01/16: Parigi

[jmausolf]

Comment below with questions or thoughts about the reading for this week's workshop.

Please make your comments by Wednesday 11:59 PM, and upvote at least five of your peers' comments on Thursday prior to the workshop. You need to use 'thumbs-up' for your reactions to count towards 'top comments,' but you can use other emojis on top of the thumbs up.

As an additional reminder, please do not distribute, share, or post the reading for this week.

[policyglot]

Thank you, Dr. Parigi and Natã, for sharing this interesting application of game theory to quantitative sociological research. Two questions:

1) IRB Guidelines state that it can be ethical to not share the full details of the experiment if some level of deception is essential to arrive at the key conclusions of the research. The deception here is to make the participants believe that the AirBnB profiles they were investing in were indeed real people (and therefore players in the game). Since you state that all participants did sign IRB consent forms, I'm curious how you explained the objectives of your research (beyond the offer of the gift card)

2) You referenced your previous paper (Abrahao et al 2017) while stating that “loss aversion has the additional effect of driving those most sensitive to losses out of the market.” However, when we look at the countries involved (Canada, the US, Great Britain and Italy), we do see a preference for other WEIRD (Westerm, Educated, Industrialized, Rich, Democratic) countries dominated by the same race (Caucasian). What studies would you recommend to look into dynamics of trust between distinct races, say for American and Canadian travellers visiting India or Kenya? Would you expect the same 11 (for hosts) and 18 variables (for guests) identified by your lasso regression model to hold as much importance in this multiracial setting too?

[rkcatipon]

Thank you for sharing your work, Dr. Parigi and Natã!

Thinking about @policyglot's second point, I was curious as to why the race was not an explicit variable in your experiments? In the experiment, the host avatars were non-descriptive and race-agnostic, but your team did decide to include gender. What motivated this decision?

Secondly, the article states that:

"Previous research by Qiu et al. shows the impact that increased review count and higher ratings have on the trustworthiness of Airbnb listings. The reputation system has been shown to have such an importance on Airbnb that common social biases like homophily are relatively weak drivers of trust"

But if a host is unable to get bookings in the first place, perhaps because of their race or gender, how are they going to have generated reviews or ratings of trustworthiness?

Finally, for Dr. Parigi: as Lead Trust Scientist at Airbnb, what compromises (if any) did you have to make in order to gain access to publishable data? Did AirBnB have any specific requests regarding the content of your research?

Thank you again for your research!

[wanitchayap]

Thank you for this very interesting research!

I have a methodology-related question. How do you deal with response bias? The participation rate is only around 5% of the whole recruitment. Moreover, the rates are quite different between the hosts and the guests (higher rate from host than guest). Seems like hosts are more likely to participate in this study. Could it be because they feel more responsibility bound to AirBnb than guests? In that case, would the hosts pay attention more to the task as well?

[nwrim]

Thank you for the interesting research and a fun read!

As you briefly mentioned at the end of the Limitation section, I think the triadic relationship between provider-platform-consumer is a better representation of the trust relationship than the dyadic relationship between provider and consumer. For example, I saw two Uber advertisements on Youtube just today claiming that all their drivers are background checked before the first drive and all trips are insured (Yes, I sadly do not have Youtube Premium). Using this prior information (about the platform's policy), my propensity to trust Uber drivers might be higher.

Building on this, I guess there are two ways to increase the propensity to trust - One will be to increase the individual level propensity (UI design, etc) and the other will be to increase the platform level propensity (Brand image, etc). Which do you think is the more effective strategy? Or do you think these two are fundamentally intertwined with each other so that platforms should work on both simultaneously?

[liu431]

Thank you for the interesting talk.

You described the observation as the correlation between engagements and propensity to trust others, instead of causation. I'm wondering if you have any research plans on the causal questions, such as directions of relationships and potential confounds?

Also, I'm curious to know based on the current observations, would the executives at Airbnb be convinced to implement the new design and features? When implementing an important design change, what other factors are considered besides experiment results?

[anqi-hu]

Thank you for sharing your work with us! I was curious to see your justification for not separating trust between users from trust between a user and the platform, until the very last paragraph, where you addressed that such a distinction was absent. How do you plan to separate these two? On a related note, to what extent do you think the participants of this experiment self-selected based on pre-exisitng trust in the platform?

[zeyuxu1997]

Thanks for sharing your work with us. The experiment is interesting in both designs and results. I wonder if you have considered other factors, like frequency of using the platform, number of years as a user and average grade on the other side of past trades, etc. I think those factors will affect the first investment in the game, since new users or users with unsatisfying experience are likely to invest less because of lack of trust in the platform.

[WMhYang]

Thank you very much for this interesting topic!

My first thoughts after reading are similar to the first point mentioned by @liu431. Since this paper does not really touch the causality issue, I wonder where should we start if we are going to tackle the problem. Moreover, I am also curious that with the explanatory model built in the paper, in other words with correlation instead of causation, could we obtain some insights on how the platforms or other parties could do to enhance people's propensity to trust others?

My second interest is regarding the matching between the guests and the hosts. Could the framework be extended so that we can figure out hosts and guests of what trust level are more likely to match together? In other words, given the propensity of a specific host, whether he/she is more likely to have a low propensity guest or a high propensity guest? And conversely, given the propensity of a specific guest, whether he/she is more likely to choose a low propensity host or a high propensity host?

[hihowme]

Thank you so much for your presentation in advance! The design of the online platform is becoming an important issue recently since it is becoming more and more popular. Therefore, the work of trying to find the right way to build a bond of trust. I am wondering would another party join this game to help build the trust? For instance, we choose UChicago because US NEWS ranked us on 3rd(just joking :)). We would choose a restaurant based on it's YELP stars. In this case, would that be possible to make some review about the house, or the hose salient? For instance their Facebook page or something.

[ydeng117]

Thank you for the presentation! In the paper, you defined trust only as of the speculation on the security provided by either hosts or customers. However, when making decisions on trust at these sharing platforms, people's expectation may not only include safety but also consists of quality of services, aesthetics of the products, communication experiences, etc. These factors may also affect the points which your participants would like to invest in the game you designed. Would confining the definition of trust challenge the accuracy of your model? How does your model could adjust these factors?

[anuraag94]

Thanks in advance for your presentation. I enjoyed your dicsussion of the literature examining practical proxies for trust. My question revolves around its philosophical underpinnings.

According to Annette Baier's rich discussion of the concept, reliance without the possibility of betrayal is not trust. You remark that the investment game is a commonly used tool used to measure trust. However, the possibility of defection in the game-theoretic context of the investment game is qualitatively different from the type of betrayal one would guard against when sharing a home in the context of AirBnB.

How does your operationalization of the notion of trust address this qualitative difference?

[lulululugagaga]

Thanks for your presentation. It's really an interesting real-world application of computational social science in the sharing economy. My question is that at the current stage the models look good, but in what cases you may start to update and add on to other factors such as what zeyu mentioned above?

[mingtao-gao]

Thank you for your presentation in advance. Constructing a statistical model for trust propensity is very useful for today's sharing economy platforms, yet it is very challenging in dealing with the complexity of trust. My question is about the investment game your team designed to simulate how each participant would trust the Airbnb profile. Since all the profiles are generated by the research team, then how got to decide how many points to return to the participants and do participants actually receive any credits back? If the money returned is decided by the research team, do all participants receive the same amount back, and if it differs, then the credits provided shows what researchers thought such a profile would provide, which would include biases into the game setting.

[Yilun0221]

Thank you for the presentation! I do believe that privacy and trust are important issues in sharing economy platforms. My question is, how the government and the company can make sure that the employees in the sharing economy platforms will use the data legally and properly? Can people develop algorithms to do this?

[timqzhang]

Thank you for your presentation. I am really interested in this topic as it concerns with machine learning methods. One curious point is about the feature selection process to select the behavioral signals. How does it proceed? And how it guarantee the signals selected are fitting well to the true behavior mechanism.

[vinsonyz]

Thank you so much for your repsentation, Paolo and Natã. The share economy has been a trend recently due to its low cost and availability. My question is that as social scientists, how can we use the big data or conduct experiments to improve people's comfort, due to there might be some issues like trust and privacy?

[RuoyunTan]

Thank you for the presentation. I am curious about how effective this investment game approximates the real mechanism of placing trust in strangers. In real-life decisions like trusting a stranger on Airbnb, people either gain something (like a memorable experience) or lose something (like rude guests destroying your house). In the investment game that you designed, participants might win a gift card if they have high points, but they would not lose anything that has significant real value even if they trusted the "wrong" people (besides the time they spend playing this game). Users are aware of this before they decide to participate in this game, and I think it is possible that this feature of the game will influence how they make decisions in trusting people in this game. Could you share your opinions on this?

[adarshmathew]

Thanks for your presentation and granting access to this paper. I have a few questions on the modelling outcomes.

Class-specific performance of user classification: In Fig 2 of your paper, we see that the model does well in predicting Low-Trust users, but heavily under-performs for High-Trust users. What could this be attributed to?

[yongfeilu]

Thank you very much for your presentation! It's so interesting to explore people's behavioral modes in both the short and long term in the context of sharing economy. However, I have 2 questions. First, how can you strike a balance between privacy protection and your goal of property protection? Second, could you please explain more on how people's behavior shifts in the long and short term and how your design deals with that? Thank you very much!

[lyl010]

Thanks for your presentation! I can see that the measurement of trust is important in sharing economy analysis. Could you please introduce more details about the difference between the trust propensity and and the self-reported perceived trust? Is there any interesting result that people with high trust propensity and low self-reported perceived trust or vice versa? And how largely can your models explain/ predict these 'counterintuitive' behaviors? Thank you a lot!

[jsgenan]

Hi Dr. Parigi. Thank you for bringing us such an interesting experiment! I really like the mechanism design, which uses a classic game theory experiment, but in th Airbnb scenario. I noticed that, please correct me if I'm wrong, in the experiment, you are only showing the participants very limited adn vague information about their potential guest/host, but in both the explanatory model and the predictive model, you are using rich behavioral variables collected on Airbnb. Is this based on a belief that the participants only make decisions based on their propensity to trust others? But then, the behaviors on the website are reactions to specific user profiles. How would you justify the general experiment and the user-specific predictive variables?

[chun-hu]

Thank you for your presentation! Trust is definitely an important aspect if these sharing platforms, and I'm assuming that there are different behavioral models and trust patterns in short-term and long-term contexts. You detailed the explanatory and predictive models of trust propensity in long-term contexts, e.g. Airbnb, and I'm wondering how we can adjust the model to better address the relationship in short-term contexts, e.g ride-hailing sharing economy platforms?

[KenChenCompEcon]

Thanks for delivering such an interesting talk. It is thrilling to see such great work in modeling customer behavior on the sharing economy platforms. My question is: the propensity to trust is modeled on experimental data from investment games, but how to control the possibility that participants' mindset for investment systematically diverges from that for consumption behavior over online platforms? Looking forward to your speech.

[di-Tong]

Thank you for sharing this interesting project! Regarding the limitation of data, you have mentioned that western views could be an important contextual condition that underpins the results you got here and recommended building “trust intentions” spectrum based on data from different countries/regions outside North America. My questions are: What kind of value classification system will you rely on for collecting more data? What do you expect the “trust intentions” spectrum to look like?

[PAHADRIANUS]

The topic of trust has grown tremendously in the field of behavioral economics over the past decades, and in the virtual space of internet its application especially deserves our attention. I am most excited to read your thorough work that managed to build a trust approximation model based on real human reactions from airbnb. However, I have a few concerns over the methodology; please excuse my presumptuous attempt since I can be totally wrong myself. The trust level predicting model that you applied seems to me to be quite similar to the credit scoring models that are currently in widespread use by financial institutions. The training structure, the feature generation and selection processes, as well as the core regression model all bear great resemblance to how the credit scoring method works. I definitely think that importing this method fit trust prediction, but there are some technical issues: frankly, this learning method is heavily reliant on the enormity of data sizes. The data with which banks predict people's honesty often have millions of observations, in comparison to this study's thousands, which is already harvested from diligent survey work; not only does it rely on the large size, the method also has access to a great numbers of available features, on which an automated algorithm selects the best. I wonder if in this study the feature filtering was entirely up to the computer or combined with economic intuitions. In the end, the validation accuracy of the model only slightly exceeds the AUC threshold of 0.75, which is an impressive result given the size of the data, but seems not fully convincing. I definitely agree to the whole design of the project, but this kind of data gathering might exceed the ability of scholarly conducted surveys; perhaps more support from the company itself is needed.

[nt546]

Thank you for your presentation! I was wondering if you could talk more about how designs around trust could be informed through your framework. Also, I might be mistaken, however, "In identifying such features, designers can nudge users toward the “path of trust building” in order to promote the intended outcome." points towards persuasive design as an outcome. What other design outcomes do you think could benefit by measuring trust propensity?

[skanthan95]

Thank you for your presentation! In the investment game you've described: how were pictures and names generated for the synthetic Airbnb profiles? Given that another study conducted on Airbnb found that guests with distinctively African American names were "16 percent less likely to be accepted relative to identical guests with distinctively white names" (Edelman et al., 2017), racial bias could have impacted your trust propensity measure. While I understand your rationale for not including this variable in your model, implicit biases could have affected the investment game outcomes. How do you account for this?

[SoyBison]

Thank you for coming to our workshop! My question has to do with with your model's predictions. It seems like the host model in particular has a really bad bias in the medium trust and high trust regimes. Why do you think this is?

[TianxinZheng]

Thanks for presenting such an interesting work. I have a questions regarding to the methodology: users who volunteer to participate in the investment game might be more active users of the platform. If so, how do you deal with the selection bias?

[linghui-wu]

Thank you, Dr. Parigi and Natã, for bringing such an interesting topic to us. My question is about the external validity of the research design. I agree with what you mentioned in the paper:

In other words, the user journey steps we mapped for the explanatory model are not comprehensive, and may not include all the points along the journey a user goes through when using the Airbnb platform or other sharing economy platforms.

Would you consider the genre of the sharing platform makes a significant difference in people's trust-building process? Or in other words, what do you think we need to be cautious about if we are trying to extend our analysis? However, there is no denying the fact that your research sheds light on how to measure trust propensity in sharing economy platforms. I am looking forward to your presentation.

[Leahjl]

Thanks for your presentation. It is inspiring to quantify the trust propensity sharing economy platform analysis. I'm interested in your behavioral framework, which is conducted to measure individual trust propensity on the platform.Does your theory apply to other platform apart from airbnb?

[bjcliang-uchi]

It does not seem like regional-specific factors are mentioned in detail. Do you think the levels of racial segregation and crime matter to your model on trust?

[YanjieZhou]

Thank you for your presentation! It is a really impressive and sucessful attempt to quantify trust as an abstract propensity for customers to harbour. I am especially interested in its extention in the area of behavioral sciences. Do you think that this quantitative method has other useful applications in this area?

[YuxinNg]

Thank you for the presentation. You designed an online investment game to capture users’ propensity to trust other users on Airbnb. This may be problematic, because when it is just online people may tend to show that they trust others. But when it is in reality, especially when it is related to people's safety or property, people may behave differently. I am wondering if you have taken this into consideration and how did you deal with this problem. Thank you!

[goldengua]

Thanks for your enlightening paper. As a psycholinguistics, I am very interested in the way you bridge real-world economical behavior to experimental methods. I was wondering whether your experimental framework could be further developed to combine both behavioral and neural experimental data. How could eletrophysiological techniques be applied under this experimental setting?

[hanjiaxu]

Many thanks to both of you for your presentation. As pointed out by other comments, it is very interesting to see that demographic information was left out when training the model to reduce bias. Ethical concerns in the machine learning field have been receiving more and more attention. Could you please talk more about the rationale behind this choice (i.e. not including demographic info)?

[harryx113]

Thank you for sharing your work! I have a lot of respect for your work in applying computational social science in sharing-economy industry. My question is regarding to the development of trust. As we know, trust can be a dynamic evolvement from interactions, either between guests and hosts or guests/hosts and platforms. It seems that the features chosen are mainly static. Could you speak on how you think "building trust" can play a role in sharing economy?

[tonofshell]

Thank you for sharing this very interesting paper! You mentioned that you sent out emails to Airbnb users and were in contact with Airbnb, who even provided you some survey data. Were these email addresses also provided by Airbnb for profiles you selected? What suggestions would you give other researchers interested in getting access to a company's private data for their project?

[yutianlai]

Thanks for sharing your work with us. I’m wondering how you deal with selection bias in your research since users who take participation might be systematically different from others. Would it be better to narrow the description scope?

[SiyuanPengMike]

Thanks a lot for your interesting and inspiring paper. I have a question related to the data generated by the game. Even though we could view the gaming data as a good representative of the trust of a person, how could we guarantee that the volunteer who participated in the game indeed played the game seriously? If they just click the options without careful consideration, the data you obtained is meaningless. Therefore, how could you prove the credibility or the quality of your data?

[dongchengecon]

Thanks a lot for the presentation! It is really a great idea to measure users' trust propensity on the sharing economy platforms. Matching hosts and guests based on trust would absolutely increase efficiency on the trust aspect. However, the matching could be less efficient with respect to the guests' preference on the product if one more constraint is added. How would you offer advice to the company considering the tradeoff here?

[bazirou]

This topic is really useful for personal experiences. Could you please introduce more details about the difference between the trust propensity and the self-reported perceived trust? Is there any interesting result that people with high trust propensity and low self-reported perceived trust or vice versa?

[MegicLF]

Thank you for your presentation! This topic is really inspiring in the aspect of designing an experiment to measure trust. After reading the materials, I wonder if there exists any potential bias in this research design and what you did to eliminate it. One of my concerns is that your design may only work in comparing the relative propensity of trust within a special platform. For instance, Airbnb is now a well-established service provider and its users somehow transform their trust towards the company to its hosts. However, if you compare the results of Airbnb with results from an unknown but similar room-sharing economy platform, would they be quite different? What did you do to make sure your experiment would also work for other sharing economy platforms?

[chiayunc]

Thank you in advance for the wonderful presentation. In the paper, you mentioned that to your knowledge, the investment game experiment has never been conducted to construct a trust model. To what extent do you think that making investments resembles opting to host someone in their homes or stay at a stranger's place? To me personally, the amount and the form of the risks the individual perceives in these separate occasions is quite different. It's one thing to give credit. It's quite another to have one's personal safety at risk.

[JuneZzj]

Thanks for presenting. The idea of game theory involved in the trust model is really fascinating. In terms of the behavior signals mapping, besides Lasso CV implementation, have you ever tried other possible penalties? Do you think there are any drawbacks of this penalty? Thanks.

[nswxin]

Thank you for the presentation! Your research remind me of the long-distance ride sharing named Didi in China. I really like that economy platform, however, it was banned after a notorious murder. I am wondering whether there's a way to borrow your framework to reform a similar ride sharing service.

[tianyueniu]

Thank you for sharing the interesting model! The model can be applied to so many social science situations. My question is, when applied to real world, sharing economy platform such as Airbnb gives users much more information than grey profile picture (e.g. users will be able to guess host's personality, or even sexuality). In that case, do you think the features selected would still be the same? Thank you!

[yalingtsui]

Thank you for your presentation! very interesting work. I wonder how you deal with other factors that may also affect the consumers' trust built, for example, service quality, individual perception, even mood!

[ShuyanHuang]

Thank you for presenting. This model has a great potential to be applied to various sharing economy platforms. The features you use are from users who are already using Airbnb. But is it possible to measure the trust levels of potential users, based on a broader range of information? If this is possible, this kind of trust measuring models can be more widely used in areas such as marketing.

[hesongrun]

Thank you so much for the presentation! My question is how do you establish the validity of the measure of trust? It seems to be an abstract concept and hard to measure in the first place. Additionally, what approaches do you take to understand people's behaviors in a systematic way? Thank you so much!