Understanding Households and Relationships in Census Data
Finding the stories—and the people—in all those tables
The changing makeup of households and the relationships within them offer some of the most fertile areas for reporting on demographic trends in America. Over time, slow but steady changes in who lives with whom—or perhaps lives alone—reflect shifts in the country’s economy, its social norms and its choices of where and how to live.
The Census Bureau’s population counts make these trends easy to track. All you need are two things: an understanding of how the Census asks Americans about households and relationships, and where to find the right tables amid the haystack of tabulations. That’s what this post aims to help you with.
Picking the Best Data Set
First, let’s note the fundamental difference between the two chief Census products that deal with population: the every–10-years complete count, aka the Decennial Census, and the annual American Community Survey. The most recent Decennial count, taken in April 2010 with results released in 2011, is the easiest to work with because of its relative simplicity. It’s a full count of the 116.7 million households at the time, and thus data are available for geographies as big as the entire nation and each state, and then all the smaller ones—every county, city and place plus tinier geographies such as Census tracts and blocks.
In contrast, the American Community Survey is just that—a random sample of about 3 million households per year. Because it’s a sample, it has limitations. First, values in the ACS come with a margin of error that can make it impossible to be certain about some comparisons from year to year or among geographies. Second, there are three separate ACS products that vary in terms of geographical coverage and timeliness.
A one-year ACS data set is the most recent—the current release is from 2013—but because the sample is thin in small places, results are only reported for geographies with 65,000 people or more. That will exclude about three-fourths of U.S. counties. Three-year and five-year data sets have more data, allowing for more reliable estimates covering smaller geographies, but the tradeoff is that the data are less current. The Census’ guide covers all of this information in greater detail.
For reporting on families, which product to use may depend on the story you’d like to tell—and when you’re telling it. Midway through the 2010s, the Decennial Census is beginning to show its age. If you’re interested in a geography that has experienced rapid growth that may affect household composition—such as parts of North Dakota with its oil boom—the Decennial full count may not offer the most recent snapshot. In that case, you may want to look at the one- or three-year ACS data. But if small geographies are key to your story, then the full count will alleviate any issues you may run into with the ACS’ margins of error. The best option in the end may be to look at both!
Asking the Relationship Question
Once that choice is out of the way, it’s time to understand how the Census derives information on household makeup. To start, look at the Census form itself. On both the Decennial and ACS forms, the Bureau asks for information on each person living at the address. But they expect that the first person on the form is someone who owns, rents, or is in the process of buying the place. This person, in Census parlance, becomes the “householder,” and each subsequent person filling out the form is asked to note how they relate to the householder.
The ACS and Decennial Census forms provide the same categories for responses, with one exception—the ACS adds a foster child category. Although the form doesn’t indicate it, the options fall into two categories that become important when reading the data—family and nonfamily.
The family category includes anyone related to the householder by birth, marriage, or adoption: their husband or wife, children (broken out by biological, adopted, and step), parents, grandchildren, in-laws etc. The nonfamily category includes everyone else: roommates, unmarried partners, someone renting a room in the same house, and so on.
How does the Census Bureau categorize same-sex households? It follows the same guidelines. According to the Census documentation: “Same-sex couple households are included in the family households category if there is at least one additional person related to the householder by birth or adoption.”
Some examples, according to those definitions:
- Single mother with a child: family.
- A cohabiting man and woman: nonfamily.
- Two brothers living together with one of their friends (if one brother is the householder): family.
- Two unmarried women in a same-sex partnership, with one of their children: family.
Finding the Counts
The Census provides two general counts of household relationships. One is the number of people living within various household permutations. The other is the number of households themselves that fall into household types. We’ll look at examples of both.
In the 2010 Decennial Census, the household results are detailed in the product called Summary File 1 (SF1). A series of tables beginning at P18 and running through P43 enumerate household demographics with a focus on type and the ages of people in them.
(The technical documentation for the release, listing all tables and variables, is available on the Census site.)
Our first example is table P21, which has 31 variables and the unwieldy name “HOUSEHOLDS BY AGE OF HOUSEHOLDER BY HOUSEHOLD TYPE BY PRESENCE OF RELATED CHILDREN.” Don’t let that throw you. As with many Census tables, this one consists of the same data categories repeated, once for households where the householder—remember, that’s the first person listed on the Census form—was age 15 to 64 and then for households where the householder was age 65 or older. Here’s what the table looks like after I dumped it into Excel and cleaned up some formatting:
The universe for this table is total U.S. households, which as of 2010 was 116.7 million. Notice that the concept of family and nonfamily takes shape here, with both categories broken out under each of the two householder age groups.
Living Together and Alone
Under nonfamily households, we find one of the telling categories reflecting steady change in American culture: the number of households where the householder lives alone. To calculate a quick percent, add the number of households categorized as living alone from both age groups. It’s 31.2 million—about 20.2 million householders 15 to 64 and about 11 million householders 65 and older. Divide the sum by the total households to get the 2010 percent of households where the householder lives alone: 26.7%.
What’s the trend? Thankfully, the 2010 table P21 has a direct match in the 2000 Decennial Census in table P20. Grab the data, do the same calculation, and the percent is 25.8%. So, between 2000 and 2010, about 4 million living alone households were added to the U.S., edging up its piece of the pie by about one percentage point.
That may not seem like a lot, until you realize that this is the unfolding of a long-term trend that simultaneously reflects a downward tick in family households. For example, according to table P21, the percent of households that were husband-wife families with children was 21.6% in 2010, down from 23.5% in 2000.
(For all the comparable tables between the 2010 and 2000 Decennial Censuses, grab the crosswalk Excel file from the Census site.)
Table P21 lets us track several trends in how households are changing. Using the data, you can calculate:
- The percent of households that are husband-wife. As with family households, those percentages have been dropping, as of 2010 accounting for less than half (48.4%) of households.
- The percent of households where the householder has a child but no husband or wife is present. That’s been rising, up to 11.3% in 2010. Keep in mind these households may have another adult, such as a parent or a long-term partner to the householder.
Earlier, I mentioned that the American Community Survey also provides data on households but is limited in some uses because—depending on the number of years covered in the data—there may be high margins of error or limits on available geographies. Nevertheless, if you wanted to get the most recent national or state snapshot for similar household data, you can find it in the one-year 2013 ACS data set. Table B11001 has the high-level counts of family and nonfamily households and living-alone households. Table B11003 has further detail on the presence of children in married couple families.
Unfortunately, another limitation to the ACS is its limited comparability with Census 2010 and 2000 data. Some questions on the ACS and the resulting data can be directly compared with Decennial Census data, such as age and sex. For others, such as race, the Census Bureau advises comparisons “with caution” and outlines the reasons. With household types and relationships, the Bureau’s guidance is to not compare ACS with 2010 or 2000 data and stick with comparisons to older ACS data.
Finally, as social trends regarding marriage develop, the Census has attempted to keep pace by introducing data products tracking same-sex couples. But as D’Vera Cohn of the Pew Research Center noted in a recent Source post, “same-sex marriage has been difficult for the Bureau to capture accurately,” and it has acknowledged “significant error” in its estimates. Be cautious of tables such as B11009 of the 2013 ACS, which reports the current number of unmarried-partner households and breaks them down by the gender of the partners. It’s no longer a good measure of same-sex couples because, starting with 2013 ACS data, the Census Bureau includes same-sex married couples along with all married couples.
Thus far, we have focused on tables where the universe for the count is households. If we want to count the number of people who checked off the various relationship categories on the Census form, we need to turn to tables where the universe is total population.
In this case, we’ll examine Decennial Census 2010 table P29, called “HOUSEHOLD TYPE BY RELATIONSHIP.” (Its Census 2000 corollary is P27, and you can find similar data in the 2013 ACS table B09019).
Earlier, we saw that 31.2 million households in 2010, or 26.7%, were those with a person living alone. In Table P29, we find that the number of people who live alone—add the male and female lines—equals the same number. Not surprising, and it should match because a live-alone household is one person.
But we have fine-grained details here that hold interest. In 2010, in family households there were 2.1 million adopted children, 4.2 million stepchildren, and 7.1 million grandchildren. The number of grandchildren in family households was a 32% increase over 2000. Some of that change is part of trends in multigenerational families due to economic distress; some is as a result of immigration.
Strategies for Finding Stories
Now that you have a broad understanding of how Census counts households and relationships as well as some of the data points in the results, here are some ways you can bring it home to the places you care about:
- Slice the data by geography: Since the 2010 Census is a full count, you can easily pull data for your state, county, towns and Census tracts. Calculate percentages and make comparisons. Where in your area is the highest percentage of households that are people living alone? Why? Are there places where married-couple families cluster? Again, why?
- Slice the data by race and ethnicity: In SF1 data, many tables are available with counts by race and Hispanic status. Again, calculate percentages and compare at a local level. If you’re in a location with strong immigration, how have households changed during the last 10–20 years?
- Slice the data over time: If you cover a fast-growing (or fast-declining) part of the nation, compare counts and percentages using the 2010, 2000 and 1990 Decennial Censuses or several years of the ACS. If you combine this with geographical slices you can find places that have changed the most over time.
All these can be a fertile place for story-hunting through the data, and doing the work now will get you a head-start on looking ahead to the results of the 2020 full count. And that’s not too far away
Thanks to Paul Overberg of USA TODAY for his insight into several aspects of this story.
Product, code, data, journalism. ISTJ. Author of Practical SQL from @nostarch. Now: @QuestexLLC Prev: @usatoday @documentcloud @PokJournal