Since I write about it often enough and recently came across some relevant figures I thought I'd share the hard numbers on the relative costs of means-testing vs. universal anti-poverty schemes. Knowing generally that means-testing is more expensive has always been enough for me, the argument against means-testing involves many relatively small effects balanced against the rather large one of less payments to wealthier people, but it always helps to have numbers.
These numbers come from Public Spending and the Poor and specifically from literature review sections in a few of the individual essays, such as "Two Errors of Targeting." Studies of the US found that administrative costs of universal programs are 2.5% compared to 12 percent for two means tested programs (367). Costs for specific programs in 1985 ranged from 13.5% for AFDC), 13.5% for food stamps, 8% for SSI, and 4.8% for Medicaid (Toward Quantifying the Trade Off, 483). It should be noted that a study on poor country didn't find differences in cost between means-testing and other programs, but this is not necessarily generalizable to richer countries.
My takeaway from this is that administrative costs alone do not justify universal programs over means-testing, of course I never though they did. Once other costs of means-testing, such as work disincentives, the relatively higher costs of under-utilization (there is a welfare benefit whenever a child gets healthcare, for instance, and a welfare loss if a child doesn't get healthcare that would have prevented a disabling condition that is greater than the cost in raw dollars and cents of providing that healthcare which means there is a relatively greater cost for each person not covered that should be than for each person covered that wouldn't qualify under means-testing), stigmatization, political impacts (there is considerable evidence that universal programs are correlated with other efficient social and economic policies, countries with less means testing than the US are more likely to have more efficient tax systems, for instance, pretending these policies are separable despite the contrary evidence makes for poor policy debate), etc. that all way against means testing. All of these individual impacts are too small to justify the increased expense of universal provision, taken together they argue strongly against means testing.