2023年5月30日北大太阳集团所有网址16877行为科学与政策干预交叉创新团队2023春季学期第四期（总第十三次）分享会顺利举行。本次分享会邀请到香港中文大学（CUHK）商学院决策、运营与科技学系助理教授李易珊带来两篇关于隐私与个人信息使用偏好的最新研究分享。

分享人 The Speaker

Professor Yi-Shan Lee is an Assistant Professor of Department of Decisions, Operations and Technology at The Chinese University of Hong Kong. She received her PhD in Economics from the University of Zurich. Prior to that, she received her MA and BA in Economics from the National Taiwan University.

Her research interests lie in the areas of experimental economics, microeconomics and decision sciences, with a particular focus on critical issues in the information economy. Overall, her research aims to extend our understanding of economic decisions involving goods and services to tradeoffs involving the modern usage of information.

分享会 The Seminar

On May 30th, the GSM Behavioural Science in Action Seminar Series welcomed Yi-Shan Lee, an assistant professor of of Department of Decisions, Operations and Technology at the Chinese University of Hong Kong (CUHK) Business School. Lee's recent research focuses on the acquisition and utilization of information in small networks. Within this field, she is particularly interested in studying behavioural biases in endogenous information usage and variations in biases across communities with different characteristics. In the seminar, Lee shared with us two of her recent studies on individuals’ personal information usage preferences.

Research 1: Rationality in Revealed Privacy Preferences (with Roberto Weber)

Lee pointed out that in the modern economy, personal information can be regularly exchanged for service or money. The ability to use information as a new currency is an important modern economic behavior, yet we have very little understanding of how individuals actually utilise information. The first research shared by Dr. Lee examines whether people can rationally manage personal information sharing. Findings from this project will provide evidence for deciding whether to intervene with individuals' information privacy using legislative means.

Testing rationality in privacy choices involves substantial challenges. It is impossible to exhaustively test privacy-related rationality in real-world contexts, as many factors that may contribute to irrationality are not unique to privacy trade-offs. To isolate privacy decisions from such factors, the study creates an experimental situation where participants’ choices are bounded to simple trade-offs between two privacy items. With the participants fully informed of the consequences of their sharing decisions, the setting drastically reduces the information cost of making an informed and rational choice. As a result, findings from this research should be regarded as an upper bound of individual’s privacy sharing rationality – if a subject fails to manage her personal data sensibly in this experimental setting, it is unlikely that she will do so in real life.

The study uses participants' body fat composition and intelligence score as the two privacy items to be shared. At the beginning of the experiment, each participant undergoes an intelligence assessment and a body composition test, and two reports (each including assessment information and a photograph of the participant) are shown to the participant in private.

Privacy attitudes

Depending on their sharing preference when no trade-off between the two pieces of information is imposed, participants are classified into four types. Those who wish to show both pieces of information to as few viewers as possible are the “private” attitude type (68%). Those who wish to show both pieces of information to as many viewers as possible are the “public” type (10%). The rest are those who wish to keep one piece of information public and the other private – the ones who wish to have their intelligence score public are named the “IQ+” type (14%), and the reverse are the “BF+” type (8%).

Rationality

The notion of rationality in participants’ privacy choices is determined through the General Axiom of Revealed Preference (GARP). Methodologically, this means the researchers observe participants’ choices and infer about their preference ordering from the observed choices. In this part of the study, the researchers impose a trade-off between the two pieces of information – to show one piece of information to fewer people, it must be that the other piece of information is shown to more people. The researchers manipulate two variables: the relative trade-off “price” between the two pieces of information (i.e., if the participant wishes to show her body composition report to one less person, the “price” is the increased number of viewers her intelligence report would be shown to, which may be either above or below 1), and the level of sharing (i.e., the number of viewers the body composition report would be shown to if the participant’s intelligence report is kept completely private). Participants are asked to decide on their preferred sharing ‘bundle’ in 20 different scenarios.

For the sake of readability, if a participant indicates that she would like her body composition report to be shown to x viewers and her intelligence report to y viewers, we denote this as a sharing bundle (x, y).

For a participant to be considered “rational,” she should be a utility maximiser with consistent preferences across circumstances. Economically, this means a rational participant’s preference revealed by her actions should be represented by a well-defined, complete, and transitive preference ordering. Intuitively, think of a participant who, when having to choose between (2, 10) and (3, 6), preferred the latter; when having to choose between (3, 6) and (4, 2) in another scenario, preferred the latter; but when having to choose between (2, 10) and (4, 2) in the third scenario, preferred the former. Then this participant is not someone whom economists would call rational.

The study finds that sixty-three percent of subjects act as utility maximisers when allocating privacy levels, which is comparable to that found in previous studies involving non-privacy choices. Conditional on participants’ privacy attitudes, rational individuals make up 72.79% of the private type, 50% of the IQ+ type, 35.29% of the BF+ type, and 40.81% of the public type.

Extension: Monetary Translation

In this stage, a set of bundles (made up by bundles selected by the participant in the previous stage and some unchosen “perturbed” bundles) are shown to a participant. The participant asked about the lowest monetary price they would accept for each bundle to be implemented, which is termed their “willing to accept” (WTA) value for the bundle. The computer randomly draws a price from the uniform distribution between 0 and 40 as the “willingness to pay” by the experimenters. If the randomly drawn price is higher than the participant’s WTA value, the participant receives the amount indicated by the random price in reality, and the bundle is implemented. Otherwise, the participant does not receive any payoff from this part, and the reports are kept private.

What makes a sensible choice in this stage? Consider a linear trade-off situation between the two pieces of information at the price of 1/2 and an overall level of 10. This would allow the following bundles: (10, 0), (8, 1), (6, 2), (4, 3), (2, 4), and (0, 5). For a sensible participant who faced this choice in the previous stage and had chosen (0, 5) as her most preferred bundle, her WTA values for any of the other four “perturbed” bundles in this part must be no lower than her WTA for (0, 5).

It was found in this stage that monetary values can indeed capture the privacy attitude at the group level – for a set of uniformly pre-set bundles, the private-type group of participants has a significantly higher average WTA than the public-type group.

At the individual level, the experiment found 47% rational types and 65% irrational types made choices that violated the notion of sensibility described two paragraphs above. However, the magnitude of such violation is 2.6 times more severe among the irrational participants than the rational ones.

In other words, individuals who are inconsistent when engaging in privacy tradeoffs exhibit substantially more costly preference reversals when pricing the sharing of their personal information. This may imply that solving privacy issues by monetizing personal information can have distinct monetary welfare consequences for people with different degrees of rationality in their underlying ability to make sensible trade-offs involving personal information.

Extension: Real Life Behaviours

The researchers also provided incentives for participants to provide verifiable information relating to their real-life privacy behaviors. Information like the number of information items shown to strangers on their social media profile and the number of active social networking accounts is collected to elicit the participant’s real-life privacy attitude. Meanwhile, the number of categorised memberships and club memberships in use, and the use of locational services by digital service providers, are used to measure the participant’s preference in the exchange of privacy for money/services.

After controlling for individual characteristics, the results show that individuals who are public types in the experiment share significantly more personal data on social networks than private types. Meanwhile, privacy attitudes have no impact on the exchange of personal information for money and services; however, it is likely that the irrational type tends to trade their personal information for money or services more often than the rational types.

Research II: Personal Information and Information as Public Goods (with Josie Chen)

The second study examines individuals’ information usage decisions when interactions with other decision-makers are involved. In the digital economy, a new form of public good contribution arises – by sharing our digital usage analytics, users can help improve the product/service. Such improvements benefit all users, even those who have chosen to retain full control over their privacy. In this study, the authors consider three types of public good contribution: personal information, others’ information (i.e. information in general), and money, and examine if individuals use the three types of contribution differently in the public good game. The study compares the public-good generating process and social welfare between the information economy and manufacturing economy.

The Experiment

At the beginning of the experiment, participants are randomly assigned roles either as a personal data provider or a data user. Personal data providers fill in 20 personal details while data users copy and paste 20 personal details of others. The information provided or copied will be verified at the end of the experiment with a 50% probability.

Then, the public game is iterated with either money, personal information (for data providers only), or others’ information (for data users only) as the unit of the game. In each round, each player is endowed with 20 units of contributable items, which can be in the form of tokens, own information, or others’ information. Players are grouped according to their endowment type, and everyone decides how many units of their endowment they wish to contribute to a public “pool.” At the end of the round, the sum of the contribution in the pool is doubled and evenly distributed to all players in the game, regardless of their contribution.

Later, the experimenters elicit the data users’ WTA for their personal information. Collection of participants’ individual characteristics and the data verification process also take place in this stage.

Observed Contribution

Theoretically, the equilibrium of the public good game is a symmetrical one where no one contributes anything. However, it was observed that most participants chose to contribute a noticeable amount to the public good pool.

Overall, we found two effects consistent with previous findings in the literature on public good games. Firstly, conditional cooperation. That is, people contribute positively but modestly in the first round to see other players’ public contribution attitudes. They stop contributing if others are non-contributors and contribute more if they are happy with the group’s level of contribution. The second effect is the endgame effect. As the game approaches the final round, the opportunity cost of being uncooperative reduces as their counterparts have fewer rounds to respond to their uncooperativeness, so participants start to gradually decrease their level of contribution. The two effects are prevalent in most public good games.

Own Information vs. Others’ Information

People contribute more when using their own personal information than when they have to share other people’s information. In the experiment, significantly higher levels of contribution were found among the personal information group than among the others’ information group.

Others’ Information vs. Money

People use others’ information as if it were money, shown by similar levels of contribution when using others’ information and money.

Own Information vs. Money

A significantly higher level of contribution was found among the personal information group than the money group. However, the WTA values elicited indicate that this “oversharing” of personal data is not caused by a low valuation of people’s own information. The finding highlights the generosity when using one’s own personal information even when it is monetised.

For more information on Lee’s research, please check out her website. More information on the first study can be found here.