This is the 1000th Supply & Demand post, beginning September 15, 2008.
Something unusual has been happening with the food-stamp program, now known as SNAP, for Supplemental Nutrition Assistance Program. Between 2007 and 2012, spending on SNAP more than doubled, adjusting for inflation and population growth.
Paul Krugman and others attribute essentially all of the SNAP spending growth to the depressed economy. They have the general direction right – a more depressed economy will cause unemployment and antipoverty programs to spend more – but have missed the single largest factor increasing program budgets: program rules that are more generous now than they were in 2007.
Veterans benefits, Supplemental Security Income, Medicaid and Temporary Assistance for Needy Families all experienced a depressed economy, too, but they somehow managed through it without doubling their spending. Veterans benefits increased the most among these – 49 percent beyond inflation and population growth – compared with 110 percent for SNAP. (These data, which exclude administrative costs, can be found in the Bureau of Economic Analysis’ National Accounts Table 3.12.) Even state unemployment benefit spending, which is directly linked to layoffs in the economy, increased “only” 24 percent beyond inflation and population growth.
Peter Ganong and Jeffrey Liebman of Harvard have recently found (see Table 2 in their paper) that seven or eight changes in SNAP eligibility have spread across the states in recent years. They have examined county-level data on SNAP participation and other variables in order to estimate quantitative importance of some these rules. They find that between 2007 and 2011, new eligibility rules by themselves added 3.4 million people to SNAP enrollment and naturally tended to increase SNAP spending.
Perhaps 3.4 million seems small for a program that enrolled 26 million people before the recession. However, at the same time, SNAP began to pay more generous benefits to people who enrolled. Although changing benefit formulas is not part of Mr. Ganong’s and Professor Liebman’s paper, the new formulas would have increased SNAP spending more than 25 percent even without any new enrollment. Combined, the spending impact of enrollment and benefit rules is remarkable.
The chart below reports two estimates of the sources of SNAP spending growth: the one on the right, which builds on the Ganong-Liebman enrollment findings, and the one on the left, based on enrollment results I obtained earlier using somewhat different methods. (The Ganong-Liebman paper does not attempt to measure the combined effect of new benefit and eligibility rules between 2007 and 2011). The vertical axis measures the increase in SNAP program spending between 2007 and 2011, measured in 2007 dollars per American per year. All Americans are in the denominator – not just those who participate in SNAP – so that more participation in SNAP increases spending measured this way.
The total increase is $112 per person per year. Part, but not all, of the $112 can be attributed to more generous benefit formulas and more inclusive eligibility rules. That part is shown in red. My estimates say that, without a depressed economy, inflation-adjusted SNAP spending per capita would have increased $77 because SNAP rules changed. Using the enrollment estimates of Mr. Ganong and Professor Liebman together with the changes in benefit formulas suggests the increase would have been $53.
The remaining or unexplained spending increase is potentially attributable to the depressed economy, although it could be attributable to changes in the conduct of the SNAP that have not yet been quantified. For example, the Department of Agriculture has perennially attributed some of the increase in program participation to its outreach efforts – that is, advertising, promotional and other activities that encourage eligible people to join the SNAP program. Mr. Ganong and Professor Liebman note that enrollment itself may react to more generous benefits, as high benefits are likely to have encouraged more households to participate. These are effects that should be included in the red area in the chart but have been left as part of the blue “unexplained” area because of the lack of quantitative estimates.
The United States had a food stamp program before the recession that automatically included more households as circumstances put their incomes near or below the poverty line. The newest estimates suggest that going back to the 2007 SNAP program rules would annually save taxpayers at least $53 per American – that’s $212 for every family of four – and put SNAP spending back in line with spending on other antipoverty programs.
The forces of supply and demand suggest that the machines of the future will continue to be significantly different from people.
Anthropomorphic robots are prominent in science fiction. Novels like Mary Shelley’s “Frankenstein,” and films like “E.T.,” “Star Wars” and “The Terminator” feature machines with one head, two eyes and many other features of humans. But those robots were created by authors to entertain audiences, and not by investors to produce some other kind of economic value.
At first glance, it would seem that the state of technological progress is all that limits the creation of machines closely resembling humans. Especially after the recent recession, people are concerned that technological progress is moving at a pace that will soon permit machines to put wide swaths of the human population out of work, and not just displace workers from one industry to another.
I disagree. Powerful economic forces will push the machines of the future to be different from people, and to complement workers rather than mimic what they do.
Machines are expensive to design and manufacture, and most of the people directing the creation of machines have an eye on the rate of return: the economic value of a finished machine as compared to the costs of creating it. The more profitable investments will, by definition, be in machines with a higher rate of return.
The earth is already occupied by quite a number of people, and a machine like Frankenstein or C-3PO might find itself with seven billion competitors.
Take my example last week of baby-sitter robots. A baby-sitting machine with a high rate of return might, as one commenter suggested, be one designed to help children during fires and other disasters. Or a machine to care for children at night. These are examples of child care tasks with less competition from people than the ordinary baby-sitting tasks.
Human help is, of course, not free, and economizing on the cost of a baby sitter or an automobile driver is a reward for creating a machine to do those tasks. But the amount of the reward is commensurate with the amount of wages that people earn in the task.
Machines that drive human workers into unemployment, rather than into another industry where the human workers will be productive, will serve only to drive down workers’ wages and thereby drive down the machine’s value. Machines that, instead, help people to be more productive will find it much harder to saturate their own market.
Rather than finding an intelligent and tireless robot in your office chair, expect the machines of the future to help workers, not harm them.
Many parents treasure their children and feel the benefits outweigh the time and costs of having children. Many other adults decide not to have their own children, and the time costs are sometimes a factor in that decision. (To be sure, the influences are complex; a study by Satoshi Kanazawa of the London School of Economics, which suggests that women with higher IQs are less likely to have children, made waves in the blogosphere in recent days.)
The time costs of child care are also a factor limiting teenage pregnancy. Teenagers are encouraged to complete high school and higher levels of schooling, and students’ parents, teachers and counselors – if not the teenagers themselves – understand that teenage motherhood takes time away from schoolwork and thereby makes academic success less likely.
The world would be very different if children did not need so much time. More people, perhaps especially teenagers, would have children if children did not require so much time and attention, especially from their mothers. People who already have children despite the cost might have more of them if they expected each child to require less time.
As the time costs of children limit population growth, the population would be likely to grow more rapidly if those costs were somehow reduced, whether you think that such growth is a good thing or a bad one.
If each child required less parental time, you might expect parents – especially mothers – to use their time on other things, like work more outside the home, pursue their own schooling or leisure activities. But it is possible that people would spend more of their lives caring for children and less time on those other things because they would be having more children.
Wealthy people have already had some of these opportunities, because they can afford numerous baby sitters, nurses and tutors. But technological progress may one day reduce child-care costs for the general population.
Because robots and other machines take on a number of tasks formerly done by people – even playing chess – and are expected to do others like drive cars, perhaps we should expect that robots will some day take care of children, too.
People today may believe that it would be inhuman or immoral to leave young children at home alone with a robot or to drop them off at a day care center staffed by machines. But economic and technological changes of the past have already transformed child-rearing attitudes and practices: take test tube babies, working mothers, screen time, fast food or children with their own telephones.
There is little need to worry that machines will take over all aspects of child rearing. People will always have a comparative advantage over machines, even if machines could in principle be better at just about anything. For the same economic reason that the world can produce more by assigning some tasks to unskilled people and other tasks to talented people, people will be doing tasks that are difficult for machines relative to other tasks.
But perhaps robots will make parenting easier and thus more popular.
update: the Mulligan link has the correct URL now
Health care and the labor market are connected because so much of the non-elderly population obtains health insurance through an employer or the employer of a family member. As the decades have gone by, Americans have been spending more and more on health care, largely through their health insurance premiums, to the point that many families cannot afford the kinds of health insurance plans in which middle- and upper-income families take part.
So it’s reasonable to wonder whether the health expenditure trends affect the amount of employment in the economy, and thereby whether policy reforms that reduce the rate of health expenditure growth might reverse some of those employment effects.
The direction of the employment effects of health care inflation is unclear, because it depends on the reasons for rising health care costs. To the degree that rising costs derive from new, valuable (but expensive) pharmaceuticals and medical procedures, rising costs may make people more attached to jobs with health benefits in order to have better access to medical innovations and to pay for them with pretax dollars.
But health economists have also pointed to less benign sources of health care inflation, including excessive malpractice penalties and a number of other health industry inefficiencies that raise employer health insurance costs without creating commensurate value for employees.
The economists Katherine Baicker and Amitabh Chandra looked at evidence suggesting that malpracticelike sources of health care inflation are economically equivalent to an implicit tax on employers (see Page 612 of their paper). Economists call it an “implicit tax” because it has many economic characteristics of a tax, even though it is not legally a tax; it reduces employee cash wages by the amount of the implicit tax, and incentive-sensitive employees respond by working less.
As the House of Representatives began to consider whether to repeal the Affordable Care Act, David Cutler of Harvard testified about the Baicker-Chandra results and asserted that the Affordable Care Act would reduce average health care costs by about 5 percent by 2015, reduce the health care cost implicit tax on employers and thereby increase nationwide employment more than it would have grown had the Affordable Care Act not been enacted.
He also organized and signed an economists’ letter to Congress asserting that “repealing the Affordable Care Act would produce job reductions of 250,000 to 400,000 annually.” The Affordable Care Act was cutting employer costs and Congress needn’t worry that it would contract the labor market, they wrote.
Neither Professor Cutler’s testimony nor the economists’ letter mentioned that the Affordable Care Act also creates explicit taxes on employers, subsidies for layoffs and various implicit taxes on employees with many of the same economic characteristics as taxes on employers.
Other advocates of the Affordable Care Act dismiss the act’s work disincentives as negligible, because incentives supposedly have little effect on employment and hours worked. At first glance, it might seem that we have a case of dueling experts, and that we’ll never know which effect dominates. But that first impression would be incorrect, because each effect cited above – like the employer mandate or the health care cost reduction – is a tax effect, and simple arithmetic is all that is needed to determine the direction of the combined effect of all of the tax-like provisions.
After I sent Professor Cutler a draft of this post, he responded: “When I was giving my testimony, I was excluding the vast bulk of policies that will affect part-time work, job choice, etc., because I wanted to focus on the overall cost issue. I don’t think you can do this right unless you include all the effects.”
He agreed with me that readers of his testimony and letter might get the wrong impression that “repealing the Affordable Care Act would produce job reductions” refers to the act as a whole, when it fact it refers to the cost-reduction provisions by themselves. (He also said that neither his testimony nor my calculations quantify the effect of the health care law on job mobility and on the health of the work force, and that he believes these two employment effects to be large. I will return to those issues in a later post.)
Furthermore, Professor Cutler told me he left out explicit and implicit tax effects because he believed (and still believes) them to be less than the cost-reduction effects, and because he “didn’t have a way to add them all up,” referring to the various effects. Since Professor Cutler’s testimony, I have shown how most of the effects can be added together because cost reduction is a tax effect comparable to the tax effect of the employer mandate, the tax effect of the subsidy for layoffs and so on (see also the methodology in Chapter 3 of my book “The Redistribution Recession”).
Begin with Professor Cutler’s (probably optimistic) estimate that the act will reduce employer health costs by 5 percent as of 2015. Because of the special payroll and income tax treatment of employer health insurance, 1.5 of those five percentage points of savings will accrue to government treasuries, leaving 3.5 percentage points of health care savings for employers and employees. Americans spend about 18 percent of their gross income on health care and 82 percent on other things: saving 3.5 percent on health care is like saving 0.6 percent on their total budget.
(In principle, the government treasuries could use their savings to cut marginal tax rates or increase them less than they would have. But they could also use the savings to pay for additional assistance programs that erode work incentives, so I take the middle ground and assume that the government savings by itself has no effect on marginal tax rates.)
Some cost-reducing provisions in the Affordable Care Act, such as the tax on “Cadillac” health plans or the Independent Payment Advisory Board, may reduce value received by employees at the same time that they reduce employer costs, and therefore affect employment less than cutting implicit employer taxes does. The implicit tax cut effect associated with the act’s cost reductions (as estimated by Professor Cutler) is therefore somewhere in the range of 0.3 to 0.6 percentage points, with the 0.6 percentage point case representing the extreme where none of the cost-saving measures reduces employee value.
The Affordable Care Act’s explicit taxes on employers, subsidies for layoffs and implicit taxes on employees, together amount to a five or six percentage point addition to the average marginal tax rate on labor income (this includes the fact that many people will not take part in programs for which they are eligible, the tendency of the act to move people off means-tested uncompensated care and the fact that the act implicitly taxes unemployment benefits, as I noted in testimony before the Human Resources Subcommittee of the House Ways and Means Committee). By these calculations, the tax effects that Professor Cutler left out are about 10 times greater than, and in the opposite direction of, those he conveyed to Congress.
Professor Cutler projected that the Affordable Care Act’s cost reductions by themselves will increase employment in 2015 by about 400,000, or about 0.3 percent of total employment (see Figure 2 in his testimony). If his estimate of the cost-savings channel is accurate, and I am right that the overall labor market effect of the act is about 10 times larger (in the other direction) than the cost-savings channel, we might then expect the act to contract the 2015 labor market by about 3 percent rather than expand it.
As time goes by and additional research results become available, it increasingly appears that even the experts failed to fully appreciate the labor-market-depressing effects of the Affordable Care Act at the time it was passed.
NY Fed's Economic Data Release Calendar