Lab Meetings
2020
Fall Series
Unless otherwise stated, lab meetings will be held at 10:00 am via Zoom (links posted below).
12/15 - Ziv Epstein and Ben Tappin
https://mit.zoom.us/j/96040964213
Ziv. Many experiments studying misinformation use survey methods to evaluate user discernment on news headlines. But how does this artificial survey context compare to the contextual experience of scrolling through social media? To explore this question, we built YourFeed a hypothetical newsfeed environment that mirrors the design of social media. With N=633 participants from MTurk, we randomly assign participants to rate sharing intentions on a fixed set of headlines either via Qualtrics or YourFeed. We find that sharing rates are substantially lower in the YourFeed context, but discernment rates are the same. We also find that CRT and political knowledge were strongly positively correlated with discernment on Qualtrics, but not correlated with discernment in YourFeed. Finally, we discuss an upcoming experiment looking at the impact of affective content, and interface “skins” in user discernment on YourFeed.
Ben. Cues from political party elites (e.g., Donald Trump) reliably influence public opinion about policies. Why is this the case? More specifically: what do people infer from the cue that causes them to update their opinion? In an ongoing project Dave and I are conducting experimental tests of a plausible theory: party elite cues communicate the partisan group norm, and the inferred group norm is what causes people to update. In the talk I will flesh out the theory/supporting evidence and show the design and results we have so far. We are seeking feedback about which direction to go in next with the project.
12/1 - Erez Yoeli and David G. Rand
https://mit.zoom.us/j/97609186175
Erez. Harnessing Observability to Encourage Lawyers to Take Up Probono Cases: Evidence from a Large-Scale Field Experiment. We teamed up with four large law firms to encourage N lawyers in N offices around the world to take up more probono cases. Together with volunteers from the law firms, we co-designed a simple observability intervention: stickers that lawyers could place on their office doors when they met certain thresholds. We'll be presenting the results of our preliminary analyses. These suggest that lawyers who were already engaged in probono cases did not change the number of hours they devoted to probono work, but that lawyers were more likely to take up new cases.
David. Dave will present pilot results for a study of accuracy nudges and Covid-19 misinformation sharing in 11 countries that we are gearing up to launch. Feedback would be much appreciated!
11/24 - Christie Newton and Jeremy Z Yang
https://mit.zoom.us/j/95457798744
Christie. Validating a comprehensive thinking style measure. There are currently more than 20 measures that claim to assess thinking styles. We propose that many of these thinking style measures are derived from similar underlying theories – dual process theories (even if unstated by the researchers). We addressed this issue by correlating items from 14 thinking styles measures with the Cognitive Reflection Test (CRT; Frederick, 2005). We reduced the number of items by selecting the items that had the highest correlation with the CRT. Next, across six studies, we systematically narrowed down the items and tested the underlying factor structure. This revealed that a four-factor correlated structure was best: Actively Open-Minded Thinking about Evidence (AOT-E), Dogmatism (AOT-D), Preference for Intuitive Thinking (PIT), and Effortful Thinking (ET). Predictive validity for the resulting 24-item Comprehensive Thinking Style Questionnaire (CTSQ) was established using a host of variables (e.g., epistemically suspect beliefs, bullshit receptivity, empathy, moral judgments, among others). The CTSQ also correlated with performance on cognitive tasks including the Alice Heim Group Ability Test, CRT, and heuristics and biases. The CTSQ, or the individual subscales, can be used in place of popular thinking style measures, including scales like the Need for Cognition scale, but also behavioral measures like the CRT.
Jeremy. First Law of Motion: Influencer Video Advertising on TikTok. This paper develops an algorithm to predict the effect of influencer video ads on product sales. We propose the concept of “motion” as a summary statistic that captures to what extent the advertised product is shown in the most engaging parts of a video. We estimate pixel-level engagement as a saliency map by fine-tuning a deep 3D convolutional neural network on video-level engagement data. We locate product placement by matching product images to video frames with an object detection algorithm. Motion is then defined as the inner product between pixel-level engagement and product placement, or engagement-weighted advertising intensity. Analogous to a fundamental law in Newtonian mechanics, motion (sales conversion) is generated when the object (product placement) is impressed with a force (content engagement) in the space and time (video). We validate the algorithm with analysis of 40,000 influencer video ads on TikTok, the largest short video platform in the world. We leverage variation in video posting time to identify the causal effect of video ads on product sales. Videos of higher motion are indeed more effective in driving sales. This effect is sizable, robust, and more pronounced among impulsive, hedonic, and lower-price products. We trace the mechanism to influencers’ incentives to promote themselves rather than the products. We discuss how product sellers can use motion to screen video ads pre-launch in a scalable way and also as a novel contractual lever to mitigate the agency problem for greater ROI.
11/17 - Nadia Brashier and David Rand
https://mit.zoom.us/j/96150399371
Nadia. Repeating claims makes them seem truer. This illusory truth effect stubbornly emerges over long delays, among intelligent people, for claims explicitly tagged as “false,” and in the face of contradictory advice. I will present two experiments that test whether monetary incentives for accurate judgments can reduce the illusion.
David. Using the illusory of explanatory depth to combat misinformation campaigns. I will present an experiment in which we test whether having participants explain the evidence they know of regarding election fraud or Covid-19 infection reduces their incorrect beliefs about election fraud being widespread or Covid-19 concerns being overstated.
11/10 - John Ternovski
https://mit.zoom.us/j/91334627028
Making America Hate Again: The Impact of Trump’s 2016 Victories and Rhetoric on Hate Music Listenership. This study analyzes time-series data to isolate the impact of Trump’s 2016 victories and other highly publicized Trump-related events on hate music listenership. Music consumption is particularly important for hate groups, as it has been used extensively by neo-Nazis and white supremacists to organize and recruit youths (Messner, Jipson, Becker, & Byers, 2007; Corte, & Edwards, 2008, Shekhovtsov, 2013). Indeed, some scholars claim hate music scenes are a primary mechanism for new member recruitment (e.g. see, Corte & Edwards, 2008). Many pundits and political experts have remarked that Trump’s xenophobic rhetoric has revitalized and normalized hate groups in America. Our research design captures a salient metric of subscription to far-right ideologies and measures how this variable changed because of public events involving Trump.
10/27 - Cathy Xi Chen and Antonio Arechar
https://mit.zoom.us/j/92566276534
Cathy Chen. The Illusory Truth Effect and Social Media Sharing. I am going to present an ongoing project related to prior exposure and sharing decisions on social media. The illusory Truth Effect suggests that prior exposure to a statement increases the likelihood that participants will judge it to be accurate (Schwartz, 1982 etc.). Pennycook, Cannon, & Rand (2018) demonstrates that prior exposure increases perceived accuracy of fake news. Based on these findings on truth judgment, this project examines the effect of prior exposure on sharing decisions on social media. In a study, we found evidence that prior exposure decreased sharing intentions. Next, we are examining the underlying mechanisms behind this relationship from different angles, and we would really appreciate your comment and discussion for that in the lab meeting!
Antonio Arechar. Promoting Public Engagement Against Corruption. Corruption is one of the most widely discussed issues in Mexico. However, there is not much evidence of successful mechanisms to prevent it. Here we implement a field study on Twitter showing that simple nudges promoting social norms against corruption can lead people to participate and denounce it more often than they naturally would.
10/20 - Ziv Epstein and David Rand
https://mit.zoom.us/j/93731628480
Ziv Epstein. Harnessing accuracy prompts to reduce misinformation sharing online. A growing body of work suggests that prompting accuracy can increase sharing discernment. Across five nationally representative surveys, we compare the relative effectiveness of a suite of different ways of inducing people to think about accuracy. We also examine heterogeneous treatment effects to identify the subgroups who respond most to the prompts. We find that all treatments successfully increased discernment, except for socials norms treatments which were ineffective. We find significant mediation on the perceived accuracy of only sharing accurate information, as well as traits that are associated with higher truth discernment.
David Rand. David will give his thoughts on how to frame and write papers for general science journals.
10/6 - Jerry Zhang
https://mit.zoom.us/j/91810317421
Title: The Dynamics of Political Motivated Reasoning
(I will mostly present the idea and experimental design with limited data.)
This project examines whether motivated reasoning could affect people's advice-taking from the opposite party and what cognitive mechanism contributes the existence or non-existence of political motivated reasoning. We first examine whether there exists motivated reasoning in advice-taking through exogenous variation of incentive level. Then we examine how providing pro-own-party feedback vs pro-opposing-party feedback could affect people's judgement of the party's competence both in the short run and in the long run. In the long run (three weeks after initial experiment), we examine whether people could recall the feedback. We find that people are more likely to recall the pro-own-party feedback. We try to examine whether people are engaging in selective recall or simply forgetting the negative feedback. If people indeed forget about the positive feedback of the opposite party, that explains the no motivated reasoning hypothesis. Since memory is the basis for prior, if the unfavorable feedback is forgotten, the belief of the opposite party will never update dynamically and people keep on believing the low quality of the opposite party.
9/29 - Deen Freelon
https://mit.zoom.us/j/99626163669
Hashtag heroes vs. disinfo dystopia: The left, the right, and the truth about social media activism
Deen Freelon, UNC-Chapel Hill
Abstract: Recent scholarship has generated two distinct impressions of US-based social media activism, one for the ideological left and one for the right. For the left, the dominant mode of engagement is hashtag activism, which entails coordinated online and offline protest campaigns linked by hashtagged slogans. The right channels its priorities through a densely networked, hyperpartisan media ecosystem that makes frequent use of disinformation and other false claims. The respective empirical records underlying these portrayals are very solid, yet questions remain about how exclusively these strategic repertoires cling to ideological fault lines. In particular, there appears to be little extant research on either conservative hashtag-based activism or on left-leaning disinformation. A comprehensive understanding of social media activism demands further explorations of these possibilities, especially in the critical areas of mis- and disinformation. Computational research methods are especially well suited to such investigations.
Suggested reading: https://science.sciencemag.org/content/369/6508/1197
9/22 - Cameron Martel, Mohsen Mosleh & Eaman Jahani
https://mit.zoom.us/j/93029615456
Cameron Martel and Mohsen Mosleh. Cameron and Mohsen will be talking about several ongoing projects. They will be discussing (1) progress on assessing the role of social relations and partisanship on correction engagement, (2) the effects of corrections on subsequent content sharing, and (3) the role of partisanship and social tie formation online as a micro-foundation for the formation of echo chambers. Mohsen may also discuss future study ideas relating to the effect of racial bias on making connections on professional social networking sites.
Eaman Jahani. We study the network effects on inequality or the mechanisms that can increase inter-group differences. In this talk, we will introduce an experiment to analyze the repeated diffusion of a piece of information that originates from random sources and access to which increases one’s utility. The repeated nature of this diffusion process could amplify individual or group differences in utility over time. Multiple players are laid out in a network and in each time period some of them randomly receive the approximate location of a gold mine on a grid, according to a probability distribution that depends on their group membership. Players try to maximize their utility by collecting the gold over all rounds. The goldmine is a rivalrous good, as sharing it with others reduces one’s potential gain, but sharing it with others might still be a good idea for reciprocal sharing in the future. By changing the network structure and testing various inter-group differences, we investigate how network structure affects the dynamics of cooperation and the final inter-group utility differences.
9/15 - Gordon Kraft-Todd & Bence Bago
https://mit.zoom.us/j/97755370782
Beliefs Underlying Americans' Resistance to Wearing Protective Masks during the COVID-19 Pandemic. Gordon Kraft-Todd. Why aren’t Americans wearing masks to protect themselves and others from the COVID-19 pandemic? To understand who isn’t wearing masks and why, we conducted a series of studies in which we asked participants to self-report mask wearing, followed by questions about their mask-relevant beliefs (in addition to other secondary measures and demographics). We find consistent evidence for a three factor solution of mask-relevant beliefs: inefficacy, inconvenience, and freedom. We employ structural equation modeling to understand how our mask-relevant beliefs factors mediate the effect of demographics on mask wearing; e.g. we find that our freedom belief factor fully mediates the effect of political conservativism on self-reported mask wearing.
Corrections of Misinformation on Social Media, and Related Topics. Cameron Martel and Mohsen Mosleh. Cameron and Mohsen will be talking about several ongoing projects. They will be discussing (1) progress on assessing the role of social relations and partisanship on correction engagement, (2) the effects of corrections on subsequent content sharing, and (3) the role of partisanship and social tie formation online as a micro-foundation for the formation of echo chambers. Mohsen may also discuss future study ideas relating to the effect of racial bias on making connections on professional social networking sites.
8/31 - Adam Bear & Ben Tappin
https://mit.zoom.us/j/97569823872
Workshop on Bayesian reasoning and statistics
8/24 - Aida Murati, Eli Kramer, Nate Sirlin & Vincent Schaffer
https://mit.zoom.us/j/99272140860
Aida Murati, Eli Kramer, Nate Sirlin & Vincent Schaffer will present what they worked on during their time as summer Research Assistants.
8/17 - Alex Coppock.
https://mit.zoom.us/j/97710752579
Alex Coppock (Yale Political Science) will present on two research papers. The first reports the results of 53 advertising experiments in the 2016 election (paper co-authored with Seth Hill and Lynn Vavreck) and the second (with Don Green) describes a theory of (the lack of) dynamic constraint in political attitudes. In sum, when a treatment changes a target attitude, nontarget attitudes (even “related” attitudes) do not dynamically change in response. # Paper 1 The Small Effects of Political Advertising are Small Regardless of Context, Message, Sender, or Receiver: Evidence from 59 Real-time Randomized Experiments Evidence across social science indicates that average effects of persuasive messages are small. One commonly-offered explanation for these small effects is heterogeneity: persuasion may only work well in specific circumstances. To evaluate heterogeneity, we repeated an experiment weekly in real time using 2016 U.S. presidential election campaign advertisements. We tested 49 political advertisements in 59 unique experiments on 34,000 people. We investigate heterogeneous effects by sender (candidates or groups), receiver (subject partisanship), content (attack or promotional), and context (battleground versus non-battleground; primary versus general election; early versus late). We find small average effects on candidate favorability and vote. These small effects, however, do not mask substantial heterogeneity even where theory from political science suggests we should. During the primary and general election, in battleground states, for Democrats, Republicans, and Independents --- effects are similarly small. Heterogeneity with large offsetting effects is not the source of small average effects. # Paper 2 Do Belief Systems Exhibit Dynamic Constraint? if a change in one opinion causes a concomitant change in a related opinion. While an enormous literature is dedicated to the study of static constraint (the extent to which individuals hold political views that ``go together''), dynamic constraint is rarely studied, especially using experimental research designs. We offer a new formalization of the theoretical argument that suggests an identification strategy for detecting dynamic constraint. We present evidence from survey experiments conducted with convenience samples of both the mass public and of political elites. Our results indicate that even among respondents whose belief systems are highly constrained in the static sense, a change in one attitude need not precipitate changes in related attitudes. These experimental results affirm and extend Converse's thesis about the limited extent of dynamically constrained ideological thinking in the mass public. The lack of dynamic constraint among our elite sample raises the question of how they come to hold political opinions that are constrained in a static sense. We present an experiment that suggests a potential explanation: elites may be more likely to be chided for expressing inconsistent positions.
8/3 - J. Nathan Matias & Julia Kamin.
https://mit.zoom.us/j/97409515448
Influencing Human and Machine Behavior with Citizen Behavioral Science. Because online platforms observe and intervene in the lives of billions of people, many have come to expect that they should address enduring social problems including online harassment, misinformation, and many others. How can we work toward a world where digital power is guided by evidence and accountable to the public? Online field experiments can test governance ideas while also developing and validating scientific understanding of human and machine behavior. In this seminar, we will hear from J. Nathan Matias and Julia Kamin of the Citizens and Technology Lab at Cornell University, who organize citizen behavioral science alongside online communities of up to tens of millions of people. In one study with a large science discussion community on Reddit, CAT Lab tested the effect of a social norms intervention on unruly behavior and newcomer participation. In another study, a news discussion community tested the second-order effects of encouraging crowd-sourced fact-checks on the behavior of ranking algorithms that promote or demote articles from unreliable sources. After discussing these studies and process of citizen behavior science, we expect to spend the remaining time discussing ideas for possible future research.
7/27 - Juan Palacios.
https://mit.zoom.us/j/99397525165
Pandemics as coordination games: Experimental evidence from COVID-19 in China. Approaches currently available to control the spread of the respiratory syndrome coronavirus 2 rely heavily on individual behavior. The adherence of citizens to socially desirable behaviors is key to avoid health, social and economic costs. However, there is scarce empirical evidence on how individuals behave in a pandemic. Here, we provide experimental evidence of how individuals behave for the first five weeks in the aftermath of a lockdown, and how they interact with their social surroundings in high and low infection risk spaces respectively. We document that individuals take the presence of neighbors in a public space as a signal of risk reduction, mimicking their behavior when they learn their actions. We document the role of risk preferences as an important mediator. Risk averse individuals are less likely to follow their neighbors when they learn that they are back on the streets. We also document the relevance of social preferences as a mediator of social distancing behaviors. We document that individuals with weak social preferences do not limit their visits to public spaces when they learn that fewer neighbors than they expected are doing so. We introduce two manipulations of risk and social preferences. We provide evidence that the precautionary actions of businesses can serve as instruments to reduce perceived risk and increase visits. Finally, we provide suggestive evidence that experimentally increasing the saliency of precautionary measures by neighbors to avoid the spread of the disease might lead to the correction of behavior of participants with weak social preferences, encouraging them to adjust their behavior to converge to the norm.
7/13 - 3-minute updates on lab members’ work.
https://mit.zoom.us/j/94886711577.
7/6 - Jillian Jordan.
https://mit.zoom.us/j/96891399282
I will be seeking feedback on a paper investigating the "Virtuous Victim effect", whereby people rate victims of immorality as morally good. The paper went through peer review and was rejected, and I am working on revisions before resubmitting to a new journal. I will specifically be seeking feedback on (i) which reviewer concerns are shared by the group and (ii) what the group thinks would be most helpful for improving the paper. I have copied the abstract from the paper below, and a link to the working paper is available here: https://psyarxiv.com/yz8r6/
Humans ubiquitously encounter narratives about immoral acts and their victims. Here, we demonstrate that these narratives can influence perceptions of victims’ moral character. Specifically, across a wide range of contexts, victims are seen as more moral than non-victims who have behaved identically. Using 13 experiments (total n = 8,358), we explore this Virtuous Victim effect. We show that it is specific to victims of immorality (i.e., it does not extend equally to victims of accidental misfortune) and to moral virtue (i.e., it does not extend equally to positive nonmoral traits). We also show that the Virtuous Victim effect can occur online and in the lab, when subjects have other morally relevant information about the victim, when subjects have a direct opportunity to condemn the perpetrator, and in the context of both third- and first-person victim narratives. Finally, we provide support for the Justice Restoration Hypothesis, which posits that people see victims as moral in order to motivate adaptive justice-restorative action (i.e., punishment of perpetrators and helping of victims). We show that people see victims as having elevated moral character, but do not expect them to behave more morally or less immorally—a pattern that is consistent with the Justice Restoration Hypothesis, but not readily explained by alternative explanations for the Virtuous Victim effect. And we provide both correlational and causal evidence for a key prediction of the Justice Restoration Hypothesis: when people do not perceive incentives to help victims and punish perpetrators, the Virtuous Victim effect disappears.
6/29 - Rahul Bhui.
https://mit.zoom.us/j/97891451958
Attention constraints and learning in categories. When different stimuli belong to the same category, learning about their attributes should be guided by this categorical structure. For example, investors might efficiently spend their limited time and effort learning about shared factors which commonly affect all stocks in the same industry, and neglect idiosyncratic factors which only affect individual stocks. Although this mechanism is thought to contribute to major financial anomalies, it is not known whether rational principles account for such biases in attention allocation. Here, we demonstrate how an adaptive response to attention constraints can bias learning toward shared qualities and away from individual differences. In three preregistered experiments using an information sampling paradigm with mousetracking, we find that people preferentially attend to information at the category level when idiosyncratic variation is low, when time constraints are more severe, and when the category contains more members. While attention is more diffuse across all information sources than predicted by Bayesian theory, there are signs of convergence toward this optimal benchmark with experience. Our results thus indicate a novel way in which a focus on categories can be driven by rational principles.
6/22 - Jerry Zhang.
https://mit.zoom.us/j/96763492602
I will have two topics for my presentation. 1. Motivated Reasoning on Racial Identity We look at whether people engage in motivated reasoning when taking financial advice from Black vs White agent. The set-up follows my previous talk on political motivated reasoning: whether offsetting disutility from trusting the other party with higher incentive could induce more rational advice-taking. I will share some recent results. 2. Does asking people what "Defund the Police" means affect people's support for the "Defund the Police" protest? The short answer is "Yes." If people are asked about what "Defund the Police" means to them, they tend to show less support for the protest. I would like your opinions on whether this is something interesting that can be turned into a research.
6/15 - Ben Tappin.
https://mit.zoom.us/j/99788515640
I will talk about two projects that are in their early stages. In the first, Dave and I study the geographic distribution of political polarization across contemporary America using recently-collected survey data from ~155,000 American adults. The project has a descriptive component: estimate the geographic distribution of political polarization across America; and a theoretical component: assess whether the variance in the geographic distribution is explained by prominent theories of political polarization. We seek both technical feedback (on the Bayesian multilevel modelling), and feedback on the theoretical contribution. In the second project, we flesh out and test a novel hypothesis for why people's policy opinions are influenced by cues from political party elites (i.e., the causal mechanism). The hypothesized mechanism has to do with social norms. Some previous studies find evidence consistent with the hypothesis, but the designs are flawed and potentially introduce bias to the results. We seek feedback on our (hopefully improved!) design.
6/8 - David G. Rand.
https://mit.zoom.us/j/95051388064
"I'll discuss the model we developed for cognitive constraints on consideration of preferences, as applied to sharing fake news." [see p. 26-29 here for background].