Unintended consequences and HIV testing

I’m trying to blog more actively, and I figure that a good way to start is to write small pieces about research that I read.

A new working paper by Yang et al. (2021) suggests that interventions that aim to improve awareness about HIV/AIDS can backfire by reducing testing rates.

They start with a theoretical argument. People at higher risk are more likely to get an HIV test. Others might see the decision to get an HIV test as a signal of high risk, and hence choose not to interact with them (stigma). The level of stigma increases if HIV is perceived as more transmissible. Anticipating this, people at higher risk become less likely to take a test as perceived transmissibility increases.

This is an incomplete model in many ways. For instance, a negative test – if that information is made public – could cause people to stigmatize someone less than the average person even if they’re at higher risk, because they aren’t HIV-positive right now. That is one incentive to test. Another is that people might internalize some of the costs of transmissibility – so if, for example, you’re in a relationship, then more transmissibility might increase the benefit of knowing your test result, so you know if you need to take precautions. In such environments, it’s also possible that others can tell whether you’re at high risk through ways that aren’t just testing information.

To help fix some of these problems, they look at a specific PEPFAR-supported program called Força à Comunidade e Crianças in Mozambique. It is “a community-level program that implements home visits to households, as well as complementary interventions in communities and schools.” They test the program’s efficacy in an RCT involving 3,700 households among 76 communities. Half the communities were in a treatment group and half were in a control group, and, among the former, a subset was given stronger encouragement to participate. Using administrative data on HIV tests, the study compared households that received stronger encouragement from within the treatment group and households in the control group on the likelihood of receiving a test two weeks after the endline survey.

They found that “the FCC program has a negative effect on HIV testing: -10.9 percentage points, relative to a base of 26.3 percent in the control group.” Indeed, treated respondents became more likely to believe falsehoods about transmission. But this data doesn’t prove the theoretical argument the paper outlines. They used a different strategy to do that: “Immediately after the endline survey, our research staff then randomly assigned households to a set of ‘minitreatments’ aimed at encouraging further HIV testing, or a minitreatment control group. The different minitreatments provide HIV-related information, seek to alleviate concerns about HIV-related stigma, and provide additional financial incentives for HIV testing.” They find that minitreatments that improve information or reduce stigma can help offset the negative effect of the treatment on testing.

The basic results are here (from table 2, page 40):

This is from Table 2 of the paper linked above, showing the results, finding a negative effect of the treatment, and that minitreatments can help offset this effect.

This doesn’t distinguish between minitreatments, so here’s a table looking at individual minitreatments (from table 5, page 42):

This is from table 5 of the paper on page 42, comparing different minitreatments. The anti-stigma and financial incentive minitreatments all have positive effects, and all the minitreatments except one help offset the negative effect of the treatment.

Perhaps unsurprisingly, the financial incentive – the high-value coupon – has the largest effect on coupon redemption, but the anti-stigma treatment also seems to have a small positive effect. The other minitreatments seem negative in terms of testing, but they do seem effective at offsetting the negative effects of the treatment.

This research also raises some potential ethical concerns about RCTs – is it ethical to subject people to treatments that could be negative? I’m not compelled by the ethical objection to RCTs here, since this is a program that exists anyway and information about how we respond to it is useful. I’ll also note that HIV testing is only one variable, so we can’t automatically conclude the treatment is negative.

I found this paper an interesting example of well-intentioned policies not aligning incentives well enough.

Minimum wages in a monopsony

A well-known result in microeconomics is that a monopsony in a labor market produces a deadweight loss (assuming it does not wage-discriminate), leading to the counterintuitive result that some binding minimum wage levels might increase both employment and wages. 

A graph illustrating the welfare costs of a monopsony.
The deadweight loss of a monopsony. CC BY 2.5 (https://creativecommons.org/licenses/by/2.5/), SilverStar at English Wikipedia.

(I promise we won’t be talking about this graph at all. It’s just for the aesthetic.)

When initially learning economics, I found it hard to grasp this concept intuitively. I recently consulted Perloff (2020), Varian (2014), Landsburg (2013), and other textbooks searching for an answer that explained this without mathematics or graphs and didn’t find anything – so I thought I would try to articulate my understanding of this. I’m not sure this explanation is correct: it’s just my attempt at roughly translating the math to English, and I would appreciate if anyone could correct me on inaccuracies. 

A monopsonist is a single consumer of a good or service, often in a market with many suppliers. In a labor market, that is a single employer, with many workers willing to supply their labor. 

In a competitive market (i.e., many employers and prospective employees), standard economic theory suggests that imposing a binding minimum wage would decrease employment, because firms would lay off workers to compensate for the increased cost of labor. The empirical research is mixed on this question (and indeed, on the question of whether labor markets are competitive or monopsonistic/oligopsonistic). Here’s a good post on this subject by Trevor Chow.

A monopsonistic market – as you may have guessed – is more complicated. So let’s (re)build a model. I’ll outline some assumptions first. This is a single labor market – for example, a labor market in a particular type of worker (someone who does the same job). The employer pays every employee of this type the same wage. This seems reasonable – since workers do not have an incentive to accurately report their reservation wage (the lowest wage at which they are willing to work), a monopsonist does not have a metric by which to wage-discriminate. Furthermore, there are tangible barriers on many blue-collar employers from wage discriminating – including unions, legislation, fear of backlash, common labor contracts, and so on.

Back to a competitive labor market for a moment. In a competitive labor market, with many employers competing for workers and many employees competing for jobs, the wage in the market will end up (1) being the same and (2) equalling the marginal revenue product of labor (MRPL, which is the marginal increase in revenue from hiring an additional worker). (1) is true because if one employer pays a higher wage, every other employer will lose out on workers and be forced to raise their wages. If one employer pays a lower wage, they will either lose access to their workers (forcing it to raise its wage) or operate at a higher profit than others (forcing others to lower their wage). In the end, wages will normalize to be about the same. This is the Law of One Price. (2) is true because an employer is making a loss if the wage is significantly above the MRPL, while if the wage is significantly below the MRPL, the worker will find an alternative job (since there are many employers competing). 

This isn’t true in a monopsonistic labor market. If workers have no (or few) alternatives, then an employer will pay a wage less than the MRPL – because they can. This is a simple syllogism: the profit-maximizing wage for a firm with no competition is less than the profit-maximizing wage for a firm that does compete for workers, and we’ve established that the profit-maximizing wage is the MRPL for a firm that does compete. Now, remember: the MRPL is how much the firm values the worker (approximately) – it is how much a marginal worker would bring in. But a firm has chosen a profit-maximizing wage less than how much it values a marginal worker. That suggests that there are workers willing to work for a wage just incrementally below their additional revenue product, but who are not willing to work at the monopsonistic employer’s current wage. The monopsonist has a choice. Either they raise wages for everybody (since wages have to be the same, per our assumptions) to get these people to work, or they don’t hire some workers (hiring whom would give them a surplus) because it would mean raising the costs of keeping everyone else. Hence, many monopsonists would choose not to hire these workers.

This is a deadweight loss. There is a trade to be made that, on its own, would benefit both parties – the workers would get a job a bit above their reservation wage and employers would hire people a bit below how much they value their work. However, the implication it has for the wages of other employees (whom a monopsonistic employer can afford to pay below the MRPL) means the employer doesn’t maximize a profit when it hires everyone willing to work at below the employer’s value for a worker. 

You can now see how a minimum wage set at the MRPL would likely increase employment for such a monopsonist. If the government sets a binding minimum wage at the MRPL for each worker, then a monopsonist would have to pay all its existing workers the MRPL. That means hiring these new workers does not mean raising everyone else’s wage – so the monopsonist makes the decision on whether to hire a new worker solely on whether the wage they’re willing to accept is less than the value they offer the monopsonist. Since there are some workers who meet this condition who were not hired earlier, they are hired now.

This also means a minimum wage set a bit above the MRPL has an ambiguous effect on employment. It has two counteracting effects – one increasing employment and one decreasing it, and which one of these effects dominates depends on how high the minimum wage is set.