New York's anti-poverty experiment 

 The mayor of New York, Michael Bloomberg, announced today that the city is proceeding with its plan target poverty using cash incentives for school attendance, medical checkups and the like.  The first phase of the plan is an experimental test of the efficacy of the incentives.  From the  NY Times :  

 Under the program, which is based on a similar effort in Mexico but is believed to be the first of its kind in the nation, families would receive payments every two months for meeting any of 20 or so criteria per individual. The payments would range from perhaps $25 for an elementary school student’s attendance to $300 for greatly improved performance on a standardized test, officials said.

 Conceived as an experiment, the program, first announced last fall and set to begin in September, is to serve 2,500 randomly selected families whose progress will be tracked against another 2,500 randomly selected families who will not receive the assistance.  

 Now, I think most of us in the social science statistical community would be very much in favor of this kind of evaluation.  In fact, the degree to which these kinds of designs are becoming the standard for policy evaluation is an impressive change from the way projects were evaluated even twenty years ago.  Gary King and several graduate students here at IQSS have been working on the evaluation of a similar project in Mexico involving the roll-out of Seguro Popular, a health insurance scheme for low-income Mexicans. 

 On the other hand, the political scientist in me wonders if (when?) we are going to start to see pushback from those being experimented on (or, more likely, from the interest groups that purport to represent them). The image of 2,500 families randomly selected to  not  receive benefits probably doesn't do much to help the cause of people (like me) who would like to see more of this.  How can we in the statistical community make these kind of randomized field experiments more palatable (beyond saying, "you need to do this if you want the right answer")?