I am very new to census analysis/survey analysis so please forgive my ignorance.

I am interested in MOST data/information about the U.S.A. population. Demographic information like breakdown by age, race, sex as well as things like health-care coverage, income, poverty levels, unemployment rates, etc. I'm also interested in these things on a national, state, and substate level.

And then I'm also interested in these values on a year to year basis to see how the United States has changed.

For now I think I am mostly interested in the aggregated tables and summaries for these values not so much in the microdata. But I suppose this could change in the future especially as I grow more comfortable and learn more about how to properly analyze survey data.

To me it's very chaotic and hard to get a sense of the landscape for accessing this data as it is contained in a variety of different surveys behind a variety of different APIs and search tools.

In any case if anyone could provide any tools or suggestions for helping navigate the myriad amount of resources on the topic of the population of the United States, that'd be great.

For now, my real question is... if I wanted to get a sense of how the U.S.A. population is changing year to year in regards to age, sex, and race, should I be using CPS (Current Population Survey) data, ACS (American Community Survey) data, Decennial census data, or PEP (Population Estimates Program) data? And then additionally, which should I be using for which geographic entities so to speak? I know this question is broad and probably will depend on the specifics, but any general guidance in this department would be helpful.

EDIT: I am interested in the totals for these populations not just the percentages if that makes sense.


2 Answers 2


Geography is often a key factor in determining which dataset you'll use. Are you looking for national numbers? States? Counties? In general:

If you want annual data, only need the basics (age, gender, race) then the Population Estimates Program data is the simplest to use. The data is published by state, county, and metropolitan areas, and also for large municipalities. Like all of the census datasets it's available via the American Factfinder and the census APIs, but since it's relatively small they also provide simple csv files you can download. Each sheet will provide numbers from 2010 to the latest year, and you can go back and get the previous decade in older sets of files. Right now data for all areas for 2016 is available. Data for 2017 has just been released for states but isn't available for other areas yet. The csvs are available directly from the program website:


The American Community Survey (ACS) is better if you need detailed socio-economic summary data beyond the basics of age, race, or gender, or if you need additional geographies - you can get states and counties but also smaller areas like census tracts and ZIP Codes. So if you need data on poverty, health insurance, unemployment, etc then this would be the source you need. For larger geographies that have more than 65k people you can get annual data, but for smaller areas the data are published as 5-year averages. Each estimate is published at a 90% confidence interval with a margin of error; in some cases you may need to use the 5-year averages even for larger areas if the margins of error for the estimates you're using are prohibitively large.

It's challenging to gather and analyze ACS data on an annual basis, partly because the API and American Factfinder are designed to deliver data for one particular release at a time and not multiple years. For the 5-year averages it's only appropriate to compare sets of years that don't have overlap; 2012-2016 vs 2007-2011 would be appropriate.

Some alternatives - the NHGIS https://www.nhgis.org/ provides some historical comparison tables, but is intended for doing longer-term research that covers many decades. The ACS came into existence in 2005; prior to that, detailed socio-economic data was collected in each ten-year census. The Missouri Census Data Center http://mcdc.missouri.edu/ has an ACS Trends application (in the toolbar on the right) that lets you pull together several years of ACS data. They also provide descriptive summaries of the different datasets which might be useful.

We have a few tutorials that we wrote in our lab that cover the basics of navigating the American Factfinder. There's also some explanation of the different datasets and geographies. https://www.baruch.cuny.edu/confluence/display/geoportal/Census+Tutorials

Lastly - the CPS is typically used if you need national estimates on an annual or monthly basis, or if you need or want microdata. https://usa.ipums.org/usa/.

  • Thanks so much for your response! I had a few more questions if you don't mind. 1) I see you have recommended the American Fact Finder and lots of GUI based tools. Have you used any of the Census API tools? If so, have they proven to be fairly reliable? 2) I noticed some of this data exists in a "tidy" data format but not all. Is it common for this data to not be "tidy". Is there work underway anywhere to make this data more tidy and thus making it more amenable to different types of analysis and visualization?
    – drizzle123
    Jan 5, 2018 at 6:43
  • You're welcome! 1) Most of the people I work with are using data on a small scale and have little programming experience, so GUIs are the best option for them. But if you're pulling a lot of data and know how to script then by all means, you can use the APIs. I don't have a ton of experience with them but they seem reliable. 2) US census data does meet the 4 tidy requirements listed here: en.wikipedia.org/wiki/Tidy_data. But in most cases you'll still have some processing work to do, typically aggregating or filtering rows and columns, and building sets and time series.
    – fdonnelly
    Jan 6, 2018 at 15:28
  • atcoordinates.info/2018/01/11/…
    – fdonnelly
    Jan 11, 2018 at 21:21
  • That's a great resource too. Thanks! This may be a dumb question, but is it generally recommended to stick to one dataset when doing analysis? For example, if I wanted to compare population density vs. % of people living in poverty on a county level, I should use ONLY the ACS dataset (for whatever year(s) I'm interested in) rather than using the population density estimate from the PEP and poverty estimate from the ACS. The reason i THINK it isn't ok to mix and match from the different datasets is because they are collected/calculated in different ways with different assumptions, errors, etc.
    – drizzle123
    Jan 18, 2018 at 17:01
  • Generally speaking, you don't want to mix and match variables from different census datasets because they are created with different methods and assumptions. That being said, sometimes it's unavoidable if there are several variables you need and they're only tabulated in one series and not another.
    – fdonnelly
    Jan 22, 2018 at 16:14

NHANES- National Health and Nutrition Examination surveys

If you are into R, You could install the NHANES package, it gives you data on 10000 Americans, 75 attributes, from 2009-2012 with adjusted weighting

Why adjusted weighting? - From the documentation:

"The NHANES target population is "the non-institutionalized civilian resident population of the United States". NHANES, (American National Health and Nutrition Examination surveys), use complex survey designs (see http://www.cdc.gov/nchs/data/series/sr_02/sr02_162.pdf) that oversample certain subpopulations like racial minorities. Naive analysis of the original NHANES data can lead to mistaken conclusions. The percentages of people from each racial group in the data, for example, are quite different from the way they are in the population.

NHANES and NHANESraw each include 75 variables available for the 2009-2010 and 2011-2012 sample years. NHANESraw has 20,293 observations of these variables plus four additional variables that describe that sample weighting scheme employed. NHANES contains 10,000 rows of data resampled from NHANESraw to undo these oversampling effects. NHANES can be treated, for educational purposes, as if it were a simple random sample from the American population.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.