Digital Cloning of the Dead: Exploring the Optimal Default Rule

1.1 Background and Main Results

Death, traditionally viewed as the end of oneʼs narrative, no longer signifies a complete cessation in the digital realm. A digital clone, synthesized from both digital and non-digital information of personal histories, can now potentially extend oneʼs presence beyond physical demise. As artificial intelligence (AI) advances, this simulated continuation of life introduces new legal and ethical quandaries. It raises questions about consent for digital cloning and data ownership in an age where our digital selves can persist.

Digital clones have been created in various contexts up to this point. In terms of visual clones, the movie “Indiana Jones and the Dial of Destiny” showcased the digital de-aging of Harrison Ford (Bedingfield 2023). For thought clones, generative AI chatbot services like Replika and Project December have enabled conversations with chatbots simulating the personalities of specific individuals (Jee 2022).

Digital cloning technology leverages the advancements of deep learning, which is built upon neural networks. Within this framework, large language models are designed to understand and generate human-like text. Furthermore, generative adversarial networks – a subset of deep learning – play a pivotal role in creating realistic visual and audio simulations. Alongside these, computer vision techniques enable computers to interpret and generate visual data with heightened accuracy. This potent amalgamation allows it to trace and emulate an individualʼs thought patterns, expressions, and idiosyncrasies with startling accuracy.

While digital clone technology can be applied to both the living and the deceased, this paper narrows its focus to the latter, namely, digital resurrection. For the living, there exists the potential to obtain consent to digital cloning and rectify discrepancies between the clone and the individualʼs intentions. However, for the deceased, unless clear instructions were given during their lifetime, there is no clear avenue for consent or rectification. For these reasons, digital cloning of the deceased inherently carries a higher potential for many challenges than that of the living.

Commercial services for digital resurrection accessible to the general public have already emerged, and there is potential for the market to expand in the future. Once, a company named Eternime offered a service that allowed individuals to create their digital clone while they were alive and leave it behind after their death, but they could not secure many users (Jee 2022). However, recently, a company called “You, Only Virtual” began providing a service where users can upload someoneʼs text messages, emails, and voice conversations to create a chatbot (Zahn 2023). Project December is also currently offering a paid service for the general public, enabling conversations with chatbots of deceased individuals.^[1] Additionally, in 2020, Microsoft obtained a patent to create chatbots from text, voice, and image data for currently living individuals, historical figures, and fictional characters (Harbinja, Edwards, and McVey 2023).

Digital resurrection utilizes data left behind by the deceased. This, however, might entail infringements of various legal rights, including copyright, privacy, portrait, and publicity. For instance, the data used for clone creation might incorporate the deceasedʼs copyrighted materials. Digital clones might inadvertently expose the deceasedʼs secrets or harm their reputation. Furthermore, there is the potential for unauthorized use of the deceasedʼs likeness. In most legal jurisdictions, rights to privacy and reputation do not necessarily persist after death, suggesting that an individual might not have full control over how their data is handled posthumously. Consequently, prior research has proposed harnessing the deceasedʼs expressed intentions in their will as a prerequisite for digital clone creation (Harbinja, Edwards, and McVey 2023; Roberts 2023). However, these assertions arenʼt substantiated with empirical evidence, leaving it unclear how the general public truly perceives the role of consent in digital resurrection. Additionally, it is ambiguous whether the conditions for allowing digital clone creation should demand an explicit expression of consent or merely the absence of dissent. The default rule that would be most socially desirable remains an open question.

Given the context outlined above, this paper contemplates two major issues. Firstly, we consider the extent to which the wishes of the deceased play a role in assessing the societal welfare or net societal benefits of digital resurrection. Theoretically, if the negative externalities of digital resurrection are significant, it might be conceivable to adopt a rule prohibiting digital resurrection from an ethical standpoint, regardless of the deceasedʼs wishes. Conversely, if the positive externalities of digital resurrection are pronounced, a rule permitting digital resurrection regardless of the deceasedʼs wishes could be contemplated. Moreover, if respecting the wishes of the deceased is essential while considering the externalities of digital resurrection, a rule might be considered that permits digital resurrection in certain cases contingent on the consent or dissent of the deceased. Before pondering what the optimal rule might be, we first need to explore the role played by the consent or dissent of the deceased.

Next, if the wishes of the deceased are found to play a crucial role in evaluating the societal net benefits of digital resurrection, we will consider what the optimal rule might be. Although various rules can be contemplated, for the sake of simplicity, this study will focus on examining the opt-in and opt-out rules as mechanisms to respect the wishes of the deceased.

This study addresses these two issues in the following manner. First, regarding the social net benefits of digital resurrection, we will investigate how the public perceives the societal acceptability of digital resurrection through a survey experiment. The underlying theory for this investigation is as follows: rooted in the Bayesian model of information processing, individuals form beliefs about the societal acceptability of specific cases of digital resurrection, taking into account the perceived social benefits and costs. When an individual believes that the deceasedʼs consent plays a significant role in societal acceptance, awareness of the deceasedʼs expressed intentions should update their beliefs. Thus, we predict that knowledge of a deceasedʼs prior consent will lead to higher societal acceptance than knowledge of a deceasedʼs prior dissent. Given that a majority of the population is expected to value the presence of consent in digital resurrection, we hypothesize that the deceasedʼs consent will significantly boost the societal acceptance of digital resurrection, particularly when contrasted with cases where dissent is expressed.

Next, to determine whether the opt-in or opt-out rule is preferable as a mechanism to respect the wishes of the deceased, we employ an observational research design and survey the public on the extent to which they would consent to their own digital resurrection under certain assumptions. This analysis is based on a theoretical model that assumes individual net benefits related to digital resurrection follow a probability distribution. As a secondary point of discussion, we also explore the possibility of using the consent of the deceasedʼs parents as a substitute for the deceasedʼs own consent.

In this study, we conducted a pre-post survey experiment targeting adults in the United States (U.S.), where a dependent variable was measured before and after treatments. In the experiment, participants were presented with vignettes, in which a young woman in her twenties tragically dies in a car accident. Following the accident, friends consider using a digital resurrection service offered by a company. Initially, the deceasedʼs consent towards digital resurrection is ambiguous, but her pre-mortem intention is revealed later. There were two treatment versions: one where the deceased had given consent and another where she had dissented. Participants read one of the two versions assigned at random and responded on a 5-point Likert scale about societal acceptability of the digital resurrection both before and after the treatment.

The presence of expressed consent was associated with 2-point higher societal acceptance on average on a 5-point scale compared to the presence of expressed dissent. In the consent scenario, 58 % deemed digital resurrection socially acceptable, compared to a mere 3 % in the dissent scenario. The majority of participants believed digital resurrection was socially unacceptable when the deceasedʼs expressed wishes were unknown. The effect of consent compared to expressed dissent was the largest in this group among all subgroups. This indicates that, for most people, beliefs regarding the societal acceptability of digital resurrection are weak when the deceasedʼs preference remains uncertain, and the revelation of their wish shift their attitudes substantially.

To supplement the experimental study, an observational study was conducted using the same vignette to compare the case where the deceasedʼs consent is uncertain with the case where the deceasedʼs parents have given consent. In the scenario where the deceasedʼs stance is unknown, only 11.3 % believed the digital resurrection to be socially acceptable. This percentage was 27 % when the deceasedʼs parents have consented. Respondents were also asked to imagine themselves in the position of the character in the vignette: first, if they, as the deceased, would agree to the resurrection and second, if they, as the deceasedʼs parents, would agree to the resurrection of their child. While 28 % answered that they would agree if they were the deceased, only 17 % answered that they would agree if they were the parents.

These results indicate an association between parental consent and an increase in societal acceptance when the deceasedʼs own stance is uncertain. However, the extent of this increase appears to be modest. Furthermore, in the context of the vignette presented, the finding that most respondents would be hesitant about creating a digital clone of themselves post-mortem, or of their deceased child, suggests a public sentiment: regardless of societal acceptability, many individuals personally oppose the concept of digital cloning.

This study aims to serve as a starting point for the topic and does not purport to make strong assertions or definitive conclusions regarding the optimal rule for digital resurrection. However, after comprehensively analyzing the above results, we derive a tentative conclusion that the opt-in rule might be more preferable than the opt-out rule. In other words, the default rule is that digital resurrection is prohibited. Digital resurrection is only permitted when there is an expression of consent.

1.2 Literature Review

This study primarily contributes to two streams of literature: the legal regulation of digital clones and the economic analysis of default rules.

First, examining the literature on the legal regulation of digital clones, Truby and Brown (2021) classified digital clones into several types and argued that regulations are necessary to protect people from unauthorized development of digital thought clones and the harmful use of personal data. Lees, Bashford-Rogers, and Keppel-Palmer (2021) discussed the ethical and legal issues surrounding the creation of deepfakes of the deceased celebrities, while Boothe (2022) focused on broader concerns, encompassing digital clones in general, of which deepfakes are but one form. Both Harbinja, Edwards, and McVey (2023) and Roberts (2023) propose using the testamentary expression of the deceasedʼs wishes during their lifetime as a requirement for digital clone creation. However, the former believes in an opt-out rule, where digital resurrection is prohibited if dissent is expressed, while the latter supports an opt-in rule, permitting it only if consent is expressed. Much of this literature is theoretical, with scant empirical evidence. Therefore, this study empirically analyzes the role of consent in digital resurrection, filling the gap between theory and empiricism, and contributes to the literature on the legal regulation of digital clones.

Next, when considering the literature on the economic analysis of default rules, many studies have shown that the design of default rules can lead to improvements or deteriorations in social welfare. Ayres and Gertner (1989) pointed out that not only transaction costs but also the strategic behavior of contract parties stemming from information asymmetry can be reasons for incomplete contracts and discussed the possibility of default rules improving efficiency. Sunstein (2002) discussed the potential influence of default rules on peopleʼs preferences and behaviors, analyzing the effects of changes in default rules. Bar-Gill and Ben-Shahar (2021) analyzed the role of information costs in opting out of default rules against the backdrop of the increasing importance of default rules as policy tools.

Morse and Birnhack (2022) argue that the so-called “privacy paradox” – a gap between usersʼ expressed preferences and their actual behavior when it comes to protecting privacy – persists in a changed form in the data people leave behind after death. They further discuss the potential for default rules to have a significant impact on posthumous privacy issues. This research seeks to determine whether prohibition or allowance is better as a default rule in the context of digital resurrection, thereby contributing to the economic analysis of default rules.

Section 2 delves into our theory and hypotheses, Section 3 details the methodology for hypothesis testing, Section 4 showcases the results, and Section 5 deliberates on the policy implications.

2.1 Legal Regulations

Legal regulations concerning digital resurrection may seem intricate at first glance, but comprehending the processes involved in creating and using a digital clone of the deceased can simplify the understanding.

The creation of a deceased individualʼs digital clone involves the collection and utilization of data left behind by that individual. During the data collection process, some of the data might include the deceasedʼs copyrighted works, and unauthorized use may raise issues of copyright infringement. When using the deceasedʼs data for machine learning, there is potential for copyright issues specific to machine learning. Depending on the purpose of the digital resurrection and the type of data used, legal restrictions might apply. There are relatively clear legal stipulations, such as Articles 3 and 4 of the European Union (EU)ʼs Digital Single Market Copyright Directive,^[2] which delineate when machine learning is permissible. On the other hand, regulations like the Fair Use Doctrine under U.S. copyright law (Section 107) determine on a case-by-case basis when machine learning is allowed,^[3] leaving the boundaries ambiguous. Thus, legal considerations might vary based on the purpose and methods of the digital resurrection and the jurisdiction in which it is conducted.

Furthermore, during the data collection and usage processes, the deceasedʼs secrets might be disclosed, or their likeness used without permission. A digital clone could potentially make statements or exhibit behaviors contradictory to the deceasedʼs beliefs or philosophy. Consequently, potential infringements on rights concerning privacy, likeness, and personality can arise in digital resurrection. The content of these rights may vary by jurisdiction. However, in most jurisdictions, including the U.S. and EU, these rights typically do not persist after death, with exceptions granted only in limited cases. For instance, in the U.S., some states recognize post-mortem rights of publicity, but unless it concerns commercial use of a celebrityʼs likeness, winning an infringement lawsuit can be challenging (Roberts 2023). The European Commission, on April 21, 2021, introduced a regulation proposal regarding AI, demanding explicit labeling of content generated artificially when using deepfakes.^[4] However, this regulation might not be sufficient to address all the issues raised by digital resurrection.

In this way, digital resurrection can give rise to various legal issues primarily centered on infringements of the deceasedʼs rights. To better safeguard these concerns, as proposed in prior research, rules can be envisaged that only permit digital resurrection with the deceasedʼs consent or that prohibit it if thereʼs explicit dissent. Theoretically, rules allowing or prohibiting all forms of digital resurrection are also conceivable. The most appropriate default rule largely depends on how people consider the deceasedʼs consent and dissent, not only from a legal standpoint but also considering ethical and other aspects. This is because the social benefits and costs of digital resurrection heavily depend on public evaluations.

2.2 Societal Perspectives: Bayesian Model of Information Processing

Next, we detail our theory and associated hypotheses. Our model is an application of the Bayesian information processing model (Bullock 2009). We hypothesize that an individual i holds a belief about the societal net benefit W _i of any given deceased personʼs digital resurrection, and W _i can be written using the function g as equation (1):

Here, B _i and C _i respectively represent the beliefs individual i holds regarding the social benefits and costs of digital resurrection, both of which are random variables (see the beginning of Section 2 for examples of the social benefits and costs). Additionally, A _i is a random variable that captures individual iʼs belief concerning the deceasedʼs consent or dissent. The presence of consent or dissent affects the perceived social benefits and costs, B _i and C _i .

An individual possesses beliefs about B _i and C _i prior to knowing about the deceasedʼs consent or dissent. However, upon becoming aware of the consent or dissent, their beliefs about B _i and C _i are updated, and as a result, the belief about W _i is updated. Regarding K _i , it is a vector of parameters representing individual iʼs knowledge and attitudes toward AI, influencing the evaluations of B _i and C _i .

If, generally speaking, whether a deceased person has consented has a significant impact on the societal acceptability of their digital resurrection, then only a limited number of people should possess strong beliefs about societal acceptability regardless of the deceasedʼs consent status. Many might have weak beliefs about societal acceptability when the deceasedʼs consent is unknown.

If this holds true, most people, upon learning about the deceasedʼs expressed consent, would likely update their attitudes in a positive direction regarding societal acceptability. Conversely, upon learning about the deceasedʼs expressed dissent, they would likely update their attitudes in a negative direction.

To investigate this in a survey experiment, one could adopt a pre-post design: initially asking participants about their opinions on the societal acceptability of digital resurrection when the deceasedʼs consent is unknown and later inquiring again after revealing the consent or dissent as a treatment. To elucidate the pure effect of consent or dissent, one would need to use a control group where the consent remains unknown. However, due to budgetary constraints in this research, we opt to compare scenarios where consent or dissent is made known. As highlighted earlier, most jurisdictions do not protect the deceasedʼs rights, including privacy, likeness, and personality. Given this backdrop among others, we currently know little about whether the deceasedʼs consent or dissent influences societal acceptability of digital resurrection. Thus, as a starting point, comparing these two scenarios is essential. We propose the following hypothesis.

2.3 Personal Perspectives: Implications for Default Choices

While peopleʼs evaluations concerning the societal acceptability of digital resurrection can be informative in inferring its social net benefits, the way people perceive societal acceptability and whether they would personally consent to being digitally resurrected are not necessarily linked. Even if an individual perceives digital resurrection as socially acceptable, they may not personally agree to undergo such a process, and vice versa.

The information on how many people personally consent to their own digital resurrection can be valuable in designing default rules. Even if the net benefit of digital resurrection is positive and digital resurrection is deemed acceptable, the societal welfare could vary depending on the default rule in place. Prior research suggests two types of rules: the opt-in rule, which allows digital resurrection only when there is explicit consent, and the opt-out rule, which prohibits digital resurrection only when there is explicit dissent. In this context, the opt-in rule sets the prohibition of digital resurrection as the default and recognizes exceptions when there is a consent statement, whereas the opt-out rule sets digital resurrection as the default and recognizes exceptions when there is a dissenting statement.

Let us consider this point using a simple model. The purpose of this model is entirely different from that of the model in Section 2.2, and it should be noted that the notations used in these two models have no relation to each other. Suppose every individual in society is homogeneous. Let us denote the expected personal net benefit an individual receives from their own digital resurrection as b. This benefit b has a cumulative distribution function F(·) on the full support (−∞, ∞). This can be interpreted as representing the distribution of personal net benefits in society. Let us also denote the average net benefit a relative or loved one receives from any individualʼs digital resurrection as b _f . In reality, b _f might vary across relatives, but for simplification, we assume each relative receives the same benefit. While b and b _f might actually be correlated, we ignore this correlation. Let us further define the cost for an individual to express their consent or dissent regarding digital resurrection, possibly through a will or another method, as e, and assume it is the same for everyone. The costs of expressing consent and dissent might actually differ, but we assume there is no difference. Other externalities in digital resurrection are not considered here. We present this simple model as a benchmark.

The social welfare W _I under the opt-in rule can be expressed by the following equation (2):

Under the opt-in rule, only individuals receiving an expected net benefit b that exceeds the expression cost e will consent to digital resurrection. The first term in equation (2) represents the social benefit obtained from the potential digital resurrection of individuals, which includes the net benefits of both the individuals themselves and their relatives. The second term denotes the social cost arising from the expression cost borne by individuals consenting to digital resurrection.

In contrast, the social welfare W _O under the opt-out rule can be expressed as shown in equation (3):

The first term of equation (3) represents the social benefit composed of the net benefits from potential digital resurrections for individuals and their relatives, similar to equation (2). However, in the opt-out rule, the threshold for expressing intent changes to −e. For individuals whose personal net benefit from digital resurrection falls between −e and e, it is more advantageous to remain in the default state rather than bear the cost e for opting out. In other words, under the opt-out rule, some individuals who receive a negative net benefit from digital resurrection might still be potentially resurrected due to the presence of expression costs. The second term indicates the social cost composed of the expression costs borne by individuals who express their dissent to digital resurrection.

The difference between W _I and W _O , ∆W = W _I −W _O , can be expressed as shown in equation (4):

If this value is positive, it means that the social welfare is higher under the opt-in rule than under the opt-out rule, making it advisable to adopt the opt-in rule. The first term represents the difference in social benefits, and the second term is the difference in social costs.

If a majority of people dissent from digital resurrection, adopting the opt-in rule may be socially preferable for the following reasons.

First, under the opt-in rule, individuals who would receive a negative net benefit from digital resurrection will not be resurrected, unlike under the opt-out rule. This effect is captured in the first term of equation (4). Even if an individual would receive a negative net benefit from digital resurrection, if their relatives or close ones receive a large positive net benefit, resurrecting these individuals could improve societal welfare. However, this scenario seems unrealistic. The net benefits b of the individual and b _f for their relatives or loved ones are typically expected to be positively correlated. If an individual does not desire digital resurrection, itʼs likely that b _f would be negative if those around them respect the individualʼs wishes. Furthermore, if a net benefit b for an individual is small and a net benefit b _f for those around them is high, these surrounding individuals could pay a compensation to get the individualʼs consent for digital resurrection, thus increasing b. Considering these facts, there seems to be no need to resurrect individuals who would receive a negative net benefit from digital resurrection.

Second, the opt-in rule can reduce the number of individuals expressing their wish, thereby saving social costs associated with such expressions, as indicated in the second term of equation (4). Although our model assumes, for simplicity, that expressing either consent or dissent costs the same amount e, in reality, the cost of expressing dissent may be higher than that of expressing consent. For instance, when expressing consent, one might bear the expression cost to realize a positive net benefit, whereas when expressing dissent, one bears this cost to avoid a negative net benefit. Recognizing the possibility of a negative outcome, and then taking steps to prevent it, can be more emotionally taxing and demanding than simply looking forward to a benefit. The latter might entail a higher psychological cost than the former. If this holds, in situations where a majority opposes digital resurrection, the cost-saving effect of the opt-in rule becomes even more pronounced.

Considering the above discussions, the starting point in determining the preferable rule is to first ascertain what proportion of individuals would not consent to digital resurrection.

When the majority of individuals have a low net benefit and only a minority have an extremely high net benefit, the distribution of b is right-skewed or positively skewed. This is likely representative of a real-world distribution. There are several pieces of evidence to suggest this. The company Eternime, which once offered a service to leave oneʼs digital clone after death, could not attract a sufficient number of customers (Jee 2022). Furthermore, when Microsoft obtained a patent related to digital resurrection, it faced significant public backlash. This pushed the company to announce that they had no intentions of developing technology for digital resurrection (Harbinja, Edwards, and McVey 2023). These facts indicate that the majority of people might not support the idea of their digital resurrection.

In contrast, for actors and other celebrities, the possibility of their digital clones generating commercial profit might mean they are more likely to perceive a higher net benefit. In fact, many of these celebrities have begun creating digital scans of themselves during their lifetime for posthumous use (Roberts 2023). If there is a significant number of individuals in such a group, the distribution of the net benefit of digital resurrection b might be bimodal, having peaks in both the negative and positive areas. However, considering the overall population, celebrities comprise a tiny fraction, and the overwhelming majority may likely have a negative personal net benefit from digital resurrection. Based on the discussions above, we propose the following hypothesis:

In this study, as a starting point in exploring the default rules concerning digital resurrection, we use observational research to inquire about peopleʼs opinions on whether they would consent to their own digital resurrection. As a supplementary analysis, we also ask participants to hypothetically imagine having a child and then determine, from a parental standpoint, if they would consent to the digital resurrection of that child.

2.4 Parental Consent

In many countries, there are accepted methods for parents or legal guardians to represent an individualʼs wishes in situations where the person is unable to appropriately express their preferences or die before doing so. For instance, surrogate decisions are permitted when minors or certain adults face medical decisions or other crucial choices. The question of whether parents or legal guardians should be permitted to indicate a preference on behalf of the deceased concerning digital resurrection, especially when the deceased did not or could not express agreement or disagreement while alive, emerges as a critical point of discussion. This study focuses on parental consent, examining its impact on the societal acceptance of digital resurrection.

When parents express a wish on behalf of the deceased, while there might be instances of potential conflicts of interest among the parents and the deceased, it is generally believed that parents are likely to make decisions that mirror the affectionate family relationship they had with the deceased when they were alive. Consequently, the presence of parental consent might enhance societal acceptance of digital resurrection. As described later, we will explore this point through a supplementary observational study. Hence, we propose the following hypothesis.

3.1 Survey Material

The survey is composed of three sections: questions about knowledge and attitudes towards AI, vignettes and related questions, and demographic and background questions. The full text of the survey can be found in the Appendix.

3.2 Data

In survey experiments, the frequent use of the post-only design, where a dependent variable is measured only after treatments, often stems from concerns about demand effects and consistency pressures when measuring dependent variables repeatedly. However, these concerns arenʼt necessarily underpinned by empirical evidence (Mummolo and Peterson 2019). On the contrary, the post-only design can compromise the statistical precision of treatment effect estimates (Clifford, Sheagley, and Piston 2021). In this study, to probe the predictions of our theoretical model, we find it imperative to gather participantsʼ perspectives on the social acceptability of digital resurrection in two distinct contexts: when the deceasedʼs preferences regarding digital resurrection are ambiguous and when there has been a clear expression of the deceasedʼs consent or dissent. Given these nuances, this research employs a pre-post design, which measures a dependent variable before and after an intervention.

Power analysis suggested that to achieve 80 % power with an assumed small to medium effect size, a sample size of around 200 would be adequate.

Data were collected in August 2023 using Prime Panels from CloudResearch, a widely recognized online crowdsourcing platform in the social sciences (Chandler et al. 2019). This yielded a sample of 222 adult U.S. residents. The average time it took respondents to complete the survey was 5 min. The survey included two comprehension questions (Q5, Q10), and those who answered them incorrectly were excluded from the sample. The proportion of those who passed all comprehension questions was 65 %. Prime Panels amalgamates several general online survey panels. Participants agree to some form of compensation before registering with each panel, and the amount of compensation varies by panel.

3.3 Regression Models

To analyze the impact of deceased individualsʼ consent and dissent on the social acceptability of digital resurrection, while taking into account other independent variables, we employ regression analysis. The regression equation for social acceptability before the intervention is as follows:

Here, the dependent variable PRE _i represents the assessment of individual i regarding the social acceptability of digital resurrection (Q6). The symbol x _i is a vector of independent variables comprising the scales for AI engagement, Market trust, Harm predictability, and Religiousness (Q1–Q4, Q18). The symbol z _i is a vector of control variables that includes gender, age, education, income, marital status, and presence of children (Q12–Q17). The symbol ε _i represents the error term.

Next, the regression equations for social acceptability after the intervention are as follows:

The dependent variable DIF _i in equation (6) represents the difference in participantsʼ evaluations of the social acceptability between post-intervention and pre-intervention. The symbol TRE _i is a dummy variable that takes on a value of 0 if the vignette of expressed dissent is assigned, and a value of 1 if the vignette of expressed consent is assigned. This essentially represents the relative treatment effect of consent compared to dissent.

The estimation equation (6) is used to estimate the treatment effect based on the overall sample. In contrast, equation (7) estimates the treatment effects by categorizing the entire sample into subgroups based on the pre-intervention evaluations (PRE _i ), which can be Unacceptable (1,2), Neutral (3), or Acceptable (4,5). The symbols T R E i u , T R E i n and T R E i a are dummy variables representing the treatment effects for the respective subgroups.