Week 4: Introduction to Unit 3

Week 4: Introduction to Unit 3

“Introduction to Unit 3 … The Basics of Experiments … Analysis and Interpretation of Results … Lab Experiments … The Spectrum from Lab to Field … Field Experiments … Debate … Skills and Knowledge”
(Source URL)

Summaries

  • Unit 3 > 3.0 Introduction to Unit 3 > 3.0.2 Unit 2 Debate Debrief
  • Unit 3 > 3.1 The Basics of Experiments > 3.1.1 A Visit to the Behavioural Lab
  • Unit 3 > 3.1 The Basics of Experiments > 3.1.3 The Building Blocks of Experiments
  • Unit 3 > 3.1 The Basics of Experiments > 3.1.5 Types of designs
  • Unit 3 > 3.2 Analysis and Interpretation of Results > 3.2.1 The Analysis of Experimental Data
  • Unit 3 > 3.2 Analysis and Interpretation of Results > 3.2.2 Regression
  • Unit 3 > 3.2 Analysis and Interpretation of Results > 3.2.4 ANOVA Effects
  • Unit 3 > 3.3 Lab Experiments > 3.3.1 Does the Use of a Credit Card Increase Spending?
  • Unit 3 > 3.3 Lab Experiments > 3.3.2 Does the Routine Use of Credit Cards Influence Spending Decisions?
  • Unit 3 > 3.4 The Spectrum from Lab to Field > 3.4.1 A taxonomy of studies
  • Unit 3 > 3.5 Field Experiments > 3.5.1 Does Restricting the Right to Withdraw Funds Help People Save More?
  • Unit 3 > 3.5 Field Experiments > 3.5.2 Will People Work More on High-Wage Days?
  • Unit 3 > 3.6 Debate > 3.6.1 Debate 3

Unit 3 > 3.0 Introduction to Unit 3 > 3.0.2 Unit 2 Debate Debrief

  • DILIP SOMAN: Debate question that was posed to you was about consumer credit.
  • Is consumer credit a blessing or a curse? And like in the past, we had five experts who were n the panel.
  • They both talked about why consumer credit is a blessing.
  • Avi made the point, a fundamentally important point, about the fact that consumer credit allows you to smooth consumption over time.
  • If you didn’t have credit, you would not be able to make big purchases like houses or even small things because you would not have money in your pockets at the right point in time.
  • So consumer credit is critically important because it allows us to manage our consumption patterns over time.
  • He talked about the convenience aspect of it, the fact that with a credit card, you could spend in Germany or Hong Kong or wherever you went.
  • Finally, also the idea that when you use a credit card, you actually don’t they a period of time.
  • Min Zhao came and she didn’t disagree with anything that Avi or Ron said, said but she made a very important point which was the fact that people aren’t very good at managing credit.
  • So she talked about the fact that people aren’t good at keeping track of their expenses.
  • The more they use credit, the less likely they are to remember how much they spent and that they’re overconfident in their ability to pay back.
  • One other very important point she said in the end and it’s also been discussed in many recent books is that credit cards can contribute be perhaps one of the biggest drivers of an impulsive society.
  • I buy a lot of books on Amazon.com and Amazon has my credit information.
  • So consumer credit does have the ability to cause impulsivity.
  • We then had Professor Andrew Ching who talked about the hidden costs of consumer credit, in particular, credit cards.
  • So he made the point that merchants have to pay fees when they accept credit cards and as the result of those fees, somebody has to cover them.
  • So there’s one consistent theme that comes up through, not just your comments, but also the comments of our panelists, which is that consumer credit is a tool.
  • Very fundamentally, the use of consumer credit and in particular credit cards is a mental accounting story.
  • We talked about the idea that you need to set up these accounts and then you need to meter your purchases.
  • It’s very hard because the credit card process is very different.
  • It doesn’t leave a strong memory trace in your head. And a lot of people simply forget the number of items that they have purchased using consumer credit.
  • The pain associated of making the payment with credit is much lower than the pain with cash or with checks.
  • Credit cards don’t let you do that because now it’s this one account called credit card bill.
  • With consumer credit, and in particular with credit cards, your credit card billing cycle doesn’t coincide with your planning cycle.
  • Can you think of credit cards where the billing cycle begins on the first and ends on the 30th? They’re extremely rare.
  • So these are all fundamental problems with using consumer credit.
  • With consumer credit, there are a lot of fees.
  • A lot of consumers will complain that they didn’t know about those fees.
  • Now if you actually open up your credit card statement and look at the back, most cases, the back of your statement will disclose every single fee.
  • A number of years back, President Obama in the United States signed the Card Act.
  • This is a concept that is hard for many consumers to understand.
  • It’s actually many times worse because you compound your interest over time.
  • So that’s what the Card Act mandates that providers of consumer credit do.
  • Now if you think about an app that you could put on your phone that curates all these text messages and lets you keep track of exactly how much you have spent, that actually overcomes the limitations of the metering problem that we just talked about.
  • As long as you are able to manage the information, as long as you’re able to keep track of the mental accounting, as long as you can use tools and devices to help you do that, then the consumer credit can be absolutely fantastic because it lets you smooth consumption over time.

Unit 3 > 3.1 The Basics of Experiments > 3.1.1 A Visit to the Behavioural Lab

  • DILIP SOMAN: If you talk to any behavioral economist anywhere in the world and you ask them what their typical day looks like, most will tell you that they spend a fair bit of time in a lab like this one.
  • I’m standing at the entrance to the behavioral lab at the Rotman School of Management at the University of Toronto.
  • Doing experiments in the lab is almost the bread and butter of the existence of any behavioral economist.
  • All of the great theories you’ve heard about, all of the cool phenomena that you’ve read about, were all born in a lab like this one.
  • What exactly is a lab, and how does a lab function? And what is an experiment? To find out more, we’ll talk to Julie Huang, who is the manager of the lab at the Rotman School of Management.
  • Julie, what exactly is an experiment? JULIE HUANG: Well, an experiment is basically a tool that researchers can use to establish whether or not there’s a causal relationship between one particular variable and an outcome that they care about.
  • In experiments, the outcome that they care about is called the outcome variable.
  • At its most basic, all experiments compare two groups, the treatment group and the control group.
  • DILIP SOMAN: If you talk to any behavioral economist anywhere in the world and you ask them what their typical day looks like, most will tell you that they spend a fair bit of time in a lab like this one.
  • I’m standing at the entrance to the behavioral lab at the Rotman School of Management at the University of Toronto.
  • Doing experiments in the lab is almost the bread and butter of the existence of any behavioral economist.
  • All of the great theories you’ve heard about, all of the cool phenomena that you’ve read about, were all born in a lab like this one.
  • What exactly is a lab, and how does a lab function? And what is an experiment? To find out more, we’ll talk to Julie Huang, who is the manager of the lab at the Rotman School of Management.
  • Julie, what exactly is an experiment? JULIE HUANG: Well, an experiment is basically a tool that researchers can use to establish whether or not there’s a causal relationship between one particular variable and an outcome that they care about.
  • In experiments, the outcome that they care about is called the outcome variable.
  • At its most basic, all experiments compare two groups, the treatment group and the control group.
  • In the control group, the independent variable will not be present, and then the researcher will compare both groups’ responses on the outcome variable.
  • There are lots of different kinds of experiments that can be run.
  • So in these cases researchers, instead of deliberately manipulating the independent variable, they’ll just look in the world and find naturally occurring groups that just have to happen to different based upon the variable of interest.
  • In lab studies, what we can do is we can bring participants into the lab.
  • Then we can really tightly control their experiences, and we can make sure that the control group and the treatment group only differ based upon the independent variable.
  • DILIP SOMAN: What are the kinds of things that researchers would typically manipulate in an experiment in a lab like this one? JULIE HUANG: Well, there are lots of things that can be manipulated in a lab like this.
  • DILIP SOMAN: Tell us about one of your own experiments, something that you would consider perhaps a favorite experiment.
  • So basically what we did is before all participants come into the lab, we’ll flip a coin.
  • DILIP SOMAN: Could you show us a couple of interesting rooms in this lab? JULIE HUANG: So sometimes researchers will want to get participants to sample food items or beverages, and that’s why we have a space like this, a kitchen where people can store the materials for those studies.
  • This is the control room where the research assistant running the experiment sits and manages the experiment.
  • I know how many conditions there are for experiments, and I randomly assign participants to those conditions.
  • JULIE HUANG: Experiments in which the data collection is done by computer are done in rooms like this one.
  • One of our Ph.D. students, Joon, is getting ready to run an experiment.
  • JOONKYUNG KIM: In my experiment, I have two conditions.
  • In this condition, the screen appears in a font that is easy to read. But here in the second condition, the screen appears in a font that is difficult to read. DILIP SOMAN: Now that we’ve seen the insides of a lab and we know what experiments are, it’s time for us to open the hood and take a closer look at the design of experiments.

Unit 3 > 3.1 The Basics of Experiments > 3.1.3 The Building Blocks of Experiments

  • DILIP SOMAN: In this segment, we’ll ask ourselves what the building blocks of experiments are and get to know the key terminology as we think about designing experiments.
  • Before we do that, let’s think about why experiments are done in the first place.
  • Any major phenomena, like decision biases or strategies used for making decisions, are typically documented by doing a series of experiments over time that converge towards the same conclusion.
  • Second, experiments are often run to develop a theory to explain the phenomena that have been documented, as we discussed earlier.
  • Third, we often do experiments to study the effect sizes of different phenomena.
  • A fourth reason for running experiments is to reconcile and test across theories that make conflicting predictions.
  • Finally, we often do experiments to test for the efficacy of what we call in behavioral economics as nudges or behavior changing interventions.
  • So these are five good reasons for why experiments are typically conducted.
  • Every experiment studies the relationship between one variable, that we call a cause, and it’s consequence, which is often called an effect.
  • In a nutshell, every experiment tries to show that a particular cause results in a specific effect.
  • The per day expense goes up to a certain level, let’s say it’s $25 or $20, then the pennies a day effect might backfire because now the consumer cannot recall anything else they do that routinely cost them $15 or $20 or $25 a day.
  • Now, let’s think about some other elements of an experiment.
  • In the kinds of experiments that we often run in behavioral economics, a control conditioned represents a condition where the intervention, or the nudge , that you’re trying to test is absent.
  • Now, oftentimes in an experiment we could have multiple causes or multiple treatments that are employed simultaneously.
  • A condition refers to the collection of specific tasks and stimuli that the experimenter presents to a participant.
  • In one of the experiments we saw earlier, price framing is a factor that has simply two levels, an aggregate level or an at a pennies a day level.
  • We could run an experiment in which a ticket that has been purchased for a basketball game has been obtained for free, for $10, for $20, or for $30. In this case, the price that has been prepaid for the ticket is the treatment in question the factor that we are studying.
  • In experiments, you could study the effect of a cause on the attitudes or on the strength of preference or on the choices that participants make.
  • What is a manipulation? It is any aspect of a process or a task that the experimenter changes across experimental groups.
  • So any element of the task that is changed by the experimenter across groups is called a manipulation.
  • A couple of other terms to keep in mind when we think about experiments.
  • What’s a background variable? Any set of variables that are given in an experiment and are not manipulated are background variables.
  • Oftentimes, the location, at which data has been collected, or the ethnicity of the participants or the gender or the education level, all represent background variables that are given in the context of an experiment.
  • We have to make sure that we interpret the experiments in the context of the background of that experimental setting.
  • In particular, care must be taken in order not to generalize too much from a given set of background variables and push the results of the experiment into other conditions of background variables.
  • It is important to replicate the experiment across different sets of background variables, so that the researcher has greater confidence in the results.
  • If you make the opposite conclusion, if you make the conclusion that the payment mechanism actually changes how much people spend, that would be a confound or, what is called, a bias, as it happens in the experiment.
  • Experimenters or research assistants who actually collect the data are typically blinded to the cause of the experiment, so that we minimize the chance of any allocation biases, minimize the chance that they aim to allocate participants to certain conditions that they think would push the results as they expected.
  • Across multiple experiments, the use of different dependent measures will strengthen the belief that the experimental results all converge towards the same conclusion.
  • One could triangulate by changing methods across experiments, by using multiple sets of background variables and finally, even by changing the data analysis techniques across different groups of experiments.
  • Triangulating across multiple experiments that all try and test the same basic idea, the same basic cause and effect relationship, will increase the confidence in the conclusion that the researcher comes up with.

Unit 3 > 3.1 The Basics of Experiments > 3.1.5 Types of designs

  • What we did in that experiment was we observed people doing laundry over a 40 day window, such that in the first half of the window, the machines accepted quarters.
  • Something has happened in the environment, in this case, the management of the laundromat, that has changed the way in which people do their laundry, or, in an experimental sense, the focal task.
  • Does an experiment like this conclusive prove that if you move away from coins to cards that people with change their laundry behavior, and that the change in the laundry behavior is attributable only to the change in payment mechanism? Keep in mind here the cause is the movement away from coins to a prepaid card.
  • In a control condition, you would have the same 40 day window, pretty much the same location, and the same kinds of people doing their laundry.
  • In the control, rather than switching from coins to a prepaid card, you would continue using coins throughout the entire 40 day window.
  • Now, let’s go back and think of another phenomenon that we looked at in the first couple of weeks, the pennies a day pricing phenomena.
  • John Gourville showed in his research that if you in fact take an aggregate price- $365 a year- and convert that into a dollar a day price, people are much more generous with their money.
  • Half the people who were randomly assigned to the aggregate condition were asked to make a donation of $365 a year.
  • The second part of the population, which was randomly assigned to the pennies a day condition, were asked to make a donation of $1 a day, which equals $365 a year.
  • All that John Gourville did was he compared the mean willingness to donate in the aggregate condition with the pennies a day condition.
  • In this case, the cause is the fact that you move away from an aggregate framing to a pennies a day framing.
  • Remember we said that pennies a day works wonderfully well if you are talking about $1 a day or $2 a day.
  • The moment you get into $20 a day or $25 a day, that argument falls apart, because people are now not able to compare 25 with other items that they spend $25 a day on.
  • What you see along the rows are two conditions in which we compare the pennies a day strategy with the aggregate strategy.
  • What you want to show is that pennies a day outperforms the aggregate framing when the amount is low.
  • When we looked at the example of the laundromat, we essentially tracked people over a 40 day window, but it was the same group of people, people that lived in that apartment building that kept visiting the laundromat.
  • Essentially, the same group of people were exposed to two treatments, the cash condition and the prepaid card condition.

Unit 3 > 3.2 Analysis and Interpretation of Results > 3.2.1 The Analysis of Experimental Data

  • Let’s start off by a simple experiment in which you have two conditions.
  • Might have some mean, x, and there would be some people who answer it more than x and others that answer less than x. Likewise, under the pennies a day condition you might have a different mean.
  • Let’s call it y. You have some people that answer it more than y, and some that answer it less than y. The question that you want to ask is, is y the likelihood of purchase significantly greater than x, the likelihood of purchase under the aggregate condition.
  • Are these means different? If you put these two distributions on the same scale, you might end up seeing something like this.
  • This is y. This is x. The question is, is so why statistically larger than x? In any experimental analysis, the null assumption- or what we would call a default assumption- is that there is no difference between x and y. That in fact, both of these distributions are drawn from the same a normal distribution and have the same basic mean.
  • The t-score, which is an index off how different these means are, is computed by roughly looking at a ratio between the difference in means and the variability.
  • The t-test of differences and means provides the likelihood that the null is false, or that the null hypothesis is rejected.
  • Let’s think about a world where we move away from two conditions to an experiment that has many conditions.
  • Again, the null is that the means of all of the groups are equal.
  • The alternative hypothesis is that not all of these means are equal.
  • If you find no support for the null- if you reject the null- you come to the conclusion that the means across your conditions are different from each other.
  • Let’s keep in mind that this analysis doesn’t say how or which means are different.
  • ANOVA is going to look at the difference between the mean of the condition and the overall mean.
  • It looks at the same variable, willingness to pay, and it will look at the difference between that value of each participant in a given group and the mean of it’s group.
  • Let’s look at a case where you have three conditions.
  • We call them condition A, B, C. The average of A is xA.
  • The average and C is xC, and the overall average is x. Between condition variation compares xA with x, xB, with x, and xC with x. Within group variation- or within condition variation- we’ll compare xA with each of the x’s in group A. The overall f statistic is a ratio of the between group variation, divided by the within group variation.
  • The larger the f, it would essentially say that the variation across- or between the groups- is greater than the variation within each group or each condition.
  • Therefore the greater the f, the more likely are you to reject the null and prove that in fact, there is a difference in means across your experimental groups.

Unit 3 > 3.2 Analysis and Interpretation of Results > 3.2.2 Regression

  • DILIP SOMAN: Now, let’s look at a second technique for using data analysis of experimental results, and that technique is called regression.
  • When do we use regression? Regression is particularly useful under two conditions.
  • First, there are a large number of variables that we believe might cause the effect, and second, when these variables don’t come in neat categories, but they might instead be what we call continuous variables.
  • Absolutely any numerical dollar value could be your credit limit, and that’s when regression is particularly useful.
  • What does regression do when we have data from several such people? What it tries to do is it tries to find a line that best expresses the relationship between x and y, and that line might look something like this.
  • What is regression actually doing? Regression is selecting a line such that the sum of the squares of these errors is minimized.
  • That is what we call OLS, the ordinary least squares approach to regression.
  • Now, once we have a line like this, we can actually make predictions about y based on x. So, we say, well, what happens if x is a certain value? And that value is there.
  • We can throw up a line, and we can make a prediction about what the y is going to be.
  • This regression line is written out in the form of a simple equation, which we will talk about in a minute.
  • Together, our regression line can be captured by this equation, y equals to a plus bx.
  • If we had a lot of x’s, a lot of variables that we believe explain y, then our regression line looks a little bit more complicated.
  • This is the simplest form of the regression line.
  • Now, as we said, if there are a large number of x’s, then our regression line is going to look something like this.
  • Let’s say we change the credit limit, and we simply ask people, will you buy a certain product, or will you not buy it? Or in general, it could be multiple levels, if we ask people to choose option A, option B, option C. The intuition behind the regression is still going to be the same, but the kind of analysis that we’re going to use is going to be called a variation of this.
  • It’s going called a logit- l-o-g-i-t- regression.
  • We use a logit regression when our x’s are still continuous, but our y, which is a dependent variable, is a categorical variable.
  • Regression is a very handy when you have complex, continuous data, lots of variables that all explain one particular y. Regression is a fantastic tool to use.
  • In the next few units, we will actually see a lot of examples off experiments that have used both the regression approach, as well as the ANOVA approach, in reaching their conclusions.

3.2 Analysis and Interpretation of Results > 3.2.4 ANOVA Effects

  • In this case, one factor, which is the framing of the payment, has a main effect, consistent effect on the outcome, which in this case is willingness to donate.
  • Now, let’s look at a second experiment that we talked about.
  • We looked at pennies-a-day framing versus aggregate framing.
  • On the left-hand side we see the data for their low-amount condition.
  • What you see here is a replication of the first study, namely, people in the pennies-a-day condition actually were much more willing to donate than people in the aggregate condition.
  • In that condition, we see a reversal of the basic effect.
  • In other words, we find that people who saw the pennies-a-day condition, actually, were less likely to donate than people who saw the aggregate condition.
  • The effect of one variable, in this case the pennies-a-day framing, on the willingness to donate changes as a function of the presence of a second variable, in this case whether the amount is high or low.
  • Every time you have a situation where the effect of one variable on the final outcome changes as a function of the presence of a second variable, we’re going to call that an interaction effect.
  • A main effect is when a given variable has a consistent effect on the outcome variable.
  • An interaction effect is when the effect of variable one changes as a function of the presence or absence of a second variable.
  • Let’s think through the notion of interaction effects a bit more closely.
  • Let’s imagine we take one condition, which is the pennies-a-day framing versus a second condition, which is the aggregate framing.
  • So what we now have is a fully-crossed design, two products times two levels of price framing.
  • On the right-hand side for product two, we see the same exact effect.
  • In this case, there is no attraction effect but simply a strong main effect of the way in which you frame the amount, pennies-a-day versus aggregate.
  • When the amount is high, the aggregate frame dominates.
  • In this case, the presence of the second variable has reversed the original effect resulting in an interaction effect.
  • Let’s say in one condition we showed people who were monthly-wage earners and put them into conditions.
  • In a second group, we took people that were daily-wage earners and then again, did the same thing all over again with them.
  • For those people, we expect that there would be no difference between framing as pennies-a-day verses framing as an aggregate expense.
  • On the right panel, what we would find is that there is no difference between the two types of price framing.
  • This is an interaction effect, but one in which the effect- in which the basic effect disappears when you introduce a second variable.
  • So there are many ways in which you would get interaction effects.
  • An effect would be stronger or weaker as a result of a second variable.
  • The second thing that you need to keep in mind is to look at how we think about capturing interaction effects in a regression setting.
  • In a regression setting, let’s imagine you have two variables, x1 and x2.
  • If you ran a model like this and came up with a regression coefficient which was significant in b3, then you would say that you had a significant interaction effect in a regression sense.

Unit 3 > 3.3 Lab Experiments > 3.3.1 Does the Use of a Credit Card Increase Spending?

  • Does the use of a credit card increase spending? In their paper, the idea was to test the notion that the willingness to pay increases when people pay using a credit card rather than using cash.
  • The factor was the payment mechanism used, and participants were put in either the credit card condition or the cash condition.
  • The first item was a pair of tickets for the Boston Celtics and Miami game, the second item was a pair of tickets for the Red Sox Toronto game, and the third item was a consolation prize that had one Celtics banner and one Red Sox banner.
  • Now, in the credit card condition, participants were told that the payment would be made by credit card, and they were asked to write down their credit card details, the card type, the expiration date, and so on and so forth, much as they would need to do if they ordered any product or service using a credit card.
  • The data analysis technique was simply a comparison of means of the two groups, the credit card and the cash group, and the results were as follows.
  • In this graph here, the gray bars represent the average of participants who were in the cash condition, whereas the red bars represent the average for people who were in the credit card condition.
  • So what did this experiment show? Using a simple two condition design, Prelec and Simister were able to show that purchasing a product with a credit card rather than cash increased the amount of money that people are willing to pay.

Unit 3 > 3.3 Lab Experiments > 3.3.2 Does the Routine Use of Credit Cards Influence Spending Decisions?

  • DILIP SOMAN: Do people that use credit cards in making routine payments differ in the way they make purchase evaluations than people that routinely use checks? That was a question that I was trying to address in a paper that I published in 2001 in the Journal of Consumer Research.
  • The goal, as I said, of that paper was to test whether the routine use of credit cards influences future spending and to identify the mechanisms that underlie these differences.
  • The first was payment mechanism in which participants were either in a credit card condition or a check condition.
  • All participants were told that they had a credit card.
  • In one set of conditions they were told that the credit limit was $3,000, in a second set that it was $8,000.
  • They were randomly assigned to a total of eight conditions that were formed by completely crossing two payment conditions with two feedback conditions with two credit limit conditions for a total of eight different experimental conditions.
  • They had bank accounts as well as a credit card with a given limit.
  • They actually made “Payments,” in inverted quotes, for 12 of these expenses either by writing 12 checks or by signing 12 credit card receipts.
  • They were in a credit card condition, they would see a counterfoil pretty much that you would see when you pay by credit card anyway.
  • Keep in mind, every participant either paid for everything by check or everything by credit card.
  • We found a significant main effect of payment mechanism, feedback, and credit limit.
  • In other words, people were more likely on average to make a purchase in the credit card condition than the check condition.
  • They were more likely to make a purchase when they had a high credit limit as compared to a low one.
  • This line represents the case where the credit limit was $3,000, and there was feedback provided to both the check and the credit card users.
  • It turns out under those circumstances, the difference between the check and the credit card was really low, and it was statistically not significant.
  • Under those substances, you’ll see that there is a dramatic difference between the purchase intention for check and the purchase intention for credit cards.
  • Now if you go from this green line to one of these lines here, where in fact you now have the same level of credit limit but you provide feedback, you notice that the purchase intention declines.
  • When you go from that condition down here where now, you not only provide feedback, but also reduce the credit limit, we get down to our condition here, which is not significantly different.
  • So in somewhat we concluded with this research was that there were two factors that seemed to be driving the effect of credit cards over checks.
  • The second was the fact that credit cards didn’t give people feedback because they were- unlike checks where you wrote down the number, the opportunity to rehearse the price paid was very low for credit cards.
  • Paying for a series of past expenses with a credit card verses check increased the purchase intention for an additional discretionary product.

Unit 3 > 3.4 The Spectrum from Lab to Field > 3.4.1 A taxonomy of studies

  • DILIP SOMAN: Thus far we’ve seen a number of different kinds of experiments.
  • We’ve seen experiments in the lab, we’ve seen some in the field, we’ve seen fully-crossed designs, we’ve seen designs that are simple before and after designs, we’ve seen experiments with you students as participants, we’ve also seen experiments that use adults and people in the marketplace as participants.
  • Let’s think for a moment about how to organize the different kinds of experiments that we’ve seen.
  • I want to organize them into three sets of experiments.
  • In particular, there are two kinds of lab experiments that we can do.
  • You could bring people into a lab, you could ask them to make hypothetical choices that have no real consequences and then analyze their hypothetical choices or their hypothetical judgments that they come up with.
  • The experiment about consumer credit and the fact that people tend to spend more using credit cards as opposed to a check was an example of a hypothetical experiment.
  • In essence, in that experiment, people were responding to a hypothetical set of circumstances.
  • You could also have experiments in which people come to the lab and make real choices that have real monitoring consequences.
  • The experiment we saw with Drazen Prelec and Duncan Simester, where they asked students to bid for tickets to a Boston Celtics game was one such example.
  • The second set of experiments that I want to talk about are what we call natural experiments, or experiments using archived data.
  • The world often conspires to create experiments all around us.
  • There could be all kinds of other natural experiments.
  • Sometimes the natural experiments happen without us even knowing that those experiments have happened.
  • So in some there are two kinds of experiments that the world conspires to create for us.
  • Or their experiments in which the work conspires to create conditions for you, but you might need to go and ask people questions or observe behavior to document the effect of that intervention on people’s behavior.
  • Finally, there are two sets of experiments that we call field experiments and they differ in the scale.
  • Both in field experiments, as well as what we call randomized controlled trials, the researcher actively goes out into the field and they come up with an intervention or a manipulation to look at the effect of a predictive variable on people’s actual decision making.
  • In their experiment they had about 10 factors that they manipulated with about two or three levels each.
  • Now if you put all of these sets of experiments together, you see a spectrum that goes from the very top where you have simple lab experiments where people make hypothetical choices to the very bottom where you fairly complex randomized controlled trials with lots of things being manipulated and with data collection being done in the field.
  • You make sure that what you are studying in this experiment is real, because at the end of the day, you’re documenting real effects on real people making real decisions.

Unit 3 > 3.5 Field Experiments > 3.5.1 Does Restricting the Right to Withdraw Funds Help People Save More?

  • DILIP SOMAN: Does restricting the right to withdraw funds help people to save more? That was the topic of a paper by Nava Ashraf and her colleagues published in the Quarterly Journal of Economics in 2006.
  • The goal of this paper is to evaluate the effectiveness of what is called the commitment savings account.
  • The paper reported one experiment, and we’ll talk about that experiment in this module, but the goal was to test the effectiveness of the commitment savings account.
  • In the marketing condition, there was a face-to-face meeting with the marketer who encouraged them to save, and then offered them the same regular savings account that they saw in the control condition.
  • Finally, in the treatment condition, they again met a marketer who encouraged them to save, but in this case, they were offered the commitment savings account.
  • Now, let’s spend a minute talking about the commitment savings account.
  • Deposits can be made by using a lockbox or an automated transfer from a primary checking or savings account.
  • Other individual characteristics collected included gender, current savings balance, income, and so on and so forth.
  • Now, the data analysis methods used to analyze these data were a regression analysis, and in particular for the choice data, it used Probit, which is a specific form of a Logit regression, and for the savings data, the OLS approach, or the optimal least square approach to regression, but in general, it was regression either in a Probit format for the take up rate or the OLS for the savings account.
  • The take up of the commitment savings product went up with impatience.
  • When it comes to the results on the savings amount, again, keep in mind that it was analyzed by using a regression that had a constant.
  • In other words, the intercept and a couple of coefficients that was significant, so one of the coefficients that was significant for the actual savings six months out was the commitment treatment.
  • People in the commitment treatment were much more likely to save more than people in the other conditions.
  • So what was the conclusion? Females with hyperbolic preferences are more likely to open the commitment savings products.
  • The average of savings balances for people in the treatment group increased by 81% compared to the control group.
  • So in fact, the commitment device does indeed help people save more.

Unit 3 > 3.5 Field Experiments > 3.5.2 Will People Work More on High-Wage Days?

  • In some cases, some days are high-wage days on an average, while others are low-wage days.
  • Why is it difficult to get a taxi in New York City on days in which it rains or in which there’s a convention in town? But you would expect that the general demand for taxis is high, and therefore the average wage rate for a taxi driver is also high.
  • You would expect that taxi drivers would choose to work longer hours on days in which the going is good.
  • The data came in the form of a trip sheet- a sequential list of trips that a driver took on a given day.
  • Next time you sit in a taxi, you might notice that the driver actually has a clipboard where he or she writes down where you left, where you ended up, how much you paid.
  • There were three types of cab drivers in the data set.
  • They were daily fleet drivers who lease their cabs from a fleet company for a 24 hour window.
  • There were lease drivers who lease the cabs weekly or monthly.
  • Finally, there were owner drivers who owned their own medallion-bearing taxi, and drive it.
  • The first data set was provided by a fleet company, and had 70 trip sheets from daily fleet drivers.
  • The second one had 1,044 trip sheets, daily fleet drivers, lease drivers and owner drivers.
  • Whereas the third one had 712 trip sheets just from lease drivers and owner drivers.
  • The labor hours and daily income of cab drivers were collected from these trip sheets.
  • The negative suggests that when the wage rates were high on an average, when the going was good, cab drivers actually drove less.
  • In conclusion, wages were correlated within days, but they seem to be uncorrelated between days, and the wage elasticities were negative.
  • These results here suggested that cab drivers make labor decisions on a daily basis instead of allocating labor and leisure across multiple days.
  • Because the average wage rate is high and taxi drivers quit sooner.

Unit 3 > 3.6 Debate > 3.6.1 Debate 3

  • ITAMAR SIMONSON: Well-designed lab studies have clear advantages over field studies.
  • Unlike field studies, well-designed lab studies can tell you why an effect that you observed happens, how it happens, and under what conditions it’s either weaker or stronger.
  • Field studies, on the other hand, you often wonder about whether they generalize, whether there are other causes for what you see.
  • It’s true that field studies make for good conversation topics and kind of intriguing.
  • Now, field studies may look more real, but actually, well-designed lab studies that capture the essential characteristics of reality can be even more helpful for predicting what happens in the real world.
  • Well, with lab studies, you can include the right controls and test it rigorously and find out exactly under what conditions you have the effect or you do not.
  • If, on the other hand, you were to rely on field studies, you probably will not have the right controls and you would be wondering whether indeed there is a general tendency to compromise or is it just some coincidence that’s reflecting other factors.
  • If you have to choose, it’s clear you have to go with lab studies.
  • OWAIN SERVICE: I firmly believe that field studies add more insight than lab studies.
  • So if lab studies have their place, it’s really to inform the design of the field study, which, by my reckoning, makes lab studies good but not quite as good as field experiments.
  • Now, the reason that I’m giving you these two examples is that we could have learned these insights from lab studies by testing variations of these messages to people that we had, for example, paid to participate.
  • ALISON XU: Well, in my opinion, I absolutely believe that lab studies can provide better insight into consumer behavior.
  • It allows us to do so, because we can isolate the independent variables from other variables and study the effect on the dependent variable of interest.
  • Second, lab studies also allow precise control over other variables that you want to keep constant over different conditions.
  • Finally, because lab studies have well-established manipulations and dependent measures, it allows you to easily replicate.
  • What good is it if I find an effect in a controlled lab environment when I don’t know if it will hold outside of the lab in the so-called real world? What if we have regular people of all kinds of ages, races, with different jobs, not just a fairly homogeneous group of young students? What if we have people who are in their unique situations, rather than students in the lab that know that they are being observed, that know they are being studied, that might try to guess what the experimenter wants from them and act accordingly? In the lab, everything is stylized.
  • Isn’t the real world ultimately what we want to effect, what we need to study, what we need to understand? If so, the best insight is gained by field experiments.
  • Running studies in the lab help you to get around these issues of self-selection by randomly assigning subjects or participants to different conditions.
  • The second reason that I think lab experiments are better than field experiments is that you can go to the limit and study what can really happen rather than what does happen in the field.
  • Whether these conditions really exist in the field or not is not of concern, because we’re trying to study extreme situations of what can happen.
  • It doesn’t mean that all teenage violence is a result of TV exposure or exposure to violent programming on TV. The third reason I think lab studies lead to greater insight than field studies is because they allow us to study phenomena that are really stretched out in reality in a very limited amount of time.
  • In behavioral economics or decision-making research, we often want to study events that are spread out over time.
  • In the Gourville and Soman paper where they study sunk cost effects and payment depreciation, it would be really hard to study this effect in the field, because payment depreciation requires a time lapse between the sunk cost and the effect.
  • So this alternative explanation, I think, really makes field studies problematic in that you can’t have a valid inference of what caused the effect.

Return to Summaries List.

(image source)

Leave a Reply

Your email address will not be published. Required fields are marked *