Greg Mankiw has an interesting post describing the absurdity of how poverty is measured in the US, especially in regards to the EITC:
…the Census omits the income from the EITC when computing the poverty rate. As a program to reduce measured poverty, the program is, by assumption, doomed to failure.
Of course, this is not really a problem with the EITC but, rather, a problem with the measured poverty rate. It makes no sense to evaluate poverty with a statistic that ignores the effects of one of the largest and most rapidly growing anti-poverty tools we have. But that is what the official statistics do.