Understanding the 2024 American Community Survey 1-Year Estimates

Estimated reading time: 9 minutes

Each year, the U.S. Census Bureau conducts the American Community Survey (ACS), providing a detailed portrait of our nation and communities.

If you've ever received the ACS, you know it asks about more than 40 topics, including income, housing costs, commuting, and education. Responses to these questions are vital to understanding the demographic, social, economic and housing characteristics of our communities.

Not everyone receives the ACS, though. It just goes to a sample - or randomly selected portion - of addresses each month. The responses are combined to paint the statistical portrait.

What that portrait looks like each year is ultimately affected by:

• Who is in the sample.
• Who responds.
• How many people each respondent represents.

This blog discusses how these elements come into play in the 2024 ACS 1-year estimates and how they may contribute to differences observed in some estimates compared to prior years.

For example, every year the survey uses "population controls" to align the ACS estimates with official population estimates. The 2024 population estimates for the overall population and certain subpopulations significantly increased after we improved a key component of the estimates. This increase is reflected in the ACS survey estimates, too, and interacted with other methodological elements discussed in this blog, which contributed to year-to-year change in the ACS results.

Who is in the Sample

First, it's important to remember that the ACS is a sample survey. The ACS samples approximately 3.5 million housing unit addresses and about 150,000 residents of group quarters throughout the 50 states, the District of Columbia, and Puerto Rico each year.

We strive to draw a sample that can be used to represent the whole population. The estimates derived from the sample need to reflect the entire population as accurately as possible.

You can get a sense of how close we're getting to the true value for the overall population by looking at the margin of error published next to each estimate.

For example, if an estimate is 200 and the margin of error is ± 10, that means there's a 90% probability that the true value for the population is within 200 ± 10. In other words, we're 90% confident it's between 190 and 210.

We encourage data users to be mindful of the margins of error when focusing on any particular estimate. The margin of error is the key to understanding the degree of uncertainty in the estimate.

Who Responds

We ask households selected in the sample to respond to the ACS online or by mail (and we send reminders). If they don't respond, we send field representatives to a sample of the nonresponding households to try to gather their responses in person or by phone. This follow-up operation is called Computer-Assisted Personal Interviewing (CAPI).

One way the ACS controls costs of data collection is by identifying households to stop interviewing in person or by phone. In 2024, we improved our method to have a limited, data-driven approach to stopping cases:

• For the first two weeks of the CAPI operation, we attempt to interview every household in the CAPI sample to ensure each housing unit has an opportunity to respond.
• During the operation's final two weeks, we prioritize the remaining cases, focusing our resources on collecting data from sampled housing units that, based on administrative data, are less represented among those who have already responded.
• The remaining cases are still eligible to respond online or via the paper questionnaire, and a small portion do.

It is possible that the CAPI respondents have slightly different characteristics than if we had not prioritized some of the cases. While this prioritization only impacts a small number of cases (less than 1% of the monthly ACS sample), some impacts on the estimates are possible.

Along the way, one of the things we evaluate is whether the people who respond have different characteristics from those who don't. For example, if only retired individuals respond, then statistics on labor force participation would not be accurate because they would not be factoring in households with people who are working.

This concept is called "nonresponse bias" because it can bias the estimates if there are major differences between the two groups.

The ACS has had lower response rates in the 2020s than in the 2010s, which can increase the risk of nonresponse bias.

To evaluate nonresponse bias for the ACS, we look at administrative data for households in the sample (when available) and compare how certain characteristics of the people who responded might be different from those who didn't.

At the national level, the characteristics of the 2023 and 2024 ACS respondents did not differ much. We provide some additional details on potential nonresponse bias in the American Community Survey Research Note: Guidance for Comparing the 2024 ACS 1-Year Estimates. This research note also includes some additional statistics describing how nonresponse bias changed at the state level.

Representing Others

Next, it's important to understand that because the ACS is a sample survey, each responding individual and household in the sample represents many others in the overall population. For example, we take an individual's response and multiply it by a "weight." From these weighted responses, we calculate survey estimates that are representative of the overall population.

To figure out how many people each respondent should represent, we develop weights based on the sample design, response and nonresponse, and the population and housing unit controls.

The controls are independent estimates of:

• Age.
• Sex.
• Race.
• Hispanic origin.
• Total number of housing units.

These characteristics within the ACS sample can differ from known population levels because of sampling variability and differences between people who respond and those who don't.

Using the controls aligns the ACS estimates with official population estimates for those characteristics and enables us to account for coverage error in the ACS estimates.

Each year, we update the population controls using the latest estimates from the Census Bureau's Population Estimates Program (PEP). Annually, PEP releases a new time series of estimates from the current year back to the most recent census. We refer to each new time series as a "vintage." Each vintage may have revisions to earlier periods in the time series to include more up-to-date data inputs and methodological improvements.

There were several changes to the PEP estimates between Vintage 2023 and Vintage 2024. Most notably, the Census Bureau improved the methodology used to calculate net international migration - a key component of the annual population estimates.

The improved methodology more accurately reflects the net increase in migrants to the United States between 2022 and 2024. As a result, there was a cumulative increase of 5.196 million people for the 2024 ACS 1-year estimates. Using these updated controls helps us improve the overall accuracy of the ACS estimates.

The increase in net international migration is reflected in the 2024 ACS 1-year count estimates of people across topics. This will be especially apparent when comparing to prior ACS years.

• For example, in our ACS comparison profile table on economic characteristics (Table CP03), you might notice that the civilian labor force increased by nearly 4 million people.
• Some of this increase might capture changing economic conditions over time, but it may also reflect the large year-to-year population increase due to both demographic changes and updates to the methodology.

The population increase may also be impacting estimates beyond just population counts.

• A good example would be the average household size, which is a function of both population size and the number of occupied housing units.
• When looking at the comparison profile for housing characteristics (Table CP04), you might notice an increase in average household size for your state, county or place this year - despite the decreasing trend over the past few years.
• Some of this change might reflect shifting living arrangements. The change could also be a function of the number of people significantly increasing, while the number of housing units had the usual steady growth.

When looking at counts of people across the ACS tables, especially when comparing to 2023, it is important to consider the increase in population because of the higher level of net international migration. While this new population total for 2024 uses an improved methodology and more accurately reflects the total U.S. population, we encourage users to take this increase into consideration when comparing subpopulations over time.

Also, keep in mind that the ACS is designed to measure characteristic distributions. Whenever possible, we generally recommend comparing population characteristics such as percentages, means, medians, and rates rather than estimates of population totals.

Quality Checks

We're meticulous about checking quality and providing accurate results.

We analyze the data by comparing it against past benchmarks to identify outliers or unexpected results that need to be further examined. We triple-check data collection and processing, analyze the sample and the potential for nonresponse bias, and verify the weighting process.

As we do with all our data products, we've investigated unexpected differences in the 2024 estimates and determined that they are not due to data collection or processing errors.

As usual, we encourage data users to be mindful of margins of error published alongside the ACS estimates. The margins of error provide information on the uncertainty embedded in the estimates related to sampling variability and should guide how you use the estimates.

We also encourage data users to be mindful of the potential interplay of the other factors discussed in this blog, as they may contribute to unexpected differences in the estimates compared with recent years.

Conclusion

Our nation is dynamic. Our goal with each release of ACS 1-year estimates is to produce the most accurate estimates possible. We know communities across the country rely on them to make decisions about funding and services.

We strive to be clear about what we're measuring and what the numbers represent. We continuously work to improve our methodology and are open about how that may affect the estimates.

In summary, the 2024 ACS 1-year estimates will reflect:

• Changes in the population.
• Who is in the survey's sample.
• Who responds and how they're different from those who don't.
• The weighting methodology, or how many people each response represents.
• Changes in the population controls, particularly the methodological improvements to better reflect the large net increase in immigrants and total population.

Be mindful of these elements and the margins of error, and remember to focus analysis more on percentages, distributions, rates, means, and medians instead of on count estimates when possible. Where available, use the population estimates for count estimates.

For more information on the topics discussed in this blog, please check out the 2024 ACS 1-year research note.