External Incentives vs. Intrinsic Motivation

In economics, we have the rational choice model, which assumes that people are rational, self-interested, and they respond to incentives. The incentives are usually a monetary reward or punishment. If you want to change people’s behavior, just change the incentives.

But reality is not always that simple. Steve Levitt, the author of the book Freakonomics, once mentioned an interesting story about how he potty trained his toddler daughter Amanda. After Amanda's mom got frustrated at the results even after she had tried all the methods she read from the books, Steve decided to take over and handle this as an economist: let incentives to work its way. He promised Amanda that every time she went to pee in the potty, she got a bag of M&M's. It worked perfectly! Well, for the first couple of days. Eventually, this incentive scheme backfired: Amanda would go to the potty, trickle several drops, ask for a bag of M&M's, and go to the potty again, trickle several drops and ask for another bag of M&M's, and more potty, and more M&M's. Incentives can backfire, even in the case of a toddler.

Most social scientists agree that people have the intrinsic motivation to abide by norms, to care for others and to behave virtuously. Even economists agree that intrinsic motivation plays a role. But given intrinsic motivation, incentives should also work, right? Adding external incentives can only reinforce the behaviors driven by existing intrinsic motivation stronger, isn't it obvious? Well, the obvious may not always be true. Behavioral economists start to find more and more evidence that monetary incentives can crowd-out people’s intrinsic motivation (Frey and Jegen, 2001; Bohnet et al., 2001; Fehr and Gächter, 2001). Proposing pecuniary payment, for instance, reduced Swiss citizens’ willingness to host a nuclear waste facility (Frey and Oberholzer-Gee, 1997). Imposing a fine on parents who arrive late to pick up their children from a day care center actually increased the number of parents arriving late, and removal of the fine did not decrease the number (Gneezy and Rustichini, 2000b). Offering a monetary compensation for blood donors reduces the supply of prospective blood donors, especially among women (Mellström and Johannesson, 2008).

The assumption of self-interested individuals may be a self-fulfilling prophecy. Everyone who is in a marriage knows that the best way to kill a  marriage is to consider the other person as lazy, unhelpful, and ungrateful. Doing that will make the your partner act that way. When you consider others as selfish egoists, it is very likely that they would start to act like one. Imposing incentives in situations like this can impair people’s intrinsic motivation.

If you want to change people’s behavior, you should not rely exclusively on rewards, regulations or punishment. People, like water, can always find cracks in any set of regulations. Sometimes, and more than often, the most effective way to change people's behaviors is to resort to their intrinsic motivation like ethics and norms.

When you write a research paper

at least keep these three things in mind:

1. Identify your key idea

2. Make your contributions explicit

3. Use examples

Flu shots! They help a lot?!

Nowadays, many people are taking flu shots each year to prevent themselves from getting flu. Indeed, as I am going on the job market, one piece of advice I got is taking a flu shot. Well, the reasoning is looking for a job as an economist takes a lot of time and energy, and endurance -- It is a marathon, not a sprint -- and the least thing you want is getting sick during that long demanding process. Sure, I'll take a flu shot.

However, have you ever wondered, is it really effective to take a flu shot? I haven't thought about this question, as I have taken it for granted, a flu shot, of course, should prevent you from getting a flue, isn't it clear enough? Until the other day, I talked to a labor economist, and he started to mention how people usually take a correlational relationship as a causal relationship, and then he brought up the issue of the flu shot.

Economists (well, at least some of them) are known for being suspicious. They cast doubt on a lot of seemingly obvious conclusions: going to college brings you a higher salary? Not necessarily. People who go to college tend to earn a higher salary anyway, even without going to college. Did better economy or better policy cause the crime rates to fall by 50% in the early 1990s? No. It was because of Roe vs. Wade in 1973 and birth control (if you don't believe this, check the  book Freakonomics for more information. And still, you don't have to buy this. It is just something shown by analyzing the data).

How about flu shot? Well, what we observe from the data is that people taking flu shots are on average healthier. Is this better health, however, caused by the flu shot? It is really hard to say. Data also shows that people taking the flu shot tend to have fewer car accident, tend to have better marriages, tend to have better salary... If we just look at the data, you would claim that flu shots are the best invention ever -- it is the panacea for all troubles!

The main issue is, the relation shown by the data is just correlational. We cannot conclude that A causes B simply because A happened before B, or A and B happened together. People taking flu shots are those who take better care of themselves, and probably more responsible, and hence have better marriages, and fewer car accident. The labor economist claim that, without an experiment or a good instrumental variable*, we can never prove the effectiveness of flu shots. 

Again, several days later, I talked with an instructor from the department of health and kinesiology about flu shots, and asked what he thought about them. He then mentioned how experiments in the lab show the strain of viruses can be killed by the vaccine contained in the flu shots. I didn't ask for the details about how the experiments are conducted, and whether that results can be carried over from the lab to the human body, but it dawned on me that medical (natural) scientists do have a different approach in tackling a problem. Instead of straining their brains looking for a good instrumental variable or randomly assigned people to take the flu shots and check the effectiveness, they go directly to controlled experiments to study the virus. Nice and easy.

Well, that said, I will continue taking flu shots, if not for their effectiveness, then at least to prove that I am a responsible person.

BTW,  the health and kinesiology instructor did tell me the most effective way to prevent disease: wash your hand, thoroughly. He also mentioned a statistic that you may not be comfortable to hear: about one third of the people do not wash their hands after using the bathroom.

I will not continue to argue whether people who wash their hands are more responsible people, but I do have  the gut feeling that washing your hands might be more effective to prevent disease than taking a flu shot. 

*An instrumental variable  is something used to establish a causal relationship from a correlational relationship. Come and take my econometrics class if you want to know more (https://sites.google.com/site/huanrenzhang/teaching/econ360sp13)

Research Heroes: Richard Thaler

Repost from http://indecisionblog.com/2013/01/15/research-heroes-richard-thaler/

I wish someone had told me at the beginning of my career how to go about combining economics and psychology. There was no road map. Mostly trial and error with lots of errors.

I most admire academically… My greatest inspiration came from Kahneman and Tversky, my mentors who became my friends and collaborators. Danny is still a source of inspiration, going full speed at 78. They were a fantastic team because of their complementary skills, but they were both perfectionists in their different ways. I owe my career to them directly, but in some ways so does the entire field.

The best research project I have worked on during my career… This is like asking a parent to name his favorite child. Not fair. So I will fudge and name more than one. I think mental accounting is probably my best “idea”. Save More Tomorrow is my most important practical application. The book Nudge has reached the widest audience and had the most impact. But truth be told my favorite single paper is the one with Cade Massey on the NFL draft called the “Loser’s Curse”. The great thing about being an academic is that you can write a paper on anything. Certainly the paper that was the most fun to work on (or not work on) was the one with Eldar Shafir called “Invest Now, Drink Later, Spend Never”. Eldar and I didn’t work on it for a solid week in Venice one year. It is about the mental accounting of wine consumption. I still devote a lot of time to that problem!

The worst research project I have worked on during my career… There is nothing in print that I would want to take back. I have abandoned lots of projects. I believe in ignoring sunk costs. As I tell my MBA students: “Ignore sunk costs. Assume everyone else doesn’t.”

The most amazing or memorable experience when I was doing research… I gave a talk on the idea for Save More Tomorrow to a large (>500) group of 401(k) plan administrators in 1996 or so, thinking that at least one one of them would think enough of the idea to try it. Then nothing happened for several years. Very frustrating. Then out of the blue Shlomo Benartzi told me that someone he knew had tried it without even telling us, and the results were fantastic. That was exciting because once we could show people that the idea worked, it was (relatively) easy to get others to try it. Now it is used by millions of people, but we had to get the first employer to try it or we would still be wondering if it would really work.

The one story I always wanted to tell but never had a chance… No such thing. But I am putting all those stories into a book I am working on, so stay tuned. The working subtitle is “The Stories of Behavioral Economics”.

A research project I wish I had done… My phd thesis was on the value of saving a life. The idea was to estimate how much you had to pay people to get them to accept a small increase in risk. So, I did an econometrics exercise regressing wages on occupational mortality rates. But the really clean study to do, as suggested by my buddy Richard Zeckhauser, would be to get people to play Russian Roulette, with a machine gun with many, many chambers (say 10,000). Then tell people there are 5 bullets in the gun, how much would you pay to remove one, or accept to add one. For some reason, no human subject committee has ever been willing to approve this project. Can’t imagine why! (Before I get into trouble, this was intended as a joke.)

If I wasn’t doing this, I would be… Less happy. I feel lucky to have found a way to make a living that is so much fun to do. Who knows what else, but I did think about going to law school instead of economics graduate school. I don’t think I would have been a great lawyer though. I suffer from a diplomacy deficiency.

The biggest challenge for our field in the next 10 years is… I see two. First, JDMers need to learn to get out of the lab some of the time (and journal editors need to encourage such risky activity by applying appropriately different standards to field experiments). The stuff we study is too important and useful for it to be limited to the lab.

Second, I fear that the science-lab model in which increasing numbers of grad students are added to shrinkingly important papers in order to supply graduate students with enough publications to go on the job market. I think this trend stifles creativity and does not encourage students to do enough thinking on their own. More generally, I think psychologists are just publishing too many small papers. Look at the number of papers Kahneman and Tversky wrote that created and defined the field we now call judgment and decision making. The judgment stuff was really 3 papers plus the Science recapitulation. Then came prospect theory. Four blockbusters that led to a Nobel Prize. Not enough for tenure these days! Amos had a line about people that he felt wrote too many papers: “he publishes his waste basket”. I don’t think he would approve of the current state of affairs.

My advice for young researchers at the start of their career is… Work on your own ideas, not your advisor’s ideas (or at least in addition to her ideas). And spend more time thinking and less time reading. Too much reading leads people to think of small variations on existing studies. Admittedly my strategy of writing the paper first and only then reading the literature (or, more likely, letting the referees tell me what they think I should have read) is an extreme one, but it is better than trying to read everything. Try writing the first paper on some topic, not the tenth, and never the 50th.

David Romer's Rules for Making It Through Graduate School and Finishing Your Dissertation

"Out in Five"

---

• Don't clutter up your life with other activities; just write.
• Don't carry out a thorough and comprehensive search of the literature; just write.
• Don't attempt to make sure that every page you write shows the full extent of your professional skills; just write.
• Don't write a well-organized, well-integrated, unified dissertation; just write.
• Don't think profound thoughts that shake the intellectual foundations of the discipline; just write.
• If you don't have a paper started by the spring of your third year, be alarmed.
• If you don't have a paper largely drafted by the fall of your fourth year, panic.
• Have three new ideas a week while you are getting started.
• Don't try to game the profession, work on what interests you.
• Good papers in economics have three characteristics:
  • A viewpoint.
  • A lever.
  • A result.

Panel Data Analysis -- Random Effects vs. Fixed Effects

y_it=beta x_it + a_i + u_it

The unobserved factors affect the dependent variable consist of two types: a_i (constant over time), and u_it (varying over time). a_i is called an unobserved effect or a fixed effect. u_it is called the idiosyncratic error or time varying error.

Fixed effects estimation uses a transformation (time-demeaned data) to remove the unobserved effect a_i. Fixed-effects transformation is also called the within transformation. A pooled OLS estimator uses the time variation in y and x within each cross-sectional observation. It is based on the time-demeaned variables and is called the fixed effects estimator or the within estimator.

In the fixed effects model, any explanatory variable that is constant over time for all i gets swept away by the fixed effect transformation, therefore, we cannot include variables such as gender or a city's distance from a river. The fixed effects estimation is adequate if we want to draw inferences only about the examined individuals.

When we assume that the unobserved effect a_i in the above model is uncorrelated with each explanatory variable in all periods Cov(x_it,a_i)=0 , we have a random effects model. This model is adequate, if we want to draw inferences about the whole population, not only the examined sample.

The fixed effects estimator subtracts the time averages from the corresponding variable. The random effects transformation subtracts a fraction of that time average, where the fraction depends on sigma_u^2, sigma_a^2, and the number of time periods, T.

In practice, it is usually informative to compute the pooled OLS estimates. Comparing the three sets of estimates can help us determine the nature of the biases caused by leaving the unobserved effect a_i, entirely in the error term (as does pooled OLS) or partially in the error term (as does the RE transformation.) Remember, however, the pooled OLS standard errors and test statistics are generally invalid: they ignore the often substantial serial correlation in the composite errors, v_it=a_i + u_it.

One can used Hausman test. A failure to reject means either that the RE and FE estimates are sufficiently close so that it does not mater which is used, or the sample variation is so large in the FE estimates that one cannot conclude practically significant differences are statistically significant.

Using FE is mechanically the same as allowing a different intercept for each cross-sectional unit. FE is almost always much more convincing than RE for policy analysis using aggregated data.


Panel data analysis can also be used to analyze clustered data. Depending on the nature of the clustered data, FE or RE can be used.

xttobit -- Linear Regression Model with Panel-level Random Effects for censored data

xttobit fits random-effects tobit models. There is no command for a parametric conditional fixedeffects model, as there does not exist a sufficient statistic allowing the fixed effects to be conditioned out of the likelihood.

Panel data analysis has three more-or-less independent approches:

1. Independently pooled panels: there are no unique attributes of individuals within the measurement set, and no universal effects across time

2. Random effects model: there are unique, time constant attributes of individuals that are the results of random variation and do not correlate with the individual regressors. This model is adequate if we want to draw inferences about the whole population, not only the examined sample

3. Fixed effects model (or first differenced models): there are unique attributes of individuals that are not the results of random variation and that do not vary across time. Adequate if we want to draw inferences only about the examined individuals. Also known as "least squares dummy variable model"

y_{it}=x_{it}beta + u_i + e_it

u_i is the random effects

e_it is the individual effect

. xi: xttobit Contribution Period Belief AmountInvested Cutoff RandomDraw realized econm

> ajor econyear if Treatment==4, ll(0) ul(20)

Random-effects tobit regression                 Number of obs      =       320

Group variable: Subject                         Number of groups   =        16

Random effects u_i ~ Gaussian                   Obs per group: min =        20

                                                               avg =      20.0

                                                               max =        20

                                                Wald chi2(8)       =    372.07

Log likelihood  = -643.86478                    Prob > chi2        =    0.0000

------------------------------------------------------------------------------

Contribution |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]

-------------+----------------------------------------------------------------

      Period |  -.3714078   .0859276    -4.32   0.000    -.5398229   -.2029927

      Belief |   .8871966   .1082666     8.19   0.000      .674998    1.099395

AmountInve~d |   .4972489   1.099413     0.45   0.651    -1.657561    2.652059

      Cutoff |  -.1031474   .0144602    -7.13   0.000    -.1314888   -.0748059

  RandomDraw |   .0298226   .0128297     2.32   0.020     .0046769    .0549683

    realized |    1.19617   1.115385     1.07   0.284    -.9899439    3.382284

   econmajor |  -.1037135   4.025793    -0.03   0.979    -7.994123    7.786696

    econyear |   -1.11204   1.565532    -0.71   0.478    -4.180426    1.956345

       _cons |   4.846538   4.027966     1.20   0.229    -3.048129    12.74121

-------------+----------------------------------------------------------------

    /sigma_u |   4.153189   .9255023     4.49   0.000     2.339238     5.96714

    /sigma_e |     4.6622   .2652641    17.58   0.000     4.142292    5.182109

-------------+----------------------------------------------------------------

         rho |   .4424506   .1130577                      .2400784    .6614899

------------------------------------------------------------------------------

  Observation summary:        94  left-censored observations

                             185     uncensored observations

                              41 right-censored observations

The output includes the overall and panel-level variance components (labeled sigma e and sigma u, respectively) together with rho, which is the percent contribution to the total variance of the panel-level variance component.

When rho is zero, the panel-level variance component is unimportant, and the panel estimator is not different from the pooled estimator. A likelihood-ratio test of this is included at the bottom of the output. This test formally compares the pooled estimator (tobit) with the panel estimator.