SciFri
 FEEDBACK

 

Last week's topic: the Year 2000 census.

We asked whether or not the government should be allowed to use statistical sampling methods to help make a head count in the upcoming census.

Here's what some of you had to say...

---

Irving I. Kessler

Your program on sampling for the 1990 census was surprisingly deficient in that it failed to provide the listener with an opportunity to formulate an objective view of the subject. The principal shortcoming was that little, if any, time was devoted to any alternative view of the subject, apart from the participants' endlessly repeated notion that statistical sampling represented the only viable alternative to the notorious political maneuverings of the Republicans in Washington.

Although the host attempted twice, but unsuccessfully, to address the potential limitations of sampling for the census, the listeners were invariably treated to such nostrums as [1] the "fact" that statistics is a science and, therefore, presumably immune from political manipulation; [2] the "fact" that the Census Bureau has been working for forty -five years to develop its sampling scheme and that, therefore, the scheme is now faultless; and [3] that statistics and statistical sampling have by now been so perfected as a science that they can be expected to resolve the age-old problem of census error.

The use of statistics and statistical sampling is, of course, based in science but it does involve the use of statistical models i.e. guessing, an inherently non-scientific action. Discrete assumptions must be made about the nature of any model before a statistician attempts to apply it, for example in a census. In addition, if a valid recensus of 1% of the U.S. population is conducted, the results will be applicable to the whole country but will not be applicable to particular states, cities or other smaller subdivisions. Therefore, if one hopes to generate accurate census figures for use in our cities [where most enumeration errors occur], the sampling scheme used may not permit this. I would have enjoyed having the participants address such an important issue.

---

Don Saklad

Why not do both? Will the same data be collected anyway?

---

Chris Miller

So, let me see if I got this right.

Your guests are arguing that sampling is a better method for counting people. Their proof is that sampling shows that counting isn't working. Isn't this similar to the following argument: We should start weighing things using my new improved SuperScale. Because the SuperScale says everything else is not accurate. How do we know we're missing any of our population, exclusive of using the Sampling process? That's the only way it can be proven that the counting method is wrong.

---

Pamela Grennes

As a researcher who frequently does survey type research, I completely understand the Census Bureau's use of this technique. As you said in the dialogue, many people are unresponsive to surveys and the only way to have any kind of data is to do estimates. Even when there is an offer of payment subjects don't always respond. I would like to see what kind of suggestions the republican congressmen would like to propose to solve this problem. Shall we spy on them?

---

Mike Pica

Yikes, give me a break. If people cared to be counted, they could figure a way. If people are suspicious of their government and country, are annoyed by the intrusion and paranoid, then they'll not be counted. And in the end, unless there's a paranoid checkbox, do we really want them counted?

Everyone has an angle it seems. I figure if people want not to be counted then let them not be counted. I mean, per reductio ad absurdum, they have a right NOT to be counted.

I say, give 'em what they're asking for!

---

Tim Evans

I am currently pursuing a master's degree in urban planning at Rutgers, and I previously worked as a mathematical statistician at the Census Bureau for 5 1/2 years. I am thus able to look at the issue from the point of view of both a data user and a data provider.

Planners depend on accurate data to make their decisions, so I support any technique that will improve the accuracy of federal statistics. And having worked at the Census Bureau I can vouch for the spirit of impartiality and professionalism that pervades everything they do. The statisticians at the Census Bureau really are in the best position to say how best to develop an estimate of the population of the US (and make no mistake, even a "complete" enumeration is still an estimate --- It's simply impossible to count everybody). And Census employees are generally quite proud of their non-political status and resent the intrusion of politics into how they conduct their business.

But it's not just Census statisticians who favor sampling. Other bipartisan or nonpartisan professional groups (the American Statistical Association, the National Academy of Science's Committee on National Statistics, etc.) have expressed support for sampling. This support arises not out of any political agenda but out of the fact that they are trained professional statisticians who recognize an opportunity to increase the accuracy of the census while simultaneously reducing the cost. Why do we trust "experts" on so many other issues, but not on how best to conduct the census?

Part of the problem is that most politicians don't understand sampling, or at least claim not to understand it. I find it ironic that some of the same politicians who openly question the accuracy of a census count based partially on sample results can turn around and put so much faith in opinion polls, which are based on samples that are selected much more hastily and with a lot less care than what the Census Bureau is proposing. Also, the "long form" Census data --- e.g., questions about household income, journey to work --- are based on sample data. The Census long form is mailed only to one in six households, and yet no one in the past has questioned the constitutionality of this use of sampling. Why, then, is sampling for nonresponse follow-up suddenly viewed as a departure from scientific techniques?

It's my suspicion that many Republican congresspeople understand perfectly well that the use of sampling would produce a more accurate count, particularly of many traditionally hard-to-count groups (inner-city residents, renters, minorities, etc.) who tend to vote Democratic. They simply don't WANT a more accurate count, for political reasons. But the Census should NOT be a political issue. The decision on how to conduct the census should rest with the experts who are best qualified to make that call, namely the trained professional statisticians at the Census Bureau.

---

James Rose

I can see why some congress-people would want the census to be an actual count. If sampling would underestimate the population by 10% or even 1.8 %, then this could cause an under-representation of the citizens of the U.S..

---

Judy Ladaika

It seems to me, from my limited knowledge and understanding of Statistics, that sampling would provide not only a more accurate measurement, but a less expensive one. There are folks in rural areas and inner cities that would be better accounted for through sampling in the Census 2000.

---

Neil McGuire

Having a degree in engineering, I understand statistical sampling and have no problem with the technical end of the process. However, I firmly believe that a statistical census would only inflate the population giving an advantage to those who want only hand outs and government assistance. They are not tax payers, but tax users. Although they are entitled to benefits of citizenship, we are not required to go to extraordinary lengths to help those who do not want to help themselves.

This is like voter registration. We do not need more registered voters, we need more people who will exercise their responsibility to become informed about the issues and candidates, and vote accordingly. We do not need more people who cannot take the time to fill out a census form.

A democracy requires responsibility.

---

z

I would support sampling, but I have a considerable concern about President Clinton using it to commit fraud. I'm willing to bet that most Republicans oppose it for the same reasons as well. It's not so much that they fear losing their seats because of a more accurate census; they fear losing them due to a fraudulent one.

---

Trigger

Mr Flatow and the guests refer to those persons who were "missed" in the '90 census. In the opening, Mr. Flatow gave the error at 2% of the population. How do we know they were missed, and how do we know how many they are if no one counted them?

---

Schmitt

the problem w/ stat sampling, as i see it, is that the formulae used will ultimately be decided by POLITICIANS not STATISTICIANS.

---

Wayne R. Sulak

The question is one of political power. We must keep in mind that the constitution is a political document, not a scientific one. Right or wrong the constitution uses the term "actual enumeration" If we are a nature of laws, we are required to follow the constitution or amend it. enumerate is defined as "name one by one"

Instead of assuming the public is ignorant or misguided, we should assume the public simply has respect for the constitution.

I understand statistics reasonable well and trust its results. On the other hand we did not change from indirect to direct election of US Senators by simply deciding it more accurately represented popular will. We amended the constitution because it was necessary to do so.

---

Jack Webb

It's no wonder that people don't want to be sampled........ The fish story is enough to turn me against it! It cannot be explained in a clear, precise manner. And if you can't understand the method, then you can't trust the outcome.

---

Matt Wandishin

Regarding the sampling technique, it makes sense that the technique would be very accurate for sampling something like balls in a bag where each ball is equally likely to be picked. But what is done to account for the fact that, to use the fish example, some fish are better at not getting caught in the net? Won't this still lead to an undercounting?

---

Eric Reynolds

We shouldn't use sampling when taking the census. The census should be an exact count with NO margin of error. If the 1990 census was off, then the census takers simply did not do their job. When you start sampling, you open the floodgates of politics and fraud.

---