Kristen Wyatt/AP - Laura Fritz, 27, left, with her daughter, Adalade Goudeseune, fills out a form at the Jefferson Action Center, an assistance center in the Denver suburb of Lakewood in July 2012. Fritz grew up in the Denver suburbs in a solidly middle class family, but she and her boyfriend, who has struggled to find work, are now relying on government assistance to cover food and $650 rent for their family.

The decline of extreme poverty — defined by the World Bank as living on less than $1.25 a day, which is derived from the average poverty line in the world's poorest countries — in recent decades has been nothing short of remarkable. As Howard Schneider noted here  last week, not only has the percent of the world's population living in extreme poverty been cut in half since 1990, but it's set to be halved again in the next two decades:

But it's worth keeping in mind that even the richest countries haven't completely eradicated extreme poverty. That's the key takeaway from the work of sociologists Kathryn Edin (Harvard) and Luke Shaefer (Michigan), who for the past few years have been trying to nail down the incidence of extreme poverty in the United States. Their latest research is set to be published in the journal "Social Service Review" next month. They use a slightly different measure, defining extreme poverty households as those living on less than $2 a day per person; that's also a World Bank measure, derived from the average poverty line in the developing world, rather than the average in very poor countries.

The results are astonishing. Using data from the Survey of Income and Program Participation (SIPP), a Census program that tracks samples of tens of thousands of households across 2 1/2 to 4 years, Edin and Shaefer estimate that in 2011, 1.65 million U.S. households fell below the $2 a day per person threshold in a given month. Those households included 3.55 million children, and accounted for 4.3 percent of all non-elderly households with children.

That sounds unbelievable, but Shaefer explained that the research was prompted by qualitative work Edin did in households like this. "Kathy and I were talking and she mentioned that she felt like she was going into more and more homes where there was really nothing in income," he said. "They were surviving on food stamps, or in some cases on nothing at all." So they set about investigating whether or not these households showed up on the SIPP. And they did.

So how do these families survive? The safety net is a big part of the story. If you take food stamps, housing subsidies and refundable tax credits like the Earned Income Tax Credit (EITC) and the Child Tax Credit (CTC) into account, the number of households in extreme poverty is 613,000, or 1.6 percent of non-elderly households with children (compared to 1.65 million and 4.3 percent without transfers). So taking government aid into account reduces the extreme poverty population by 62.8 percent. Then again, there are limits to treating aid as equivalent to cash. As Shaefer said, "You can't eat a housing subsidy."

Civil society also plays an important role. Shaefer noted that follow-up qualitative work he and Edin have done shows that many of these households rely heavily on homeless shelters and soup kitchens, as well as public goods like libraries. "We see people relying on the availability of libraries, since you have to have Internet access to have any chance of getting a job, as most hiring is now done online," he explained.

There's a possibility that some of the results Edin and Shaefer generate are the result of underreporting income, a problem they note in the paper. While SIPP picks up just about every kind of cash income, it doesn't pick up in-kind transfers. "If my uncle doesn't give me cash but buys me groceries, we're not going to pick that up," Schaefer said. It's also likely that households underreport aid like food stamps out of embarrassment, and don't report money earned from illicit activities. "Underreporting is an issue, but SIPP is the best data we have," he concluded.

Another reason underreporting likely isn't the whole issue is that Edin and Shaefer find a large increase in extreme poverty between 1996 and 2011 (see the above chart). In 1996, only 1.7 percent of households were in extreme poverty. By 2001, that had grown to 2.3 percent, and then to 3.0 percent in 2009 and 4.3 percent in 2011. Taking transfers into account, the 1996-2011 increase was from 1.1 percent to 1.6 percent. "Even if you assumed that income was being underreported at 30-40 percent, that still means we have this dramatic increase in people living under, say, $6 a day," Shaefer said. It's still pretty concerning."

It's more concerning still when you note that most of the increase took place before the 2009 recession, suggesting it's a longer-term trend not totally dependent on the business cycle. "I definitely think there's a cyclical component to it, but that can't explain the dramatic increase and the fact that it never fell down to the previous levels after welfare reform," Schaefer said.

What seems to have happened is that, since welfare reform, the amount of aid devoted to low-income households didn't decline, but it shifted from the desperately poor to the working poor. "For EITC, we're spending $55 billion, whereas the government never spent more than $20 billion on TANF. It's not right to say that we've retrenched our aid to the poor," Schaefer said. "What's true is that we've shifted a lot of the aid from the people at the very bottom to the people right at the poverty line."

You can see that in the chart above. In 1996, AFDC/TANF was keeping a whole lot of families out of extreme poverty. Over time, though, TANF's role diminished as welfare reform took effect. EITC more than took its place, but Edin and Shaefer's work suggests that it did so at the expense of the most desperately poor.