Behavioral Economics and Labor Market Policy ∗
Carsten S. Nielsen
†Alexander Sebald
‡February 2016
Dansk resum´ e
En typisk økonomisk analyse er, enten implicit eller eksplicit, baseret p˚a en beskrivelse af hvad der motivere folk og hvilken viden de besidder. For eksempel, en analyse af de økonomiske effekter af en ændring i dagpengesystemet vil typisk baseres p˚a beregninger af hvordan dagpengemodtagers motivation p˚avirkes, samt en stillingstagen til hvilken viden han/hun besidder om ændringen. Normalt baseres s˚adanne analyser p˚a en antagelse om, at folk handler i deres egen økonomisk interesse, og at de besidder evnen til korrekt at transformere den information de f˚ar om til viden. Dette er gode simplificerende antagelser.
Indsigter fra psykologi peger dog p˚a, at disse antagelser ofte er for simple. Folk er ikke kun motiveret af deres egen økonomiske interesser, men i høj grad ogs˚a af sociale interesser.
Ydermere, s˚a er vi underlagt naturlige kognitive begrænsninger—vores hjerne har sim- pelthen ikke ressourcerne til at kapere al den information vi modtager. Adfærdsøkonomi er en gren af økonomi som, n˚ar det er nødvendigt, erstatter de typiske antagelser med nogle der er mere psykologisk naturlige. Værdien af den adfærdsøkonomisk tilgange er ikke kun at den kan forbedre præcisionen af vores forudsigelser, men ogs˚a at den kan foresl˚a nye værktøjer, som politiske beslutningstagere kan gøre brug af.
I de seneste 30-40 ˚ar har adfærdsøkonomer kortlagt en række afvigelser fra de an- tagelser, som typiske økonomiske analyser baseres p˚a. Disse afvigelser kan inddeles i tre
∗We would like to thank Pelle Guldborg Hansen and Simon Lamech for very helpful discussions during the writing of this review. This review is a consultancy work prepared for and financed by the Danish Agency for Labour Market and Recruitment.
†Department of Economics, University of Copenhagen, Øster Farimagsgade 5, Building 26, DK-1353, Copenhagen K, Denmark. Phone: (+45) 3532-4839. Fax: (+45) 3532-4932. E-mail: carsten.nielsen@
econ.ku.dk. Web: https://sites.google.com/site/carstennielsen/.
‡Corresponding Author: Department of Economics, University of Copenhagen, Øster Farimagsgade 5, Building 26, DK-1353, Copenhagen K, Denmark. Phone: (+45) 3532-4418. Fax: (+45) 3532-3064.
E-mail: alexander.sebald@econ.ku.dk. Web: http://www.econ.ku.dk/sebald.
kategorier: Kognitive begrænsninger, begrænset selvkontrol, og grænser for egeninteresse.
Kognitive begrænsninger referer til det fysiske faktum, at der er grænser for, hvilke in- formationer vores hjerne kan forst˚a og fortolke. Tilmed vil konsekvenserne af mange af de valg vi træffer først realiseres i fremtiden. For eksempel, nytten af at skrive en god jobansøgning vil først realiseres n˚ar, og hvis, man bliver kaldt til samtale. I disse situ- ationer søger folk ofte øjeblikkelig tilfredsstillelse—det tager tid og kræfter at skrive en god ansøgning, s˚a vi vil hellere lave noget andet—p˚a bekostning af det langsigtet m˚al—at f˚a et godt arbejde. Med andre ord, vores selvkontrol er ofte begrænset. Sidste, men ikke mindst, s˚a træffer folk ofte valg, som ikke kun baseres p˚a deres egen økonomisk interesse.
For eksempel, s˚a ”straffer” ansatte nogle gange deres arbejdsgiver ved at yde mindre, ogs˚a selvom det har økonomisk konsekvenser for dem selv. I afsnit 2 gennemg˚ar vi de vigtigste afvigelser indenfor hver af disse kategorier.
Det skaber nye udfordringer for politiske beslutningstagere, at folk afviger fra de an- tagelser typiske økonomiske analyser bygger p˚a. En grundlæggende, men ofte implicit, antagelse er at folks præferencer bliver ”afsløret” af deres valg. Det betyder, for eksem- pel, at hvis vi observere at folk ryger for meget, s˚a m˚a det være fordi, de foretrækker det. En naturlig konsekvens af s˚adan en analyse vil være, at foresl˚a en højere afgift p˚a cigaretter, s˚a det bliver dyrt at ryge, og folk ændre deres præferencer. Men hvad nu hvis hovedparten af rygerene ikke udviser selvkontrol? Disse folk vil ryge fordi de nyder det i øjeblikket, men senere vil de fortryde det p˚a grund af sundhedsrisikoen. S˚adan en inkonsistens besværliggøre den politiske beslutningstagers arbejde—for hvilken af de to skal tolkes som den sande præference? Afsnit 3 gennemg˚ar nogle af de mest populære retningslinjer, som er blevet foresl˚aet til at løse s˚adanne situationer. Mest prominent af disse retningslinjer er ”libertarian paternalism,” hvor det foresl˚as at man ”nudge’er” folk mod det de ville have gjort, hvis de havde været rationelle, uden at gøre det vanskeligere for dem at gøre som de vil. Et eksempel kunne være advarselsmærker p˚a cigaretæsker.
Form˚alet med disse advarselsmærker er, at gøre folk opmærksomme p˚a sundhedsrisikoen forbundet med at ryge, før de tænder cigaretten. Der er mange fordele ved s˚adan en
”nudging” tilgang, men tilgangen kan ogs˚a være for restriktiv. Hvad nu hvis det for nogle folk er helt rationelt at ryge? Givet at det er et politiskm˚al at f˚a disse folk til at holde op, s˚a vil ”nudging” alene ikke virke. I s˚adan en situation kan det være fordelagtigt at blande
”nudging” med mere gængse økonomiske policy-instrumenter, s˚asom en ekstra afgift p˚a cigaretter, s˚a længe at der tages højde for de adfærdsøkonomiske effekter.
I afsnit 4 præsenterer vi en række nye værktøjer, som kan bruges af politiske beslut- ningstagere n˚ar folk vedvarende afviger fra de antagelser, som typiske økonomiske analyser
baseres p˚a. For eksempel, s˚a vil m˚aden hvorp˚a alternativer præsenteres have stor betyd- ning. Dette syntes næsten trivielt. Fordelen ved en adfærdsøkonomisk tilgang er dog, at det bliver helt kart hvorn˚ar og hvorfor det har en betydning. Listen af værktøjer er ufuldkommen, men best˚ar af de virkemidler, der er fundet mest robuste. De værktøjer vi præsentere er ikke gængse økonomiske policy-instrumenter, og skal derfor forst˚as i en ”nudging”-kontekst. Dette afsnit er tænkt som en praktisk inspiration til politiske beslutningstagere.
Afsnit 5 beskriver den metodiske tilgang adfærdsøkonomi bygger p˚a, nemlig eksper- imentel metode. Det vil være forkert at tro, at eksperimentel metode kun kan bruges indenfor videnskaben. Metoden har i høj grad ogs˚a en praktisk berettigelse. Man kan, for eksempel, test og m˚ale effekten af et politisk tiltag i det sm˚a, før det rulles bredt ud.
I afsnittet diskuterer vi ikke kun mulighederne, men ogs˚a begrænsningerne, ved at bruge eksperimentel metode. Vi gennemg˚ar ogs˚a forskellige typer eksperimenter, der kan være relevante for politiske beslutningstagere. Mere specifikt, s˚a fokusere vi p˚a eksperimenter der opst˚ar natuligt, designet eksperimenter udført i et laboratorium eller p˚a internettet, og designet eksperimenter implementeret i den virkelige verden. Dette afsnit er vigtigt, ikke kun fordi det giver en nødvendige metodemæssige viden, men ogs˚a fordi afsnittet kan inspirerer til at tænke i disse baner.
Til sidst, iafsnit 6, gennemg˚ar vi som cases en række videnskabelige studier vi memer er relevante for beskæftigelsespolitik. For eksempel, s˚a ser vi p˚a hvordan forholdsvis sm˚a tiltage, s˚asom at sende informationsbrochure til arbejdsløse, kan motivere de mest udsatte arbejdssøgende. Derudover gennemg˚ar vi ogs˚a mere anvendte studier, s˚asom dem gennemført af det britiske Behavioral Insight Team.
De tre første afsnit er nødvendigvis mere abstrakte og teoretiske af karakter. Hen- sigten er at give en intellektuel forst˚aelse for, og indsigt i, hvordan adfærdsøkonomi kan hjælpe politiske beslutningstagere. De sidste tre afsnit er af mere praktisk og anvendelig karakter. Vi h˚aber at disse afsnit kan inspirerer embedsmænd og politikere, der arbejder med beskæftigelsespolitik, til at tænke i adfærdsøkonomiske baner.
1 Introduction
All economic models implicitly or explicitly describe what motivates people, what in- formation they posses and how they process the information they receive through their environment. Standard economic theory is mostly based on the premise that people act in their material self-interest and process information correctly, i.e. according to statis- tical rules. Based on findings in experimental economics and psychology showing that people’s judgments and behavior are greatly affected by cognitive and motivational bi- ases like choice overload and self-control, behavioral economics challenges this view and substitutes – wherever necessary – the standard assumptions concerning human behavior with a more descriptively accurate model.
The aim of this review is to give guidance and inspiration to policy makers and regu- lators interested in making use of behavioral economics in the context of labour market policies. To this end, this review summarizes and explains in part 2 the most important of these cognitive and behavioral departures from standard economic theory. In part 3, we describe and discuss the ‘Nudge Agenda’ in the context of public policy. Policy makers and regulators trying to make use of behavioral economic insights are often con- fronted with debates concerning the paternalistic nature of policies that try to help and enable people to overcome their cognitive and motivational biases. This section is thus also meant to provide policy makers and regulators with a short insight into the Nudge Agenda’s philosophical basis: ‘libertarian paternalism’. The following section – part 4 – presents different instruments and tools that can be used to help people affected by be- havioral or cognitive biases to take better decisions. Those tools and instruments include, for example, such things as frames, default options and commitment devices.
In the subsequent section – part 5 – we present and describe the methodological approach which underlies a lot of the studies investigating the nature and impact of our cognitive and behavioral biases: experiments. We discuss what experiments are good for, what types of experiments exist as well as their advantages and disadvantages. This section is important because it not only provides the necessary methodological knowledge to better understand the different findings in behavioral economics, but also shows how experiments can be used to study ex-ante the likely effect of the introduction of a certain policy. Lastly, in part 6 we review many studies that have investigated different cognitive limitations and motivational biases in relation to the labor market and search behavior.
We also mention studies – for example, from the UK Behavioral Insight Team – that investigate policy changes that took into account such biases to increase the effectiveness of existing rules and regulations. The first three parts are necessarily more abstract
and theoretical - providing more of an intellectual background for policy makers and regulators that are unfamiliar with behavioral economics. Whereas the last three parts are more applied in that they discuss different studies that have been done. Certainly one of the main aims of the last section is to provide some inspiration for regulators and policy makers concerning the question: What could be done to inform Danish regulators and policy makers to improve people’s labor market decisions and job prospects?
2 Behavioral Economics
Most behavioral theories modify one or two standard assumptions in the direction of greater psychological realism (Rabin,1998; Camerer and Loewenstein, 2004;DellaVigna, 2009). We will in this paper follow a convention in behavioral economics and classify psychological departures into three broad categories: limits on cognition, limits on self- control, and limits on self-interest. These three psychological departures from standard economic theory have important implications for how people perceive the choices the have.
2.1 Limits on Cognitive Resources
We do live in a complex world and are continuously confronted with an overwhelming amount of information which we have to evaluate before we make our decisions. At least since Simon (1955), economists have acknowledged the fact that human cognition is limited, and have qualified their view on people’s capacity to comprehend and process complex information. Economist have proposed theories of bounded rationality in which people perceive, store and retrieve information accurately, but do not collect all potentially available information because it requires a costly cognitive effort (Stigler, 1961; Akerlof and Yellen,1985b,a).
Theories of bounded rationality suggest that people will make decisions according to standard economic theory if information is costless. But this does not seem to be the whole story. We judge information quickly and intuitively before controlling it. This judgment is automatically biased towards the salient features of the environment we act in (Kahneman, 1973). Unfortunately, what draws our attention is not always what is most necessary or needed for sound decision-making. We often underweigh critical and overweigh other less relevant pieces of information.
What does it mean that ‘we judge information quickly and intuitively before controlling it’ ? To differentiate between automatic and controlled information processing,Kahneman
(2003, 2011) suggest using the intuitive terms ‘System 1’ and ‘System 2,’ borrowed from Stanovich and West(2002). These authors use the term ‘system’ as a label for collections of cognitive processes that can be distinguished by their speed and controllability. System 1 quickly judges information, while System 2 slowly controls these judgments. It is said that System 1 implements ‘automatic’ processes, which are fast and which occur with little or no feeling of effort. For example, you cannot prevent yourself from knowing that
2 + 2
is equal to 4. System 1 evolved to solve problems of evolutionary importance rather than to respect logic. We are born prepared to perceive the world around us, recognize objects, orient attention, and avoid losses. Consider for a moment the illusion in Figure 1. Most people perceive the middle circle to be smaller when placed among the larger
Figure 1: Ebbinghaus illusion.
circles, and larger when places among the smaller circles. The central circle is the same size for both positions, but it appears to change depending on what we place next to it.
The illusion is not just a curiosity, it mirrors the way System 1 is wired: our impressions are always relative to some reference level or point. Reference dependence holds true not only for physical things–circles and consumer goods–but also experiences such as jobs and vacations and short-lived things as emotions, attitudes, and points of view. We compare our job to other jobs, our last vacation with other vacations, and our happiness with the happiness of others.
Judgments made by System 1 represent inputs to the cognitive processes controlled by System 2. The operations of System 2 are slower, serial, effortful, and deliberately
controlled. For example, it demands cognitive effort to know that (2 + 2)·(2 + 2)·(2 + 2)·(2 + 2)
is equal to 256. Most people automatically solve the addition problem but experience slow thinking as they (based on the judgements made by System 1) address the problem of multiplication. The multiplication rule learned in school is first retrieved from memory and then implemented. Carrying out the computation is effortful–it is costly to hold information in memory.
The division of ‘labor’ between System 1 and System 2 is highly efficient because it minimizes effort and optimizes performance. This arrangement works well as System 1 is usually good at what it does, but the swiftness and intuitive way in which it works also leads to biases in judgements and decision making.
2.1.1 Rules of Thumb
System 1 makes quick and intuitive judgements by using ‘rules of thumb’, i.e. mental short-cuts. However, the use of rules of thumb can lead to systematic biases. This insight, first suggested by Tversky and Kahneman (1974), has changed the way we think about thinking. Their original work identified three heuristics–or rules of thumb: availability, representativeness, and anchoring.
Availability. The availability heuristic describes people’s tendency to consider informa- tion that quickly comes to mind as being more likely, more relevant, and more important.
In other words, information that is easily imagined will be more available to System 1 than information that is difficult to imagine.
For example, an employee who works in proximity to the employer’s office is likely to receive more critical performance evaluations than his colleague sitting down the hall simply because the employer’s System 1 will be more attentive to the nearby employee’s errors (Bazerman and Moore, 2013). Similarly, an unemployed will base his assessment of the probability of getting a job on his recollection of the successes and failures of his recent job search. This rule of thumb will often lead to accurate System 2 judgments.
But sometimes it will also fail, mostly because of the fact that memory is drawn towards other factors that can bias beliefs about certain events.
In a beautiful demonstration of availability,Bazerman and Moore(2013) asked people to rank, by order, the following causes of death worldwide: war and civil conflict; star- vation; cancers of the trachea, bronchus, and lung; chronic obstructive lung diseases; and
respiratory infections. Most people ranked the causes in the order listed above. However, this was a mistake. The leading cause of death was respiratory infections that caused 3.5 million death according to WHO, while war and starvation caused 182,000 and 418,000 deaths, respectively. Dramatic deaths such as those resulting from war and starvation tend to get more press coverage than common afflictions such as lung diseases and res- piratory infections. The availability of dramatic stories biases our System 1 towards the first two causes over the last three. As a result, people often underestimate the likelihood of death due to lung diseases while overestimating the commonness of death by war and starvation.
Representativeness. When making judgments people tend to look for information that corresponds to previously formed images or stereotypes. The idea is that when asked to judge if an object A belongs to some category B, then System 1’s rule of thumb is to look for how ’representative’ A is of B.
Employers often use the representative heuristic when hiring new people. They may predict an applicant’s possible performance based on an established category of people that the candidate represents for them. If an employer thinks that the best salespeople are like to be extrovert, ex-athletes, or men, for instance, then the employer will favor those sorts of people for a sales job (Kling et al., 2011). In some cases, the use of the representativeness heuristic offers good first-cut approximations, drawing System 2 to the best options. Other times this rule of thumb can lead to serious errors. For example, it can cause an employer to engage in race or gender discrimination that he would consider disgraceful if he were conscious of it.
The most famous demonstration of such a bias involves the case of a hypothetical woman named Linda (Tversky and Kahneman,1983). Tversky and Kahneman told peo- ple participating in an experiment the following: ‘Linda is thirty-one years old, single, outspoken, and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice and also participated in anti- nuclear demonstrations.’ Following this, participants were asked to rank eight possible categories for Linda from most likely to least likely. The two crucial categories were ‘bank teller’ and ‘bank teller and active in feminist movement’. Most people responded that it was less likely that Linda was bank teller than a bank teller and active in the femi- nist movement. Such an answer is a logical mistake. Linda is, of course, more likely to be a bank teller than a feminist bank teller, because all feminist bank tellers are bank tellers. This judgment error steams from System 1’s use of the representativeness heuris-
tics: Linda’s description seems to be closer to the stereotype of a bank teller that is active in the feminist movement rather then simply the stereotype of a bank teller.
Anchoring. Anchoring simply means that our judgments are influenced by unconscious mental ‘anchors’. In a classic demonstration of anchoringTversky and Kahneman(1974) showed people the outcome of a spin of a wheel of fortune that could range between 0 and 100, and then had them guess whether the number of African nations in the UN was greater than or less than the outcome. Although the result of the spin was random, peoples’ guesses were strongly influenced by the spin of the wheel. For example, among people who got the number 10, the median estimate was 25% African countries in the UN. Among those who got the number 65, the median estimate was 45%. Even though people were aware that the anchor was random and unrelated to the task, the anchor had a dramatic effect on System 1. Paying participants according to their accuracy did not reduce the magnitude of the anchoring effect.
New graduates sometimes experience the effect of anchoring when receiving offers from the place at which they worked as student assistants (Babcocket al.,2012). When hiring, employers know the wage of a student assistant. Inevitably, this wage will act as an anchor and influence the offers that the student receives. This is despite the fact that a student assistant’s pay may only be weakly related to his future performance. A better valuation of a student assistant’s future performance would probably be what he could earn elsewhere. In addition, after having accepted a job, the future wage typically only increases as a percentage based on current salary. This implies that a new graduate who negotiates aggressively upfront will obtain a higher start pay, which then will serve as an anchor for future years’ wages. But the tendency to negotiate may be unrelated to his performance on the job. For example, evidence suggests that women are less inclined to bargain in situations such as wage negotiations than men (Bowles et al., 2005).
2.1.2 Reference Points and Loss Aversion
Overwhelming evidence has identified another pervasive feature of System 1: people judge elements of their environment relative to reference points. For example, we use reference points to categorize choice options in terms of gains or losses that are potentially associated with them and - in many situations - we are more averse to losses than attracted to same- sized gains. This is, we are loss averse. In a seminal demonstration of this Kahneman et al. (1990) randomly gave mugs worth about $5 each to one group of people. Those given the mugs reported the price at which they were willing to sell (using a procedure
that ensured reliable reports). Those not given a mug had to state the price at which they were willing to buy. These ‘sellers’ and ‘buyers’ faced precisely the same choice between money and mugs, but their reference point differed. Those given a mug treated having the mug as their reference point and considered not having a mug to be a loss, whereas the reference point for those who did not get a mug was: not having a mug. In one of Kahneman et al. (1990)’s experiments, the median value placed on a mug was $3.50 by buyers but $7.00 by sellers–a ratio of two. Many researchers have replicated this finding in many different contexts. The example also illustrates whatSamuelson and Zeckhauser (1988) call a status quo bias, a preference for the current state of affairs: the disadvantage of selling the mug loom larger than its advantage.
The key question is of course what determines the reference point. A promising candi- date is expectations: what people expect affects how they feel about what actually occurs (K˝oszegi and Rabin,2006). In the example above it can be argued that people expect to keep their endowment (i.e., ‘mug’ or ‘no mug’) and that this expectation is what affects their valuation. This way of thinking about the reference point has been very successful in many fields of economics. Numerous examples exists. Here we will give two that are relevant for labour market policy.
Reference points based on expectations have been used to explain how the number of hours people work changes with their hourly wage. The standard economic prediction is that people optimally trade-off work for leisure by working more when wages are high and less when wages are low. However, among taxi drives, who are perfect for analyzing this relationship,Camerer et al. (1997) find that as the hourly wage increase drives work fewer hours. The special thing about taxi drivers is that they choose how many hours they would like to work each day. Typically, they lease their taxis for a fixed amount of money and drive for as long as they like within a given period. In addition, their hourly wages varies every day due to weather, the day of the week, holidays, etc. Although rates are fixed, taxi drives will earn more on good days because they have more customers per hour. The negative relationship between hourly wage and hours worked can be thought of in terms of drivers having a daily reference earning based on expectations. On good days when wages are high, drivers will reach their reference earning quickly and quit early.
While on bad days when wages are low they will drive longer hours in order not to lose earnings relative to their reference point.
Such reference points can also explain why unemployed sometimes are slow and even reluctant to accept jobs they objectively should take (Babcock et al., 2012). According to standard economic theory this causes a problem because unemployed should return to
work when they get a job offer which pays more than their objectively reasonable reser- vation wage. However, in reality unemployed tend to remain on unemployment insurance for inefficiently long periods. A reason for such behavior is that some unemployed base their reference wage on the expectation that they will get the same wage in a new job as their old job paid, independent of whether or not this is realistic given the prevailing labor market conditions. If they take a job paying less than this reference wage, they feel a loss. The aversion to realize such a loss will lead them to be reluctant to accept job offers below their reference wage.
2.1.3 Confirmation Bias
In addition to the previously mentioned aspects of System 1, a lot of research suggests that once an expectation has been formed or a judgment has been made, System 1 has the tendency to be inattentive to new contradicting information. People seem to be more attentive to information that is coherent with their expectations and previously formed judgements. For example, once you become convinced that one type of job is more lucrative than another, you may not sufficiently attend to evidence suggesting this hypothesis is flawed. This tendency is called confirmation bias.
To get a sense of how the confirmation bias works, consider Bazerman and Moore (2013) example of judging whether marijuana use is related to unemployment. In assessing this, most people typically try to remember several marijuana users they know of and recall whether these people were unemployed. However, a rational analysis would require you to recall four groups of people: marijuana users who are unemployed, marijuana users who are employed, unemployed who do not use marijuana, and employed who do not use marijuana. However, our everyday decision-making commonly neglects this fact. Instead, we intuitively use selective data when forming expectations. Our focus on selective data or a single possible cause of an effect such as unemployment may lead us to neglect alternative causes and conclude that the association is stronger than it is in reality. Thus, we may falsely conclude that marijuana has a stronger association with unemployment than it does in reality.
The confirmation bias leads to overconfidence, in the sense that people on average believe more strongly than they should in their favored explanation (Rabin and Schrag, 1999). Consider, for example, an employer who has narrowed the field of candidates for a job down to two persons: Ann and Bob. The employer may believe either that Ann is the better candidate or that Bob is the better candidate. Assume that the employer initially (before receiving any further information) believes that both possibilities are equally likely.
During the job interview the employer collects information that helps him to identify who is the better candidate. If, after receiving one or more pieces of information, the employer believes that Ann is probably the better candidate, the confirmation bias may lead him to erroneously interpret his next piece of information as supporting this expectation.
The idea is that if the next piece of information is against Ann (thus supportive of Bob), then the employer who believes strongly that Ann is the better candidate is likely to misinterpret this piece of information as being supportive of Ann. Therefore, the employer’s new belief that Ann is the better candidate may be stronger than is warranted.
The notion that the employer is likely to believe ‘too strongly’ that Ann is the better candidate corresponds to the commonly held intuition that the confirmation bias leads to overconfidence. The confirmation bias also underlines the commonly held belief that first impressions matter most.
Overconfidence, of course, has many facets. People might be overconfident with regard to their knowledge and skills, with regard to their skills relative to the skills of others and might also think that they have more control over future outcomes than they actually have.
As will be discussed also in the last section 6 such types of overconfidence might impact people’s labor market decisions and search behavior. For example, if an unemployed is overconfident regarding the effectiveness of his or her search effort, then he might search too little which ultimately leads to dissatisfaction and long then necessary unemployment spells.
2.1.4 Choice Overload
We have seen some of the mistakes System 1 can make when we intuitively interpret our environment and evaluate information. The evidence clearly demonstrates that people are prone to judgement errors even when choosing among only a handful of alternatives.
Errors only get worse as the number and complexity of decisions increases (Jacoby,1984;
Iyengar and Lepper,2000). As more decisions are required and more options are available, the challenge of making the right decision becomes increasingly difficult.
With many decision, the consequences of making errors may be trivial. But with some, the consequences of error may be quite severe. We may make bad investments because we cannot comprehend the consequences of investing in the various complex possibilities.
We may choose the wrong insurance because we do not have time to read all the fine print. We may go to the wrong school, embark on the wrong career, all because of the complexity with which options were presented to us. When getting involved in more and increasingly difficult decision, we may be forced to make many of those decisions with
inadequate reflection - an effect which is called choice overload.
Choice overload relates to a general finding in psychology, namely that people dislikes difficult choices so much that they sometimes want to avoid choosing at all and stick with status quo (Tversky and Shafir, 1992). For example, selecting a health insurance plan is often too complex and people too often stick with the default plan (Chetty et al.,2013).
Having too many choices can lead to a decrease in people’s motivation to choose and may make people less satisfied with the choice they actually make (Scheibehenne et al.,2010).
2.2 Limits on Self-Control
The last two psychological dimensions we discuss in this section are not related to how people process information but rather related to people’s motivation, i.e. their preferences.
One element of people’s preferences regards their weighting of costs and benefits that occur at different points in time. Often we have to take decisions that involve consequences that occur at different points in time, e.g. a cost today and a benefit in the future. The weighting of current benefits or costs vs future benefits and costs is captured in people’s time preferences.
Experiments analysing people’s time preferences have shown that people have a taste for immediate gratification. We procrastinate on tasks such as writing a job application that involves immediate costs and delayed rewards, and do things such as browsing the internet that involves immediate rewards and delayed costs. Standard economics models such tastes by assuming that people discount future utility at a constant rate. An im- portant feature of such an assumption is that it implies that a person’s preferences are time-consistent: A person feels the same about a given tradeoff today and tomorrow.
However, our short-term tendency to pursue immediate gratification is often incon- sistent with our long-term preferences. Today we may feel that it is best to not eat too much tomorrow, but when tomorrow comes we tend to overeat. Today we think it is a good idea to send a job application tomorrow, but when tomorrow comes we tend to put it off. More generally, when considering trade offs between two future moments, we give stronger relative weight to the earlier moment (Frederick et al., 2002). For example, peo- ple have been found to seek immediate gratification when selecting pricing plans for gym memberships (DellaVigna and Malmendier, 2006). Expensive monthly plans make sense only if people intend to go to the gym a sufficient number of times because this makes the average daily cost of the monthly plan less than the price of a day pass. However, it turns out that people who choose the monthly pass attend the gym to few times during the month to make it worthwhile. People receive immediate gratification months or weeks
in advance when they decide on the days they want to use the gym but then, when many of those days arrive, they decide that the cost is too high, and they would rather not go.
The failure of people to display time-consistent preference is an example of the gen- eral tendency of bounded self-control. Translating intention into action seems to involve behavior that standard economic theory does not allow. People sometimes do things that they do not want to do or fail to do things that they wish they had done. In the case of a gym membership, people may engage in procrastination, failing to take actions that they intended to take. Conversely, time-inconsistent preferences can lead people to cave into temptation and take actions they did not intend. For example, people prefer junk food and trashy movies, while stating a preference for healthy food and high-brow films when making plans for later (Read et al., 1999; Milkman et al., 2010).
Peoples’ ability to exhibit self-control depends not just on the context of choice but also on their visceral state. First, the emotional state of individuals can induce lack of self-control. For example, stress can make quitters more likely to start smoking again (Shiffman and Waters,2004). A similar effect may result from other visceral states, such as hunger or fear (Loewenstein,1996). Second, people can have a difficult time predicting their future selves at the time of forming their intentions. In particular, people tend to display a ‘projection bias’–a tendency to project their current preferences onto their future selves (Loewenstein et al., 2003). So, for example, when ordering by catalog, people are more likely to return orders for cold weather gear when orders are made on unusually cold days. At the time of placing the order, people are thought to project a desire for such gear that does not persist when the order arrives and the weather has improved (Read and Van Leeuwen,1998).
There are of course alternative explanations that we have not mentioned yet. For example, failures of self-control can be thought of as a result of discount rates changing over time (Laibson, 1997) - often referred to as hyperbolic or present-biased preferences.
They can also be thought of as a result of conflict between mental processes by which people plan and those by which they act (Thaler and Shefrin, 1981). Alternatively, they might be thought of as a result of a decision-making process in which self-control demands willpower and willpower is costly to exercise (Loewenstein and O’Donoghue,2004). These alternative models and hypotheses are closely linked to the fact that cognition is limited.
2.3 Limits on Self-Interest
The assumption that people only do things out of self-interest is often a correct sim- plification that is often useful in economics. However, people sometimes also choose to
punish others who have harmed them, reward those who have helped them, or to make outcomes fair even if it is costly for them. A convincing demonstration of such behavior is the ’gift-exchange’ experiment (Fehr et al., 1993). In a typical gift-exchange experiment the employer has a pot of money, typically $10. Out of this, he can give some amount between $0 and $10 (denoted X) as a ‘wage’ to the employee. The employer will keep the remaining amount of money for himself. The employee can then accept or reject X.
If the employee rejects the wage X, then both receive $0. If the employee accepts the salary X, then he has to exert costly ‘effort’ by giving an amount Y between $0 and X back to the employer. A higher effort increases the employer’s profit but is costly to the employee. According to standard economic theory, a self-interested employee should choose the lowest feasible effort (Y = $0) and never reject X, and a self-interested em- ployer should anticipate this and offer nothing (X = $0). However, in most experiments the employers pay a substantial wage and employees pay back by exerting effort, which is surprising because everything is anonymous. Interestingly, the effort varies positively with the wage as if the employer feels an obligation to reciprocate by making the outcome fairer.
Evidence for the importance of fairness in actual labor markets has been documented beyond the gift-exchange context. For example, Krueger and Mas (2004) examined the impact of a labor conflict at a Bridgestone-Firestone plant on the quality of the tires they produced and found that union workers reciprocated the lockout by lowering the quality of their work. The union workers went on strike in July 1994 and were replaced by replacement workers. The union workers were gradually reintegrated in the plant in May 1995 after the union, running out of funds, accepted the demands of the company. A final agreement was not reached until December 1996. It was found that the tires produced at this plant from 1994 to 1996 were ten times more likely to be defective than usual. The increase in defects did not appear to be due to lower quality of the replacement workers–
the number of defects was higher in the months preceding the strike (early 1994) and in the period in which the union workers and the replacement workers worked side-by-side (end of 1995 and 1996).
Another example comes fromBandieraet al.(2005) who examined personnel data from a fruit farm in the United Kingdom that paid its workforce according to a relative incentive scheme and then switched to piece rates. Under relative incentive scheme, workers’ daily pay depended on how many pieces of fruit they picked as well as the average amount coworkers picked on the same field. In contrast, under piece rates workers’ pay only depended on how many pieces of fruit they picked themselves. To identify whether workers
had fairness concerns, the amount of fruit picked by a worker under the relative incentive scheme was compared to the amount he picked under piece rates. The comparison was revealing because under the relative incentive scheme individual effort has a negative effect on coworkers’ pay, whereas under piece rates individual effort has no effect on others’ pay.
The difference in the amount of fruit a worker picked under the two schemes, if any, then provides evidence on whether and to what extent workers takes into account the negative effect they imposed on their coworkers pay. It was found that, after the change to piece rate, the productivity of each worker increased by 51.5 percent. Workers clearly reciprocate coworkers low effort by exerting low effort themselves in the relative incentive scheme. The result was not due to a change in incentives because the flat piece rate was on average lower than the relative-pay piece rate, which would only have contributed to lowering, rather than increasing, productivity after the switch.
By large, people will often behave in a way that conforms to community norms (Elster, 1989). For instance, results suggest that in addition to an intrinsic preference for fairness, people also has a desire to be perceived as fair by others. For example, people given a flyer are less likely to dispose of it by littering in environments that have been manipulated to be relatively free of litter (Cialdiniet al.,1990). Direct messages that indicate to people that most other people behave in a certain way have been found to promote conformity to that behavior (Goldstein et al., 2008). For example, showing people how their consumption of energy compared with that of their neighbors–framing above-average energy use as undesirable–was found to reduce energy consumption (Ayres et al., 2013).
3 The Nudge Agenda and Public Policy
Standard economic theory judges policy-objectives based on welfare analysis and rational- ity. When people are rational their choices will reveal their preferences. These preferences can then be aggregated across society to account for social welfare, which is the objective policy-makers should maximize.
Unfortunately, when people are bounded rational (e.g., face cognitive as well as moti- vational challenges) their choices do not necessarily reveal their preferences. The reason is that choices need no longer be logically consistent and the precept that they coin- cide with people’s preference breaks down. Take the case of cigarette smoking. People choose to smoke cigarettes today because they really enjoy it. At the same time, some expect to regret this decision tomorrow because smoking causes lung cancer. Such incon- sistencies posse a problem when choosing the policy-objective; which of the two should
be interpreted as being the ‘true’ preference? In general, when implementing policies, policy-makers must in these situations make judgements about how to weigh the various conflicting preferences that people appear to hold. This task is not an easy.
The least intrusive way to solve this predicament has been labelled ’libertarian pater- nalism’ byThaler and Sunstein(2003). Libertarian paternalism is a relatively weak, soft, and non intrusive type of paternalism because certain choices are not prohibited or made significantly more difficult–it does not involve coercive actions like forbidding options or significantly changing economic incentives. Still, the approach is paternalistic, because people are not free to do as they like as a purely libertarian approach would prescribe.
Rather, policy-makers attempt to move people in the direction they would have gone had they been rational, without making it more difficult for people to do what they want. For example, trying to make people aware of the long-run cost of smoking by putting warning messages on cigarette packaging counts as a libertarian paternalistic intervention. Such an approach does not limit the choice of somebody who rationally choose the number of cigarettes he wants to smoke, weighing the current benefit against the future cost. Impor- tantly, a libertarian paternalistic policy should not affect smokers who are rational and move non-rational smokers in the direction they would have gone had they been rational - a direction that for some is abstention. Policies that follow this philosophy have been labeled ‘nudges’ (Thaler and Sunstein,2008).
A paradigmatic nudge is defined as: ‘any aspect of the choice architecture that alters people’s behavior in a predictable way without forbidding any options or significantly changing their economic incentives’ (Thaler and Sunstein, 2008, p. 6). Nudging does therefore not reduce people’s freedom of choice. When nudging policy-makers should design the context, or choice architecture, so that people are steered towards welfare promoting decisions that they would have take in the absence of e.g. cognitive limitations and self control problems. A very clear example of a nudge which helps people to overcome their self-control problems is the canteen example hinted at above. If a manager of a canteen would like to help people in eating healthier food he can rearrange the order in which the dishes are presented in such a way that people have to pass by the fresh and healthy dishes first before being directed to the fattier dishes. In this way, the manager does not restrict the choice set of people in any way, but designs the choice environment in such a way that people with self control problems will unconsciously fill their plates first with healthy food leaving only little space for the fattier dishes at the end.
It is easy to understand why nudging is attractive to policy-makers. At a small or no cost they can change the context in which people choose and by doing so steer them
in the wanted direction. So even if the effect of a nudge is not substantial, it can be more cost-effective than other costly policy interventions. The popularity of the nudging concept has also been supported by the fact that it does not violate the fundamental tenets of the standard economic theory. The kind of paternalism implied by nudging seems acceptable to a policy-maker with a libertarian view as it does not reduce peoples’
freedom. It is also consistent with a belief in the power of competition, markets, and incentives, and a crucial but limited role of government. The attractiveness of nudges has initiated ‘nudging units,’ to encourage evidence-based behavioral policy-making first in the United Kingdom in 2010 and then in the United States. More recently, initiatives have also been seen in countries such as Australia, Canada, Denmark, France, Germany, Israel, Netherlands, New Zealand, Norway, Singapore, South Africa, Turkey and Saudi Arabia (Madrian,2014).
Nudges, however, might be to limited in scope and applicability because of the fact that some people rationally choose against policy-objectives. For example, a policy-maker might want to change the behavior of those who rationally choose to smoke. Camereret al.
(2003) suggest ’asymmetric paternalism’ as a slightly more intrusive form of regulating behavior that goes beyond nudging. Asymmetric paternalism, like libertarian paternalism, advocates that policy-makers should aim at moving non-rational people towards what is rational. However, this form of paternalism also allows, by use of coercive actions, to impose a little harm on those who are rational as long as the welfare benefits are higher than the costs. An example of an asymmetric paternalistic policy, which is not libertarian paternalistic, is the ‘sin tax’ imposed on cigarettes. Putting a tax on cigarettes hurts rational smokers who do not exhibit self-control problems and rationally choose the number of cigarettes they smoke. On the other hand, the tax is beneficial for smokers who would otherwise smoke too many cigarettes due to insufficient self-control and later regret their choice. An asymmetric paternalistic policy is thus allowed if it improves welfare.
That is, the policy should benefit the non-rational smokers more than it hurts the rational smokers.
Furthermore, despite the wide acceptance of nudges as behavioral policy-making tools there is some well-founded criticism of how using labels such as ‘nudging’ limits the way we think about possible interventions. Here we mention two major points. First, the fact that nudges narrow the focus on choice architecture may cause us to overlook other important policy tools. For example, another way of moving people in the direction of rationality is to foster people’s critical thinking and their capacity to decide (Binder and Lades,2015).
This strengthens the cognitive aspect of decisions, promotes self-empowerment, and seeks
to make people independent of external influences. People are helped to engage in mental strategies, thus debiasing their decisions. Improving people’s critical thinking is not an attempt to directly change people’s behavior, but rather to improve the process used to reach a decision.
Second, the nudging literature’s postulate of not significantly changing people’s eco- nomic incentives may in many situations be too restrictive (Bhargava and Loewenstein, 2015). Economic incentives can be a powerful motivator of interest to the policy-maker.
For example,Laceteraet al. (2014) conduct a field experiment that studied peoples moti- vation to donate blood when economic incentives where varied. They found that donation rate increased with the size of the economic incentive. However, they also found that peo- ple who were unaware of the economic incentives when they showed up to donate blood, subsequently gave less relative to the control group who were not given any compensa- tion. This example highlights the importance of taking behavioral effects into account when considering changing monetary incentives. According to standard economic theory increasing the economic incentives should always make people more motivated. However, behavioral economics has found that providing small economics incentives for behaviors that usually tend to carry some level of personal/intrinsic reward can reduce motivation and lead to lower effort. In a famous field experiment Gneezy and Rustichini (2000) showed that when a daycare provider started issuing fines to parents for picking up their children late, the number of late pick ups actually increased. Evidently, attaching a price to late pick up legitimized the behavior in the mind of parents. Thus, in situations where economic incentives are a potentially cost-effective approach to changing behavior, behav- ioral economics can inform us how to design economic incentives to make them maximally efficient.
These two points make it obvious why it is the broader concept of behavioral eco- nomics (which encompasses the nudging agenda) which should be taken into account when designing and evaluating policy interventions. As will also be seen in the context of the cases presented in section6, behavioral insights can help to design better policies that more effectively achieve the chosen policy objectives - an issue not only but of particular importance for labour market policies. Danish policy makers and administrators should, for example, use behavioral insights to optimize the presentation of the information that unemployed get and the timing in which they receive the information. It is important to take behavioral insights into account to help people overcome their cognitive biases and self-control problems that might hamper their search activities. Furthermore, behavioral insights should be used as well to help administrators and case workers to optimize the
support they give to unemployed as it is often their expertise and judgements which are decisive in motivating the unemployed to stay active in searching for a new job.
4 Tools for the Behavioral Policy-Maker
What are the tools that policy makers and regulators have at their disposal to help people overcome their cognitive and motivational biases? In this section, we consider a selection of behavioral ‘tools’ that are relevant for policy-makers to know. The list is not intended to be exhaustive, but instead highlights the tools that we believe are supported by the strongest scientific evidence.
4.1 Framing
The specific way in which a decision situation presents itself (e.g., the way a specific type of information is presented, the specific labels used to describe the choice options, the specific language used to describe the decision situation) is usually referred to as the
‘frame’ of a decision situation. It is often possible to present the same decision situation using different frames. Standard economic theory suggests that frames, i.e., the way a certain type of information or alternative is presented or the language or labeling used, should not matter for the decisions taken. A lot of research however suggests that frames matter for behavior (Kahneman and Tversky, 1984).
Framing has shown to e.g. influence the reference points that people use in judging the value of their choice options. Framing is thus a very powerful policy tool especially when taking into account that people are loss averse. A natural consequence of this combination is the possibility of influencing the behaviour of people by changing whether people perceive a given outcome of their choice as a gain or a loss - through changing their reference point.
Consider the following two examples. First, when searching for a job and thinking about an appropriate wage, the reference point might be given by the wage received in the last job (DellaVigna et al., 2015). I.e. accepting a job which pays less that the previous job might be difficult because it entails the realization of a loss. Thus people might be reluctant to accept those offers. Second, at the time of job loss, the reference point of an unemployed might be his previous wage, which is significantly higher than the unemployment benefit. The unemployed, therefore, finds the new state of unemployment particularly painful given the loss in income relative to the reference point. This causes
him to search hard for a new job at the beginning of the unemployment spell. Over the course of unemployment, however, the reference point shifts as people adapt to the lower benefit level. This might cause the unemployed’s search effort to decrease over time. As the end of the unemployment insurance benefits draws near, the unemployed anticipates the loss in income due to the exhaustion of the benefits and searches harder again. If the unemployed does not find a job before the unemployment insurance ends, he once again slowly adjusts to the new lower benefit level. An obvious policy implication taking this loss averse behavior into account is that an unemployment insurance scheme that relies on multiple declining steps may increase job search effort. In fact,DellaVigna et al.(2015) found that introducing a new step in the Hungarian unemployment insurance reform speeded up employment, while being revenue-neutral from the perspective of the government.
Framing need not be relative to a reference point or include gains and losses to have an impact. Simple labeling can also frame situations in different ways and in this way impact people’s behavior. For example, Saez (2009) studies the impact of framing a financial incentive to open an Individual Retirement Account (IRA) at the time of tax filing, when incentives are framed either as a match or as a tax credit. This study was motivated by a presumption that the Saver’s Credit, a feature of the US tax code designed to encourage lower-income households to save, was largely ineffective because people did not understand tax credits and did not know what they are missing out on. Saez (2009) found that framing the incentive as a match was indeed more effective; doing so resulted not only in more people opening an IRA but it also increased contributions. Framing by labeling is thus also an important policy tool.
How money is labeled also impacts how it is spent. In a policy context, the labeling of government transfers affects how money is spent even if, in reality, the money can be used for anything. For example, Kooreman (2000) finds that the tendency to buy children’s clothing is 10 times larger out of an income labeled as a ‘child benefit’ than out of other income sources. The labeling of income as a ‘child benefit’ is believed to create a moral obligation to spend that money on their children. Similarly, Benhassine et al. (2014) found a significant impact on school enrollment of a transfer program in Morocco that labeled income as ‘child education,’ although the funds could be used for other purposes.
These results suggest that careful consideration should be given to the labels attached to any policy intervention. For example, the labeling of unemployment benefits may impact behavior. As Madrian (2014) observes: in the United States, unemployment benefits are referred to as unemployment insurance, a label that reinforces a recipient’s
status as unemployed; in contrast, in the United Kingdom, unemployment benefits are referred to as a job seeker’s allowance, a name that emphasizes a recipient’s attachment to and activity in the labor force.
4.2 Default Options
A default option is an option that obtains if nothing is chosen. A good example of a default option comes from Johnson and Goldstein (2003)’s study on organ donation. In many countries the default is not to be an organ donor; people must opt-in to be a potential organ donor at death. In these countries, the fraction of individuals who sign up to be organ donors is relatively small. Other countries have a system where the default is to be an organ donor; people must opt-out if they do not wish to be organ donors. In these countries, the fraction of people who opt-out of organ donation is extremely low. In fact, Abadie and Gay(2006) showed that actual organ donation rates were 25−30% higher in opt-out countries relative to opt-in countries.
Another domain where the default option is important is savings. In the United States, for example, savings plan participation rates are substantially higher when the default option is ‘automatic enrollment’ in the savings plan and people must opt-out if they prefer not to save. In the first study of the impact of automatic enrollment on savings outcomes, Madrian and Shea (2001) document a 50% increase in savings plan participation for newly hired employees at a large corporation that switched from an opt-in to an opt-out automatic enrollment regime. Other subsequent studies document similar participation rate increases (Choi et al., 2004, 2006).
Default options might also be applied in the domain of job search assistance. (Spin- newijn, 2015a), for example, finds that people underestimate the benefits of searching for a job. Policies might thus seek to promote enrollment in job assistance activities beyond what people would choose on their own. Agencies can default people into such services, for example, by worker profiling upon signing up for unemployment insurance.
Default required job search assistance may help people to overcome their tendency to procrastinate.
However, when there is substantial differences in people’s preferences, specifying a default option may be difficult because any default is unlikely to align well with preferences for more than a small minority of people. If this is the case, then it may be useful to require people to actively choose. In the savings domain, Choi et al. (2003) compare the outcomes in an employer-sponsored savings plan before and after employees were required to make an active choice about whether to participate. They found that when
not requiring people to make a choice, only 41% of newly hired employees enrolled in the savings plan. In contrast, when required to make an active choice 69% enrolled.
4.3 Simplification
The complexity of a task may generate slow execution of desirable behaviors; it may even stall them. An example that has received much attention is the simplification of the process of applying for college financial aid in the United States (Madrian, 2014).
Until recently, the gateway to financial aid, the FAFSA form, was eight pages long and included over 100 questions. As a consequence, a sizable fraction of eligible students did not even bother to apply for financial aid. Bettinger et al. (2012) conducted a field experiment designed to test the effect of personal guidance through a streamlined version of the FAFSA form. They found that this approach to simplifying the aid application process increased the fraction of targeted families with high school seniors who apply for college financial aid by 16%; it also increased the fraction of children who attended college by 7%. The US Department of Education has subsequently implemented this simplified process.
Evidence from a broad range of evaluations suggests that job search assistance pro- grams are typically effective at helping people to find a job more rapidly by simplifying the task, making it less effortful (LaLonde, 1995; O’Leary, 2004). This seems to be true for most programs and for most people who use them, also those most at risk. Optimal job search, for example, requires considering information about job market conditions and how they match with own qualifications in a way that most people find difficult.
Job search assistant help and tools, both online and in one-stop career centers, can help simplify and streamline the search experience of job seekers (Babcock et al., 2012).
Finally, a more heavy-handed approach to reducing the cost of complexity is removing choices. People cannot comprehend situations where they have too many choices. Effects include procrastination, avoidance, dissatisfaction, reliance on heuristics, and potentially mistakes. When making choices people screen options to form a consideration set from which they then choose. What ultimately matters is not the total number of options available but rather a limited number of options that people can keep ‘in mind.’ Searching for options takes time and effort. For example, job searchers have been found to prolong their search in the face of many opportunitiesIyengaret al.(2006). Search time and effort can be made smaller by reducing the total number of options available. Job-websites can, therefore, help job seekers by narrowing down opportunities and ease search effort.
4.4 Commitment Devices
Commitment devices represent a category of tools that can help people execute their pref- erences when they are likely to cave into temptations that generate short-run benefits that are outweighed by long-run costs. Job search is an example of such. As mentioned before, people have a tendency to procrastinate on job search activities. Introducing commitment devices to overcome this procrastination problem could potentially be efficient.
Babcock et al.(2012) suggest using monitoring of job search effort as such a commit- ment device. They propose two monitoring options. First, that unemployed should have frequent contact with an unemployment agency to demonstrate active job seeking. Sec- ond, the unemployment agency should monitor the unemployed’s job acceptance strategy.
Monitoring is to some degree already a part of most unemployment insurance systems.
For example, to be eligible for benefits in most systems, the unemployed must be actively looking for work and must be available to start a job immediately. However, in most systems there is also some period at the beginning of an unemployment spell where there is minimal monitoring. Job seekers are in this period typically allowed to restrict avail- ability to jobs in their occupation or on the basis of pay (Paserman, 2008). This period enables, at least for a while, rational job seekers to pursue job opportunities without being restricted in their freedom to choose as they like.
Several US states have also experimented using re-employment bonuses programs as commitment devices, modeled on the successful Illinois Re-employment Bonus Experiment (Woodbury and Spiegelman,1987). A typical re-employment bonus program could involve a bonus equal to 10 weeks of unemployment benefit to workers who found a job within 13 weeks and then where able to hold that job for 13 more weeks (Paserman, 2008).
Evidence shows that applying such commitment devices has a sizable effect on job search success (Meyer, 1995), especially the re-employment bonus program has been shown to be successful.
Research in psychology has identified a lack of planning as another barrier that im- pedes individuals from executing on their preferences (Gollwitzer, 1999; Gollwitzer and Sheeran,2006). Without a plan for implementation, people who face competing demands for their attention are prone to forget what it is they wanted to do (Madrian,2014). En- couraging people to form a plan to carry out their intentions has been shown to increase the attainment of desired goals in a variety of relevant policy domains. One effective way of doing so is by prompting people. For example, (Nickerson and Rogers,2010) evaluate the effectiveness of prompting people to make a concrete voting plan by asking them a series of questions: (i) ‘Around what time do you expect to head to the polls on Tues-
day?’, (ii) ‘Where do you expect you will be coming from when you head to the polls on Tuesday?’, (iii) ‘What do you think you will be doing before you head out to the polls?’.
They find a 9% increase in voter turnout among voters from single-voter households, who they posit are less likely to have other support mechanisms in place to encourage voting.
A natural complement to planning aids is the provision of reminders to follow through on a desired course of action. Reminders can take a variety of forms. For example, reminder letters or SMS are among the most cost-effective ways to encourage immu- nization, increasing immunization rates by 8% on average (Briss et al., 2000; Szilagyi et al., 2000). Reminders have also been effective at encouraging savings. Karlan et al.
(Forthcomming) provide evidence from three different field experiments on the effect of providing reminders, either text messages or letters, on savings in Bolivia, Peru, and the Philippines. They found that reminders increased the total savings by 6%, and increased the likelihood that people achieve their savings goals by 3%. Similarly, in a savings field experiment conducted in Chile, Kast et al. (2012) found that people who received text message reminders saved substantially more than people who did not. A combination of planning aids and reminders could be an effective way to encourage, for example, more active job seeking for workers who have lost a job (Madrian, 2014).
4.5 Social Norms
A final category of behaviorally informed policy tools derives from the observation that people are not only self-interested but rather evaluate outcomes in a social context, that is, in terms of what others around them are doing and the judgments that others may pass on their behavior. From a policy perspective desirable or undesirable behavior can be increased, at least to some extent, by drawing public attention to what others are doing (Thaler and Sunstein,2008).
For example, Allcott (2011) and Allcott and Roger (2014) found that that sending consumers home energy reports, which contain a social comparison element significantly diminishes home energy consumption. Similarly, Gerber et al. (2008) found that voter turnout was higher when people were led to believe that their neighbors were informed about whether they voted or not, andGerber and Rogers(2009) showed that voter turnout was higher when people were led to believe that expected voter turnout would be high rather than low.
On the labor market, for instance, the behavior of unemployed may be influenced by a social norm to work (Stutzer and Lalive, 2004). Typically, unemployed are sanctioned by social pressure from other members of their community and feel internal pressure to
comply with the norm to work. Stutzer and Lalive (2004) found that the stronger the norm, the more quickly unemployed people find a new job. The reason is that unemployed will not enjoy leisure to the same extent if the norm to work is strong in their environment compared to if the norm is weak. As a consequence, the effort unemployed put into searching for a job, as well as the probability that they accept a job is higher when the norm is strong. Moreover, they showed that unemployed who have not yet found a job were systematically less satisfied with their life when the norm is strong than when the norm is weak.
When considering using social norms as a policy tool it is important to differentiate between non-selfishness and social pressure, because it has very different effects on people’s preferences and therefore also policy considerations. An interesting example comes from the case of charitable giving and door-to-door fund-raising (DellaVigna et al., 2012). If giving is due to non-selfish behavior, then people derive positive utility from giving. If, instead, giving is due to social pressure, then people get a negative utility from giving but still give some because they have a disutility cost from saying no. Hence, potential donors will seek fund-raisers if giving is due to non-selfishness, but will avoid them if giving is due to social pressure. DellaVigna et al. (2012) test this prediction by comparing a standard door-to-door fund-raising campaign to a campaign where the day before the fund-raising visits people were notified by a flyer on the doorknob. The households that got the flyer responded in a direction consistent with social pressure: compared to the first campaign, the share of the households opening the door to the solicitors were 10−25% lower. A policy that aims to at increasing charitable giving by appealing to non-selfishness will, by large, not be as effective as one that uses social pressure.
5 Experimental Methods
The previous sections have–among other things–presented (i) the cognitive and motiva- tional challenges that impact people’s judgments and decisions and (ii) possible policy tools that can be used to help people overcome these challenges. In this section the focus is shifted towards a methodological aspect. In how far a particular frame, commitment device or default helps people to make decisions that are in their interest is in the end an empirical question. One tool that can be used to analyze the effectiveness of certain policy changes is ‘experimentation’. In this section we will explore the important, underlying aspects of experimentation in social sciences, present the different types of experiments and discuss their advantages and disadvantages.
Good policy analysis explores thecauses of certain social phenomena and causal con- sequences of different policies and regulations before they are implemented as well as in retrospect. The importance of causal relations for policy analysis and advice can best be exemplified by looking at two examples (i) the case of minimum wages and (ii) the case of obesity and heart attacks. Standard economic theory has long advocated that minimum wages might lead to unemployment and are bad for economic growth. In recent years, more and more empirical studies have found that minimum wages might not necessar- ily lead to lower growth and unemployment. In fact, there are many studies that claim that a generous minimum wage legislation might actually help economic progress. Many politicians have take this as evidence to argue for higher minimum wages whereas others remain more cautions. The debate between these two groups is ultimately connected to the question whether minimum wagescause unemployment and hamper economic growth or whether minimum wages cause higher growth and prosperity. Example (ii) concerns the relation between obesity and heart attacks. There seems to be more heart attacks among obese people. Physicians, politicians and health-care regulators usually take this as an argument for implementing policies that incentivize people to diet so as to reduce the likelihood of heart attacks (see, for example, the case of the Danish Fat Tax in 2011).
Also here the important part of the argument is the allegedcausal relation between obesity and heart attacks.
This shows, causation is at heart of any policy dispute and debate. Unfortunately, however, causal links underlying social phenomena are usually unobservable and hidden below the complexity of the real world. When collecting and analyzing real world data we are usually able to identify mere correlations. However whether these correlations necessarily imply causation often remains an open political dispute.
What is correlation and what is causation? Two variables are said to be correlated if observing the value of one allows a statement about the value of the other. Correlation between two variables can both be positive as well as negative. When two variablesX and Y are positively correlated, for example, then a high value ofX also means a high value of Y. Correlation is symmetric in the sense that high values ofX might imply a high value ofY and at the same time high values ofY also imply high values ofX. However, the fact that we observe a highY in connection with a high X does not necessarily imply thatX causes Y. For example, there could be an unobserved third variable Z that both causes X and Y at the same time, but there is no causal link between X and Y. Different to correlations, causal relations are necessarily asymmetric because when X causes Y then a highX will imply a high Y, but not vice versa.
