Research News

The Costs of Imperfect Census Data

IPR statisticians advise on balancing costs, accuracy


The 2010 U.S. Census cost between $10–$15 billion. Is it possible to trim the costs for the next one in 2020?

Each decade, the U.S. Census Bureau carries out its constitutional duty to count “the whole number of persons in each State.” In doing so, it produces a snapshot of the U.S. population used for vital electoral and funding allocations, albeit one that comes with a significant price tag. With 2010 costs estimated at  $10–$15 billion, is it feasible to trim costs for the next one in 2020?

Cost-cutting is always possible, said IPR statistician Bruce D. Spencer, “But the question is how much accuracy do you get for that money, and is the accuracy sufficient for the uses?”

Bruce Spencer

Spencer, who has conducted census cost-benefit analyses since the late 1970s, underscores the critical role of census accuracy: The information collected guides how billions of dollars in federal funds are disbursed in programs from Medicaid to Pell Grants. Plus, the census fixes each state’s official population, which serves as the basis for allocating each state’s seats in the U.S. House of Representatives. Since each state receives one additional Electoral College vote for each of its congressional seats, “errors in the census could conceivably affect the outcome of a presidential election,” Spencer noted. Importantly, census data also serve to calibrate the results of all national sample surveys.

Zachary Seeskin

In a recent IPR working paper, Spencer and IPR graduate research assistant Zachary Seeskin modeled census accuracy profiles, using four different error distributions. A 4 percent average error in state population estimates, they find, could result in 9–14 of the 435 seats in the House going to the wrong states—and $60–$80 billion of federal grants potentially being misallocated. 

In addition to their U.S.-based work, Spencer and Seeskin also conducted a cost-benefit analysis in South Africa, described in another IPR working paper.

Like the United States, South Africa is required by its constitution to conduct a decennial census. Given that most previous South African censuses were legislated and designed under apartheid, a more recent, post-apartheid Statistics Act from 1999 adds key safeguards to protect the rights of all South Africans, including a required census every ten years and an optional mid-decade census.

By modeling two alternative sets of midyear population estimates, Spencer and Seeskin helped researchers at Statistics South Africa advise the prime minister that they could forgo the mid-decade 2016 census, saving 3 billion South African Rand (approximately $208 million), by employing less costly methodological improvements, for instance, expansion of a population survey (to a sample size of 1 million persons), better collection of official birth and death statistics, and doubling the size of the research group that produces annual population estimates for non-census years.

The two research projects are related, Seeskin noted, because they both study the consequences of census data quality for important uses of the census. These studies allow statistical agencies “to guide decision-making for the census by quantifying the benefits from more accurate census data,” better equipping agencies to evaluate their choices and costs.

Bruce D. Spencer is professor of statistics and an IPR fellow. Zachary Seeskin is a doctoral candidate in statistics, IPR graduate research assistant and a U.S. Census Bureau Dissertation Fellow. For more information, read their related IPR working papers.

Photo Credit: frankieleon (Flickr)