The False Claim That Higher-Income People Are Driving SNAP Costs, Part 2

November 21, 2012 at 12:40 pm

Part 1 of this post explained that SNAP (food stamp) spending has risen significantly in recent years mainly because of the recession and a temporary benefit increase that policymakers enacted in response to it.  Now let’s turn to the claim by George Mason University’s David Armor and Sonia Sousa that a fifth of SNAP recipients have incomes above twice the poverty line.  The most reliable data indicate that this claim is far off base.

SNAP administrative data — which are based on actual, verified case records and are much more reliable than the data that Armor and Sousa use — show that only 0.1 percent of SNAP participants had incomes above twice the poverty line in an average month of fiscal year 2010.  (Under longstanding SNAP rules, a tiny number of households with elderly or disabled members can qualify for a very small SNAP benefit with income at this level.)

Armor and Sousa’s much higher estimate reflects their use of data from the American Community Survey (ACS), which significantly overstate SNAP receipt among higher-income people.  The reasons why are technical but important:

1.  The ACS measures people’s income over the entire year and asks whether their household received SNAP at any time during the year.  By contrast, the SNAP administrative data look at people’s income in the month that they actually received SNAP.

The fact that some people have incomes above twice the poverty line for the year as a whole, yet receive SNAP for part of the year when their incomes are significantly lower, shows that the program is working as intended.  Some people lose their job during the year and turn to SNAP for the months when they have less income.  Others may receive SNAP at the beginning of the year while out of work but then find a job, no longer need help, and leave the program.

2.  The ACS does not distinguish between SNAP recipients and non-recipients who live with them. As a result, Armor and Sousa incorrectly count as SNAP recipients some higher-income individuals who do not get SNAP benefits.

Specifically, the ACS data do not distinguish between SNAP participants living in a house or apartment and other individuals who lived in the same residential unit but were separate economically and did not receive SNAP themselves.  Nor do the ACS data exclude people who lived in the same unit as SNAP participants at the time of the survey but not when the latter actually were receiving SNAP benefits.  (The ACS simply asks whether anyone living in a house or apartment received SNAP at any point during the year; where the answer is yes, Armor and Sousa count all people who were living in the residential unit at the time of the survey as SNAP recipients.)

Because of these weaknesses in the ACS data, Armor and Sousa treat some non-SNAP recipients as though they received SNAP benefits.  They also count the income of everyone who lived in the same house or apartment as a SNAP recipient as though it went to someone who received SNAP benefits.  Essentially, they count as income going to people on SNAP the income of all individuals who were living, at the time of the survey, in the same house or apartment as anyone who received SNAP at any point over the year as a whole.

Many SNAP households “double up” with other people from whom they are separate economically and with whom they do not share costs for food or other items.  It’s also common for people who lose income or experience a marital break-up to move in with relatives or friends over the course of a year; people who received SNAP at some point during the year may not have been receiving benefits at the same time they were living with other people who had higher incomes.  Such situations can produce anomalous results in the ACS data.

For example, if an individual received SNAP in April and May and then moved in over the summer with middle-class cousins who do not receive SNAP, the ACS data would treat all of them — including the middle-class cousins — as SNAP recipients and (because of the cousins’ income) show them all as having income over twice the poverty line.

SNAP provides benefits only to groups of people who buy and prepare food together and share in the costs.  Accordingly (and appropriately), SNAP administrative data do not include the incomes of people who live in the same house or apartment as SNAP participants at some point during the year but are separate economically and do not themselves receive SNAP assistance.

3.  The ACS is simply a survey, usually conducted by mail, with no verification of the accuracy of the answers it receives. The ACS cannot compel survey respondents to be careful and complete about reporting their income; it differs in this respect from SNAP, where eligibility workers must interview applicants and verify each household’s income.  The caseworkers require documentation, check wages against employer records, and run regular data matches to verify income.

Moreover, the majority of households listed in the ACS as having received SNAP benefits and having income above twice the poverty line either did not actually report their earnings level or did not report receiving any SNAP benefits; Census staff in effect took their best guess as to these households’ income levels or whether they received SNAP.  This is known as “imputation.”

Specifically, when households skip questions on the ACS about their wage, salary, or self-employment income or SNAP benefits, Census staff use established procedures to guesstimate the missing information, based on responses from similar households chosen at random.  As a result, the ACS identified some people as SNAP recipients who did not report receiving any SNAP benefits; the ACS also attributed more income to some SNAP recipients than the recipients themselves reported.

Imputation of missing data is a common tool in survey research and is reasonable in small doses, as long as one does not place too much weight on the supposed accuracy of the imputed data.  The data generated through such techniques can be useful when no better source is available.  But, in this case, a much more reliable source is available:  the SNAP administrative data, based on verified data on household income levels.

Our preliminary analysis of the ACS and other Census data suggests that if you remove the imputed data from the ACS, take into account the differences between monthly and annual income, and limit the analysis to people who actually are part of a SNAP assistance unit, most of the SNAP participants whom the ACS lists as having incomes above twice the poverty line either would no longer show up as receiving SNAP or would no longer be shown as having incomes over twice the poverty line.

As we have said elsewhere, SNAP is working as intended to reduce and alleviate hunger and to lift families out of poverty or lessen the severity of their poverty.  Especially now, with many low-income Americans struggling to make ends meet, this would be a very different country without SNAP.

Print Friendly

More About Dottie Rosenbaum

Dottie Rosenbaum

Rosenbaum is a Senior Policy Analyst focusing primarily on federal and state issues in the Food Stamp Program as well as issues that involve the coordination of food stamps and other state-administered health and income security programs, such as Medicaid, TANF, and child care. In addition, Rosenbaum has expertise on the federal budget and budget process.

Full bio | Blog Archive | Research archive at

Your Comment

Comment Policy:

Thank you for joining the conversation about important policy issues. Comments are limited to 1,500 characters and are subject to approval and moderation. We reserve the right to remove comments that:

  • are injurious, defamatory, profane, off-topic or inappropriate;
  • contain personal attacks or racist, sexist, homophobic, or other slurs;
  • solicit and/or advertise for personal blogs and websites or to sell products or services;
  • may infringe the copyright or intellectual property rights of others or other applicable laws or regulations; or
  • are otherwise inconsistent with the goals of this blog.

Posted comments do not necessarily represent the views of the CBPP and do not constitute official endorsement by CBPP. Please note that comments will be approved during the Center's business hours. If you have questions, please contact

 characters available