What the quality of bureaucracy can tell us about government effectiveness

Guo Xu, Associate Professor at the Haas School of Business, joined the O-Lab Faculty Network in May 2024. Xu joined Haas in 2017, where he researches the organization and development of personnel in both public and private-sector organizations, working at the intersection of political economy, economic development, and economic history. Much of his recent research has examined aspects of bureaucracies, exploring topics including the interplay of political ideology and bureaucrat performance, how large shocks influence norms in the civil service, and the role of bureaucracies in economic development.

We sat down with Guo and discussed key takeaways from his work on bureaucracies and personnel, how he has leveraged data to inform multiple investigations into economic history, and where he’d like to take his research next.

---

How did you become interested in economics research? Were there any particular policy problems or phenomena that you witnessed that initially caught your attention?

I'm a trained development economist, so the question I always had was: why are some places rich, and others poor? When you travel around, you notice these huge income differences – what's driving that? During my undergraduate studies, it was all about institutions – there were all these papers on colonial legacy, and the impact stemming from differences in institutions that were set up. I found that really fascinating. At the same time, I wanted to dig deeper, which led me down the rabbit hole of looking at public organizations and their personnel.

Your primary research fields of interest are political economy, economic history, and development economics. Can you describe your specific research interests within these fields, personnel being one of them?

My work is really at the intersection of these fields. The questions I have in mind tend to be development questions. There's no denying that you need some basic set of public goods for a country or society to function. So a natural question is, “How can you improve public service delivery to stimulate economic development and growth?”

The intersection with political economy comes into play when thinking about policy. Sometimes you know what the right policy choices are, and yet they are not implemented for political reasons. And even if policies get implemented, there remains a huge amount of variation in the quality of implementation. You might see the same policy being rolled out within the very same country - and yet there will be differences depending on how effective the local administration is.

The historical part comes in because these studies often rely on natural experiments such as big reforms or large-scale changes in governance that happened in the past. What's very appealing with the study of history is that you can rewind time and study big events after the dust has settled.

Much of your recent research centers on studying bureaucracy, with evidence from the US Federal Government. Can you describe your interest in studying workplaces of the public sector, as compared to the private sector?

A lot of the toolkits that I use are also used in the study of private sector organizations. Instead of studying senior bureaucrats, you study CEOs and managers. There are many similarities, but what I find particularly appealing about the study of public organizations is simply that they're so big and important for the lives of many. In the U.S., for example, the federal government has been the biggest civilian employer for a long time. The public sector is a huge part of the economy.

Could you go over some of the key takeaways or main findings that you've found from looking at the public sector?

There's often this notion that management in the public sector is different from the private sector. But most personnel practices in the private sector also apply in the public sector. For example, some might say that monetary incentives in the public sector don't work, or that paying people more is not necessarily a good thing – you might crowd out highly motivated people. But most of the research seems to suggest that those differences are not really there once you look closely. Many personnel policies that we've seen in the private sector seem to work, to some extent, in the public sector as well.

To give you an example, there’s typically a notion that bureaucracies are very rigid, and can't really motivate people, because you don't have the same type of systems - performance pay or up-or-out promotions - in place as in the private sector. In a study where we looked at the Indian federal bureaucracy, which is particularly rigid, we see that incentives matter quite a bit, and that there are always margins for incentives to kick in. In that particular setting, it turned out it was not about providing monetary incentives, but promotion incentives that allow individuals to reach the most prestigious positions - the “glittering prizes” - before the mandatory retirement age. This implicit retirement incentive governs how people exert effort throughout their career. There are many indirect ways of providing incentives – for example, by transferring employees to more desirable locations or more prestigious postings. Even rotations – you don't have to promote people, you can rotate them across different jobs – can get you quite far.

I expect that one key advantage of doing research on the US government is the availability of data, but in the case of your paper on federal employment segregation, you undertook an extensive digitization project. Can you describe what this process looks like? It seems like you've been able to use that data to study a couple of different questions.

One major part of doing economic history is trying to unearth old data. At that time, we wanted to study the economic costs of the introduction of racial segregation within the federal government under Woodrow Wilson. We were lucky to find these personnel records that go far back in time. 

Getting historical data into a workable format is a lot of work – back then, we relied on optical character recognition, and then processed the data in a semi-automated way. We did a lot of manual work too – we had teams of people check the data line-by-line. That was only a couple of years ago - I’m curious to see how AI will help speed up these digitization efforts.

Back then, however, it was a pretty big investment in terms of data collection. It took us more than a year to digitize and clean it all. Luckily, we were also able to use the data to analyze other questions. One question that we got from the study of federal employment segregation was: “Why are there so many black civil servants in the first place?” It seemed very unusual because of the ongoing discrimination at that time. That led us to study the introduction of the Pendleton Act a couple of decades earlier.

Before Pendleton, the President could appoint public servants at will - everyone served “at the pleasure of the President.” The Pendleton Act put in controls by introducing rule-based selection, and employees couldn't be fired at will anymore. We studied that rollout, and looked at whether it actually changed the composition of workers and their performance. The idea was that the introduction of rule-based selection would remove potential biases, increasing the entry of individuals from underrepresented groups. One surprising thing we found was that the Act didn't do much to change the demographic makeup of the civil service. So what we thought might be the primary effect of the Pendleton Act turned out to be much less important.

But it turned out that we still saw large performance gains, which we traced to the fact that the Act was reducing disruption in the organization. In the old system of appointments, with every new presidential election, the President would replace existing civil servants with their own political supporters, creating huge layoff cycles every 4 years. With service protections in place, these political cycles began to diminish and ultimately disappear. That, in turn, made the organization more stable, creating a civil service that more closely resembles the stable careers we associate with public sector work today – you can transmit knowledge more, and there's more on-the-job learning. We found that this increased the performance of the public sector, as measured through reductions in postal delivery errors and increased cost efficiency.

We focused on post offices for the study for several reasons: first, the Pendleton Act was gradually rolled out across post offices, allowing us to compare post offices with newly introduced civil service protections to those without. Second, rich administrative data on post offices was plentiful, allowing us to apply modern statistical methods to estimate the effects of the reform. And of course, the post office was a crucial infrastructure through which communication flowed, connecting far-flung areas of the country.

Initially, we expected the reform to have a large impact on getting individuals from underrepresented backgrounds into the government - but it turned out that the question of whether exams would increase the amount of Black employees in the civil service was not that clear, even in the historical record. It was a case where doing empirical analysis yielded some interesting surprises.

How do you see retrospective research as informing future policy?

It's perhaps a little cliché to say, “Oh, you can always learn from the past” - but that’s true. We can’t run experiments all the time, and many policies and events take a long time to fully unfold, making them difficult to study. In that sense, history offers useful episodes and lessons we can look at from a more distant perspective. 

That said, I think economists tend to be wary of extrapolating, unless we fully understand the underlying mechanisms through which certain effects unfold. All empirical research is retrospective and in a sense historical. Historical work is not about the age of the data, but taking a deep dive - a careful study of the context and institutional setting. This also helps to get a sense of how comparable different events in time are and the extent one can extrapolate.

Personally, I get a lot out of studying history, because it's a great way to learn – when I moved to the US, I didn't know much about US administrations. The good thing about this job is that you can choose what you study, and that's very enjoyable.

What kinds of new questions have your research projects raised that you see as important areas for future projects?

The nice thing about studying public organizations is that you can study different functions of the government. One dimension I'm hoping to do more work on is the study of environmental regulation – a big challenge that is tightly linked to implementation. There are federal environmental programs that are implemented in different places, with a lot of heterogeneity. You have that in the US, and there’s a similar structure in India, so there's a lot to be looked at!

Cailin Slattery on cost drivers of US transportation infrastructure

Assistant Professor Cailin Slattery joined the faculty of Haas School of Business in the Fall of 2023. Her work focuses on the relationship between firms and local governments, and on the role state and local incentives play in driving local economic development. 

Slattery is also one of the leaders of O-Lab’s new initiative on US Transportation & Infrastructure Development, which is building a new body of research on what kinds of policies and strategies governments can deploy in efficiently and equitably bringing large-scale infrastructure and transportation projects to fruition. As part of this initiative, Slattery is leading new work on the drivers of transportation costs in the United States. In particular, she and her co-authors are examining how much of the cost of road maintenance and resurfacing costs are influenced by the competence of the managing agency staff. The research builds on work she conducted with co-authors Zach Liscow and Will Nober to survey procurement practices across all 50 states to understand what those working on procurement and maintenance on the ground think about what’s driving the costs, as well as what the actual variation looks like across states. 

We recently sat down with Cailin to learn more about her research and the potential lessons for policymakers that could result from her ongoing work into the drivers of high costs in US road maintenance and resurfacing projects. 

---

Cailin, your work focuses on identifying what’s driving up costs in US road projects - before we get into it, it would be great if you could provide some background on what we know more generally about the costs associated with infrastructure development in the United States. How does the United States compare to other high-income countries?

You're starting off with a hard question – we don't know a ton. There's really good research on transit costs across countries – subways and light rails – but not on resurfacing and road building, which my paper looks at. The US does look like an outlier, in terms of how much it costs to build a mile of subway in New York, versus in Paris or in Rome.

We focus on roads because it’s the bulk of the work that state DOTs are doing – every state is maintaining their roads and their bridges. But not many states at this point are building new roads, and not all states have big transit projects. We wanted to study a type of project that was common across states, so we took the bread and butter, resurfacing, as our case study.

Do we know anything about how infrastructure costs are related to different regulations at the state or federal levels, or to different political processes?

If you look at highway infrastructure, my co-author Zachary Liscow has a nice paper on the cost of road building over time in the United States. It ends in 1980 or 1990, because that's when we stopped building a lot of new roads, but the authors find a significant increase in cost per mile between the 1950s and the end of the century, even after controlling for all the variables they can – things like weather and geography. They also show that increases in labor costs don’t explain the bulk of the road building cost increase. Their main interpretation of this is that citizen voice is making it more expensive to build roads; there were a series of regulations in the 1970s that made it easier for people to get involved in the government process. That’s not necessarily a bad thing – neighborhoods can organize to prevent a road being built near them, or require more environmental review - but that greatly slows the process, which indirectly makes it more expensive.

Your most recent paper - Procurement and Infrastructure Costs - focused on the role of the procurement process in driving costs. You built a large database of state-level procurement rules and interviewed those working on projects on the ground. Can you start by giving a high-level overview on how the procurement process for US infrastructure projects work? And how this could increase costs?

The basic way that it works is that each state DOT (who manage the majority of infrastructure projects)writes up a plan about which projects they want to complete in the next couple of years. They submit that plan, with a budget, to the federal government as an application for federal funds for infrastructure. They'll get some federal funds based on that, which are earmarked to those projects in the proposal. So a lot of money is coming from the federal government going to the state level to build and maintain infrastructure, and then states have their own sources that they complement this funding with.

At the next stage - the procurement stage - the DOT says, “We have this money to resurface this part of the highway – we're going to put out a call for bidders.” For the most part, the government is not the one building and maintaining infrastructure – the private sector is. 

 The state’s question is, “How do we find the right contractor for the job, at the lowest cost to the government?” So they'll run an auction, where all the prospective contractors look at the project plans and submit bids privately. The government then looks at all the bids; generally, they pick the lowest bid, and that's who will build the project. Along the way though, there are many inputs that can drive costs.

For example, before a project is even posted, engineers at the DOT estimate how much it's going to cost, and give very detailed plans to the contractor – because if the contractor doesn't know exactly what you want them to do, how are they supposed to bid on it? And the more detailed the plans, the better – because as you provide more details to the contractor, they have a better idea of exactly how much it's going to cost them, and they can make a more informed bid on the project. If the contractor has an incorrect expectation of what the project is going to look like, they break ground and things are not as they seemed; the contract has to be renegotiated, which causes delays and increases the project’s cost. So putting plans together is a big part of the job of the DOT, and is potentially where you could see costs increase.

Another decision that DOTs make is how to advertise the projects. How do you get people to bid on your project? Are you actually having all the contractors in the market that could potentially do the project? 

 Once the contractor wins the contract, they actually have to complete the work. Now, the DOT has to oversee and manage the private contractor on the project. Here, communication is really important – for instance, being quick to respond to changes that might need to be made to the plans or to the contract. Anything to avoid delays is potentially something that can keep costs down. Anything that has to do with permitting and traffic rerouting is where the DOT is going to get involved. So it's not simply, “we have a procurement auction and someone wins.” When someone wins, there's months or years of repeated interaction as the project is actually constructed. 

What were some of your biggest findings from the survey and the review of state-level rules?

The motivation for the data collection exercise was that we couldn't say, in the aggregate, how much differences in procurement processes— how states procure projects — explain differences in costs. At the end of the survey, we had around 80 variables that we correlated with costs, – and very few of them showed any significant correlation. For instance, how do states differ in deciding how to throw out a bid? It doesn't seem like [they do] at all. The way I think about it now is that there are a lot of rules in the United States, and a lot of standardization across states.

But the [variables] most correlated with costs [from survey and administrative data] were all about the capacity of the DOT, or the competition for the contract. So I think of it as inputs into the procurement process that matter – the capacity of the DOT, as measured by employment, employee quality, and things that are correlated with those two things, like plan specification.

Another factor seems to be competition in the market – how many contractors there are actually bidding on projects.

What I liked about the survey was that it featured free responses. So you can ask the DOT official, or ask the contractor, “What do you think drives costs?” “What do you think of your DOT colleagues?” “What do you think of the DOT procurement process?”

They're very descriptive about these issues. There are many state DOT engineers that are super knowledgeable on all of the rules. Because there’s so many rules, you have to be there for a while to understand all the ins and outs, and what to do in any given situation. If you're new, then you struggle with that. If a bunch of people retire, and we can't fill those spots, then engineers face this knowledge gap and have to learn from scratch. 

Does California’s unique position as a state, and specific infrastructure procurement processes, present any challenges for external validity?

I think the capacity story is a universal one. We know from the survey that a lot of states are dealing with this. I talked to some industry groups when I was in New York, when we first started this project and, and they were saying the same things, even coming from outside of the DOT. So I don't think it's a California specific issue – it's just that California is kind enough to share detailed data with me.

Your new paper goes beyond correlations and focuses on how much the capacity of DOT staff impacts costs. It seems difficult to isolate the effect of specific staff members on the overall costs of these projects. Can you explain a bit about how you’re doing this?

Our plan is to do a fixed effects analysis, which is pretty standard in the literature. You see it applied to studies of teachers, judges – you know, how do you measure a teacher’s value added? Well, you look at all of the students a teacher has, and then students switch teachers, and you can measure what the teacher effect is. So the idea is to do that with the engineers who manage the projects. 

 This type of fixed effects analysis has been done in other procurement settings, mostly in developing countries, to try and understand how much a bureaucrat matters. What is interesting about our setting is that we have two levels of bureaucrat. We know who the person is that is managing the project – and they’re supposed to be managing the project from the inception of the project to when it is completed. They manage putting together the plans, doing the environmental review, dealing with permits – all this stuff that happens outside of the actual building of the road. But for each project, we also know the engineer on the ground, who’s managing the road building. Having these two layers in our setting will help us understand what matters more for driving costs.

Are there particular methodological challenges that come with studying infrastructure projects in this way? If so, how has your paper dealt with these challenges?

 What’s difficult in our setting, which we're still struggling with, is that in any project, things can go wrong that are unobservable to the researcher – like, maybe it just rained for three weeks straight, and that's why it costs so much – not just because the engineer was doing a bad job managing the project. Examining many projects that the same engineer has worked on is an attempt to control for that. But if that's happening all of the time, it adds a ton of noise to the data, and that makes it a lot harder to measure the true value of the engineer. 

The  descriptions we have of the projects are also pretty short. There's a “project type” category, but I wish we knew more about exactly what each project entailed to be able to, to the best of our knowledge, say we're comparing apples to apples here. So that's something we're still working on.

Addressing this involves a lot of going back and learning more about how Caltrans categorizes projects – going into project plans and seeing what they look like, to see how much variation there is across types of projects.

The second part of your project investigates the impact of a program Caltrans has developed to increase staff capacity in local project management. Can you talk about the motivation behind this program, and describe some potential effects of a more or less centralized management process for these types of projects?

Caltrans has a program, the Local Assistance Program, which aims to help localities who have their own Departments of Public Works with some of their projects. Caltrans is interested in knowing how much that helps. One aspect they're interested in studying is having a Caltrans engineer oversee a project, instead of a local engineer. It's not clear in theory what effect that would have.

You might think the local engineer has local expertise, and maybe has a better idea of what the exact specifications need to be – and you'd think that the state isn't going to add much value there. On the other hand, if you think the state engineers have more experience just in general, maybe they've seen a particular type of project before that a local engineer hasn’t, it could help reduce costs. It's an interesting empirical question they want to study.

You can find two very similar looking projects, but one is managed by a state engineer, and one stays local, and compare them. We do that for many projects, and then we could dig into the mechanisms of why getting state support might help or or hurt. I could come up with good reasons in both directions, so I think that that's what makes it interesting. Maybe what we'll find is that sometimes it really works, and sometimes it really doesn't. I think it's a more general question in economics – how centralized do we want to do things? How centralized should we procure, not even roads, but say, pharmaceuticals? If the government was going to procure pharmaceuticals, you might think, the bigger the buyer the better, because they have a lot of buying power and can achieve a lower cost. But with roads, the amount of knowledge it takes, and the fact that there's supervision on the ground – all of these things make it much more complicated.

If procurement rules don’t seem to have a substantive effect on the cost and timeliness of infrastructure projects, what other policy tools might states use to address these issues?

 I think that there are creative policies to be had on the labor market side of how to recruit high quality employees. It used to be that being an engineer for the state was a prestigious and well-paying position, but now the private sector dominates, in terms of wages, and state benefits are not what they used to be. That's more of a structural change related to government jobs versus the private sector.

The other thing that we haven't talked about as much is the contractor's market. Contractors complain there's not enough firms for competition on the subcontracting market. DOTs complain there's not enough competition on the primary contracting market. I think in many cases, firms exist that do this type of work, but perhaps don't know how to apply for government contracts and go through the process. In some situations, the forms that firms have to fill out are hundreds of pages long. One avenue for policy might focus on ways that the state can grow the market of contractors that take on public sector work. There are contractors out there that don't compete for government contracts, and we want to understand why. That’s a  place where there's potential for more research.

Adam Leive on the effects of work requirements in SNAP

Over the last several years, there has been renewed attention to how the United States promotes economic security for low-income families. During the height of the COVID-19 pandemic, expansions to the Child Tax Credit - combined with direct cash assistance and increased unemployment insurance - demonstrated the powerful capacity of the government to deliver robust support that reduced poverty and protected families’ ability to access food, housing, and other basic needs. The immediate and dramatic impact of those policies have lent additional momentum to efforts to strengthen the safety net by increasing our investments in nutritional support and tax transfers.  

While the evidence continues to grow on both the short and long-term benefits of programs like the CTC, the Earned Income Tax Credit (EITC), and the Supplemental Nutrition Assistance Program (SNAP), there is ongoing debate around who is most deserving of accessing the safety net, and how strictly aid should be tied to work. 

Adam Leive, Assistant Professor at the Goldman School of Public Policy, has produced valuable evidence to inform the debate around the impact of work requirements in safety net programs. His 2023 paper, “Employed in a SNAP? The Impact of Work Requirements on Program Participation and Labor Supply” evaluated how the introduction of work requirements in Virginia’s SNAP program affected overall recipients’ employment and earnings, as well as how it affected their participation in the program itself. 

The team found that the introduction of work requirements for adults 50 and under led to no increases in employment or earnings, but did lead large numbers of people to leave the SNAP program entirely. Leive and his research team also found that homeless populations were disproportionately impacted by the change. 

In this O-Lab Q&A, we sat down with Adam Leive, to discuss his findings, the background behind SNAP work requirement policies, his experience collaborating with Virginia’s Department of Social Services, and his perspectives on how research can inform policy.

Source: Social Security Administration

Can you start by talking a bit about the motivation for your study? The question of if (and how much) to make public benefits conditional on work has been around for a long time. Can you give a little background and talk about how this debate has evolved?

To start with the pros and cons, proponents [of work requirements] have been really concerned that providing government assistance ends up discouraging work. So their argument for policies like work requirements is that by incentivizing work, people develop a stronger attachment to the labor market, and then eventually people will earn enough through their job to get by without government assistance.

On the other hand, critics see work requirements as a policy that prevents people who are vulnerable from receiving assistance, ultimately weakening the safety net. So opponents argue that if the reason that people aren't working is something other than the work requirements, then this policy is just going to cut people's benefits without improving their labor market outcomes.

In the US, the safety net has been heavily tied to employment, both incentivizing employment as well as tying benefits to employment. The biggest expansion in terms of work requirements was in 1996 under president Clinton. The official name of that bill was the "Personal Responsibility and Work Opportunity Act," which is better known as welfare reform, and the major change that that bill introduced in terms of work requirements was to make sure that people who are in SNAP who are childless adults and didn't have dependents or disabilities were working for some amount of time in order to receive benefits for more than just a few months.

I think the other thing that's worth mentioning with the '96 reform is that [it] received bipartisan support. Today, work requirements are much more divided, with Democrats largely being against, and Republicans being in favor. In terms of how the debate has evolved, I think there's been a view that some of these conditions can be harmful, and are now something that are not as widely supported the way they were 30 years ago.

What kind of evidence do we have on work requirements, and what new contributions does your study make to this literature?

There's actually been relatively little research on work requirements, when you think about how politically fraught they are. One challenge with looking at the 1996 reform is that so much else changed along with [the work requirements] that it was difficult to tease apart the work requirements from everything else that was going on.

Source: Grey, Leive et al. 2023

There were studies from the Welfare to Work experiments in the 70s and 80s that were really good in that they were randomized controlled trials (RCTs), but they also bundled work requirements with other types of labor market policies, and for the same reasons, also struggled to distinguish the effects of work requirements from some of these other policies. 

More recently, and around the same time we wrote our study, there were other studies on SNAP that used different sources of data that also focused on able-bodied adults without dependents, or ABAWDs. These studies tended to find mixed results on participation and labor market outcomes.

Perhaps the best evidence on work requirements comes from Arkansas's experience with Medicaid. One state implemented work requirements in Medicaid for a roughly 9 month period in 2018, and the researchers studied what happened in Arkansas and surrounding states using survey data — they found that the Medicaid work requirements, which were much more onerous than SNAP work requirements, led to large drops in program participation. People also didn't realize what was required to meet them — many people were already meeting them, but just weren't reporting that they were. That experience was a case study of bureaucratic obstacles, as well as the work requirements themselves.

Source: Grey, Leive et al. 2023

Our study examined Virginia - one state - but used detailed administrative data to try to provide a new, focused look on the topic. We took advantage of the fact that work requirements had been suspended during the Great Recession. Like many states, Virginia suspended them and then reintroduced them well after the recovery, in October of 2013. So, we studied what happened before and after work requirements resumed. Importantly, we could study outcomes at the individual level, using administrative records on people's participation in SNAP and their earnings in Virginia. We linked – through a partnership with the Virginia government – these detailed earnings records with SNAP participation records, and that allowed us to see what happened in a world without work requirements and in a world with work requirements.

At a high level, what we did is we took advantage of the fact that people under age 50, if they don't have dependents or disabilities, have to meet work requirements, whereas those over age fifty don't. So we compared outcomes for those just below age 50 with those just above age 50. The only difference between these people should be their age when work requirements came back. What we then found was that people just below age 50 had much larger reductions in SNAP participation. We looked at people who were on SNAP after the work requirements resumed, and much higher rates of that group left SNAP compared to those just above age 50 who didn't have to meet work requirements. And yet those large participation declines did not translate into any improvements in earnings or employment on average. So if a major goal of the work requirements is to improve labor market outcomes – we didn't see that. We also found that people who are homeless experience disproportionately large reductions in SNAP participation, so they were especially impacted by the policy.

Effect of work requirements on SNAP retention, 18 months after work requirements

Source: Grey, Leive et al. 2023

This is such a political issue. How do you think about the prospect for your findings to meaningfully inform policy in Virginia or elsewhere? Are there enough legislators who are motivated by research like this?

I wish I knew what politicians were thinking, or motivated by, or how they make decisions. But I do think there's an important role for research in terms of generating evidence that can inform these debates. There are often so many studies out there on any particular topic, but I think a very small number of studies have the potential to influence policy. And the studies that do, I think, are really high quality, often RCTs – if it's not an RCT, it has really compelling evidence. So my focus is generating that kind of research, where there may still be limitations, but it's much more with things like external validity, rather than "I don't believe the study itself," or the internal validity of the research. I see value in trying to produce those types of studies that can have an impact, and the way that I try to go about doing so is by testing some main theory or some motivation, for these work requirements. But often what the research does is rule out some explanations, rather than to “rule in” a particular  explanation – that's what the social sciences are well equipped to do. 

So what we've done is say, "Okay, work requirements don't lead to large improvements in employment or earnings. We can feel pretty confident statistically about this, for this population." If we saw a big improvement in employment and a decline in participation, that would arguably be seen as a good thing, or at least it would be consistent with the intent of the policy. So our results indirectly suggest that maybe there are other barriers people are facing. For example, people don't have a stable job that offers hours that don't fluctuate. Other people might have challenges getting transportation to a job. What our study would point to is that looking at those barriers may be important for future work.

Another motivation that is sometimes discussed is that imposing some kind of hurdle can be efficient from an economic perspective, because only the people who need benefits the most will go through the hassles of overcoming those hurdles. So if the government has a limited amount of resources to provide as part of safety net benefits, then these hurdles can perhaps better target assistance by getting people to self-select into the policy. That has been more of a theoretical point that some economists began making in research decades ago. And our results that homeless people are disproportionately screened out, I would say, is strong evidence against that theory. If the most vulnerable people face a higher cost to overcoming these barriers, even if they have higher benefits from the safety net, they might be screened out.

Your paper finds that the reestablishment of work requirements in Virginia disproportionately caused homeless individuals to leave the program. What role does SNAP play in supporting homeless people, and in what direction should safety net debates shift in light of the homelessness crisis?

Homeless people are some of the most vulnerable people in society, and SNAP and other safety net programs can provide lifelines for them. So I think efforts to make it easier for people who are eligible to receive and maintain their benefits are extremely important. I'm not an expert in this particular aspect of SNAP, but I think that making sure that information is communicated via different modalities, like alternatives to mailing addresses, can be important, especially when benefits require a lot of reporting from the client side. In terms of the mechanics of benefit calculations, there's a shelter deduction for people without stable housing that can allow them to increase their benefits, and that's important for people who have some income. People who don't have any income are getting the max benefit, but many homeless people are working, so shelter deductions can help increase those benefits.

Taking a step back on this issue of homelessness, safety net policies that provide assistance for healthcare, for utilities, for rent can be important for preventing homelessness, but those deal with the problems caused by a lack of affordable housing. So ultimately, I really think the issue is much more about increasing the supply of affordable housing as the crucial policy lever. SNAP and the safety net are very important, but we should also be focused on making upstream investments in increasing housing supply. 

Virginia State Capitol building. Source: faungg via Flickr.

How would you describe your experience working with the Virginia state government? How did the collaboration arise? Has the partnership offered any insights into working with governments that might be useful for researchers conducting similar work?

Overall, it's been an absolutely fantastic experience. Before coming to Berkeley, I spent 6 years at The University of Virginia’s Public Policy School and I wanted to produce research on topics relevant to the local community and the state. I'm very lucky to have great partners in Virginia – and as someone who's in a policy school, it's kind of a dream come true. You get to work with government collaborators to study a question that they're interested in answering using tools that you have specialized training in. I was very fortunate that my counterparts in Virginia were people who understood research and its value, and were interested in results regardless of what the outcome was, and who also had the training to be able to articulate why certain data was necessary. There were a lot of “asks” that had to be made in terms of trying to get more information from various government officials, or understand certain parts of a program. To have partners who have themselves been involved in research, many of whom have PhDs, is really helpful because they can advocate for the importance of research, as there can understandably be skepticism about an outside scholar coming in to study a program. I think what you learn is the importance of finding someone who has the decision-making power to say yes, in terms of helping you get the access you need to do the study, and who's genuinely interested in the answer, regardless of what it is.

Do you have any plans to examine how work requirements affect people in the long term? If not, what avenues could you expect other researchers to take?

I'm continuing to collaborate with Virginia's Department of Social Services to help them design ways that the program can best support the SNAP population, and simultaneously improve economic outcomes for recipients. Some outcomes that I think are important to look at in the context of work requirements, that we don't know much about, are food insecurity and health, the most fundamental aspect of what SNAP is supposed to be about. Taking a broader lens, on the benefits side, I think it's possible that some of the effects on health might be very important for long-term outcomes.

Tax Uncertainty, Gig Worker Preferences, and the Impact of Outside Options on Wages

In Fall 2020, O-lab welcomed Sydnee Caldwell as one of its newest affiliates. To introduce her work, we have spotlighted a few of her recent projects, along with their potential implications for understanding the gig economy and the low-wage labor market.

Sydnee Caldwell joined the UC Berkeley faculty in July 2020 as an Assistant Professor in the Department of Economics and the Haas School of Business. Her appointment marks a return for Sydnee, who also attended Cal as an undergraduate, earning a dual bachelor’s degree in applied mathematics and economics in 2008.

Her recent research focuses on different forms of worker compensation and on how uncertainty around the Earned Income Tax Credit (EITC) affects behavior and outcomes for low-wage workers. Prior to returning to Berkeley as a faculty member, Sydnee served as a post-doctoral researcher at Microsoft Research New England. She earned her PhD in economics from MIT in 2019.

Tax Refund Uncertainty: Evidence and Welfare Implications (with Scott Nelson & Daniel Waldinger)

In this 2021 working paper, Caldwell and her colleagues focus on how low-income individuals understand and predict the amount of earnings they are likely to receive through the Earned Income Tax Credit (EITC). The EITC is one of the country’s most powerful anti-poverty programs, and for those low-income families who receive it, the credit can make up a substantial portion of their total annual income. Accurately anticipating the size of the EITC is therefore a crucial part of household budgeting and decision-making for these families. Due in part to the complexity of the tax code, however, it can be very difficult for filers to anticipate their tax liabilities and credits.

To understand the scale of this problem, as well as potential policy solutions, Caldwell and her colleagues used a combination of administrative tax data, credit bureau data, and survey data on individuals’ tax refund expectations. The data was collected through the Volunteer Income Tax Assistance (VITA) tax preparation centers in Boston to gather information about individuals’ expectations of their refund amounts. A low-income tax filer’s expectations were then compared to the actual refunds they received. When asked to predict their EITC refund amount within a $1,000 interval, the authors found that a quarter of individuals were “not certain at all” that their prediction would be correct.

Their research suggests that uncertainty impacts how people choose to consume and borrow; those who were more uncertain going into tax season tended to repay less of their credit card debt after they received the refund, for example. The refund uncertainty distorts recipients’ consumption-savings choices and causes a 10 percent loss in the the value of the EITC.

Uber vs Taxi: A driver’s Eye View (with Joshua Angrist & Jonathan Hall)

As the gig economy continues to grow, understanding worker compensation models and worker preferences will be critical to crafting fair rules and payment structures. In this paper in the American Economics Journal in Applied Economics (forthcoming 2021), Sydnee Caldwell and her co-authors partnered with Uber to compare drivers’ preferences under two different potential compensation models. In the first, drivers pay a fixed payment independent of their earnings; this compensation structure is the standard model for most conventional taxi drivers. In the second model, rideshare drivers pay a portion of their earnings per ride, a commission that Uber charges drivers for every ride drivers provide; this second model is the standard model for most “rideshare” programs. Caldwell and her colleagues find that a taxi-like model is (a) more efficient, and (b) results in higher net wages for drivers, since drivers keep any income they make in excess of the fixed payment.  To explore if Uber drivers would prefer the taxi-like option, the authors partnered with Uber to offer drivers an opt-in taxi-like contract where they could pay a fixed payment instead of the usual Uber commission fee per ride. Sydnee says, “we find that drivers who would have earned higher wages under the taxi-like model don’t opt in. Our explanation for this is that drivers are loss averse.” In other words, drivers are averse to the scenario where they cannot make enough trips to cover the fixed payment and will give up the possibility of higher earnings in favor of greater flexibility to work as much or as little as they want. The authors find that drivers were often uncertain about how much they were going to work in a given week, which likely contributed to their aversion to the fixed payment since they were unsure about their ability to drive enough to cover the lump sum cost. 

The results of this paper will be important in understanding and creating rules to govern “gig jobs” and other emerging labor markets where workers can choose how much to work. There is a serious ongoing policy debate about how gig workers should be classified---whether as independent contractors or as employees---and how they should be compensated. “When we started the project,” Sydnee said, “we expected to see that drivers who drive a relatively high amount of time would prefer the “taxi-like” scheme, and that drivers who drive a relatively low amount would not. However, we found that…these taxi-like schemes are perhaps less popular than you would expect, even among drivers who would have benefited a great deal.”

Outside Options, Bargaining, and Wages: Evidence From Coworker Networks (with Nikolaj Harmon)

In Caldwell’s job market paper, she sought to understand the extent to which changes outside of an individual's labor market impact the wages that they may earn at their current job. For example, would a change in a firm competitor’s demand for workers, or a change in their salary ranges, influence wages at one’s current employer?

Using employer/employee data from Denmark, Caldwell and her co-authors find that consistent with economic theory, demands for labor at one firm will induce greater firm-to-firm mobility and will drive up wages at competing firms. These increases were driven largely by information sharing, and were most pronounced among those in the top half of the skills distribution. Access to information about increases in demand for workers at one firm is critical for workers in another firm to leverage the information and increase their wages at their current firm. An interesting finding of this paper - and one which Sydnee hopes to explore further in the future - was that within each skill group, women benefit less than men from information about outside options. Caldwell said there are many reasons why women might be getting a smaller bump from outside options, including the potential that they are less inclined to negotiate or are punished for negotiating. These options are the subject of a growing literature in economics, and questions Caldwell hopes to look into more in the future.

The Impacts of Raising the Minimum Wage

Can a higher minimum wage reduce “deaths of despair?” And who benefits most when the minimum wage goes up?

Michael Reich has spent much of his career researching how minimum wage policies impact low-income Americans. His work has changed policy, having been cited by former President Barack Obama in a State of the Union speech and contributing to labor reforms across the country. Most recently, he helped advise New York City in regulating how Lyft and Uber drivers should be paid.

Over the past year, Professor Reich has co-authored two new studies examining how minimum wage policies affect specific groups of low-income Americans. One report, co-written with UC Berkeley colleagues William Dow, Anna Godoey and Christopher Lowenstein, asks whether increases in the minimum wage and the Earned Income Tax Credit (EITC) can reduce the numbers of “deaths of despair” – from alcohol, drugs or suicide – among low-wage workers. The team found that higher minimum wages do in fact lower deaths from alcohol and suicide but do not lower the number of drug-related deaths. The study also found that a higher EITC does not significantly reduce deaths of despair as much as minimum wage increases do.

Another study with Godoey examined how increasing the minimum wage to a $15 an hour level impacts low-wage workers at sub-state levels such as counties. The study didn’t just look at the effect of an increase in the absolute wage rate, but also at the effect of raising the relative minimum wage, or how the minimum wage in an area compared to its median wage. The report found that higher minimum wage levels, even in low-wage areas, don’t raise unemployment among low-wage workers even when the relative minimum wage was above 80%. The study also found that those who benefitted the most from increasing the minimum were less-educated workers living in areas with high relative minimum wages – in general, poorer regions with lower overall incomes.

These studies are particularly timely, as momentum has grown around efforts to raise the federal minimum wage to $15, and with nearly all of the 2020 Democratic presidential candidates supporting such an increase.  They also build upon a long and influential body of research on the minimum wage coming out of UC Berkeley, from David Card’s widely-cited work to Claire Montialoux and Ellora Derenoncourt’s more recent contributions on the role of the minimum wage to promote racial equity.

In February, the Opportunity Lab discussed some of this recent work on the minimum wage with Professor Reich at the Institute for Research on Labor and Employment (IRLE), where he serves as the co-chair. Here is an edited record of that interview.

What are the innovations of your latest research on relative minimum wage increases?

Michael Reich: I’ve been studying low-wage labor markets for most of my career. My past work with other co-authors has been influential in turning around the academic community, the economics community, on the effects of the minimum wage on unemployment and showing in a more rigorous way than any before that the moderate minimum wages we had in effect through the 2000s and the policies since then have not really had a significant effect on unemployment. Theory tells us that minimum wage increases should hurt the number of jobs, but that’s a very incomplete model partly because there are all kinds of frictions in labor markets - the cost of hiring people, the cost of employers and employees looking for a good match. We want to look for the effects of a policy on a whole economy and not just in one labor market. People are mainly affected by the minimum wage in just a few industries - restaurants, retail, childcare, elder care, janitors; there can be adjustments in prices in those industries.  If the price in restaurants goes up a little bit, you would not expect jobs to have to disappear. Heads of lots of burger companies have said, “I would rather sell more hamburgers than fewer. If I have more workers I can sell more burgers.”

Laws raising the minimum wage to $13 go much higher than what previous research has looked at, so in our paper on minimum wage effects in low-wage areas, we were concerned with saying what happened when the relative minimum wage is much higher than what we’re used to, not 35% or 40% but 50%, 60% even 80%. If the minimum wage went up to $15 federally in the whole U.S., which was the bill the House (of Representatives) passed, then the ratio of the minimum wage to the median wage would be 80% in Alabama. That sounds scary to many people including economists. How could an economy withstand such a policy effect without a negative effect on employment? We looked at sub-state levels and what we found is that when we just looked at the counties where the ratio of minimum wage to median wage was 83%, the highest quartile of the most affected counties, the minimum wage did not have an effect on unemployment.

“Even in counties – not just in Mississippi or Alabama but some in Fresno or rural parts of many states – where the minimum wage relative to the median wage was 80%, it did not have a negative effect on employment

The big take-away is that even in counties – not just in Mississippi or Alabama but some in Fresno or rural parts of many states - even when the minimum wage relative to the median wage was 80%, it did not have a negative effect on employment.

There are other effects besides the employment effect. One of the outcomes we look at is child poverty and we find that an increase in the minimum wage has a big effect on child poverty and that’s as important as whether it has an effect on jobs.

How has the low wage labor market changed?

Reich: The federal minimum wage is still stuck at $7.25 where it’s been for 11 years. That’s a decline of 30% in real terms. That’s pretty shocking to have a minimum wage that goes down over time in real terms. On other hand, 24 or 25 states are raising the minimum wage above the federal level and in some states, cities can do that as well. In California, it’s over two dozen cities. The cities are showing the way, showing it’s possible. The sky didn’t fall and that gives more courage to state level policymakers and same with the state to the feds.

How has the labor market changed over time? Over 10 and 20 years, it’s changed from being much less a youth market, so the stereotype that many opponents have is that most minimum wage workers are young people in high school, in their first job, gaining skills and discipline and you don’t want to discourage that by raising the minimum wage but in fact the composition of the low wage labor force has changed. Ninety percent of low wage workers are now over 25, quite a few are women and quite a few have children. It means that the importance of raising the minimum wage is greater for families and for their children.

Are there any reasons behind the changing demographics of minimum wage workers?

Reich: My sense is there’s a decline in the real value of the minimum wage. People are more desperate, adults are more desperate and need to take jobs. Something like 10% to 20% of the fast food work force have college degrees. On the other side, with teens, they’re staying in school longer so they’re not taking jobs after school. They are going to summer programs so there’s been a real fast decline in teen employment.

What about Uber drivers and others working in the gig economy? How much are they earning in relation to the minimum wage?

Reich: We found that drivers in New York City, quite a few working full time are making less than the minimum wage. They’re making more like $12 an hour after expenses. There’s just a lot of people willing to work for less because they’re desperate.

Do minimum wage increases impact people earning just above the minimum wage?

Reich: When the minimum wage goes up, we do see increases in pay for people who are above the new minimum because you want to keep some equity. If you’re an assistant manager of a fast food store and you’re paid $15.50 an hour and the new employee gets $15, the employer will want to raise your pay to keep some kind of equity because you’ve been there a while. That effect tends to evaporate just a couple of dollars above the new minimum wage but it still has an effect on reducing wage inequality.

What were the advantages of using more localized data in your study?

Reich: In each state, there are well-off and less well-off counties. There’s a big difference between Fresno and San Francisco in minimum wages. In most states, there’s one minimum wage in the entire state so this gave us more variation to work with. Particularly, all states that don’t go above $7.25, they weren’t playing a role in other people’s studies because there was no minimum wage change there but because we were using the ratio (of minimum wage to median wage), which is affected by the median wage, we could exploit the variation that existed, say, between Philadelphia and rural Alabama.

The second study taps into the larger question of rising death rates in the U.S. What is the impact of the EITC or minimum wage on that?

“Ninety percent of low wage workers are now over 25, quite a few are women and quite a few have children. It means that the importance of raising the minimum wage is greater for families and for their children

Reich: Deaths of despair just cried out to be looked at because of public awareness of how U.S. life expectancy rates are in decline and the big jumps in death rates especially among mid-wage people and not just in rural areas. The longer-term outcome is wages haven’t been rising very much for men.

They’ve been rising for women but not enough. Our question isn’t as much ‘What’s causing deaths of despair but what can we do about them?’

We found really positive effects of both the minimum wage and the EITC on reducing suicides especially among women. We did not find many effects on drug overdoses. We used many tests to make sure we’re looking at causal effect and not correlation.

What are some of the key policy implications of this research?

Reich: Minimum wage is not a panacea. Housing costs are rising really fast. The minimum wage isn’t going to have an impact on that. When the labor market is tight anyways and there’s any loss of jobs and immigrants are being scared away to work in restaurants or landscaping or construction, I would think the minimum wage isn’t so binding. Many restaurant workers are paid well above the minimum wage because employers have to keep people.

The big policy context is that there are many states that have passed high minimum wage laws, that Democrats are pretty united among presidential candidates to raise the minimum wage from $7.25 to $15 but that would be phased in over 15 years. The entry level wage is higher in many places than the minimum wage. The entry level wage is well above the $7.25 minimum wage, more like $9 for very recent high school grads. So the jump to $15, if it was phased in and starting from an entry level wage, is not all that steep as a lot of people think and it does have these positive effects including on health. It’s an issue that the Democrats and Republicans are divided on and it’s going to come up as an issue in the fall and it’s on the ballot in Florida, where there’s an initiative to go to $15. We’re concerned about inequality, and the minimum wage is the most effective tool in terms of the number of people who are covered.

“We’re concerned about inequality, and the minimum wage is the most effective tool [for reducing it] in terms of the number of people who are covered.