In what became a viral chart in economics blog/twitter circles, Bard College economist Pavlina Tcherneva used data from the World’s Top Income Database to show that the top 10 percent of tax filers have received an increasing share of income growth during expansions over time. Manhattan Institute sociologist Scott Winship objected to several parts of her analysis and responded with his own figure. The back-and-forth raises quite a few questions about how to best measure income trends. But one in particular stands out: do we look at households or tax units?

When people file taxes with the U.S. Internal Revenue Service, they do so on behalf of their tax unit, which includes individuals or married couples that file together along with any dependents they declare that earn below the filing threshold, which was $6,100 in 2014. These tax-unit data are the basis of Tcherneva’s figure, drawn from the underlying data in the World’s Top Income Database compiled by University of California-Berkeley economist Emmanuel Saez, Thomas Piketty of the Paris School of Economics, and several other researchers. The IRS tax data do not capture people who make too little to be required to file taxes.

Other data sources, such as the U.S. Census Bureau, use households as their unit of analysis. Households include everyone that lives under the same roof, whether immediate families, extended families, or other living arrangements, such as roommates or boarders.

For most people, their tax unit and household are the same. But there are cases where a household would include multiple tax units. When an elderly parent moves in with an adult child with kids filing taxes separately to form a three-generation household there would be multiple tax units exist in one household. The same is true when a child moves back in after college or an unmarried couple moves in together. These are groups that we probably want to think about together because they rely on each other economically. On the other hand, a group of unrelated adults living together as roommates would also count as a household.

We might prefer tax units to households for income trend analysis because household structure is not independent of economic circumstances. For instance, adult children move back in with their parents (Gallup reported that 14 percent of 24 to 34 year olds lived with their parents in 2013 and a Pew Research Center study showed this to be increasing). Similarly, homeowners sometimes take in boarders for economic reasons. In each of those cases, the tax-unit data captures the deteriorating economic situation associated with these types of living arrangements.

In contrast, household data captures more earners under one roof, which increases household income. Imagine a group of three graduate students living together where each makes $25,000. They would show up as three different tax units but one household making $75,000. Now image that they all got $5,000 raises and decided to get places on their own. This would show up as a 20 percent increase in their incomes by tax units, but their household incomes would have dropped by 60 percent.

These are not insignificant concerns because there are roughly seven million multi-person nonfamily households, out of a total of about 115 million households, according to the Census Bureau.

Tax units are not perfect measures either. One flaw with using tax units is that dependent children that make more than the filing threshold are separate tax-filing units but generally counted as part of their parents’ household. And as mentioned earlier, the tax data do not capture incomes of very low earners.

The debate about how to best measure income trends clearly requires explaining the benefits and drawbacks of which measure is being used. A variety of factors go into determining which unit of analysis to look at and often the choice isn’t clear cut.