Monday 31 March 2014

Is sex like driving? HIV prevention and risk compensation

The title of this post is the title of a paper last month in the Journal of Development Economics (ungated version here), by Nicholas Wilson of Reed College, Wentao Xiong of Harvard, and Christine Mattson of the University of Illinois, Chicago. I posted a couple of weeks ago on offsetting behaviour, so I thought I might address this paper as well.

There has long been some concern over the extent of offsetting behaviour that may result from new health interventions that reduce the risk of HIV transmission. In public health circles, this offsetting behaviour is known as "risk compensation", and basically it works the same way as I described earlier. If a new health intervention or health technology (e.g. adult male circumcision as in the Wilson et al. paper, or a hypothetical HIV vaccine as in this paper by myself and others (ungated earlier version here)) reduces the risk of acquiring HIV infection, then rational (or quasi-rational) people face lower costs of risky sexual behaviour. They will respond by engaging in more, or riskier, sexual activity (e.g. more unprotected sex). This reduces the impact of the health intervention, making it less effective overall (in terms of its reduction in HIV transmission in the population).

The authors investigate the impact of adult male circumcision on subsequent risky sexual behaviour using a randomised controlled trial (RCT) among men in Kisumu, Kenya (in some future post, I aim to talk about the wonderful world of RCTs in economics, but for now if you are interested I suggest you read this article from The Economist last year). One interesting and important aspect of this study is that they investigated the effect of risk compensation depending on whether the respondent believed that male circumcision was effective in reducing their risk of HIV infection (since those who don't believe it is effective won't engage in risk compensation). In other words, they compare the effect of circumcision on the 'believers' with the effect on the 'non-believers' (as well as a control group of uncircumcised men). I initially thought this was a good innovation, but on reflection I' not so sure (more on that later in the post).

Surprisingly (or unsurprisingly, depending on what literature you have already read on the topic, if any), the authors find no evidence of risk compensation. In fact, they find quite the opposite effect:
The results of our empirical analysis suggest that the behavioral response to circumcision among believers net of the response among non-believers was a reduction in risky sexual activity. That is, we find what appears to be a behavioral response that is the opposite of the risk compensation hypothesis.
How to explain this result? The authors suggest that the marginal cost of risky behaviour actually increases, unlike the traditional Peltzman effect where marginal cost decreases, and this leads to less risky behaviour. Why would marginal cost increase rather than decrease? There are two components of marginal cost of risky sexual behaviour. First, there is the chance of getting infected by HIV - based on various studies, this decreases when a man is circumcised (and remember that in this study they are comparing the men that believe this with those that don't). Second, there is the cost of dying young (foregone income, time with family, leisure, and all the good things of life, etc.). The authors argue that, because circumcision reduces the lifetime chance of acquiring HIV, it substantially increases life expectancy. So, while the first effect (lower chance of infection) reduces the marginal cost of risky behaviour, the second effect (higher life expectancy) increases the marginal cost of risky behaviour and more than offsets the first effect. They also have some evidence to support this, which arises because of the way they organised their research by beliefs:
...our results indicate the existence of a behavioral response that was not due to a perceived reduction in the HIV transmission probability. Namely, circumcised males who did not believe that circumcision is effective at reducing HIV transmission appeared to increase their risky behavior.
Of course, if you believe that people are rational, there is an alternative explanation for these results. If people in the trial have complete (or near-complete) knowledge (as we might assume that rational decision-makers are), then they will know beforehand that when their risk of HIV infection decreases, they will respond with riskier behaviour (we could call them 'rational risk compensators'). That is, they foresee their own risk compensation. So, when these 'rational risk compensators' are asked: "Do you believe that male circumcision increases, decreases, or does not influence your risk of acquiring HIV?" (the actual question used in the Wilson et al. study) they may answer "increases" or "does not influence". This would erroneously categorise the 'rational risk compensators' into the non-believers category (when in fact they are believers, but also believers in risk compensation). Then, when we look at the data on the non-believers category (which includes the 'rational risk compensators'), we might observe an increase in risky behaviour within that group following circumcision (provided the number of 'rational risk compensators' is large, relative the the number of true non-believers). This story is also consistent with the data in Wilson et al., including their results that suggest that circumcision (independent of the effects on beliefs) is associated with an increase in risky behaviour.

Indeed, when the authors ignore the difference between believers and non-believers, there is some evidence of risk compensation, with men less likely to say they always use a condom or that they used a condom the last time they had sex (both indicators of lower risk behaviour). To eliminate this alternative explanation for the results, it would have been interesting to see whether the research participants believed beforehand that they would engage in risk compensation.

So, given this alternative explanation, maybe the demise of risk compensation has been greatly exaggerated.

Addendum: Risk compensation in children's playgrounds (Conclusion: If you make the playground safer, children respond by playing harder leading to more long-bone injuries like broken arms or legs).

Sunday 23 March 2014

Cornering the market for Christmas toys - an application of elasticities

This week in ECON100 (and next week in tutorials) the class is covering elasticities. So, I thought it might be timely to talk about this video. Now, I'm not a big fan of The Office, but that scene makes me laugh. Mainly because, I've thought about doing this exact thing many times. At least, I thought about it many times in the aftermath of the Great Christmas Bakugan Crisis of 2008 (my son, like many others, missed out on Bakugans, and had to wait all the way until his birthday - oh, the injustice!).

Here's how it works. Dwight does some market research, and identifies the Christmas season's hottest toy: the Princess Unicorn doll. He then whips around the local stores and buys up all of the available stock in town, before raising the price and turning a profit. What Dwight is relying on is a change in the price elasticity of demand of parents. When time horizons are long (long before Christmas), parents have time to shop around. This means that their price elasticity of demand is going to be lower (more elastic). They're not willing to pay a high premium for the Princess Unicorn. And, because people tend to procrastinate, that gives Dwight the opportunity to buy the stock.

But as Christmas approaches, time horizons for Christmas shopping get shorter. With less time available for shopping around, parents' price elasticity of demand increases (less elastic, more inelastic), and they become willing to pay a higher premium for the Princess Unicorn. Voila! Profits to be made for the entrepreneurial Dwight. But only if Dwight has market power - he needs to have some control over the price. Which is guaranteed if he is the only seller of Princess Unicorns in town.

Unless... If Princess Unicorn dolls are available online, then there is a perfect substitute available for Dwight's Princess Unicorn dolls (a Princess Unicorn bought online is the same as one purchased in a store). The availability of a perfect substitute reduces the parents' price elasticity of demand for Dwight's Princess Unicorns, and eliminates the premium they are willing to pay. This leaves Dwight penniless, sad, and with a large pile of dolls.

OK, maybe it's not all bad news, since delivery time means that the Princess Unicorn bought online is not a perfect substitute for one bought from Dwight, if buying online means it will arrive too late for Christmas morning. But clearly that reduces Dwight's window of opportunity (since he can only sell at a premium after the delivery window for Christmas has passed) and his profits (since there are substitutes available, albeit imperfect substitutes).

[HT: I was reminded of this scene on Dirk Mateer's excellent website last year]

Sunday 16 March 2014

Why are weddings so expensive?

This Washington Post blog post by Caitlin Dewey caught my eye, talking about a US$99 wedding dress. This lies in stark contrast to weddings in general which are terribly expensive (these ones are totally out of hand). Why?

Let's start with the simple explanation, and let's stick for the moment with wedding dresses (rather than wedding venues, catering, flowers, and other costs). If this was a story about supply and demand, the high price could be caused by high demand, or low supply. I'm not convinced there is high demand for weddings - the number of weddings is declining over time. Of course, we should consider demand in comparison to supply. There could be low supply because of barriers to entry into the wedding market, stopping potential suppliers from entering the market and driving the price down. Again, this seems unlikely unless there are wedding-dress-specific tailoring skills that are in short supply (this suggests not). You might think that wedding venues may plausibly have barriers to entry, but there are plenty of beautiful places that could become wedding venues if the price rises enough. So, the price here is not a result of a simple supply-and-demand story.

Dewey's blog post talks instead about signalling. But signalling by whom, to whom, and of what? Let's take a step back and think about the purpose of signalling.

Signalling is a solution to a problem of asymmetric information. This happens when one party (the informed party) has private information that the other party (the uninformed party) doesn't know, and (importantly) the informed party uses that information to their advantage and to the detriment of the uninformed party. The classic example that we use in ECON100 and ECON110 is the used car market. Sellers know the quality of the car, but buyers don't. Since buyers don't know whether they are being offered a good car or a lemon until after they have bought it, sellers can easily misrepresent the car as being good quality even if it is a lemon.

Crucially, asymmetric information is only a problem if it leads to market failure. In the used car market example, since buyers don't know the quality of the cars in the market, they have to assume that any car on offer is low quality. This lowers the amount that they are willing to pay for a car, and drives the good quality cars out of the market (since sellers of good quality cars can't convince buyers of the quality of their cars, and buyers aren't willing to pay enough to buy them). The market for good cars collapses (of course, the market has developed mechanisms that deal with this market failure, such as test drives, pre-purchase inspections, etc.). We call this an adverse selection problem, since those that select to remain in the market are those with the lowest quality cars (when at least some buyers want those with the highest quality cars, not the lowest quality). The description of these 'markets for lemons' was what George Akerlof won the Nobel Prize in Economics for (the original paper from 1970 is here (gated on JSTOR) or here (ungated)).

Signalling is one way that markets have adapted to deal with adverse selection problems. With signalling, the informed party finds a way to credibly reveal the private information to the uninformed party. There are two important conditions for a signal to be effective: (1) it needs to be costly; and (2) it needs to be more costly to those with lower quality attributes. These conditions are important, because if they are not fulfilled, then those with low quality could still signal themselves as having high quality. Sticking with used cars as an example, offering a warranty on the car is a good example of signalling. It is costly (since if the car breaks down, the seller must pay the cost of repair), and it is more costly to those with low quality cars (since they are more likely to break down).

As an aside, asymmetric information isn't a problem if it doesn't lead to market failure. For instance, the formula for Coke Zero is a closely guarded secret (though it almost wasn't), which Coca Cola knows but consumers don't. However, this information asymmetry doesn't lead to market failure because Coca Cola isn't using that information to the detriment of consumers (as far as we know!).

Back to weddings. Is there asymmetric information here that leads to market failure, and will signalling be effective? Dewey's blog post notes two types of signalling - couples signalling to their guests, and the wedding industry signalling to couples. Let's start with the second of those.

The wedding dress maker (or other part of the wedding industry, but let's stick with wedding dresses) knows the quality of their dresses, but the couple does not. So, in theory low quality dress makers can misrepresent themselves as high quality dress makers, and the couple wouldn't know until the big day when the dress falls apart. So, high quality dress makers need some way of distinguishing themselves from the low quality dress makers, through signalling. Does making the dress more expensive constitute an effective signal of quality? In theory price shouldn't act as a signal, because it doesn't meet both of the conditions above (and also because there are more effective signals of quality than price). Raising the price may entail some opportunity cost (through lost sales), so it may be costly. But, it is not more costly to low quality dress makers. So, in theory at least, price should not be a signal of quality. But as we know, consumers are not fully rational and it turns out that they do use price as a signal of quality (see this 1983 paper by Asher Wolinsky (gated on JSTOR) as an early example, or this more recent paper by Maarten Janssen and Santanu Roy (ungated)). So, the high cost of weddings might arise because high quality wedding dress makers (and wedding venues, caterers, florists, etc.) are trying to signal their quality by having a higher price.

Why are couples willing to pay such high prices for wedding dresses? It may be because they are signalling as well. They are trying to reveal two items of private information to their wedding guests (friends, family, etc.): (1) the quality of their relationship; and (2) their social status.

Starting with (1), guests don't know the quality of the relationship that is about to be formalised, but the couple does (hopefully!). Does this create market failure? That is, can the couple take advantage of this information asymmetry to their advantage and to the detriment of their guests? Maybe, if we consider wedding gifts. Guests would probably give less valuable gifts if they believed the marriage wouldn't last (i.e. if the marriage is low quality), than if they thought it would last a long time (i.e. high quality). So, if guests can't be sure about the quality of the marriage, then they may assume the marriage is lower quality and buy less expensive wedding gifts (or no gift at all) as a result. So, high-quality couples need to find some way of signalling their quality, and this may be through the cost of the wedding. This may be an effective signal, because it is costly (obviously), and more costly to low-quality couples since they may expect to marry more than once over their lifetime. So, lower quality couples may be less willing to spend a lot on their wedding than high quality couples.

What about (2)? This isn't an adverse selection problem at all, since there is no market that will fail. However, there is still signalling here - the couple may want to signal their social status to the community. Higher social status is linked with wealth, which means that couples with high social status are likely to be able to afford a more lavish wedding celebration than couples with lower social status. This is of course conspicuous consumption (where spending is intended as a way of maintaining or attaining social status). And, there is at least some evidence to support this (gated, here is an earlier ungated version) - even though the evidence is from India, it doesn't seem much of a stretch that there is something similar at play in a lot of weddings in the western world as well.

So, there you have it. Weddings are most likely costly because of signalling - the wedding industry signalling couples about their quality, and the couples signalling wedding guests about the quality of their relationship and/or their social status.

P.S. I have neglected the role of marriage as a signal from one partner to another. See Chapter 8 in this book for the theoretical background to this idea.

[HT: Marginal Revolution]

Saturday 8 March 2014

Safer cars and offsetting behaviour: NASCAR edition

Classes started this week, and tutorials start next week. In ECON100 and ECON110, among other things at the start of the paper we cover the role of incentives in determining people's behaviour, and how changing incentives can change behaviour (for some people), for better or worse.

Sometimes (more often than most people think) changes in incentives have unintended consequences. One of the most famous of these in economics is the Peltzman Effect, described by University of Chicago economist Sam Peltzman in the 1970s. Using U.S. data, Peltzman showed in this paper (JSTOR gated) that mandatory safety devices on cars, such as seat belts, do not reduce traffic deaths, and actually increase the number of non-fatal car accidents. How can we explain that?

Rational (or quasi-rational) drivers weigh up the costs and benefits of driving faster. The benefits include less time wasted on the roads (an opportunity cost - you give up some time you could spend doing something else). Moreover, the marginal benefits probably decrease the more a driver speeds (because opportunity costs increase the more time is wasted). The costs of driving faster include an increased risk of a serious car accident - this cost is made up of two parts: (1) the probability of a serious accident occurring; and (2) the health and other costs of the accident itself. The marginal costs increase as speed increases, because both the probability of an accident and its seriousness both increase.

If they are optimising, the driver will choose to drive at the speed where the marginal benefit (MB) of driving faster is exactly equal to the marginal cost (MC0). This occurs at S0 in the diagram below. At this point, driving a little bit faster entails a higher additional cost than the benefit they would receive (which is why they will drive no faster than S0).



When safety devices are installed in cars (e.g. seat belts, air bags, crumple zones, side impact bars, etc.), this changes the incentives that drivers face. These safety devices by definition make faster driving safer, and so they lower the cost of driving fast (since the seriousness of the consequences of an accident are reduced). So, in the diagram above, marginal costs are lower (MC1). This increases the optimal driving speed to S1, which increases the chances of an accident. What we observe then is a greater number of accidents in total (more drivers driving faster or more recklessly), but the number of traffic deaths might not decrease at all (more accidents, but the seriousness of each accident is less). In addition to Peltzman's paper, other more recent papers have demonstrated this as well (see here or here, the first one is ungated).

So, increasing car safety leads to what economists term "offsetting behaviour", because the actions of drivers act to offset the benefits of increased safety. One way of dealing with offsetting behaviour in this situation, mischievously suggested by the economist Gordon Tulloch (or by Armen Alchian, according to Steven Landsburg in his book The Armchair Economist, which I read not long ago (the first edition, not the new second edition)), is to attach a large spike to the steering wheel, aimed squarely at the driver's heart. Then they won't offset safety by driving faster!

Anyway, the point of this post was to briefly link to this new research by Andrew Maness, which uses data from 662 NASCAR races from 1994 to 2013. Maness finds that offsetting behaviour is alive and well in NASCAR:
As driver-safety improves by 10%, competitors' recklessness increases by as much as 3.8%. More concretely, NASCAR's combination of the HANS device, SAFER barrier, and Car of Tomorrow results in a 3.6% growth in wrecked vehicles.
So, just like drivers in general, professional racecar drivers offset increases in safety by driving more recklessly. And not just NASCAR drivers, the same effect is observed in Formula One (gated).

[HT: Marginal Revolution]