open the door to political tampering with the census in the future...The outcome of the enumeration process cannot be directly affected in such a say", NYT July 26, 1991) Statisticians favoring adjustment point out that ``virgin data" notion is misleading; the result of the enumeration procedure itself is only an ``estimateg" (Eriksen, Kadane, Tukey, JASA, 1989). What do you think?
SOURCES
News Stories (chronological)
Sun, Marjorie. Plan to assess census undercounting dropped. Science 239 (January 29, 1988), 456-457.
Commerce Dept. cancels plans to statistically adjust counts. Barbara Bailar, head of statistical research at the Census Bureau, whose group formulated the plan, resigns in protest. Claims decision was politically motivated; implementing plan would have resulted in more money and votes in areas heavily black and urban.
Bailar, Barbara. The miscounting of America Washington Post, March 6, 1988, pC3.
Excellent pre-1990 overview. How do we know there is an undercount? Demographic analysis and case-matching. Proposed coverage-measurement survey for 1990. Timing issues for congressional reapportionment, allocation of federal aid. Dealing with criticisms of adjustment proposals.
Reinhold, Robert. ``Census is Prodded to Adjust Figures." The New York Times March 9, 1990.
Roberts, Steven V. An all-American snapshot: how we count and why. U.S. News &World Report, April 2, 1990, p. 10.
Includes some brief historical notes on census undercounts since 1790.
Why this census sags in the middle U.S. News &World Report, April 23, 1990, p14-15.
In addition to traditionally undercounted minorities, the middle class is not cooperating with this census either.
The best laid plans.... (1990 Census) (American Survey). The Economist 315 (7652), April 28 1990, p28.
Martz, Larry with Glick, Daniel. Apathy and foul-ups foil the census. Newsweek, April 30, 1990, p23.
Early reports of compliance with census were dismal (63%mailed in compared with the 70%anticipated). Discussion of possible causes: too much junk mail; difficult forms; govt. miscues on deadlines, computerized lists, etc.; ``Big Brother"; something to hide.
Hamilton, David P. Census adjustment battle heats up. (statistical adjustment of its decennial count of the American population), Science 248, (May 18 1990), 807-808.
Includes a simple graphic on the E-sample (original enumeration) and P-sample (post-enumeration sample), and the calculation of the adjustment factor.
Barringer, Felicity. Head of Census defends urban tallies. The New York Times Sept. 12, 1990, pA12(N), pB6(L). Garland, Susan B. Money, power, and numbers: a firestorm over the census. Business Week, 17 Sept 1990, p45. The Census Bureau needs some new math (editorial). Business Week, 17 Sept 1990, p164.
Highlights need for adjustment, and calls on the government to explain in plain English what is going on and why.
Secondhand reports were used to count millions in census. (House census and population subcommittee investigation of 1990 census count). The New York Times, Feb 2 1991, p10(L).
Congressional investigation shows that 7.3 million were counted by second-hand reports. Includes breakdown of total census figures by Barbara Bryant.
The 1990 census, state by state (Census Bureau report) (statistics on Minnesota and Wisconsin) (illustration). The New York Times, Feb 2 1991, pA10(L).
The 1990 census, state by state (Census Bureau report) (statistics on Alabama, Hawaii, Nebraska, Rhode Island) (illustration). The New York Times, Feb 9 1991, p9(L).
The 1990 census, state by state (statistics on Illinois and the Washington D.C.) (illustration). The New York Times, Feb 12 1991, pB6(L).
The 1990 census for Connecticut (Census Bureau report) (illustration). The New York Times, Feb 15 1991, pA13(N), pB2(L).
The 1990 census, state by state (Census Bureau report) (illustration). The New York Times, Feb 20 1991, pA10(N), pA18(L).
The 1990 census, state by state (Census Bureau report) (illustration). The New York Times, March 9 1991, p9(N).
Barringer, Felicity. Two million blacks not counted, head of census panel asserts. (Thomas C. Sawyer). The New York Times, March 12 1991, pA13(N),pA18(L).
Barringer, Felicity. Federal count of homeless is far below other figures. (U.S. Census Bureau's methodology expected to be questioned by National Law Center for Homelessness and Poverty, other advocates). The New York Times, April 12 1991, pA14(N), pA11(L).
Gonzalez, David. New York's homeless recall a no-count night. (Bureau of the Census' attempt to count homeless in New York, New York). The New York Times, April 14 1991, p16(N), p27(L).
Barringer, Felicity. Federal survey finds census missed 4 million to 6 million people. (Census Bureau survey). The New York Times, April 19 1991, pA8(N), pA1(L).
Results of post-census survey: populations is 252.9-254.9 million (compared to 248.7 by Census). Implications of statistical adjustment: California and Georgia would each gain a congressional seat; Montana would keep its second seat, which would be lost based on Census; Wisconsin would lose one seat; Penn would lose three instead of two; Mississippi might lose one of its five seats. Includes data map showing %undercount by state.
Barringer, Felicity. Making repairs on a faulty census count. (limitations of 1990 census survey). The New York Times April 21 1991, Sec. 4, pE7(N).
Not the best census ever. (Commerce Department says 1990 census left many uncounted) (editorial). The New York Times, April 25 1991, pA16(N), pA24(L).
Barringer, Felicity. Census revisions would widen political gains of 3 big states (Northeast and midwest might face ebb in power). The New York Times, June 14 1991, pA1(L).
As expected, biggest undercounts are among underclass, concentrated in large cities. But geographically, most are in West and South. Includes table of proposed corrections by state.
Levine, Richard. Census adjustment means more for New York. (1990 undercount adjustment survey's impact on Metropolitan New York City). The New York Times, June 14 1991, pB7(L).
The populations of NY, NJ and CT would indeed grow under adjustment, but less than the national average of 2.1%undercount. So these municipalities might still stand to lose. (It turns out that NYC would gain at the expense of NY state.)
Adjusting for injustice. (making adjustments in the 1990 Census undercount) (editorial). The New York Times, July 13 1991, p14(N), p20(L).
Barone, Michael. Playing census politics. U.S. News and World Report. July 15 1991, p31.
Barringer, Felicity. Decision today on adjusting the census. (flaws in 1990 census may be remedied by using statistical formulas. The New York Times, July 15 1991, pA11(N), pA12(L).
Statisticians disagree on feasibility of adjustment. A panel of 8 non-governmental experts was convened. The four members appointed by the Commerce Dept. opposed adjustment; the others favored it!
Barringer, Felicity. Commerce Dept. declines to revise '90 census counts; 5 million people missed; critics say administration's ruling will hurt minority groups and big cities. (Commerce Secretary Robert A. Mosbacher announces that the 1990 census counts will not be adjusted; most of undercount involves minorities). The New York Times, July 16 1991, pA1(N), pA1(L).
Lee, Felicia R. Already struggling, big cities are hit hard. (impact of decision not to adjust 1990 census for undercount on municipal revenues). The New York Times, July 16 1991,pA16(L).
Skerry, Peter. Sense on the census. (government will not adjust 1990 census for undercount). The New York Times, July 18 1991, pA19(N), pA21(L).
Adjustment would be a mistake. Cites NRC study that undercounts affect fund allocation to a relatively small degree. Argeus that isaffected minorities would be even less likely to participate if they felt they could be counted statistically, and would ignore community leaders' calls to stand up and be counted.
Just a traditional census. (estimates of undercounts consistent with past results). U.S. News &World Report, July 29 1991, p10.
Table of undercounts since 1940. (Completely ducks issue of differentials by ethnic groups).
Passell, Peter. Can't count on numbers (the elaborate effort to adjust the census by using statistics adds up to ambiguity). The New York Times, August 6 1991, pA1 (L).
The experts are split on the merits of readjustment. Good accessible discussion. Passell, Peter. Federal gold, census rainbow. The New York Times, August 7 1991, pD2 (L).
Even if they winreadjustment lawsuits, plaintiffs may not happy. First, total aid is fixed, so more people does not directly translate into more dollars. Winners are areas where adjustment exceeds the average of 2%. Likely winners: Western states, Sun Belt cities, DC. NYC not a big winner; NY stat, NJ, Illinois, Cleveland and Phila would lose!
Barringer, Felicity. U.S. finds more flaws in census undercount. The New York Times, August 25 1991, p23.
A General Accounting Office report released 23 Aug makes explicit the fact that the undercount of 5.3 million is a net figure which reflects both omissions and double counting. The census actually missed 9.7-15.5 million, and double-counted 4.4-10.2 million.
Barringer, Felicity. Democrats in Congress revive doubts on accuracy of census. (Democrats want Robert A. Mosbacher, Commerce Department to make public adjusted set of census figures). The New York Times, Nov 20 1991, pA11(N), pA22(L).
Bush officials to give census tapes to Congress. (1990 census count adjustments). The New York Times, January 10 1992, pA14(L).
Commerce Dept. will give to House subcommittee the computer tapes with data on more than 3 million areas. Mosbacher had originally refused, saying that it would disrupt redistricting efforts; House had the tapes subpoenaed.
Discussion articles
Bounpane, P. A. and Jones, T. A. Automation of the 1990 U.S. census. Chance 1 (1), (Winter 1988), 28-36.
The wider use of computers was one reason statisticians felt adjustment, which had been rejected in 1980, was now feasible. Description of PES scheme.
Goldfield, Edwin D. What we expect to learn from the 1990 decennial census. Chance3 (2), (Spring 1990), 13-19.
Good overview of demographic ideas.
Exter, Thomas. Inside the information industry: the survey that measures the census.American Demographics 12 (11), (November 1990), p12-14.
Includes sidebar by Feinberg on ``the most elusive Americans."
Fienberg, Stephen E. An adjusted census in 1990? Chance 3 (2), (Spring 1990), 23-25.
Fienberg, Stephen E. An adjusted census in 1990? The judge rules and the PES begins. Chance 3 (3), (Summer 1990), 33-36.
Freedman, David. A. Adjusting the 1990 census. Science 252, May 31 1991, 1233-1236.
Description of demographic models and PES plan (dual system estimator). Details how the adjustment would actually be implemented in terms of individual records. Conclusion: present state of the art cannot support adjustment in 1990. (Can't have an adjustment process prone to 10%errors when trying to fix a 1-2%undercount.)
Gleik, James. Why we can't count. New York Times Magazine 54, July 15, 1990, 22-26.
Subtitle: ``Scientists who can count fish and stars with precision, say the Government still counts people with its toes." Hazards of one-by-one enumeration. Discussion of capture-recapture ideas, pitfalls of selection biases.
Keyfitz, Nathan (1979). Information and allocation: two uses of the 1980 census (with discussion). The American Statistician33, 45-56.
Norman, Colin. Who should count in the 1990 census? Science 243 (489), (Feb. 3 1989), 601-602.
The 1980 census is estimated to have missed 1.4%of the total population, but undercounted blacks by about 5.9%(in some areas 18-24%of black males may have been missed). New York, Houston, Chicago and LA have filed lawsuits. Also discusses including illegal aliens in the count (estimated 2 million total in 1980 count probably gave congressional seats to NY and CA that might have gone to GA and IN).
Rossi, Peter H., Wright, James D., Fisher, Gene A., and Willis, Georgianna. The urban homeless: estimating composition and size. Science235 (March 13 1987), 1336-1341.
Empirical study of the homeless population in Chicago. Includes a ``shelter survey" (probabilistic sample of persons spending the night in shelters) and a ``street survey" (complete enumeration of persons encountered between midnight and 6:00 a.m. in a probability sample of Chicago census blocks).
West, Kirsten K. and Fein, David J. Census undercount: an historical and contemporary sociological issue. Sociological Inquiry 60 (2), Spring 1990, 127-142.
Statistical Articles
ASA Technical Panel on the Census Undercount (1984). Report of the ASA technical panel on the census undercount (with comment from census bureau). American Statistician38 (4), 252-260.
ASA Technical Panel on the Census Undercount (1985). Correction. American Statistician 39 (3), 241.
Eriksen, Eugene P., and Kadane, Joseph B. (1985). Estimating the population in a census year: 1980 and beyond (with discussion). JASA80, 98-131.
Eriksen, Eugene P., Kadane, Joseph B. and Tukey, John W. (1989). Adjusting the 1980 census of population and housing. JASA84, 927-943.
Freedman, D.A., and Navidi, W.C. (1986). Regression models for adjusting the 1980 census (with discussion). Statistical Science 1, 1-39.
Mulry, Mary H. and Spencer, Bruce D. (1991). Total error in PES estimates of population (with discussion). JASA86, 839-863.
Wolter, Kirk M. and Causey, Beverly D. (1991). Evaluation of procedures for improving population estimates for small areas. JASA86, 278-284.
Survey Methodology, June 1992. Contains a special section on the undercount. We have not seen this yet, but Laurie has ordered copies.
General
Landwehr, James M., Swift, Jim and Watkins, Ann E. Exploring Information from Surveys and Samples. Palo Alto, CA: Dale Seymour Publications, 1987.
Section VIII ``A Capture-Recapture Method, pp. 66-74. Simple capture-recapture model, with good analysis of assumptions. Very helpful, at an elementary level.
Moore, David S. Statistics: Concepts and Controversies. NY: W.H. Freeman, 1991.
Intro to Part I describes the Chicago Homeless Study. Chapter 1, Section 6 describes stratified random sampling.