Terry Gerton Let’s start this conversation at a fairly high level. The Department of Commerce issued an order last month dealing with how federal statistical agencies protect confidential information. What’s this order all about and what prompted it?

Paul Schroeder So this order essentially says that to protect confidential information, the Department of Commerce, which includes Census and the Bureau of Economic Analysis, is limited to either coarsening or suppressing data. Suppressing data means it’s simply not released at certain levels, and coarsing means that it’s released at very high levels of analysis. So think states instead of county level data. We don’t know what the impetus for this order was. There has been a lot of controversy around what they call noise infusion, and particularly differential privacy, which we’ll probably get into a little later. But this is the order that came out, and what our statement did was have a real problem with the way it was developed and the way that it just came out without any public input.

Terry Gerton And how does this relate to the president’s executive order on restoring the gold standard of science? There are certain statistical standards here. Is this compliant?

Paul Schroeder I would say it’s not, largely because science should be a transparent and reproducible enterprise. And issuing orders like this without really cause or justification really threatens that trust and integrity of federal science. And that’s something that we’re very concerned about.

Terry Gerton And people rely on census and BEA data probably without realizing it. Just give us a couple of examples of broader measures or functionality that might be impacted by this kind of rule.

Paul Schroeder Sure. So there’s a lot of economic data sets that are affected. The Census Bureau itself has the economic directorate that collects a lot of economic data that a lot business owners and folks use. So for example, if you want to open up a new restaurant, say in Montgomery County, you’re not really going to care about the average disposable income of those say in Western Maryland or even the average for the state of Maryland. You want to know about your county and your neighborhood. You really want to make sure that you don’t care about what, let’s say, John Smith is spending on going out to eat, but you care about what folks in the community are doing. And that’s the problem with this order. We don’t know how it affects data sets that will give that information to folks that’s collected by the federal government.

Terry Gerton And you said you were concerned with the process by which this rule was developed and publicized. Tell me more.

Paul Schroeder Yeah, so the rule just came down. There was no notice about it. No one knew about it, including those within the agencies that it affects. I won’t say no one, but a lot of people were caught off guard in those agencies. And certainly those of us who use public data on a regular basis were caught of guard. Typically these sorts of rules that are going to affect a lot of data and a lot of research are put through the federal register. And there’s an open comment period for people to put input. There’s a discussion about it and there’s more explanation about why this rule came about, and that was all lacking in this case.

Terry Gerton Well, certainly, privacy protection and data collection is a hot topic. We hear a lot of concerns about identity theft and exposure of private data. So when you’re thinking about that, supporters of stronger privacy protections would say the government has an obligation to keep that data private. Does this rule take advantage of advances in computing and artificial intelligence that would expand privacy protections, or does it not?

Paul Schroeder No, it does not. In fact, it dials back the clock. So there’s been a lot of technology that has come out in the past 20, even 30, years that produces things like noise infusion, where we infuse a certain amount of noise into the data, which doesn’t affect the aggregate statistic or estimate very much, and also protects the privacy of those individuals who are within the data set. The issue with privacy is that people can use linking mechanisms to outside data sets, which could then be traced back to identify a certain person and their answers in the data set. That’s a real problem and that’s where the noise infusion comes in. So we’re not denigrating the accuracy of the data itself, but we are protecting the privacy of those who contribute to the data sets. So what this does is move everything back to just suppression or coarsening, which are privacy protections that have been around for literally decades.

Terry Gerton Paul Schroeder is executive director of the Council of Professional Associations on Federal Statistics. Mr. Schroeder, weren’t there measures already in place for agencies to balance the need to protect privacy with this need to publish useful information?

Paul Schroeder Yes, there’s very well documented methods and great methods that have been used for over 10 years now and were used in the 2020 census. So, differential privacy is a mathematical mechanism which is used to infuse noise where you understand how much noise you’re putting in and also what that does to the accuracy of the data. So you can mathematically calculate and find that right balance between privacy protections and having accurate data. Again, this depends on what level of data you’re looking at. But it’s a very robust and often used method of protecting confidentiality for the respondents. And it’s being used in a lot of data sets that are being collected by the census now. And the issue with this order is that we don’t know what happens to those data sets now that it’s not allowed.

Terry Gerton And your coalition also argues that this order could lead either to less privacy, less useful data or both. So walk me through that. What is the practical outcome you’re concerned about and how might an ordinary citizen Notice the difference?

Paul Schroeder The privacy protections are there in federal law. It could lead to a little less privacy. The big issue I’m worried about is having less data available. And when I say less data, I mean, less usable data. So we’ll still have data at the national level. We’ll even have it at the state level. But local communities, rural communities, these are the geographies that we’re concerned about because those are the ones that are most likely to be identifiable because there’s less people in those communities, obviously. So it’s easier to find out a single person who, say works on a farm and grows cattle versus someone who lives in the city and works in an office. So that’s the problem with this order is that certain communities will be more vulnerable to privacy breaches, but my big concern is that data is something that will be suppressed and not released because of this, and that’s a real concern for the American public since they’re paying for this data, and these data are relied on by businesses and researchers around the country to make real decisions and policy applications.

Terry Gerton A lot of your concern seems to stem from the uncertainty about implementation here. What do you know about which data products might be affected and what are the biggest unanswered questions?

Paul Schroeder So that’s the big unknown. We have had reports of certain researchers being having some of their publications pulled because they use data from the census that had used differential privacy. So for one example, for a data set is the Longitudinal Employer Household Dynamics data set. This uses differential privacy, this provides information about community relationships between home and work. And this goes down to the census block level. So this could be something that is at risk. We haven’t seen it yet, but our concern is that data sets that are using differential privacy now simply will not be updated. And we won’t see the effects of this rule in terms of data quality or data recency for a good year or so.

Terry Gerton Do you expect this approach to data privacy to spread to other critical data sets, or do you think it will be limited here to Census and BEA?

Paul Schroeder It’s a great question. What isn’t known widely is that Census actually does a lot of data collection for other agencies as well. We’re not sure if this applies to those datasets or not, because a lot them are actually done jointly with the Census Bureau. And so that’s a real concern as well, but oftentimes the way Census goes, other statistical agencies follow that lead. We haven’t seen any indications of that yet, but we’ve seen it on other guidelines and various data privacy rules. So, it may be coming. I hesitate to speculate on it, but it’s certainly possible.

Terry Gerton And what does your group advocate for action at this point? The rule is out, it’s being implemented. What do you hope happens next?

Paul Schroeder Well, we really hope people reach out to their representatives and make sure that they are aware of this change and really make representatives aware of how this is going to affect their district. Again, these data contain within it a lot of information that has policy implications, economic development, transportation infrastructure, the health of a community. These are all very important things that folks at a district level would be very concerned about. And that’s our hope is that people reach out to the representative and also to OMB itself to hope and see whether this change can be rescinded.

Terry Gerton And for folks like you who actively use this data, what measures might you have for adapting?

Paul Schroeder Unfortunately, the immediate recourse is to use older data, which is never ideal. And so that’ll work for a year or so, a year or two, but, you know, switching over to other data sets that are collected outside the federal government just isn’t an option. No one has this amount of data and this amount of granularity available to researchers, and that’s the real loss here.