Litigation over adding a citizenship question gave voice to data expert John Abowd.
The Trump administration may see him as a stalwart of the deep state, but to his colleagues at the Census Bureau, John Abowd is the agency’s chief scientist, associate director for research and methodology, and a senior executive.
Over the past six months, Abowd has been put through the ringer as perhaps the central witness in four courtroom hearings during the litigation over Commerce Secretary Wilbur Ross’s unorthodox, late-in-the-game plan to add a citizenship question to the rapidly approaching 2020 decennial census.
The Supreme Court on Tuesday is set to hear oral arguments on the rulings by district judges who found that Ross’s plan—which he presented as a way to help the Justice Department enforce voting rights—is “arbitrary and capricious,” unconstitutional or violates the Administrative Procedure Act.
Abowd, as has been reported, weighed in against the plan, saying the question had not been fully tested, would be costly and risked worsening an already expected undercount of immigrants by scaring off respondents.
The bureau’s press office, when asked for more details on Abowd’s views, referred Government Executive to the Commerce Department, whose spokesman declined to comment. Census did supply a link to a key August 2018 research paper Abowd supervised, which outlined the technical challenges of adding a citizenship question. It contains a disclaimer noting that it is among papers that “have not undergone the review accorded Census Bureau publications and no endorsement should be inferred. Any opinions and conclusions expressed herein are those of the author(s) and do not necessarily represent the views of the U.S. Census Bureau.”
Abowd declined to discuss his views on the citizenship question with a reporter, but Government Executive reviewed a transcript of nearly 700 pages of Abowd’s exchanges with attorneys at the U.S. District Court for the Southern District of New York over several days in November. The case, which consolidated complaints from 15 states and immigration-oriented nonprofits that sued the Commerce Department, was eventually decided against Ross by Judge Jesse Furman. The judge ruled that Ross’s March 2018 decision to add a citizenship question, though within his authority as Commerce secretary, was likely made in response to political pressure and not to support the Justice Department as he claimed.
The exchanges reveal Abowd as a cautious and technically sophisticated academic who, while taking a clear stand against adding a citizenship question, was tactful in criticizing fellow scholars. And he was respectful of the limits on civil servants when they’re up against political appointees higher in the chain of command.
His testimony under oath also offers a portrait of a bureaucrat stretched to the point of exhaustion.
A Single Meeting
Both the lead attorney for the plaintiffs, Dale Ho of the A.C.L.U. Foundation, and defense attorney Stephen Ehrlich of the Justice Department, drew Abowd out on his long resume. Having advised the Census Bureau since 1998, Abowd signed on as chief scientist in July 2016 after years at Cornell University working as its principal investigator for the National Science Foundation-Census Research Network. He is a fellow and past president of the Society of Labor Economists, a fellow of the American Statistical Association and the Econometric Society, and an elected member of the International Statistical Institute. Perhaps most important, the attorneys established, Abowd sits on the 2020 decennial census steering committee and operating committee, the highest decision making level.
His economic and statistical background, however, has included the actual design of only one survey, as Justice’s Ehrlich confirmed.
Abowd was enlisted by plaintiff’s attorney Ho to show that Census career staff were blindsided by Ross’s plan for a citizenship question, which was pursued by doing an end-run around norms for advance notice, field testing and pre-decision consideration by advisory committees.
“The consistent recommendation from the leadership of the Census Bureau has been not to include a citizenship question on the 2020 census, right?” Ho asked his star witness, before proceeding to unpack the timeline and reasons. “You didn’t learn that it was the Department of Commerce officials who had requested that a citizenship question be added to census, rather than the other way around, until after this litigation was initiated, right?”
“Correct,” Abowd said. “Everyone I know at the Census Bureau, including all the senior executives, were surprised by the portion of the administrative record that predates Dec. 12, 2017,” which, that reconstructed record shows, was the date that Abowd was told by acting Census director Ron Jarmin of the initiative that had been under consideration since early in 2017. Asked whether talk of adding a citizenship question was “in the political stratosphere,” in the media, Congress and among political people, Abowd replied. “It was not in the air at the Census Bureau, so to speak. Not the air I was breathing.”
Abowd was assigned to assemble a “SWAT team” of experts who worked over the Christmas holiday to draft a memo for the Commerce secretary to respond to Justice. Its basic conclusions in January 2018, according to attorney Ho: Adding a citizenship question would cause a difference in response rates between households with non-citizens and other households of 5.8 percentage points. With an estimated 126 million U.S. households, 9.8 percent of which contain a non-citizen, the total reduction in responses would be 9.8 percent, or 630,000 households, more than a million people, and a “non-response rate” more than twice as high for Hispanics.
The memo also cited focus groups on barriers for respondents in completing a census questionnaire. They showed that the current political climate made responding particularly difficult for Hispanics, who fear their information will be shared with immigration enforcement officials. The Justice Department’s approach to using administrative records to verify responses to the citizenship question may leave data gaps, the memo said.
On Feb. 12, 2018, Abowd had his sole meeting with Secretary Ross. “Let’s be clear,” Ho said, “Secretary Ross had only one meeting with the chief scientist of the Census Bureau about the citizenship question before he issued his decision memo?”
A Series of Omitted Steps
In addition, Abowd confirmed, because the Commerce Department leadership had missed the July 1, 2016, deadline for giving input to the bureau for the 2020 count, it was already too late to run the usual randomized control trial with the citizenship question. Adding the question would involve extra costs of what is called non-response follow-up efforts. When attorney Ho said the total added cost could be $82 million, the defense objected, but Abowd backed the number.
“If you had been given more advance notice, then the Census Bureau could have consulted with, for example, the Census Scientific Advisory Committee before a decision was made,” Ho asked. Abowd said there was no such discussion.
On March 26, 2018, Ross issued his memo announcing the unusual move that Census would add the citizenship question—versions of which appeared on the short-form questionnaires as recently as 1950—with the added back-up of using federal administrative records to document accuracy of individual responses.
In the aftermath, Commerce officials, led by Undersecretary Karen Dunn Kelley decided not to conduct a randomized control trial of the citizenship question despite there being sufficient time in the schedule for one, Abowd testified. What’s more, Ho elicited, a set of answers to questions raised by Commerce and Justice officials being prepared under Abowd’s supervision was altered in material ways after it left Abowd’s control, affecting the description of how changes to the census questionnaire are typically accomplished.
“You were under the impression that all of that work that you had done analyzing the effect of a citizenship question, that it mattered as far as the secretary’s decision-making process, right?” Ho asked Abowd.
“I was under the impression that it mattered in the conduct of the 2020 census, yes.”
Though Ehrlich objected and won a rephrasing, the plaintiff’s attorney asked Abowd about the propriety of a later decision by then-Attorney General Jeff Sessions to block Justice staff from meeting with Census career staff on the citizenship question.
On the impact of Ross’s overall decision on his work, Abowd said, “Many experts, including myself, would interpret that as political interference [but] I don’t personally have an opinion on that, because I viewed it as the secretary’s authority to make that determination . . . If he had made up his mind, which I don’t have any direct evidence of, I would have preferred being told in December, because all those resources could have been deployed more efficiently.”
The Administration Strikes Back
Cross-examining Abowd, Justice attorney Ehrlich acknowledged to the judge that his goal was to “demonstrate uncertainty” in the research, based, for example, on the admitted “need to impute” results in the big count that are imprecise. He established that the wording of the citizenship question—taken from a previous American Community Survey (a smaller, more frequent survey)—was selected by acting deputy Census director Enrique Lamas and communicated to Ross.
He got Abowd to agree that the citizenship question performed adequately on the ACS such that it “satisfies” the exception in Office of Management and Budget guidance that allows some questions that have “performed adequately in another survey” to be used without pre-testing. “Yes, 41 million households already have been asked that question,” Abowd said. Ehrlich noted that data proving that an undercount was likely were lacking, as Ross himself had noted.
Abowd also acknowledged that the proposal for a citizenship question had gone out for public comment in 2018 (attracting 147,000 comments). He also confirmed that two untested questions had been added in the past with OMB’s approval, one in 1990 on whether the respondent is an Asian or Pacific Islander, and one in 1970 as to whether the respondent is of Hispanic origin.
Ehrilich noted that the bureau is already planning on a lowered 61% rate of self-response (completed questionnaires that don’t require the added expense of sending an in-person enumerator to follow up), versus 64% in 2010 and 70% in 2000. “Self-response in general has been declining,” Abowd acknowledged.
But Abowd stressed that all relevant numbers are needed to forecast a net undercount. “You can’t change things in a vacuum . . . you have to account for all the co-movements.”
The government’s defense succeeded in eliciting a positive description of Secretary Ross’s grasp of the issues. During their one meeting in February 2018, Abowd said, Ross “asked a lot of questions, and those questions indicated to me that, whether he had read the memo or not, I can’t say, but he had a very thorough understanding of the arguments that were made in that memo. He asked about sampling and definitions of how weights are constructed. He asked some very sophisticated question about non-ignorable data.”
Ehrlich asked whether his team met with Justice staff on their own after the secretary’s decision. “We’ve had a meeting with career civil servants and technical specialists in the voting rights division” to talk about measures of blocks of citizens of voting age by race and ethnicity, Abowd said. But he later clarified that “it had absolutely nothing to do with the citizenship question. We were planning to do it all along.”
Judge Furman asked the defense attorney to speed up. “Don’t get me wrong, I am enjoying Dr. Abowd’s seminar on all this stuff,” he said. Ehrlich replied that such detail was needed “to show what the analysis could control for and what it couldn’t control for, and what factors this analysis gives you a good idea of and a bad idea of how much of that could possibly be related to the citizenship question.”
Toward the end of his three-day testimony, plaintiffs’ attorney Ho asked him, “Is it fair to say you would rather be working on implementing the 2020 census right now instead of attending a two-week trial over this issue?’
“Yes,” Abowd replied.
“You don’t get to make the decision yourself, right?”