Introduction
While economic policies in previous decades mainly revolved around individual support (i.e., providing money or services to low-income individuals or families), recent years have witnessed a surge in interest in “place-based” policies. Programs aimed at helping individuals escape poverty can only do so much. We also need to consider poor areas, given that only some are willing or able to move to affluent areas.
Under the Biden administration, there has been a flurry of “place-based” programs. These programs have been integrated into various legislations, including the American Rescue Plan, Infrastructure Investment and Jobs Act, and The CHIPS and Science Act. The Brookings Institute estimates that the 117th Congress has funded 18 place-based programs, totaling over $80 billion in spending.
There are two broad rationales for place-based policy. One is to take advantage of agglomeration effects to create effective regional hubs for industries. It’s not enough for a single factory or a single company to be located in an area. Broader industries are necessary so that people can move between different companies working in the same location.
The second reason mirrors general anti-poverty policy: that there are many poor people in poor regions. Not only that, but many have raised concerns that geographic concentration of poverty results in further poverty since it can result in social norms that reinforce behaviors that result in poverty. As such, people need help more; conceptually, every dollar spent goes a little further. But is this true?
Measuring impacts where they hit
One way of assessing this is a framework recently proposed by researchers at Harvard’s Opportunity Insights, called the “marginal value of public funds” (MVPF). MVPF is a way of calculating how much the next dollar spent on a program will benefit people. The MVPF is calculated by estimating peoples’ willingness to pay (how much a person would pay for the program benefit) and dividing that by the net fiscal cost of the program. The net fiscal cost includes the “upfront cost” of delivering a program andany secondary costs or benefits (for example, if a program increases wages, you would need to incorporate future tax revenues into the cost).
We recently completed a review of “geographic” MVPFs, that is, MVPF estimates that can be tied to a specific geographic location. Geographic MVPFs include estimates from a program conducted only in a single state and estimates from a study across multiple sites, where the program benefits and costs were measured separately for each site, such that you could calculate a separate MVPF for each.
The results are limited: out of 133 MVPFs in the policy impacts library, we could only link or calculate geographic MVPFs for 63. But even given that limitation, it is surprising to see little to no impact of geography on MVPFs.
For example, Figure 1 below shows MVPFs against the unemployment rate during an intervention. Because MVPFs are a ratio, a low or zero-cost program can have a very high or even infinite MVPF. As such, the graph is bounded at an MVPF of 5 (any finite MVPFs above 5 are plotted at 5) and 0 (negative MVPFs are plotted at 0), and MVPFs with infinite values (i.e., programs that would pay for themselves) are graphed separately above.
Figure 1 – MVPF by Unemployment Rate
Across all studies in our sample, there is no relationship between the MVPF and the unemployment rate.
This is true even with the inclusion of two high MVPFs with associated unemployment rates in the 30s – both from an evaluation of the Great Depression-era “Civilian Conservation Corps”. Because 30%+ unemployment is starkly higher than the rest of the sample, these two observations pull at the relationship, but still do not achieve statistical significance.
Move to opportunity (but don’t move to Chicago)
Looking at individual studies instead of the broad aggregate can help clarify what is happening. For example, one of the studies for which we calculated multiple MVPFs is Moving To Opportunity (MTO). This was a program in which families in public housing received vouchers that would pay them to move from high-poverty neighborhoods to low-poverty neighborhoods. MTO was tested as a randomized controlled trial, in which some eligible families received the voucher and others did not, which allowed us to estimate the programs’ effects. Because families were already receiving public housing (and the voucher had the same approximate market value), the only additional cost of the program was the counseling services and moving assistance.
MTO is widely considered a success. Children who were younger than 12 when they moved saw a 31% increase in their adult wages compared to children assigned to the control group. That increase made the MTO program effectively pay for itself–the subsequent increase in taxes was more than the cost of the program itself. In other words, it was a program with an infinite MVPF.
But that’s only true of the MTO program as a whole. MTO was implemented in five cities: Baltimore, Boston, Chicago, Los Angeles, and New York City. If we calculate the MVPF in each city instead of the aggregate program, we see that the infinite MVPF only happens in three. The Boston MTO program had an MVPF of 1.2 (in other words, the program’s benefit was only slightly greater than its cost). In Chicago, average adult earnings for children who moved were actually negative, with a final MVPF of -3.2.
It’s unclear why MTO was so successful in some cities but failed in Chicago. It may be attributed to a unique city feature, a difference in program implementation, or simply happenstance. If it’s something unique to Chicago, what that might be is unclear.Chicago indeed had a somewhat lower poverty rate than the other cities in 1995 (14.7%, while the other cities ranged from 17.7% to 24%).1 Still, its unemployment rate was in the middle of the pack (5.1% in Illinois, while the average unemployment rate in other sites during this period ranged from 4.7 to 7.4%)2.
Do you know the way to San Jose?
Another program we found geographic MVPFs for was JOBSTART, a vocational training program tested in the 1980s.
JOBSTART was tested in 14 sites, giving us a wealth of geographic information. Unfortunately, only two sites had MVPFs greater than 1 (an MVPF less than one suggests that people would have been better off if they just received the cost of the program in cash). The Dallas program had a MVPF of 2.1, and the San Jose program had an astonishing MVPF of 13.9. The average San Jose participant saw their annual income go up by $6,715 (a net present value of $28,741 throughout a lifetime).
Again, it’s unclear what, if any, unique factor made the San Jose program so successful. During the intervention period, California had an unemployment rate of 6.8%.In comparison, other states with JOBSTART trials had unemployment rates that ranged from 3.8 to 8, with an average of 6.9.
Because the JOBSTART program tracked the cost of delivering the intervention, we can look at the costs and benefits separately. Surprisingly, not only was the San Jose program the most effective, but it was also the cheapest, costing $2000 to deliver per participant. That’s less than half of the average cost of the program in the other sites ($4,900).
There is generally a negative (though not statistically significant) relationship between the program’s benefits and costs generally, as shown below in Figure 2.
Figure 2 – Willingness to Pay versus Upfront Costs, JOBSTART locations
Conclusion
Our data suggests that, in conducting place-based policy, we should be attentive to how the program costs shift across geographic areas just as much as benefits.
In our aggregate data, program costs increase slightly more than benefits in areas where we expect programs to be especially effective. It’s possible that areas of low economic mobility– even if they are the places that need jobs the most–might also be where it is hardest to recruit trainers or program managers. Getting trainers and managers to move from a high-opportunity area to a low-opportunity area may require paying them premium wages, which may be enough to cancel any increased benefits.
These findings do not suggest that place-based policies are ineffective, but they do indicate that they should be implemented carefully. Ongoing place-based efforts shouldn’t be designed just to improve outcomes for low-opportunity areas. They should also help us better understand why place-based programs might be effective in some areas and not others. This requires careful accounting of how costs vary over locations, and deliberately taking a broad approach such that programs are attempted across a wide range of economic conditions so we can learn what actually works.
- “State and County Estimates for 1995,” United States Census Bureau, accessed April 17, 2024, https://www.census.gov/data/datasets/1995/demo/saipe/1995-state-and-county.html. ↩︎
- Author’s calculations. State unemployment data from U.S. Bureau of Labor Statistics, retrieved from FRED, Federal Reserve Bank of St. Louis. ↩︎