Lessons from North of the Border: Why a Voluntary ACS Could Wipe Some States Off the Map

by Terri Ann LowenthalTerri Ann Lowenthal

What if we took a survey and no one answered? Or, to be more realistic, only two-thirds of us did?

That’s what happened north of the border recently. The Canadian Parliament decided to do away with the nation’s mandatory long-form survey and replace it with the voluntary National Household Survey (NHS). Statistics Canada (StatCan) reported the results of the first NHS, conducted in 2011, this week. Instead of the 94 percent response rate achieved with the 2006 mandatory long form, only 68 percent of households returned the voluntary survey. Instead of having reliable data for 97 percent of the country, only three-quarters of Canada’s localities will have a picture of their socio-economic conditions.

In abolishing the mandatory survey, conservatives decried the burden on Canadians of revealing “personal” information to the government. How ironic, then, that in order to make up for projected falling response rates, StatCan increased the number of households that received the survey, from one in five to one in three. That’s a 65 percent jump!

Now that we’ve recovered from the initial shock of a proposal (H.R. 1638) to axe just about everything the Census Bureau does, legislation to make American Community Survey (ACS) response optional might seem relatively tame, if not harmless. Think again, census stakeholders.

Rep. Ted Poe (R-TX), citing “big government at its worst,” reintroduced a bill (H.R. 1078) to let people ‘just say no’ to all or part of the survey. (See my March 20, 2013, post.) A 2003 field test of a voluntary ACS, which Congress demanded, gave a glimpse of the stiff consequences of such a significant change in methodology. Response rates would plummet, especially for traditionally hard-to-measure population groups, and costs would skyrocket (by at least 30 percent), as the Census Bureau scrambles to ensure enough response to produce accurate data for towns, small counties, rural communities, neighborhoods and smaller population groups such as veterans, people with disabilities and ethnic subgroups. The Canadian experience, the first of its kind to our knowledge, bears this out.

Congress doesn’t seem in the mood to allocate more money for good data; the Census Bureau already is reeling from an 11 percent budget cut this year (13 percent if you count the $18 million dip into the Working Capital Fund). The bureau might have to follow StatCan’s lead and put a warning on all small-area data estimates: Use at your own risk due to high non-response error. Translation: The data are flawed because some population groups are less likely to respond than others and therefore skew the representation of the sample.

More likely, we might not see any data for small areas because the bureau won’t have the money to compensate for plummeting response rates by increasing the sample size (that’s sampling error, folks) like StatCan did. Forty-one percent of U.S. counties are home to less than 20,000 people; even with a mandatory ACS, the Census Bureau must aggregate data over five years to accumulate enough responses to yield statistically valid estimates for these areas.

New York? Most counties are larger, although we’d lose information about communities and neighborhoods within counties, making it difficult for local governments and businesses to target services and investment dollars. But bye-bye to most of Montana, the Dakotas, Nebraska and Kansas, Idaho and Iowa. You can wipe half of Texas, Nevada, Wyoming and Utah, much of Colorado, Missouri, Georgia, West Virginia, Kentucky, Arkansas and Minnesota, and not insignificant portions of other states off the map. No data for 95 percent of American Indian reservations and Alaska Native areas, most elementary school districts, and more than half of secondary school districts. How is anyone supposed to make rational decisions without all of this local information?

Meanwhile, joining the list of conservative voices that appreciates the value of objective, reliable data to support decision-making is The Weekly Standard. A May 20 article calls the ACS “one of the most robust and important tools we have for measuring and understanding American trends.” Ironically, The Weekly Standard admonished the Census Bureau for deciding, because ACS content is now a zero sum game, to drop the question on how many times a person has been married, to make room for questions on use of health care subsidies and premiums that will help policymakers assess the effectiveness of the Affordable Care Act (okay, Obamacare).

Raise your hand if you remember what happened the last time the Census Bureau tried to mess with a census question on marriage? Well, before the 2000 count — when the census long form still ruled the data world — the bureau thought it might streamline the short form that everyone received, by shifting a question on marital status to the sample (or long) form. You would have thought someone proposed abolishing Mother’s Day! Very conservative Sen. Jesse Helms (R-NC), incensed at the inference that marriage was no longer a “sacred institution” — and who had been complaining for years that the census form was too long — proposed an amendment (to the Transportation appropriations bill, 106th Congress) in support of keeping the question on the short form.

So, we have some conservatives railing against the public burden of so many nosy questions, and others urging the government to keep asking how many times you’ve been married. While Sen. Helms and conservative colleagues (e.g. John Ashcroft, Sam Brownback) were fighting to save the marriage question, the same Senate went on record urging Americans to answer only the long form questions they liked in the 2000 census. Yes, I feel a census headache coming on…!!