Working With Census Data Today

by Kendrick Leong, Hawai‘i Data Collaborative

Many of us look to “Census data” to better understand the circumstances of households and communities in Hawai‘i. In my work to provide research and analysis support to local organizations, I often help teams reference several different Census products depending upon their specific need. Following the onset of the pandemic, I’ve also helped organizations understand and navigate the limitations of Census data for understanding circumstances today. In this post, I distinguish some of the more popular Census data products, run through what you need to know to work with them, and lay out some important limitations of using Census data today.

A Variety of Census Products

The Decennial Census is what most people are familiar with. This effort seeks to count every person in the United States every ten years. The first wave of results for the Decennial Census were released in 2021, and are mainly used for U.S. House Reapportionment and Redistricting, but a detailed release with the demographic information we’re used to is scheduled for later in 2022. I use this count when we need to accurately reference housing and population totals, mostly at the state level.

The American Community Survey (ACS), conducted annually, asks a sample of the population, rather than every person, a longer list of questions about more detailed characteristics of their household. This list includes housing questions such as tenure and housing expenses, economic questions such as employment and travel time to work, and race and ethnicity questions, among others. When we need to understand the detailed makeup of our population in any given year, especially at the community level, I use the ACS. However, because the ACS only asks a sample of the population, the Census Bureau works hard to make sure that the sample in any given year is as representative as possible of the total population as a whole.

The way Census does this is by releasing ACS data as two products: the 1-Year Estimates and the 5-Year Estimates. The 1-Year Estimates are limited to one year of data collection, released the year after collection ends, while the 5-Year Estimates combine five years of data collection to produce more representative and granular results. With more data collected, the 5-Year Estimates can offer more precise counts, but at the expense of encompassing a longer time period. I use the 1-Year Estimates to get a general idea of what conditions are like in one specific year and the 5-Year Estimates to compare changes over time at smaller geographic scales.

Finally, the yearly Population and Housing Estimates seek to estimate the population of the United States between decennial censuses. I sometimes use the Population and Housing Estimates when they’re released yearly in December to monitor population change. This product is the link between the less-accurate but detailed sampling of the ACS and the enormous effort for a full and accurate count undertaken by the Decennial Census.

What happened in 2020?

For us regular consumers of Census Bureau products, 2020 was an unusual year in more ways than one. First, data had to be collected for all three of the products mentioned above. Second, the COVID-19 pandemic affected the ways that the Census Bureau could contact respondents. Finally, the severe economic shock of COVID-19 lockdowns made a timely understanding of our communities more urgent, important, and harder to attain.

The American Community Survey in the time of COVID-19

We have previously written about the implications of lower response rates during COVID-19 on the quality of the Decennial Census, so I’ll focus here on how the ACS 1-Year and 5-Year Estimates were impacted.

In 2020, the Census Bureau experienced difficulties in reaching out to households for responses to the ACS. For example, COVID-related service disruptions meant survey invitation mailers had to be canceled or postponed, and follow-ups with households, normally done in person, had to instead be conducted over the phone in some areas. This led to lower response rates, which the Census Bureau did their best to account for by weighting responses differently.

Because of the various new methods the Census Bureau employed in 2020 to uphold their commitment to quality and accuracy, the 2020 ACS 1-Year Estimates were released in November 2021 as an “Experimental” data release. What this means is that data collected in 2020 were prepared in a way that differed enough from previous years that we should not make comparisons between previous years’ ACS 1-Year Estimates and the 2020 ACS 1-Year Experimental release - essentially the Census Bureau told us not to use it.

Referencing 5-Year Estimates?

But of course, noticing the many ways the world has changed since 2020, we still are looking for some way to assess that year’s impacts in our communities. The ACS 5-year Estimates, released in March of 2022, were not labeled as “experimental,” meaning the Census Bureau has confidence in the estimates provided. Although the 5-Year Estimates include data collected in 2020, they also include data collected from previous four years (2016-2020), While valid, the current 5-Year Estimates offer little insight into circumstances immediately following the pandemic because the 2020 measurement is diluted by the prior (pre-pandemic) four years.

The bottom line is that we currently do not have Census data that helps us understand the circumstances of our households and communities in the aftermath of the pandemic. The 2020 ACS 1-Year Experimental release cannot be compared to other releases, while the ACS 5-Year Estimates will continue to contain pre-pandemic data, making it hard to observe noticeable differences until 2025.

As such, I am planning to wait for the 2021 ACS 1-Year Estimates (to be released December 2022) , which were not subject to as many of the disruptions that occurred during the early days of the pandemic, as the first comparable Census estimate of pandemic-related societal effects. Some things I’m looking forward to examining in the 2021 ACS 1-Year Estimates are changes in housing affordability, possible inequities in economic outcomes, and changes in benefits recipiency (eg. SNAP and TANF).

Unfortunately, for the far more useful 5-Year Estimates, this waiting will continue until 2025, when the 2020-2024 ACS 5-Year Estimates will be released. This will be the first time we will be able to make detailed comparisons between pre-pandemic (2015-2019 ACS 5-Year Estimates) and post-pandemic (2020-2024 ACS 5-Year Estimates). Basically, we will have to let three more years pass before we are able to make reliable, community-level longitudinal estimates.

Our Reliance on Traditional Datasets

Amidst all the things we need to understand about our communities today, I often have to include the disclaimer that the most reliable and representative datasets we have are informed by data collected and prepared in prior years. This is problematic when data is needed to understand and guide change on shorter time scales.

That is not to say that we cannot rely on Census products - they are the authoritative source on household and community conditions from year to year. Instead we should work to build supportive, timely, and local data systems to complement the limitations of Census products for informing local timely decision need.