1 / 49

Laura Brandimarte

Discounting the Past: Bad Weighs Heavier than Good. Laura Brandimarte. Co-authors: Alessandro Acquisti Joachim Vosgerau. The research question. In November 2006 two Ottawa employees of a grocery store chain made admissions of theft on the message board of a Facebook group

gerald
Download Presentation

Laura Brandimarte

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Discounting the Past: Bad Weighs Heavier than Good Laura Brandimarte Co-authors: Alessandro Acquisti Joachim Vosgerau

  2. The research question • In November 2006 two Ottawa employees of a grocery store chain made admissions of theft on the message board of a Facebook group • In January 2007 they were fired after their employer found out about the message • About two years ago two users created a Facebook page for Keep A Child Alive, a nonprofit organization that helps provide drugs for people, living in Africa and India, who are affected by AIDS • There was no consequence to the creation of this page

  3. The research question • How does information related to past events and retrieved today get discounted? Does information about past events with negative valence receive more weight in impression formation than information with positive valence, as time goes by? How does information with positive and negative valence affect perceived trust? • The literature (Baumeister et al., 2001; David et al., 1997; Brickman et al., 1978) looked at the longer lasting effects of bad relative to good, but not at the possibility that two different discount factors may be applied to them

  4. The research question • Trustworthiness of a technological environment depends on: • secure physical infrastructure • enforcement of well-defined regulations and policies • how such physical and regulatory infrastructure is eventually interpreted and used by individuals

  5. The research question Weintroduce the hypothesis of differential discounting: What the literaturefocused on: Slopeof the redlineissmaller in absolutevalue at eachpoint in time Slopeof the redlineislarger in absolutevalue at eachpoint in time

  6. Hypothesis of differential discounting • Impact of info with negative valence lasts longer than impact of info with positive valence, not only because of asymmetric effects of valence, but also because of different weights – or discount rates – applied to the two types of info • This may be due to • mobilization effects (Taylor, 1991) and evolutionary theory (Baumeister et al., 2001) • negativity bias (Seligman and Maier, 1967) • negative info is more attention grabbing (Pratto and John, 1991) • Note on terminology: we use the term “discount rate” to refer to past events

  7. Testing our hypothesis: Three experiments • Survey-based randomized experiments, manipulating valence of information provided to subjects and time to which that information referred • All subjects (CMU students) received the same baseline information about a person or a company • subjects in the neutral conditions only received baseline info • subjects randomized to good or bad info conditions received same info except for one detail with either positive or negative valence, referring to a more or less recent past • Then subjects were asked to express a judgment on the person or company they just read about • that judgment, measured through various indices, represents the dependent variable of interest

  8. Testing our hypothesis: Three experiments • In order to guarantee robustness we varied several design features across experiments • DV: measures of inferred vs. direct judgment • object of judgment: people vs. company • different time frames: weeks vs. months vs. years • time manipulation: sometimes emphasized by moving the position of the relevant piece of information in the text, other times left in the same position of the text

  9. Testing our hypothesis: Three experiments • For each experiment, we estimated a difference-in-difference model described by the following equation: • DVi = β0 + β1Badi + β2MoreRecenti + β3Bad_MoreRecenti + β4Agei + β5Malei + εi • where the Dependent Variables of interest changed across experiments • This specification allows to disentangle the effects of time, valence and their interaction. Demographics are added for descriptive analysis and randomization checks • We were mostly interested in the coefficient on the interaction term, β3, which tells us whether time has a differential effect on information with positive and negative valence • We expected β1 and β2 to be positive, and β3 to be negative

  10. Experiment 1: Dictator game • Hypothetical scenario: subjects are to decide how they would split $100 with another player – the opponent • Priming subjects on what to consider a fair split: subjects are instructed that on average dictators keep for themselves 70% of the sum • Experimental manipulations of valence and time • the opponent is described as having played as the dictator in 7 previous rounds in the 7 previous weeks • allocations were all average (~ 70-30 split) except one, which was: • either generous (50-50 split, good info conditions) or unfair (100-0 split, bad info conditions) • and occurred either 6, or 4, or 2 weeks ago (old, middle and recent conditions respectively). • Neutral condition: all allocations were fair

  11. Experiment 1: Dictator game • Good old condition • Alpha’s previous decision on how to split the sum of $100

  12. Experiment 1: Dictator game Good old condition Alpha’s previous decision on how to split the sum of $100

  13. Experiment 1: Dictator game • Good old condition • Alpha’s previous decision on how to split the sum of $100

  14. Experiment 1: Dictator game • Good old condition • Alpha’s previous decision on how to split the sum of $100

  15. Experiment 1: Dictator game • Good middle condition • Alpha’s previous decision on how to split the sum of $100

  16. Experiment 1: Dictator game • Good recent condition • Alpha’s previous decision on how to split the sum of $100

  17. Experiment 1: Dictator game • Bad old condition • Alpha’s previous decision on how to split the sum of $100

  18. Experiment 1: Dictator game • Bad middle condition • Alpha’s previous decision on how to split the sum of $100

  19. Experiment 1: Dictator game • Bad recent condition • Alpha’s previous decision on how to split the sum of $100

  20. Experiment 1: Dictator game • Dependent variables: • money allocation between subjects and their opponents • fairness assessment of their opponents • We expected the slope of the line describing average allocations to be smaller in absolute value for bad info conditions than for good info conditions • Note: For all experiments, DV is not expressed in levels, but it’s the absolute difference between values in each condition and the average value of the neutral condition

  21. Experiment 1: Dictator game - Results Figure 4. Average sum that subjects chose to allocate to themselves in the dictator game • Based on pair-wise t-tests, good information allocations do not differ from the neutral allocation except for the recent condition • On the other hand, all bad information allocations differ significantly from the neutral allocation

  22. Experiment 1: Dictator game - Results • Recent vs. middle conditions: all coefficients are of the expected sign and significant • Recent vs. old conditions: the coefficients are of the predicted sign but the interaction is not significant • Middle vs. old condition: the sign of β3 is reversed but is not significant and β2 is not at all significant

  23. Experiment 3: The Company Study • Hypothetical scenario: subjects are to express a judgment on a company based on background information we provide • Neutral condition: only baseline info is provided • Experimental manipulations of valence and time: we add to the baseline information one detail with either positive or negative valence, and vary the time to which that detail refers

  24. Experiment 3: The Company Study “Here is some background information about a company called PitStop. Please review this information, and be ready to answer the questions below and in the next page. PitStop was founded in 1961 as a mechanics garage. After a while, it started to manufacture its own brake discs - discs were previously imported from Great Breaker. The production of brake discs was also followed by the manufacturing of other braking system components, which allowed the company to enter the market of racing vehicles. In the meantime, in the aim to develop in the field of motorcycles, PitStop acquired 70% of the company MagMoto, specialized in producing wheels for racing motorcycles. In 1977 [2009] PitStop received the National Entrepreneurship Award for accounting and financial transparency [was heavily fined for accounting and financial fraud]. In order to compete in the sector of brake discs for industrial vehicles, PitStop took over the company SlowDown for car brake disc machining.”

  25. Experiment 3: The Company Study • Dependent variables: • how subjects would have liked to work for the company • and how likely they would be to choose that company as their main sponsor • These questions were created in order to infer a measure of liking for the company and a sense of belonging and of agreement with its policies, which can’t be measured by typical indices, like stock prices or profitability indices • Our expectations in terms of hypothesis testing are exactly the same as for Experiment 2

  26. Experiment 3: The Company Study - Results Figure 7. Average level of liking indices across conditions in Experiment 3 • Similarly to what we found in previous experiments, the good information indices do not differ from the neutral averages except for the recent condition • On the other hand, bad information average indices differ significantly from the neutral averages

  27. Experiment 3: The Company Study - Results • The data partially support our hypothesis: while for the second index all our predictions are met, coefficient β3 is small and positive for the first index – but not at all significant

  28. Conclusions • This paper presents a theoretical background and some empirical evidence in support of the hypothesis that information retrieved today and related to good and bad events occurred in the past may be discounted according to different discount factors • With the results of three survey-based randomized experiments, we provide some evidence that bad is discounted less than good, and thus the effects of bad last longer than the effects of good, not because the immediate effect of bad is larger, but because a smaller discount rate is applied to it as compared to the one used for good

  29. Thank you! • Questions…

  30. Extras

  31. Agenda • The research question • Motivation • Our contribution to the literature • Hypothesis: differential discounting of bad and good information • Testing our hypothesis: Three experiments • Experimental design • Empirical analysis and results • Discussion • Limitations • Conclusions

  32. The research question Weintroduce the hypothesis of differential discounting: What the literaturefocused on:

  33. Testing our hypothesis: Three experiments • Several challenges in experimental design • separating the effects of good and bad: the need for baseline information • preventing confounding effects: • of magnitude and/or rarity on judgment • of valence on memory • of age on impression formation • defining an index of judgment for a company: we are not interested in a financial assessment of the company, for which we could have used stock market data, but a measure of sense of belonging and agreement with its policies

  34. Experiment 1: Dictator game - Results

  35. Experiment 1: Dictator game - Results Figure 4a. Average sum that subjects chose to allocate to themselves in the dictator game - modified

  36. Experiment 2: The Wallet Story - Results

  37. Experiment 2: The Wallet Story • Hypothetical scenario: subjects are to express a judgment on a person based on background information we provide • Neutral condition: only baseline info is provided • Experimental manipulations of valence and time: we add to the baseline info one detail with either positive or negative valence, and vary the time to which that detail refers

  38. Experiment 2: The Wallet Story “Here is some background information about Mr. A. Please review this information, and be ready to answer the questions below and in the next page. Mr. A was born in San Diego, California, where he attended elementary and middle school. As a child, he obtained his social security number and received the standard DPT vaccination. When he was 16 years old, he moved to Sacramento, California, with his family. He attended high school there and got his driving license. Just about graduation, he found a lost woman's purse containing $10,000 in cash. He reported [did not report] the discovery to the police, and the rightful owner retrieved [did not retrieve] her money. After graduation he moved to Houston, Texas where he has been living and working for the past 12 months [5 years].”

  39. Experiment 2: The Wallet Story • Dependent variables: • how much subjects liked the person described; how they would have liked to work with her (Interpersonal Judgment Scale, Byrne 1961) • and whether they considered her trustworthy (General Social Survey, World Values Survey) • We expected the slope of the line describing average measure of judgment to be smaller in absolute value for bad info conditions than for good info conditions

  40. Experiment 2: The Wallet Story - Results Figure 6. Average level of liking and trust indices across conditions in Experiment 2

  41. Experiment 2: The Wallet Story - Results • Based on pair-wise t-tests, values of liking indices in the good-old info conditions do not differ from the values in the neutral condition • The trust index is the only one for which the good-old info condition differs from the neutral one • On the other hand, all bad information average indices differ significantly from the neutral averages

  42. Experiment 2: The Wallet Story - Results • The data strongly support our hypothesis, as all coefficients have the predicted sign and are statistically significant.

  43. Experiment 2: The Wallet Story - Results Figure 6a. Average level of liking and trust indices across conditions in Experiment 2 - modified

  44. Experiment 3: The Company Study - Results

  45. Experiment 3: The Company Study - Results Figure 7a. Average level of liking indices across conditions in Experiment 3 - modified

  46. Discussion • Why should differential discounting occur? We can think of three possible explanations • asymmetric mobilization effect generated by negative events: the human organism responds more intensely to bad events than to good or neutral events at every level – physiological, emotional, behavioral, judgmental (Taylor, 1991). This mechanism itself may be reconnected to evolutionary theory (Baumeister et al., 2001) • negativity bias (Seligman and Maier, 1967) and overestimation of the probability of a person behaving unfairly given that she behaved unfairly once in the past, relative to the probability of a person behaving fairly given that she did so once in the past. The weight given to one negative episode will be larger than the weight given to the corresponding positive episode • info with negative valence is more attention-grabbing (Pratto and John, 1991)

  47. Discussion • The results of our experiments provide some evidence in support of differential discounting • Asymmetry effects can’t be the only explanation because the slope of the line describing judgment in bad info conditions is generally smaller than it is for good info conditions • Our results cannot be explained by effects of valence on memory (Kreitler and Kreitler, 1968; Skowronski and Carlston, 1987; Bless et al., 1992; Matling and Strang, 1978) • We intentionally separate the effects of good and bad, so we do something different from what the criminology literature on redemption does (Blumstein and Nakamura, 2009)

  48. Limitations • All our experiments provided hypothetical scenarios (Bhatia and Fox-Rushby, 2009). In order to make our results more convincing we plan to repeat Experiment 1 in the lab, so that subjects face real allocation decisions • Population was not heterogeneous • Limitations of survey-based experiments and reliability of our metrics for Experiment 3

  49. Conclusions • Important implications of differential discounting in the era of Web 2.0 applications • On the one hand, there is a drive to make oneself popular on social media, and one way to do it is publishing extravagant information about oneself, that may be perceived as having negative valence if taken out of context • Indeed, a pilot we ran confirms that information with negative valence is very common (examples?) • On the other hand, there is a public that processes information with positive (or neutral) and negative valence in two very different ways • If the footprint of the negative information is more persistent, then the impression and the judgment about the subject involved will be more affected by it, even if that information refers to events that happened way back in the past

More Related