Lukas Meub and Till Proeger, “Are Groups 'Less Behavioral'? The Case of Anchoring,” February 18, 2014; updated version available here.
• Do groups (as opposed to individuals) filter out irrationality and promote rationality? Do they “debias”? The broad consensus position is “yes,” though perhaps the advantage of a group is more pronounced with issues that have objectively correct answers (as opposed to being matters of judgment.)
• Experiments conducted by Meub and Proeger used groups of three players, faced with 15 questions, 5 each from the categories of: factual knowledge; likelihoods; and, price estimation. The experiments involved two conditions, the “calibration” condition and the “anchor” condition.
• The calibration condition is employed to identify appropriate anchors. The high anchor and low anchor values come from the 15th and 85th percentiles of unanchored responses. There is a small monetary incentive to guess well; payments varied (based on performance and condition) from 5 to 15 euro.
• The anchor index summarizes how much the anchors sway responses. The numerator of the index is the median difference in responses between the high anchor group and the low anchor group. The denominator is the difference between the high and low anchor values themselves. An anchor index of zero indicates that people gave the same response, on average, no matter what anchor they were exposed to. An anchor index of one indicates that the respondents, on average, just let the anchor be their response.
• The authors find that over all 15 questions, groups are less biased, with an anchor index of .34 as opposed to .52 for individuals.
• High anchors seem to matter more, but in any event, groups are less biased than individuals on the factual knowledge questions. Groups showed no advantage over individuals on the probability questions, nor for price estimation.
• Participants were asked if they perceived an anchoring problem, and the responses did not vary between individuals or groups.
Sunday, October 25, 2015
Kahneman (2011), Chapter 11, “Anchors”
Daniel Kahneman, “Anchors.” Chapter 11, pages 119-128, in Thinking, Fast and Slow, New York: Farrar, Straus and Giroux, 2011.
• Anchoring occurs when a person considers one number, before estimating another, unknown number. The estimate will tend to stay close to the original number considered, even if that original number is known to be completely unrelated to the estimated quantity.
• Anchoring might result from two (separate?) mechanisms. One is priming, the power of suggestion, which works on our rapid responses (“System 1”); the second is when we deliberate (“System 2”) over the estimate by adjusting away from the anchor, but do not adjust sufficiently.
• The asking price for a house can be a form of anchor – one that affects professional real estate agents as well as amateurs. The amount of damages requested in a civil suit also might serve as an anchor. Further, legislated maxima (caps) on damage awards might serve as an anchor, drawing upwards what otherwise might have been smaller awards.
• Instructions to deliberately counter the anchor can be effective.
• Anchoring occurs when a person considers one number, before estimating another, unknown number. The estimate will tend to stay close to the original number considered, even if that original number is known to be completely unrelated to the estimated quantity.
• Anchoring might result from two (separate?) mechanisms. One is priming, the power of suggestion, which works on our rapid responses (“System 1”); the second is when we deliberate (“System 2”) over the estimate by adjusting away from the anchor, but do not adjust sufficiently.
• The asking price for a house can be a form of anchor – one that affects professional real estate agents as well as amateurs. The amount of damages requested in a civil suit also might serve as an anchor. Further, legislated maxima (caps) on damage awards might serve as an anchor, drawing upwards what otherwise might have been smaller awards.
• Instructions to deliberately counter the anchor can be effective.
Dahl and DellaVigna (2009) on Movie Violence and Crime
Gordon Dahl and Stefano DellaVigna, “Does Movie Violence Increase Violent Crime?” Quarterly Journal of Economics 124(2): 677-734, May, 2009 [pdf].
• Experimental evidence and survey responses indicate a rise in aggressive behavior after exposure to movie violence. Dahl and DellaVigna examine evidence from an ongoing “natural experiment,” seeing what happens to reported crime after a blockbuster violent movie.
• Assaults fall on the evening (both before and after midnight) that a violent movie attracts a large audience.
• For every million people watching a strongly violent or violent movie, assaults fall by more than 1% that night; after midnight, the fall is closer to 2%. A popular violent movie reduces assaults by about 1,000 per weekend. (Still, an unseasonably cold day would lead to an even greater fall.)
• The mechanism seems to be that the criminogenic population selects into watching the movie rather than, say, drinking. Note that most movie houses do not serve alcohol. So for the evening of the movie watching, there is “voluntary incapacitation.” For the post-midnight hours, presumably there is a carry-on effect from the relatively salubrious hours in the cinema.
• Even though violent movies reduce crime by enticing violence-prone individuals into the comparatively peaceful setting of a cinema, there would be an even further reduction in crime if these same people watched a non-violent movie. The laboratory evidence isn’t wrong, but it does not capture what appears to be the most important element of the movie violence-crime connection, the incapacitation of likely criminals as they watch movies. There is no evidence for an overall cathartic effect from watching violent movies.
• Experimental evidence and survey responses indicate a rise in aggressive behavior after exposure to movie violence. Dahl and DellaVigna examine evidence from an ongoing “natural experiment,” seeing what happens to reported crime after a blockbuster violent movie.
• Assaults fall on the evening (both before and after midnight) that a violent movie attracts a large audience.
• For every million people watching a strongly violent or violent movie, assaults fall by more than 1% that night; after midnight, the fall is closer to 2%. A popular violent movie reduces assaults by about 1,000 per weekend. (Still, an unseasonably cold day would lead to an even greater fall.)
• The mechanism seems to be that the criminogenic population selects into watching the movie rather than, say, drinking. Note that most movie houses do not serve alcohol. So for the evening of the movie watching, there is “voluntary incapacitation.” For the post-midnight hours, presumably there is a carry-on effect from the relatively salubrious hours in the cinema.
• Even though violent movies reduce crime by enticing violence-prone individuals into the comparatively peaceful setting of a cinema, there would be an even further reduction in crime if these same people watched a non-violent movie. The laboratory evidence isn’t wrong, but it does not capture what appears to be the most important element of the movie violence-crime connection, the incapacitation of likely criminals as they watch movies. There is no evidence for an overall cathartic effect from watching violent movies.
Tuesday, October 13, 2015
Duflo, Kremer, and Robinson, “Nudging Farmers to Use Fertilizer…” (2014)
Esther Duflo, Michael Kremer, and Jonathan Robinson, “Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya.”
American Economic Review 101: 2350-2390, October 2011.
• The (limited) use of fertilizer seems to hold very high returns – both social returns and private returns. Why do so many farmers not apply fertilizer, when it would seem to be in their private interest to fertilize? Behavioral biases, perhaps – and in particular, procrastination.
• Some farmers are only partially sophisticated, overestimating the probability that they will be patient in the future. Calibrations suggest nearly 50% fit this category. These farmers might choose to postpone fertilizer purchase, fully expecting to buy it later, and then fail to follow through on their plan.
• Offering small, time-limited fertilizer discounts just after harvest (when farmers have cash) can significantly increase purchase by such partially sophisticated farmers.
• The use of a second dose of fertilizer has negative social returns, but farmers might administer a second dose if the price is highly subsidized. Hence a small, time-limited discount might be a better policy than large subsidies, even if the large subsidies spur fertilizer use, too.
• Small, time-limited discounts have other advantages. First, they are not that demanding upon the public purse; second, if fertilizer isn’t really such a good idea, if it has negative returns, farmers probably won’t administer it (which is the socially efficient behavior under the negative-returns circumstances). The small subsidy limits the possibility that socially-excessive amounts of fertilizer will be applied.
• The (limited) use of fertilizer seems to hold very high returns – both social returns and private returns. Why do so many farmers not apply fertilizer, when it would seem to be in their private interest to fertilize? Behavioral biases, perhaps – and in particular, procrastination.
• Some farmers are only partially sophisticated, overestimating the probability that they will be patient in the future. Calibrations suggest nearly 50% fit this category. These farmers might choose to postpone fertilizer purchase, fully expecting to buy it later, and then fail to follow through on their plan.
• Offering small, time-limited fertilizer discounts just after harvest (when farmers have cash) can significantly increase purchase by such partially sophisticated farmers.
• The use of a second dose of fertilizer has negative social returns, but farmers might administer a second dose if the price is highly subsidized. Hence a small, time-limited discount might be a better policy than large subsidies, even if the large subsidies spur fertilizer use, too.
• Small, time-limited discounts have other advantages. First, they are not that demanding upon the public purse; second, if fertilizer isn’t really such a good idea, if it has negative returns, farmers probably won’t administer it (which is the socially efficient behavior under the negative-returns circumstances). The small subsidy limits the possibility that socially-excessive amounts of fertilizer will be applied.
Mullainathan on Development Economics (2006)
Sendhil Mullainathan, “Development Economics Through the Lens of Psychology.” Proceedings of the Annual Bank Conference on Development Economics, 2006 [pdf].
• Parents claim to value education highly, but kids attend school sporadically. The problem seems to be the short-term decisions by the parents, so it might be best to target those decisions. The provision of meals in schools might help, and perhaps collecting school fees in small, regular payments rather than as one large, annual payment. Policies that increase school attendance also might improve teacher morale.
• ROSCAs serve as savings commitment devices, promoting regular, small deposits. The lottery-like, skewed payoff is a commitment not to spend the savings until there is a major purchase. Holding wealth in illiquid forms such as jewelry or livestock also helps solve commitment problems. If the access to microcredit undermines other savings commitments, then the mere profitability of a micro-lending program is not a good indicator of its social value.
• Making banks available to rural dwellers might make saving, and not spending, the default condition, providing some commitment.
• Loss aversion suggests that there is a lot to be said for protecting status quo positions in the course of reform. Existing stakeholders might be grandfathered, as a means to avoiding the imposition of losses.
• People look at the world in a biased way, and can perceive failures of reciprocity even where they do not exist. They might conform their behavior in accord with a negative stereotype.
• Development policies can be aimed at solving internal problems (commitment, self-control, bias) as well as external problems.
• Parents claim to value education highly, but kids attend school sporadically. The problem seems to be the short-term decisions by the parents, so it might be best to target those decisions. The provision of meals in schools might help, and perhaps collecting school fees in small, regular payments rather than as one large, annual payment. Policies that increase school attendance also might improve teacher morale.
• ROSCAs serve as savings commitment devices, promoting regular, small deposits. The lottery-like, skewed payoff is a commitment not to spend the savings until there is a major purchase. Holding wealth in illiquid forms such as jewelry or livestock also helps solve commitment problems. If the access to microcredit undermines other savings commitments, then the mere profitability of a micro-lending program is not a good indicator of its social value.
• Making banks available to rural dwellers might make saving, and not spending, the default condition, providing some commitment.
• Loss aversion suggests that there is a lot to be said for protecting status quo positions in the course of reform. Existing stakeholders might be grandfathered, as a means to avoiding the imposition of losses.
• People look at the world in a biased way, and can perceive failures of reciprocity even where they do not exist. They might conform their behavior in accord with a negative stereotype.
• Development policies can be aimed at solving internal problems (commitment, self-control, bias) as well as external problems.
Monday, September 21, 2015
Kremer and Levy (2008) on Peer Effects and Alcohol
Michael Kremer and Dan Levy, “Peer Effects and Alcohol Use Among College Students.” Journal of Economics Perspectives 22(3): 189-206, Summer, 2008.
• The data come from a large state university that randomly assigns some roommates. The finding: males assigned roommates who drank in high school had a lower GPA.
• By using high school characteristics, we can rule out that the GPA connection stems from common shocks to the roommates.
• For the most part, studies don’t find much support for the idea that academic characteristics affect the academic performance of roommates. Kremer and Levy likewise find no effect of academic background or social background on roommate academic achievement.
• The GPA decline for males is about .27 if the roommate drank in high school. The drop is especially bad for weak students, and for those who themselves drank in high school, and it’s much greater in the second year (though you are no longer roommates).
• Roommates who are not randomly assigned, but who choose each other, do not see falls in GPA from roommate drinking history.
• Reducing the drinking of one student can reduce the drinking of others. But collecting non-drinkers in substance-free housing will concentrate the drinking students, with a negative effect on overall GPAs.
• The data come from a large state university that randomly assigns some roommates. The finding: males assigned roommates who drank in high school had a lower GPA.
• By using high school characteristics, we can rule out that the GPA connection stems from common shocks to the roommates.
• For the most part, studies don’t find much support for the idea that academic characteristics affect the academic performance of roommates. Kremer and Levy likewise find no effect of academic background or social background on roommate academic achievement.
• The GPA decline for males is about .27 if the roommate drank in high school. The drop is especially bad for weak students, and for those who themselves drank in high school, and it’s much greater in the second year (though you are no longer roommates).
• Roommates who are not randomly assigned, but who choose each other, do not see falls in GPA from roommate drinking history.
• Reducing the drinking of one student can reduce the drinking of others. But collecting non-drinkers in substance-free housing will concentrate the drinking students, with a negative effect on overall GPAs.
Falk and Ichino (2006), “Clean Evidence on Peer Effects”
Armin Falk and Andrea Ichino, “Clean Evidence on Peer Effects [pdf].” Journal of Labor Economics 24(1): 39-57, 2006.
• Observational methods make it hard to separate out peer effects from birds of a feather flocking together, or from a shared environment.
• Here, the authors use random assignment to either a peer setting or to a solo task; the task is envelope stuffing. The existence of peer effects would lead to stronger correlation between peers than between unmatched others.
• The authors find the requisite correlation, as well as evidence that the peer effects raise average productivity (mainly by raising the output of the less productive person).
• The workers were Swiss high school students, given the envelope-stuffing job that paid more than $60 for a four-hour session. Peers worked separately but near each other. The total number of participants was but 24, eight of whom were in the solo condition.
• They find that if one peer raises output by one unit, the other person raises output by .14 unit.
• Would a non-working peer also tend to increase the output of a worker, by providing a sort of supervision?
• Observational methods make it hard to separate out peer effects from birds of a feather flocking together, or from a shared environment.
• Here, the authors use random assignment to either a peer setting or to a solo task; the task is envelope stuffing. The existence of peer effects would lead to stronger correlation between peers than between unmatched others.
• The authors find the requisite correlation, as well as evidence that the peer effects raise average productivity (mainly by raising the output of the less productive person).
• The workers were Swiss high school students, given the envelope-stuffing job that paid more than $60 for a four-hour session. Peers worked separately but near each other. The total number of participants was but 24, eight of whom were in the solo condition.
• They find that if one peer raises output by one unit, the other person raises output by .14 unit.
• Would a non-working peer also tend to increase the output of a worker, by providing a sort of supervision?
Subscribe to:
Posts (Atom)