Federal Data, Local Impact: What Cities Need to Know

City leaders are on the front lines of data use, but most lack visibility into the federal data landscape, what's available, what's changing, and how federal policy decisions affect local outcomes. This gap delays emergency response, misdirects resources away from high-need neighborhoods, and undermines AI systems that depend on accurate data and community trust.

Host Stephen Goldsmith speaks with Denice Ross,  Director of Federal Data Policy at the Federation of American Scientists, about the relationship between local and federal data, what city CDOs should prioritize, and why cities have untapped power to shape federal data policy.

In this episode, you'll learn:

• The often-hidden relationship between local data needs and federal data infrastructure
• How to identify and access the federal data your city should be using
• Why now is the time to prepare for Census 2030 and protect funding
• How community participation in data decisions prevents disparities and builds legitimacy for AI systems
• How local data leaders can advocate effectively during federal policy windows

Listen here, or wherever you get your podcasts. The following is a transcript of the conversation. 

Stephen Goldsmith:

Thank you and welcome back. This is Stephen Goldsmith from the Bloomberg Center for Cities with another podcast. We have an illustrious guest today, Denice Ross, who was all sorts of things. Director of Enterprise Information in New Orleans, Presidential Innovation Fellow, Deputy US Chief Technology Officer, US Chief Data Scientist, now Director of Federal Data Policy at the Federation of American Scientists, just to name a few. Welcome, Denice.

Denice Ross:

Thank you, Stephen. It's so great to be here.

Stephen Goldsmith:

Did I leave out any one of your 25 previous, illustrious titles?

Denice Ross:

Well, my favorite job was in City Hall.

Stephen Goldsmith:

Well, I'm glad you have your priorities correct.

Denice Ross:

Yeah.

Stephen Goldsmith:

I occasionally meet a governor who was a mayor and I expressed my sorrow for their demotions.

Denice Ross:

Exactly.

Stephen Goldsmith:

You worked in a great city with a great mayor. Let's just talk for a second to get to know you. What's a common theme, a common thread through all those jobs you've had and how do you describe your accumulating expertise in data and government?

Denice Ross:

I'm glad you asked that because I've just recently started to make sense of my career path and realized that when I was in New Orleans, I was working for a nonprofit data intermediary. And I moved to New Orleans in 2001 to publish census data. It was the first time the internet was a thing and the census data were being released and it was this opportunity to democratize the data so it could be used by communities and nonprofits to sort of chart their path toward their own destiny. And we analyzed the data and made it really easy for neighborhood folks to use. And then when Katrina happened, all of those data became instantly historical. And I found myself shifting from becoming a data analyst and a data user to also becoming an advocate for the data that we needed in order to create a more equitable recovery for the city of New Orleans. And what I've been doing since is really about public data policy and making sure that those data supply chains are robust and that they're meeting the needs of the American people.

Stephen Goldsmith:

That's ambitious. Let's deal with that for a second. So let's just start with a framework. Many of us who have spent all our time in local government recognize that most of the data we need is local, but not all of it. So how should a mayor or a chief data officer think about the importance of federal data, both what is possible today and what's missing? How do we think about that, that hierarchy of data?

Denice Ross:

There's a very complimentary relationship between the data work at the local level and that at the federal level. One nice thing that federal data bring to the picture is standards. So a local government doesn't have to invent how it's collecting crime data, for example. And because of those standards and because of the voluntary or mandatory reporting requirements for some federal statistics that local governments have, that enables local governments to compare across time, because you have continuity of data collections, and across space so you can benchmark how your city's doing compared to another city, and that's a huge gift. And in order for the federal government to be able to provide that sort of data stability, there's a tension because then the federal data aren't quite as adaptive when things change rapidly, whether it's with a disaster or a disruption to our economy like AI is causing, for better and for worse.

Stephen Goldsmith:

I have to tell you a story that's 30 years old. You've just kind of provoked me. I'm so old that when I was a district attorney, I was on an FBI local committee to revise Uniform Crime Reporting, right?

Denice Ross:

Yeah.

Stephen Goldsmith:

And I remember being in the room when the top FBI person assigned to it, in response to a question about, I think, guns, said, "We couldn't possibly let local governments and even private individuals use our data because there's too many errors in our data." And it felt to me like the reason there were errors in the data was because people weren't using it. And if you're a FBI agent or local police officer, you just assume that maybe it was right, maybe it was wrong, and you just went ahead and act. So let's talk for a second, now that you've kind of provoked my story, about data quality and data usage. Federal data and even with local data, when do you use it, how do you think about AI's access to it, and when is quality enough?

Denice Ross:

Well, I'm so glad you brought up the Uniform Crime Reporting data because that's a perfect dataset to talk about these issues with and it's the dataset that got me involved in some of the federal statistics when I was a Presidential Innovation Fellow. There are lots of types of federal collections that come from local governments. That includes the crime reporting from law enforcement agencies, public health reporting on disease surveillance, wastewater tracking, and whatnot, also school performance data that comes up from school districts and states. And each of those represents an opportunity both for that consistency across time and space when the data go to the federal government, but also for the data to start to be used immediately by the local community in order to drive better outcomes.

And I remember when I was working on the police data initiative, it was just a couple of years after Ferguson, and I was talking to a police chief and he was describing. He's like, "Well, we've got this data here that we actually use to understand what's happening in our communities and this data over here is what we report to the federal government," and so that was just extra work for him, right? So you definitely do want a reality where the data that's being reported to the federal government is also the same type of data structures that you need to make those local decisions. And we have an example from recent history where local communities were innovating on that crime data, now NIBRS data rather than UCR, and there was a gap. So it turns out that the standard did not require that police departments report non-fatal shootings, which is sort of an oversight. If you're trying to understand gun violence, you don't want to just analyze the gun violence that resulted in deaths, you want to analyze all shootings. So some local jurisdictions started just adding a question in their form about non-fatal shootings.

And then the FBI, through its deliberative standard setting process, just recently in the last six months added the non-fatal shootings category to the national standard and now it's a challenge of propagating that new standard across the nation's 18,000 law enforcement agencies. So the innovation started at the local level, got adopted at the national level, and now propagates across the nation.

Stephen Goldsmith:

That's a good example. It's also an example of a practical change that makes the data more useful. I mean, the difference between the two categories is how good the shooter was.

Denice Ross:

Totally, and how close the hospital was. Yeah.

Stephen Goldsmith:

Yeah, right. Let's stay with this for a second on the comparison side. One way to think about comparisons is housing versus housing. Another is to think about it demographically. What federal data is available and how should we think about disaggregating the data so we can understand conditions relative to the opportunities that individuals have in those neighborhoods? If you take a neighborhood in Indianapolis, you might want to compare it to other neighborhoods in Indianapolis, but you also might want to compare it to similar neighborhoods in Cleveland where the demographics are the same. So you've got both inside the city comparisons and comparisons from city to city.

Denice Ross:

That's where the magic happens, I think, with small area data. The challenge, of course, with small area data is that because of privacy concerns and also just the nature of statistical sampling, it's relatively rare that we get neighborhood level data at the city level and census data is an exception to that. Obviously, the decennial is a full count. And the American Community Survey, you get that neighborhood level geography, but it's a rolling average over five years so it doesn't give you that sort of year by year comparison as cleanly as one might hope.

There is a civic infrastructure that some cities have that can be really helpful in making those comparisons and that is the Urban Institute has something called the National Neighborhood Indicator Partnership. And when I was in New Orleans, we were one of those data intermediaries. So we took publicly available data, organized it into neighborhoods, and made it really useful and relevant for local decision makers. And what was really transformative is then once or twice a year, we'd meet up in one of the cities, I think there's 30 or 40 cities now that have this civic capacity, and we would compare notes with each other and compare methodologies. And so this was really useful after, for example, the housing crisis of 2008. Youngstown and Detroit were struggling with declining populations and lots of blighted properties. And New Orleans, of course, was just three years out from Katrina and we similarly had a blight problem, but for different reasons.

We all honed in on a specific type of dataset that was really useful for that type of benchmarking and it was a little bit surprising. It was the company that sends you junk mail. It turns out that they had the best data on addresses actively receiving mail because there's a business purpose for them to not want to deliver mail to an empty house. And so all of these different cities were using the junk mail data to track repopulation after the population declines. And then there was also a complimentary dataset that an analyst in HUD had produced. He had gotten permission from the postal service to aggregate addresses with the status of "no stat," which meant that the mail was not deliverable to that house. It's a good proxy for blight and he was able to aggregate that, I think, into census block groups. And so between those two datasets, we had a pretty good comparable dataset across time and across place to get a sense for how is the recovery from these housing crises going in these different places and what interventions might be making a difference.

Stephen Goldsmith:

These are great answers really. These are terrific answers. I love the examples.

Denice Ross:

Oh, thank you. I try to be specific.

Stephen Goldsmith:

Where would a local chief data officer or chief of staff or a person who's interested in data access, how would they find out what's available at the federal level that could assist them that's still available at the federal level? Obviously, we talked about crime, but we have some projects involving environment. Some of the data's there, some of the data's not. So how do you figure that out if you're working in a city?

Denice Ross:

When you're working in a city, you're heads down on what your community needs and what your mayor's office is looking to do. And I remember when I was in City Hall, I had very little articulation with the federal data apparatus. It was just something I took for granted, right? We've got the census data, we've got the American Community Survey data, we've got different transportation datasets and whatnot. When I talk to people who use federal data, one common theme is that they can't live without it. For example, the American Community Survey is going to be your city's best way of understanding where are the neighborhoods where there's a high proportion of households without access to a vehicle.

And then if you're making an evacuation plan, you would want to make sure that you had enough buses and you bring those buses to the right neighborhoods so that people can get on the buses and safely evacuate. And there's no replacement for the American Community Survey for that dataset and it wouldn't make sense for a local jurisdiction to try to collect that data on their own. And so that's a dataset that's used broadly by emergency managers in local governments that needs to keep flowing.

There are also lots of datasets that are underutilized that would be useful, like that obscure postal dataset at HUD that we just happened to hear about, but most city GIS departments, for example, probably don't know about that data on the "no stat" addresses that are a proxy for blighted properties. And that's part of the challenge is that we might lose some of these federal data due to diminishing capacity before we even started to use it in our city government operations. And there are some proactive things that you can do around making sure that your city is well represented as we develop our nation's denominator for Census 2030. I can talk about that in a minute.

Stephen Goldsmith:

And there's no easy inventory of what's there that would be useful and what's not, right? You just have to keep probing and looking?

Denice Ross:

There is no easy inventory. There are ways that city data leaders can find data that are useful. One is thinking about using data as sort of like having a recipe. If you're trying to solve the problem of blight or emergency preparedness or workforce development, there are different federal datasets combined with local datasets that are going to be really powerful. And networks like the Civic Analytics Network that I know you are instrumental in setting up, those types of networks can be great for sharing those recipes so that you can meet those pressing use cases. And then there's some software packages that also sort of compile those templates and will do the work once to find the data that's relevant to that issue and then sort of bring it to your doorstep.

Stephen Goldsmith:

Thanks for mentioning the American Community Survey. Every city should use that on a whole host of kind of access areas. How should a city today be thinking about the next census?

Denice Ross:

It's 2026, so we've got four years until the next decennial census, but now is the time to start getting your data ducks in a row. And this is important because census data determine federal funding. It determines whether your jurisdiction is eligible for different types of funding if you're a metropolitan or a rural community. And the project on government oversight recently identified almost 350 federal programs that rely on census data to direct more than $2.1 trillion in federal funds to state and local communities. That was in 2020. So the stakes are high and the basis of the decennial is the addresses and knowing where people are, and that is a local government responsibility. There is a program called Local Update of Census Addresses, affectionately known as LUCA. And now's the time. Ask your GIS person to tap into the LUCA process at the Census Bureau and start making sure that all of your addresses in your city are fully accounted for, including your group housing.

Stephen Goldsmith:

Great advice. You've spoken a little bit on the American Data Index and I saw you reference that as a weather forecast for government datasets. What does weather forecast for America datasets mean?

Denice Ross:

Well, you would hope that you could just take the data that's being produced by the federal government, assume it'll keep flowing, assume it will be high quality and fully supported, and go about doing your business of running your city. But right now, what we're seeing is an unprecedented degradation of federal data capacity and it takes a few different forms. One is just data elements or entire datasets that are not aligned with the administration's priorities, like on DEI or gender or climate, and the other is really just a reduction in overall government capacity with staffing losses, contracts being cut, the loss of scientific advisory committees that used to help keep federal datasets sort of up with the times, and then an increase in red tape where data collection processes that used to be routine now have to go through multiple levels of political review before the agency can move forward.

So because of all of that, what we're starting to see, and I think we'll see more of in this year, is some datasets be terminated completely. For example, the food security supplement, which is the best way that we understand childhood hunger in America and food insecurity at the state level. There's also been datasets where gender identity, for example, has been taken out, especially datasets where it's particularly relevant like the 988 crisis hotline or the National Crime Victimization Survey had some of those fields taken out as well. There have been some explicit losses.

And also, I think what we're going to see more of is datasets that used to be published in a certain month might end up being late because there just aren't the staff and the people in place to produce those in a timely way. We might see less detail, sort of a thinner report coming out. We definitely are seeing reductions in access. So it's more work now to get access to the data because those front end tools that made it easier to access the data in many cases are not being supported anymore. So it requires more technical expertise to derive value from federal data than it used to. And unfortunately, that shifts the burden to local governments.

Stephen Goldsmith:

So what is essential data that you work on?

Denice Ross:

One of my big regrets from my time in the White House most recently is that we didn't do a great job of telling the story about why federal data are so important for running a modern society and how federal datasets benefit everyday Americans. The example I gave you about everyone in a neighborhood has a seat on the evacuation bus because their city used American Community Survey data to plan ahead, that's an example of everyday American who's benefiting from federal data. So what we're doing is really telling the story about how federal data touches every corner of American lives and livelihoods.

One example that I like to tell is a few years ago, I got an email from the online grocer that I get my groceries from and they told me that the peanut butter in my pantry might have salmonella. And I was sort of creeped out by how did they know that my peanut butter might have salmonella? And a few years later, I realized that it was because of the Consumer Product Safety Commission's dataset on product recalls that's an open API, and then that online grocer like any company can just tap into it. They know I bought the peanut butter and now the federal government is telling them that it might have salmonella and I know to throw the peanut butter away. We're telling the story about, through all of these datasets, why it matters, and there's a particularly compelling angle I think for local data leaders to play a role in that.

And what we're doing at dataindex.us is we're monitoring for opportunities for public input. If a local CDO, for example, speaks up about how they need the food security supplement or the American Community Survey or some DOT dataset, these policy windows mean that there will be an outsized impact by them giving their feedback at that time because the agency is open for public input. And local data leaders have a real outsized role to play because they're closest to the citizenry, right? And if they don't weigh in, then what happens is federal agencies are career colleagues on the inside. They assume that the changes they're making to a dataset are fine or they might think that nobody cares about the dataset.

Stephen Goldsmith:

I'll remember this as the peanut butter story.

Denice Ross:

I have a PS to that. A colleague of mine bought his peanut butter in the store and so he didn't have that chain of custody digitally, and he actually got the salmonella poisoning.

Stephen Goldsmith:

Well, it certainly approves essentiality. I'll say that.

Denice Ross:

Yeah.

Stephen Goldsmith:

So how do you think about the quality of data that underlies the use of AI? It's almost back to my crime story, right? Now with AI, more people can use the data so therefore that's good and that's bad. So if you were, today, chief data officer in New Orleans or fill in the blank, how would you go about accessibility of the data so that it can be used for good purposes?

Denice Ross:

I would prioritize meaningful public participation in any AI tools that you're building on your data. And also, be really thoughtful about fitness for purpose because often data might be collected for one purpose. For example, SNAP data. States collect social security numbers just as a rough sort of identification measure, but it's never rigorously used. But if those social security numbers then go up to the federal level and are used for fraud detection, the quality of those socials isn't sufficient for that high stakes use case at the federal level where people might get kicked off of benefits. For every dataset that we're thinking about putting into an AI model, we need to think about what is the use of that model and what are the stakes and is that data high enough quality for that purpose and those stakes. And that's something that public participation, like public input, can also help really sharpen your thinking about the match between data quality and data purpose and the AI applications.

Stephen Goldsmith:

Well, that was a great example. Let me repeat it back to you with my words ...

Denice Ross:

Okay.

Stephen Goldsmith:

Just to see if I have it right....so as the significance of the use escalates, the quality of the data needs to escalate accordingly. Is that a fair summary?

Denice Ross:

Yes. Often in city government we produce data that's good enough. An example is permitting data. I know after Katrina, we took out a couple of permits. And I must have had a few false starts because there must have been three or four records from my address from me trying to get permits for one thing or another. And I could always find, ultimately, the right permit that I needed to continue acting on, but there was this debris that I had created in the process. And that was fine because it was good enough for me to find the actual permit and it was good enough if I went into the Office of Safety and Permits for the city employee to find my address and the right permit. But if I'm an analyst and I'm trying to understand permit activity in a neighborhood and I'm going to be making resource allocation decisions based on that, if three of the four permits attached to my address were false starts, that data's not high enough quality to inform that analysis.

Stephen Goldsmith:

You know too much, really. This is a very difficult interview because you know the exceptions to every rule. This is very difficult to be simplistic with you. Looking forward, it feels to me like we're at a point in time where thanks to data and generative AI, we could see cities operate at a much higher level of performance, understanding causation, anticipating problems, identifying needs. If you were speaking to others that work in the local area, maybe aspire to do the things you've done in life, what would be the one or two pieces of advice you would give them about how to make their cities better insofar as the use of data to improve the quality of life?

Denice Ross:

I would say engaging the public at every step of the way is going to be the most impactful thing you can do. And I'll give a cautionary tale of something that happened in Flint, Michigan after the water crisis there. A group of technologists from Georgia came in and they had a machine learning model that could predict where the lead pipes were most likely to be. Flint is an old city. There wasn't really any reliable data on where the lead pipes were. And with this model, when the city was digging up pipes to see if they were lead, they had a 70% hit rate. They were more efficient than they would have been if they were just randomly digging up pipes. But because these technologists had not engaged the public, the public wasn't really bought into how important this model was for prioritizing getting clean water to the areas that needed it the most.

You can imagine a city council member looking at the map of where the digs were happening and saying, "Why aren't you digging in my neighborhood?" There was a contractor change and the decision was made to dig equally across neighborhoods rather than predictively using that AI model. The hit rate went down to 15%. So the wealthier, Whiter neighborhoods were being reassured that their water was clean while the lower-income, Black neighborhoods were continuing to be poisoned by lead. And I think if there had been a participatory process where the neighborhood residents were involved in understanding the data collection, testing the water quality before and after, I think it would have been politically untenable to change the method of prioritizing the digs.

Stephen Goldsmith:

Yeah, it's a very troubling example for many reasons. We've been spending some time. We have a Robert Wood Johnson Foundation grant to look at hyperlocal environmental data and its effect on public health in various areas. And as federal subsidies for lead pipe remediation have gone down, the intensity of local remediation has gone down. But if you looked at it the way you just said it, you would still be engaged in your most high risk neighborhoods, right?

Denice Ross:

Yeah.

Stephen Goldsmith:

The homes where kids are getting poisoned every day. Better data equals better resource allocation equals higher quality of life for residents of your city.

Denice Ross:

I agree.

Stephen Goldsmith:

So whether it's lead paint or abandoned houses or other similar local issues, how do you think about the community not just as the user of the data, but as a producer of important data to augment what the federal, state, and local governments have?

Denice Ross:

I think community science is one of the best ways to establish legitimacy of data. It's tricky though. After Hurricane Katrina, with the federal data being instantly historical because 80% of the city flooded, neighborhoods collected data about housing condition on their own and that created its own data disparity on top of the preexisting demographic and economic disparities that the neighborhoods suffered from. So for example, a neighborhood that was highly resourced could get a copy of Salesforce and walk the streets with a notepad and enter it into the database and had very good visibility on the status of different properties and then could advocate with City Hall around remediation, whereas a lower resource neighborhood did not have the capacity to walk the streets and collect that or purchase the software tools for it.

So what we saw was this data disparity that was on top of the preexisting disparities, and that's just something to be cautious of. Same thing happens with 311 data, right? We've heard stories. For example, I just was talking to a Fairfax County representative and he was saying that in the Hispanic neighborhoods in Fairfax County, that their calls have gone down. And that's not because they don't have problems. They just didn't feel comfortable calling the government because of the current climate.

Stephen Goldsmith:

Let's just close with, I know you've got multiple things you're doing now, but just tell us a little bit about what you're doing and what you're trying to accomplish.

Denice Ross:

I am collaborating with a group of federal data watchers to monitor what's happening with the federal data ecosystem, identify opportunities for public input, and tell the story about why it matters. We would love to collaborate with any of our colleagues in local government. If you've got a favorite dataset, please let us know. Let us tell the story about it. We can monitor it. We can let you know. If you sign up for our newsletter at dataindex.us, we'll let you know when opportunities for public input do arise because it is literally written into the law, into the Evidence Act that the chief data officers and federal agencies need to engage with state and local data leaders around the value of federal data and how it can be improved. So they're required to get your input on how to do data better. And admittedly, those feedback loops are a little rusty right now, but we're trying to make them more effective so that folks who need the data can advocate for its continued investment.

Stephen Goldsmith:

Well, thank you. This is Steve Goldsmith from the Bloomberg Center for Cities. I'm with Denice Ross, who is one of the nation's preeminent data experts and apparently knows more obscure anecdotes than almost anybody I've met. So thank you for your contributions in multiple ways.

Denice Ross:

Thank you, Stephen.