Counting on you

Lynne Billard is University Professor of Statistics and former president of the American Statistical Association. When she heard in 1996 that some members of Congress were arguing that “even statisticians don’t believe in the use of sampling” for the upcoming 2000 census, she decided it was time to fight back. Her efforts to defend the scientific validity of statistical sampling resulted in a report to Congress and, since then, she has frequently been called on to address the issue. This summer, for example, she was invited to address the National Press Club in Washington.

Columns: Can you begin by defining statistical sampling?
Billard: Sampling for the census is the process of trying to find out about the population in such a way that the information contained in the sample reflects the characteristics of the population accurately.

Columns: Why should it be used in the census?
Billard: In theory, a person might ask why we can’t just count each person, one by one. And, in theory, there’s no problem with that. In practice, because we are dealing with such terribly large numbers and because of the time frame, it’s just impossible to count people one by one. The question, then, is: How do you get an accurate figure? Sampling is one methodology that allows you to obtain a more accurate figure than the straight counting.

Columns: Sampling is not only valid, it’s more accurate than counting?
Billard: What the census is asking is how many people were in a certain house at midnight on April 1. Roughly 65 percent of people will mail back their forms. Believe it or not, there are even some errors there--a two or three percentage-point error.
But not everyone sends the forms back, and then they have to send out interviewers. They wind up asking at a house how many people were there six or seven weeks ago, and some people really don’t have a clue. You also have so-called last-resort counting: they’ll ask a neighbor, and the neighbor will say, “Well, I think there are five people there.”
By August, there’s a huge error rate, some 28 percent in 1990.
But you can use sampling earlier in the game and have more highly qualified interviewers. Of course, there’s always sampling error, but it can be measured and controlled. That’s why sampling becomes so important. Without the sampling, the final total count will be about
2 percent off--or at least it was in 1990. The expectation is it will be much further off this time.
With sampling, we can reduce that significantly. Two percent is four million people. That’s a lot of people.

Columns: How successful have the opponents of sampling been?
Billard: The census is used for many purposes. There are many, many parts of the census that use sampling, and they have no problem with that.
Some members of Congress have said all along they don’t want sampling, but they just don’t want sampling used for redistricting, because they fear that with a complete count they might lose some seats. Republicans think the people who are uncounted--if they were to vote--would vote Democrat.

Columns: What steps did you take as president of the American Statistical Association in 1996?
Billard: Two years ago, the rhetoric from Congress was that “we don’t want sampling because even the statisticians don’t believe in sampling.” I knew that wasn’t true. So I set up a blue-ribbon panel--and it really was the superstars of the sampling world--and asked them to look at this issue. They came out with a report, and there were also a couple of reports from the National Research Council. All of these endorsed sampling, of course, which wasn’t surprising to me as a statistician. I felt it was important that we as a statistical community stand up and say, “Wait, we do believe in sampling.”
All the members of Congress got copies, and we had to do some letter writing, but by about a year ago they had stopped using the argument that statisticians don’t believe in sampling. Now their argument is that the Constitution says it has to be a one-by-one count. So, they’ve shifted it to a political argument. The reasons were always political.

Columns: Is that what you discussed at the National Press Club?
Billard: By setting up the panel, I sort of found myself on the hot seat. Newspapers came to me when things happened. I’ve also now been put on an advisory committee for the Secretary of Commerce for the census. I was asked to go to the National Press Club and talk about how sampling was used in the census. It was really exciting. I thought I’d be nervous, but once you get going it’s like a classroom. They genuinely wanted to know about statistics, and their questions were very good.