This is a really big deal

From Yglesias

Most low-income Americans aren’t poor at all by global standards, so evidence from successful anti-poverty programs in the developing world are difficult to apply to domestic poverty. That’s why it’s so telling and fascinating that a study on the cognitive downsides of poverty would find identical results in New Jersey and Tamil Nadu. Much work on domestic poverty rightly emphasizes the idea of skills and “human capital” needed to navigate a complicated modern economy. This naturally leads to a focus on education, whether in the guise of various school-reform crusades or the push to bring high-quality, affordable preschool to more households. But adults need help, too, and the perception that poor adults—as opposed to presumably innocent children—are irresponsible often leads to reluctance to treat adults as adults who are capable of deciding for themselves how best to use financial resources.

This paternalistic notion that we should be relatively stingy with help, and make sure to attach it to complicated eligibility requirements and tests, may itself be contributing to the problem of poverty. At home or abroad, the strain of constantly worrying about money is a substantial barrier to the smart decision-making that people in tough circumstances need to succeed. One of the best ways to help the poor help themselves, in other words, is to simply make them less poor.

The causal ordering here is really important.  We see this dilemma with other variables that are difficult to randomize.  So, for example, it isn't 100% clear if lack of exercise contributes to obesity or if being heavy makes one less likely to exercise.  It can make a big difference in public policy if the causal arrow reverses direction (or if there is a positive feedback loop that goes between the two variables).

In this case, there seems to be evidence that making a stingier and more complex welfare state increases the long run poverty via decreased decision making due to financial stress.  If this is borne out in other context then it totally changes the optimal policy responses to poverty.