My Two Census

Formerly the non-partisan watchdog of the 2010 US Census, and currently an opinion blog that covers all things political, media, foreign policy, globalization, and culture…but sometimes returning to its census/demographics roots.

Interesting 2010 Census Op-Ed from The Huffington Post

Michael J. O'Neil

Michael J. O’Neil

President, O’Neil Associates

Census Pick Illustrates Broader Obama Strategy

Posted: June 15, 2009 04:07 PM

A strategic retreat on the matter of employing estimation in the Census is thus one illustration of a selective approach the Obama administration has taken to many issues. Choose your battles carefully, prioritize, and do not give the opposition the opportunity to define you.

What does the testimony of a Presidential nominee to a nonpolitical federal statistical agency tell you about the strategic approach of the Obama administration? If you observe how it fits a pattern, it tells you quite a bit.

Robert M. Groves is President Obama’s pick to head the Census Bureau. From a purely technical standpoint, he could hardly have picked a more qualified candidate to head the bureau. Groves is among the most respected survey researchers in the field. In that sense, the pick resembles the choice of Nobel Laureate Steven Chu as Energy Secretary: each is a highly respected scientist and neither has discernible political leanings.

But an 800 pound gorilla loomed over any pick to head the Census Bureau: whether to use statistical estimation to adjust the results of the Census for purposes of congressional apportionment. The issue is a political hot potato: any hint of using estimation for apportionment would result in screams from the Republican side of the aisle that the Census was being manipulated for political purposes.

Why? While the Census attempts to interview everyone, complete enumeration is an especially imperfect business. And the professionals at the Census Bureau not only know this, they know the characteristics of those they miss. Poor people (particularly homeless ones), minorities, and immigrants are those most likely to be missed — over 4 million of them according to a study of the 2000 census. And these groups tend to be Democrats — and live in areas with disproportionately more Democrats. So applying statistical estimation would inevitably increase population estimates in Democratic areas, and thus result in more Democrats in Congress.

Which is more accurate, imperfect enumeration or statistical estimation? There is really no scientific dispute here. The Census conducts hundreds of surveys and produces tens of thousands of population estimates. The Census uses estimation, to my knowledge, for every one of these estimates — except for congressional reapportionment. The users of the Census want the most accurate data possible. The apolitical scientists know what is accurate and how to derive accurate estimates, and that is achieved through the application of carefully derived and tested estimation models rather than a highly flawed actual headcount with known and systematic errors.

(If you need a non-statistician’s explanation of why sampling and estimation is as accurate as an actual enumeration, consider your last blood test: did they remove and test ALL your blood?). These matters are beyond any scientific dispute. Yes, sampling is theory — the same way gravity is theory.

Groves is a professional sampling statistician. He knows all of this from a lifetime of work in this and related arcane methodological matters. If a purely scientific panel of the most eminent sampling statisticians in the country were commissioned to make recommendations in this area, Groves name would be on the top of almost any list of preeminent experts in this area. Indeed, while serving as Assistant Director of the Census after the 1990 Census, he recommended using sampling and estimation to correct for the known errors in that Census. While from a scientific perspective the issue is a complete no-brainer, Republicans in the Senate have seized on this to assert that he has a partisan agenda.

They clearly do not know the man. I do. I have known Bob Groves for over 30 years as a professional colleague. I have read his work, and even collaborated with him on a couple of papers on some of the specialized methodological issues of survey research that have been his life’s work.

And after 30 years of professional acquaintance, I can tell you that I have not a clue about Bob’s political leanings. (And those who know me for the political junkie that I am will find this astounding, for the political realm has always been my personal passion: it tends to be among the first topics I gravitate to in any casual conversation). This fact, I think, attests to Bob’s nonpolitical nature. He is not motivated politically, he is a Scientist, and one of the first order.

But, curiously, Groves, in his confirmation hearings, indicated that he had no intention to employ or advocate the use of sampling or estimation in conducting the 2010 Census. What gives?

I think this says something very revealing, but far more about the Obama administration than about Bob Groves. I have no doubt whatsoever what Bob’s private counsel would be if asked about whether applying estimation principles to the Census would increase its accuracy. Indeed, his scientific judgment on this matter is already a matter of public record. But what is interesting here is how this new position mirrors the Obama administration’s approach to dealing with many controversial matters. There is a pattern: President Obama does not want the political distraction of Republicans screaming that the Democrats have “fixed” the Census to produce a partisan result. It would not matter that as a matter of scientific certainty, such claims would be wrong; they could score political points in making the charge. (This is the type of technical issue that is difficult to explain to a statistically lay audience; many intelligent people simply won’t understand it.) Obama looks willing to forgo the congressional seats, perhaps a dozen or so, Democrats would gain in order to avoid this political distraction and pursue higher priorities. He has bigger fish to fry.

This strategic retreat resembles the back-burnering of issues such as gun control and gays in the military. Each has been delayed out of a fear that it could be divisive and derail his core agenda, especially the economy and health care reform. To pursue key objectives, he has been willing to delay action on other issues that could distract or dilute his mandate. While he has pursued many initiatives, he has carefully avoided those with the explosive potential to blow up the broader agenda. And an attempt to use estimation for reapportionment has that potential. While the scientific merits are indisputable, getting the public to understand such arcane statistical principles is a lost cause. The Obama administration has concluded that it is simply not worth the political capital to try.

One need look no further than the impact the gays in the military issue had on the early Clinton administration to see the risks he seeks to avoid. The strategy does not necessarily mean that such issues will be avoided indefinitely. But it does argue that they should be held back lest they come to define his Presidency. And, in the case of gays in the military, it allows time for proposals to percolate up from the military establishment rather than appearing to be imposed on them.

Tags: , , , ,

One Response to “Interesting 2010 Census Op-Ed from The Huffington Post”

  1. Satorist Says:

    For someone purporting to be both a professional colleague of the nominee, and a “political junkie”, this article conveys an astonishingly limited understanding both of the issues addressed and the politics surrounding them. Sadly typical fare for this half-baked site.

    A small sample (pun intended):

    “A strategic retreat on the matter of employing estimation in the Census is thus one illustration of a selective approach the Obama administration has taken to many issues.” An absurdity employed in the service of banality posing as insight. As demonstrated below, the decision to not use statistically adjust census data was made years, ago before Obama even made it to the Senate. There was no position from which he could “retreat.” Furthermore, every administration picks and chooses the issues it will tackle. It’s called prioritizing–something we all do every day.

    “But an 800 pound gorilla loomed over any pick to head the Census Bureau: whether to use statistical estimation to adjust the results of the Census for purposes of congressional apportionment.” Yes, “loomed” perhaps like its equally fictive ten-story unicorn pal, “Sparkle.” The Supreme Court ruled in 1998 that it was not legal (not unconstitutional, just not legal) to use statistically adjusted counts for the purpose of [re]apportionment. The post-2000 controversy was whether or not to employ statistical adjustment for purposes other than reapportionment–none of which were ever on the table for 2010.

    Which is more accurate, imperfect enumeration or statistical estimation? There is really no scientific dispute here.” Agreed: After trying to statistically adjust the 2000 census count, the Census Bureau abandoned the effort because it introduced as much or more error than the results of the “headcount.” Where have you been? For eight years now, the Census Bureau has stated it does not know how to use statistical estimates to adjust census data to achieve a more accurate count and has no plans to do so. That’s why operational plans for the 2010 census formulated much earlier in the decade have never included provisions for using statistical adjustment.

    The problem is not mathematical but with what is being estimated and for what purpose. Far from being a simple total count of the population; a census is a count of population by age, gender, race and ethnicity–at very small levels of geography. It is one thing to accurately estimate the population of, say, a county but it is exponentially more difficult (impossible in practical terms) to more accurately estimate populations at the block level. That is why survey data, mentioned below, are not published at sub-level geographies. Furthermore, surveys are used for limited purposes many federal funding allocations rely on decennial census data, not surveys.

    “The Census conducts hundreds of surveys and produces tens of thousands of population estimates. The Census uses estimation, to my knowledge, for every one of these estimates — except for congressional reapportionment.” 1. See above; 2. All surveys rely on decennial census data as a base–yet another reason why accuracy of the decennial census is crucial.

    “The apolitical scientists know what is accurate and how to derive accurate estimates, and that is achieved through the application of carefully derived and tested estimation models rather than a highly flawed actual headcount with known and systematic errors.” First of all, it might be helpful to clarify what the issue actually is. The “highly flawed actual headcount” was, in 2000, something like 98.6% accurate. That is “highly flawed” by practical standards of perfection common only among the blindly optimistic. In fact, statistical adjustment was more difficult in 2000 precisely because the headcount was so accurate. Secondly, no one is seriously suggesting doing away with decennial censuses in favor of estimation but, rather, using statistical estimates to augment the traditional headcount to achieve still greater accuracy. Third, “missing” four million people is itself an estimate (subject to error as everything else). It sounds like a lot until one realizes that it is roughly one percent of the total count. It is obviously better to miss fewer but it is a fallacy to believe that estimates are error-free. They can produce less accurate counts or reduce some error while wildly exaggerating others, as was the case in 2000 (the number older Asians, for instance).

    (If you need a non-statistician’s explanation of why sampling and estimation is as accurate as an actual enumeration, consider your last blood test: did they remove and test ALL your blood?). These matters are beyond any scientific dispute.” A hoary analogy, as inappropriate as it is glib, only (and only possibly) applicable if we lived in a completely homogeneous society–Austria, maybe, or perhaps a small, isolated aboriginal tribe.

    “Yes, sampling is theory — the same way gravity is theory.” More of a tool, actually, effective depending on the intended use. The number of people of Native American ancestry in the US? Sure. The number of single parent Native American women under 50 sharing a rental home with a grandparent and two independ children in block 689 of the District of Columbia, with a household income of between $47,0000 and $53,000 per year? Demonstrably less accurate.

    “[W]hile serving as Assistant Director of the Census after the 1990 Census, [Nominee for Census Director Bob Groves] recommended using sampling and estimation to correct for the known errors in that Census. While from a scientific perspective the issue is a complete no-brainer, Republicans in the Senate have seized on this to assert that he has a partisan agenda.” I hesitate to disagree with any statement containing the terms “no-brainer” and “Republican” but it bears noting that Groves was one of an (I believe) 11-member internal review panel formed to respond to an emergency in the wake of an unanticipated low response rate that precluded adequate time for non-response follow-up field work. Faced with the certitude of not canvassing households known to have not returned a questionnaire, the panel recommended estimating the population within those households–which is quite different than estimating the size of an unknowable undercount.

    “[Groves] is not motivated politically, he is a Scientist, and one of the first order. As is the entire executive staff of the Census Bureau, including those who recommended against using the results of the 2000 statistical adjustment. They were and are no more “anti-sampling” than Groves is “pro-sampling.” There really is no political bias at all at the Census Bureau, decisions are based on judgments about what produces the best data. Politics is simply too boring for mathematicians and demographers.

    “But, curiously, Groves, in his confirmation hearings, indicated that he had no intention to employ or advocate the use of sampling or estimation in conducting the 2010 Census. What gives?” Reality-based judgment, perhaps? The task of counting everyone residing in the United States “once, and only once, and at the right location” (as the bureau likes to say) requires years of planning. A fairly large group of people are even now beginning to plan for the 2020 census. Plans developed for employing statistical adjustment of the 2000 census began while GW Bush was still failing at various business ventures. Those plans involved recounting more than 3 million households, scientifically selected and weighted, and the hiring, training and deployment of workforce parallel to but separate and distinct from all other decennial census operations and staff (to elimminate bias). Different offices, different staff, different equipment, enumerators, different collection and compilation procedures, etc. The idea that Groves could assume office a mere ten months from Census Day and decide to introduce a new procedure that is delusional. That’s what makes the Republican’s suspicions so ludicrous: the train left the station long ago. For good or ill, all that can be done at this point is to try ensuring that the existing operational plan is executed as effectively as possible.

    “I think this says something very revealing, but far more about the Obama administration than about Bob Groves. I have no doubt whatsoever what Bob’s private counsel would be if asked about whether applying estimation principles to the Census would increase its accuracy. Indeed, his scientific judgment on this matter is already a matter of public record. But what is interesting here is how this new position mirrors the Obama administration’s approach to dealing with many controversial matters. There is a pattern: President Obama does not want the political distraction of Republicans screaming that the Democrats have “fixed” the Census to produce a partisan result. It would not matter that as a matter of scientific certainty, such claims would be wrong; they could score political points in making the charge. (This is the type of technical issue that is difficult to explain to a statistically lay audience; many intelligent people simply won’t understand it.) Obama looks willing to forgo the congressional seats, perhaps a dozen or so, Democrats would gain in order to avoid this political distraction and pursue higher priorities. He has bigger fish to fry.” Utter nonsense borne of a misconstruing of limited facts shallowly understood. Again, the decision was irreversible long before Iowa’s presidential primary last year. Accepting what is a fait accompli doesn’t signify an “approach” to dealing with anything, controversial or otherwise. It’s a simple recognition of reality. You may well be correct in your opinnion of Obama’s strategy but this is not evidence of it. There is no controversy at all about this–except in the inflamed and fearful minds of Republican politicians who project on others their own propensity for calculating subterfuge (WMD, torture, FISA and other denied criminality)… and, apparently, posturing proselytes.