Nobel Prize for Oliver Hart and Bengt Holmström

The 2016 Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel (aka Nobel Prize in Economics) has been awarded to Oliver Hart and Bengt Holmström, "for their contributions to contract theory". Excellent coverage of their contributions to economics are provided by Tyler Cowen (here for Hart, and here for Holmström). See also here for a more accessible summary of their work (also by Cowen), as well as this piece by Noah Smith. Note that much of their work is quite theoretical, and there are definite links to Jean Tirole (who won the prize in 2014).

My ECON100 and ECON110 students might recognise some of their work in our discussions of moral hazard, principal-agent problems, and performance pay. I just finished talking in the final ECON110 lecture about the links to the work of Nobel laureates in that paper - I wish I'd checked my emails sooner, and I could have noted this award in that lecture!

Stuck with indecision? Let the coin decide!

Quasi-rational decision makers are loss averse (we value losses more than we value equivalent gains). One of the interesting outcomes of loss aversion is status quo bias. Because changing something entails both a loss (we give up what we were doing before) and a gain (we start doing something new), the change has to make us much better off before we are willing to make the change. Status quo bias keeps us investing in projects that have little chance of success, and keeps us in unhappy relationships, horrible jobs, and so on.

So, what if we could overcome our indecisiveness by outsourcing the decision to a coin? Kind of like this:

Ok, maybe not quite like that. Tim Harford explains in a recent post:

The roll of a die or the toss of a coin can actually help us make better decisions.

There are two quite different reasons for this. The first is that by pre-committing to follow a random instruction, we can end up making decisions that we should have been making all along. The status quo has a strange hold over us. Stuck in a job that we dislike, or with a romantic partner who is anything but romantic, all too often we stick with the devil we know.

Deciding that “if the coin comes up heads, I’ll leave my boyfriend” may be the only way that some of us have to break through the inertia and make tough decisions. A 50 per cent chance of dumping the oaf is better than no chance at all.

Harford goes on to talk about this recent NBER Working Paper (ungated version here) by Freakonomics author Steven Levitt (University of Chicago). I've been meaning to write about this paper for some time - it was covered by Marginal Revolution, Jodi Beggs (Economists Do It With Models), and The Economist Free Exchange blog, all back in August, and has been on my must-read list since.

Why do a study that uses coin tosses to make decisions? Levitt explains in the paper:

What we really care about, however, is the impact on the marginal decision maker. It would not be surprising if getting a divorce would have a devastating impact on the inframarginal married person. A much more interesting question is whether divorce, ex post, will be the right choice for someone teetering on the edge of ending a relationship.

Even if one found such a group of individuals who are close to indifferent between remaining married and getting divorced, an ex post comparison of the happiness of those who do and do not make a change still would not have an easy causal interpretation, because the people who make a change will systematically differ from those who do not on many dimensions. To convincingly answer the question, a researcher would not only need to find large numbers of these marginal individuals, but also, through some sort of randomization, influence their important life choices.

That is what I do in this study. I created a website called FreakonomicsExperiments.com. On the website, individuals who are having a difficult time making a life decision are asked to answer a series of questions concerning the decision they are struggling with... One choice (e.g., “go on a diet”) is assigned to heads and the other choice (in this case “don’t go on a diet”) is assigned to tails. The outcome of the coin toss is randomized and the user is shown the outcome of the coin toss. The coin tossers are then re-surveyed two months and sixth months after the initial coin toss.

What Levitt finds is remarkable. First, people actually do follow the advice of the coin (in at least some cases). Those who flipped heads were 24.9 percentage points more likely to change than those who flipped tails (a statistically significant change). And even better, those who changed were happier afterwards. Levitt notes:

when it comes to “important” decisions (e.g. job quitting, separating from your husband or wife), making a change appears to be not only correlated with increased self-reported happiness, but also causally related, especially six months after the coin toss. Those who were instructed by the coin toss to make a change were both more likely to make the change (as noted above) and, on average, report greater happiness on the follow-up surveys... Choices on “less important” decisions (e.g. dying hair, improving posture) do not generally have a measurable impact on later happiness.

How big was the impact on happiness? People who made a change reported happiness about 0.48 points higher (on a scale of 1 to 10), or about 0.2 standard deviations. When looking at individual decisions, the results are under-powered to find much but do show that the effects appear to be largest for job quitting and breaking up. So, the next time you are agonising over whether to end that relationship or quit that job, maybe let a coin decide.

[HT: Marginal Revolution first, then others]

Flipped classrooms work well for top students in economics

Earlier this year I blogged about two AER Papers and Proceedings papers that compared online only, blended (e.g. flipped) and traditional classes. Here's what I said then:

These two papers give quite complementary findings. The Swoboda and Feiler paper found a significant positive effect of their blended learning approach (compared with face-to-face), while Alpert et al. find no significant effect of blended learning...

Now I would really like to know what the distributional effects of blended learning are. My expectation is that it will probably work well for keen, motivated students, who are almost certain to watch the lectures before class. These are the students who currently read the textbook, and additional resources (like the lecturer's blog!). These students will likely benefit most from the change in approach, and gain significantly from the interactive learning in class, which is what the blended learning approach is designed to facilitate. The less-motivated students, who may watch some of the videos before class but often don't, will not benefit as much, or may actually be made worse off by the switch to blended learning.

Which brings me to this new paper by Rita Balaban and Donna Gilleskie (both UNC Chapel Hill), and Uyen Tran (University of Chicago), published in the latest issue of the Journal of Economic Education (sorry I don't see an ungated version anywhere). In the paper the authors compare a flipped classroom model (where students watch lectures online before attending class where more active learning approaches, such as problem-based learning, are employed) with a traditional lecture-based model. There were nearly 400 students in each semester. Unfortunately, the research design is not clean because the two models were employed in different semesters. The authors demonstrate that there are no observable differences between the students, but I would also be concerned that the lecturer (who is one of the co-authors) knew that the study was being undertaken during the second semester (when the blended learning approach was used), and put in greater effort (a genuine concern for any single-blinded trial).

Notwithstanding my concern about the single-blinded nature of the approach, the results are interesting and very positive for the blended learning approach:

The values of average percent correct among all common questions for the traditional and flipped classroom formats suggest that the course redesign led to a (statistically significant) 6.9-percentage-point improvement in student performance (i.e., a difference-in-means result).

Once they control for student characteristics, the results remain similar, with an overall increase of about one-half of a standard deviation. This is quite a large impact. When they disaggregate the results by question type, they find:

Our results indicate that the flipped classroom does not differently impact performance on knowledge questions (objective 1), which require memory, recognition, and recall. We find that the flipped format significantly improves performance on comprehension questions (objective 2) by one-quarter of a standard deviation...

With regard to performance on application questions (objective 3), the flipped classroom boosts performance by 0.74 standard deviations on average...

On analysis questions (objective 4), we find gains of 0.47 standard deviations. The ability to analyze involves differentiating, organizing, and attributing.

So, as one might expect, the main impacts are on students being better able to apply their learning (it's worth noting that there were only three questions in the exam in the knowledge domain, and only three questions in the analysis domain).

Finally, the authors looked at the results by performance level (using quantile regression). They find:

that the flipped classroom had slightly different effects on students (depending on their position within the performance distribution) for different types of questions. Overall, students in the top 25 percent of the distribution appear to benefit more from the flipped format than those below the 75th percentile, although all students benefit substantially.

Which brings me back to my initial disquiet at the flipped classroom model. This paper has done little to dissuade me that it benefits the motivated top-performing students and could make things worse for the unmotivated marginal student. I wouldn't necessarily take this study as representative of the average university student. The average exam result was 80.7 percent, which struck me as rather high until I looked at the average SAT scores for the sample, and found that the average student was in about the 90th percentile on the SAT. So, even the students in the bottom of this class are relatively good students in the overall scheme of things. To add to that, the lecture attendance was over 90 percent on average. So not only are these mostly top-achieving students, they are well-motivated top-achieving students. Exactly the students I would expect to benefit from the flipped classroom. I guess I'm still waiting for the research that will convince me that the flipped classroom model will have positive outcomes for the marginal (or even the median) student that I teach.

Could your social media posts make insurance more expensive?

James from my ECON110 class pointed me to this insightful Tamsyn Parker article in the New Zealand Herald with the above title. Parker writes:

Could that Instagram image of you bungy jumping in your 20s result in having to pay higher insurance costs in the future? One insurance expert thinks so.

Michael Naylor, a senior lecturer in finance and insurance at Massey University, says people should expect insurers to mine their social media accounts in the future to determine how much they will charge for insurance premiums and if they will pay out on claims.

"People have to be aware everything they do on social media can be effectively public.

Why would insurance companies want to mine social media data to find out about us? It's because of the adverse selection problem. An adverse selection problem arises because the uninformed party (the insurer) cannot tell those with 'good' attributes (low-risk people) from those with 'bad' attributes (high-risk people). To minimise the risk to themselves of engaging in an unfavourable market transaction, it makes sense for the insurer to assume that everyone is high-risk. This leads to a pooling equilibrium - low-risk people are grouped together with the high-risk people and pay the same premium, because they can't easily differentiate themselves. This creates a problem if it causes the market to fail.

I've written about how the insurance market fails before (this comes from a post about health insurance, but it equally applies to accident insurance or life insurance - see also this post on adverse selection in life insurance):

In the case of insurance, the market failure may arise as follows (this explanation follows Stephen Landsburg's excellent book The Armchair Economist). Let's say you could rank every person from 1 to 10 in terms of risk (the least risky are 1's, and the most risky are 10's). The insurance company doesn't know who is high-risk or low-risk. Say that they price the premiums based on the 'average' risk ('5' perhaps). The low risk people (1's and 2's) would be paying too much for insurance relative to their risk, so they choose not to buy insurance. This raises the average risk of those who do buy insurance (to '6' perhaps). So, the insurance company has to raise premiums to compensate. This causes some of the medium risk people (3's and 4's) to drop out of the market. The average risk has gone up again, and so do the premiums. Eventually, either only high risk people (10's) buy insurance, or no one buys it at all. This is why we call the problem adverse selection - the insurance company would prefer to sell insurance to low risk people, but it's the high risk people who are most likely to buy. 

In order to solve an adverse selection problem, the uninformed party can try to reveal the private information - we call this screening. Mining people's social media posts could reveal to the insurance companies which people are high-risk and which are low-risk. They can then separate the two groups, and we move from a pooling equilibrium to a separating equilibrium. High-risk people will pay higher premiums, and low-risk people will pay lower premiums. Or, as noted in the New Zealand Herald article linked above:

[Michael Naylor] predicts it will be one to three years away in New Zealand and says it may not come from existing insurers but new entrants to the market who will use personal data to cut insurance premiums for less risky customers.

That could leave old-style insurers with more risky customers and the prospect of rising premiums to cover their costs.

He says the change could have implications for people who seek adventure when younger and record it all on their social media pages.

"Of course the internet doesn't die."

They may have to prove they no longer undertake those activities to get insurance in the future or sign exclusion agreements meaning they won't be covered for certain activities, says Naylor.

So, if new insurers use social media mining to offer cheaper insurance to low-risk people, you can bet the large incumbent insurers will follow suit soon after, because insuring low-risk people is much more profitable than insuring high-risk people.

So for now, before you apply for life insurance (or health insurance, accident insurance, or even car insurance), it might be best to lock down your social media accounts. Or at least delete any references to the risky exploits of your youth.

If everyone locks down their social media accounts so that insurers can't access them, things start to get interesting. Will insurers simply ask to see your past social media posts? If I was an insurer, I would. People who say 'no' to such a request are more likely to be high-risk (because low-risk people would have nothing to hide), and the insurer could price their premiums accordingly. Essentially, low-risk people could signal that they are low risk by making their past social media posts available to their insurer, and reap the reward of a lower premium. In fact, low risk people could probably do this right now.

[HT: James from my ECON110 class]