As a Census data nerd, I’ve been geeking out with the rest of my tribe during each American Community Survey (ACS) Data Users Conference. Each one has been a fun and interesting experience, with insider experiences about the challenges and opportunities in using Census data. I use this inside information for the benefit of my clients, who often want to use Census data to make better policy and operational decisions.
This year was no different. Here are some brief reflections on the things I’ve learned at this year’s conference.
- The data generated by the Bureau of the Census is important. In addition to all the ways non-government users work with ACS data, the ACS has more than 300 known federal government uses and supports many basic decisions made by elected officials and civil servants. In my experience, many cities and other local municipalities are interested in using the data Census provides, relying on it to provide insight beyond what their own internal data is able to provide. For them, Census provides the “denominator” over which their “numerator” of administrative records can be divided. I’ve seen this in my work where Census data provides an understanding of the “unknowns” better than any other data source available.
- The Census cares about engaging stakeholders. 4 years ago, Census created the position of Respondent Advocate to create greater awareness within the Bureau for the interests and concerns of respondents. We were told this wasn’t just coming from Congress, but from within the Bureau as well. I can’t imagine ignoring the concerns of respondents is great for response rates.
- The Census is trying to be responsive. In addition to engaging stakeholders, The Census is also testing various behavioral insight techniques to better engage respondents with language that is clearer and “softer” in it’s asking for compliance. They’re also looking to develop better ways of disseminating the data, moving away from text-based summary files and towards a data portal with the ultimate goal (through their data dissemination road map) of giving users a single search bar to find the data they’re interested in.
- The Bureau of the Census is ultimately a customer-driven organization. The various statistical products they provide each have their own sponsor, with Congress being the sponsor for the Decennial Census and the ACS. While this customer-driven process means that the surveys are responsive to the needs of the data consumers within government, it also means that differing interests among customers leads to differences in the products, impacting (negatively usually) how well they can be used together. Additionally, if the customer isn’t interested in something, it’s not going to be measured.
- Margins of error in the Census need to be respected. Margins of error come from the fact the ACS is a sampling of the population and are plotted at the 90% confidence interval. For particularly small geographies (like tracts or block groups), the margins can be greater than the estimates. Failing to keep these in mind can lead to decisions that aren’t supported in the data. Like most data users that aren’t rigorously trained in statistics, I struggle with how best to use them, which is why it was great to see work done to build tools for estimating data reliability. Unfortunately, these are built for ESRI and haven’t been ported to QGIS (at least not yet).
- Consolidated data files are important (and cost-saving). I’ve been following the great work of the UK Government Digital Services team in building data registers, which are authoritative lists that can be used to build various services across government. For the Census, the need for a Master Address File is clear. In order to survey the population, you need to know where they live. This need is not unique to the Census, and obviously something the US Postal Service and other agencies deal with as well. Unfortunately, each agency seems to do this work on their own (like the US DOT), potentially replicating the work of others even if they use the other products and likely wasting taxpayer money. Having a consolidated list of addresses that can be used by all of government would seem to be advantageous for everyone, including Census. And releasing this to the public free of charge (without the licensing fee USPS charges) would be advantageous to businesses as well. I was reminded the really necessary work of managing data in government is often these basic plumbing and wiring tasks rather than the sexy front-end displays.
- Visualization is important for communicating the insights from ACS data. There were many examples of how ACS data is being disseminated to policymakers, decision makers, citizens, and other stakeholders. The Maricopa Association of Governments is doing some great work in their large area of concern and thinking critically about how to do it, and Fairfax County did a great job showing what they’re doing to democratize open data and the lessons they learned in how to do it that are useful for anyone. I’m already thinking of ways to incorporate these lessons into my data visualization with Excel course for city employees.
- Stay away from quantiles. As Joel Alvarez showed with his research into the ways to show ACS data, there are clearly better and more accurate ways to display data, and quantiles are rarely it. I’ve seen that in my own work that targets a wide audience. Even technically minded people can be misled by a map that uses quantiles, even if it’s important and well-meaning.
And a bonus, the US Patent and Trademark Office building is absolutely gorgeous:
Looking forward to the next conference!