## Introduction

The COVID-19 pandemic has highlighted the importance of digital epidemiology based on public health data. (Vuong et al., 2022) Conceptually, the paradox between public health and personal privacy has come into question. In the early stages of the pandemic, public health officials faced critical problems in collecting effective information input from the public. Specifically, the issues around the donation of digital data as it relates to the context of mobile phone privacy became a major roadblock in trying to fight the COVID-19 pandemic. The question of how to effectively influence motivations on the data provided with the sub-question of the influence of monetary, versus altruistic motivations, became a key factor in understanding how to resolve the paradox.

One key data source has been human mobility which can be collected from geospatial global positioning system data (GPS) on smartphones that can be used to support efforts to understand the transmission patterns of COVID-19 and to control the effectiveness of public health interventions like contact tracing (Perra, 2021; Beria and Lunkar, 2021; Grantz et al., 2020). GPS data provided by users’ smartphones can be analyzed to obtain a verifiable record of individuals’ human mobility patterns and help predict the future disease trajectory of COVID-19 such as the identification of hotspots and the social and environmental factors that contribute to the further spread of COVID-19.

However, large concerns have been brought up about how users may not readily release their personal data, and if so they will only do it under specific conditions (Acquisti et al., 2016; Posner, 1981). This “privacy concern” (Posner, 1981) must be compensated by an additional means that would drive people to give access to their data. For social scientists, one of the key points in data donation behavior is the motivational mechanism that can be activated to incentivize people to donate their private data. During the COVID-19 pandemic, disclosing personal data can generate social benefits, and so, “data altruism”—data donated for the common goodFootnote 1 —may act as a major motivational mechanism. However, “data altruism” by itself may not be able to counteract the strong obstacle of concern for privacy, at least to be able to obtain a sufficient acceptance rate level (estimated at 60%, in Ferretti et al., 2020). Evidence had also emerged in the field of behavioral economic framing (e.g. Oullier et al., 2010), that the form of the message may have an impact, depending on the salience of the social motive for the data donation.

As there is the collective benefit of data donation, the decision involves more than the individual balance between privacy concerns and the participant’s personal willingness to give access to their data. Thus, the public authority may consider another motivational mechanism: a monetary reward, with the aim to trigger people to internalize the collective benefit of their decision. It has already been shown that a monetary reward can bias the self-interest calculation toward data donation decisions (Gefen et al., 2020). However, in the behavioral economic literature, individuals seem to be more sensitive to the loss aversion effort—greater fear of losses than to a symmetric gain. In addition, a strong complication could also come from the “crowding-out effect” of extrinsic incentives. Deci, the founder of the self-determination theory was the first to detect this kind of unexpected effect in professional behaviors (Deci, 1972): workers could do less (rather than more) when they are remunerated for some tasks that they anyway have the intention to do. Economists then progressively developed a reflection on professional motivations (Kreps, 1997) and more generally on prosocial motivations (Frey et al., 1997; Gneezy and Rustichini, 2000). Since the end of the 1990s, several studies on pro-social behavior, particularly in the health domain (either patients or professionals), have confirmed the concerns that monetary payments may crowd out intrinsic and altruistic motives, thus ultimately reducing the overall social contribution (Rothman and Rothman, 2006; Himmelstein et al., 2014). To cite the most famous example, it has been documented that blood donation behavior, initially studied by Titmuss (1971) is above all motivated by a sense of altruism, with a (generally) unanticipated crowding-out effect of monetary incentives (Mellström and Johannesson, 2008): people could give less when they are remunerated for blood donation. In other words, offering a monetary reward may deter users from donating for altruistic means. Thus the purpose of this study is to investigate the willingness to provide digital geospatial GPS data from Smartphones when users are paid versus not paid. We also sought to study the reality of a loss aversion phenomenon in the data donation domain by examining the use of a negative versus positive valence in the framing of the message.

Specifically, we aimed to evaluate people’s willingness to provide their geospatial global positioning system (GPS) data from their Smartphone during the COVID-19 pandemic based on different methods of framing and incentives. In this randomized experimental design, we test the suitability of various messages that could increase smartphone users’ willingness to provide their personal information on human mobility.

## Methods

Participants were recruited via Amazon Mechanical Turk (MTurk), an online crowdsourcing platform that is one of the suites of Amazon.com Web Services. In recent years MTurk had been extensively used in social science research. MTurk enables researchers to recruit participants to perform tasks such as filling out surveys, opinion polls, cognitive psychological studies, and other research services. MTurk rules state that participants can terminate the study by returning the task at any time, without any penalty. Participants on MTurk have a unique Worker ID, which is a semi-random alphanumeric string. Participants’ Worker IDs is associated with the study results making participant anonymous, as no identifying information including their names or address can be collected. Additionally, MTurk has several mechanisms in place to protect unauthorized access including protecting the security of information during transmission by using Secure Socket Layer (SSL) software to keep users’ privacy protected.

Users on MTurk have presented a list of potential tasks or Human Intelligence Jobs (referred to as HITs) when they log into their MTurk account. Our research study was listed on MTurk as a HIT and potential participants were proposed a compensation ($0.05—standard compensation to complete a HIT) to complete the HIT questionnaire (5 min). Once users clicked on the HIT they were directed to the online consent which provided further information about the study. If the user agreed to the consent they were directed to complete the study that was hosted on the Qualtrics platform. The first portion of the study questionnaire included questions on demographics, COVID-19 testing history, and whether they know anyone who has tested positive. Inclusion criteria included 18 years or older and owning a smartphone and what type of operating system they used (Android-Google, IOS-Apple). Users were then randomly assigned to equal arms whereby they were provided messages that related to (1) self-interest; (2) pro-social; (3) monetary motivations for contributing their GPS data from their smartphone to understand the COVID-19 pandemic. Within each arm, participants received either a message framed with either positive or negative valence (creating six questions or “groups” in total) and were asked if they would be willing to contribute their GPS data from their smartphone. • Arm 1, self-interest (+valence): We will provide you feedback on how to navigate your daily schedule in a safe way with COVID-19. • Arm 2, self-interest (−valence): We will provide you feedback on if you have been in contact with someone who has tested positive for COVID-19. • Arm 3, pro-social (+valence): It will help us identify how to re-open your community safely • Arm 4, pro-social (−valence): It will help us identify hotspots that need to be sheltered in place in your community. • Arm 5, monetary (+valence): You will receive a$5 bonus payment if you give your GPS data.

• Arm 6, monetary incentive (−valence): You will not receive a $5 bonus payment if you renounce giving your GPS data. The monetary incentive arm offered a$5 payment for their willingness to provide their GPS data. Participants in this arm that indicated that they were willing to provide their data received a $5 bonus payment, on top of the base$0.05 HIT payment. Assignment to the monetary arm was completely random and equally likely for all participants. The test of the crowding-out effect due to monetary incentives stands in the comparison between arms 5&6 with arms 1–4. The test of the positive vs. negative valence framing effect stands in the comparison of arms 1, 3, and 5 to arms 2, 4, and 6, respectively.

Individuals in all groups that mark that they were not willing to contribute their GPS data received a Vickrey auction. A Vickrey auction or sealed-bid second-price auction (SPSBA) is a type of auction where a “bidder” submits a bid without knowing the bid of the other users in the auction. This auction method has been shown to elicit more truthful values for data provision of personal internet data (Gefen et al., 2020). Thus, in our study, users were asked if they would like to place a “bid” of a selected monetary value of their choosing in exchange for their GPS data. In our study, users were blinded to other users’ bids as well as the maximum threshold bid that was deemed a payout. Here is the sentence they were confronted to

## Vickrey auction

Participants who refused to provide their GPS data after the first request was followed up with a Vickrey auction whereby users were presented with an option to place a monetary bid to be paid for their GPS data, as in Gefen et al. (2020). To analyze the acceptance of this Vickrey Auction (see flow chart (Fig. 2) for a recap of the process), we ran a second step regression on the acceptance of the Vickrey auction process. The results show a significant negative association between the monetary condition (arms 5 + 6) to provide GPS data from their smartphone by using the auction procedure. Thus, among participants who initially refuse to share their GPS data, the type of exposure to the first request influences their decision to bid a price for their GPS data. In addition, Fig. 3 shows the distribution of the monetary bid by the participants who initially (first proposal) refused to provide their GPS data. Those who had received the monetary condition as a first proposal, were less willing to give their data for a bidding price in the Vickrey auction (Table 3).

After refusing to provide their GPS data and being provided with the Vickrey auction to receive monetary compensation in exchange for their GPS data, the average compensation (monetary value) for they will exchange their GPS data in the self-interest experimental condition (arms 1 + 2) is on average $17 and for the pro-social experimental condition (arms 3 + 4) it is$21. If participants received the monetary incentive experimental condition (arms 5 + 6), the average compensation (monetary value) they will exchange for their GPS data $19, however, the latter is non-significantly different from 0 (Table 4). The distribution of the monetary value of the GPS data of participants who are confronted with the Vickrey auction procedure (bid price) is presented in Fig. 3. ## Discussion This study was made to assess persons’ willingness to provide their GPS data stored on their smartphones and to better understand how to frame messages that will encourage the provision of such human mobility data. Two specific questions were addressed: (i) the effectiveness of monetary incentives, compared to other motivations and (ii) the effectiveness of a negative versus positive valence in the framing of the message. Based on our logistic model, our result suggests that framing the use of users’ GPS data as a positive or negative incentive was not significant. A simple positive or negative framing without additional information does not significantly change the approbation rates for giving access to GPS data. In fact, a strong source of privacy behavior arises from incomplete and asymmetric information (Acquisti et al., 2015). It appears that the positive vs. negative valence framing is not enough to push away people’s concern about providing private data on their mobility patterns. When designing the experimental condition into different types of motivational messages to provide private data, self-interest motivation, pro-social benefit, or monetary benefit, we found that monetizing access to GPS data increased the proportion of participants willing to share their human mobility data. The perspective of a small$5 monetary compensation seems first to generate a higher proportion of acceptance rate (64%). These results parallel a previous study conducted by Munzert et al. (2021) whereby the addition of a small monetary value up to $5 significantly increased the likelihood of users downloading a contact tracing application to fight COVID-19. However, this study did not offer a monetary incentive in the first proposal whereas our study provided a secondary bid proposal, in the form of a Vickrey auction, allowing users to have an additional opportunity to bid for a greater monetary value for their private smartphone data. In our study, offering to indicate a monetary value for their GPS data after the first framing proposal, we found that proposing a payment through an auction mechanism could be also fruitful, with a significant additional increase in acceptance rates, varying between 23% and 7%, depending on the previous framing (self-interest, pro-social), with the monetary arms (5 and 6) having the lower additional acceptance rates at the step of the Vickey option. Munzert et al. (2021) did not evaluate this secondary effect of two monetary incentives and our study gives evidence that individuals who rejected the first monetary incentive are less likely to give up their private data. These findings show that some users estimate their mobility data as “priceless”, as they reject both the first and the second financial incentives; making it clear that their data is not for sale at any price. When using the result of the elicitation of the monetary value obtained from the bid, the average monetary valuation given by participants was US$17 when they are told their human mobility data will be used for a self-interest purpose. This value is similar to the value given by participants for public use of their internet health data (Gefen et al., 2020). The valuation of participants when the motivational message was the use of their GPS data for a collective purpose was US$21. This valuation, although a little bit higher, is not significantly different from the self-interest condition. At the end of the survey (taking into account that spontaneous data donation exists, for free), we evaluate the average cost of acceptance, per sharing accepted, US$3 for the self-interest framing to US$4 for the pro-social benefit framing, and US$6 for the monetary incentive framing. This means that, in return for a small compensation to users, it is possible for a service provider to obtain the needed GPS data. In addition, the significant difference between these average costs (p-value = 0.0104) suggests that an efficient approach to encourage smartphone users to provide their historical location data is to start with a self-interest incentive framing for the request of private smartphone data, and, for those who remain resistant, that a follow-up of a small monetary incentive will be effective. However, certain individuals, who initially rejected to provide their data, may still reject both the first framing messages and a secondary monetary incentive.

From a theoretical point of view, the findings of the study tend to invalidate the preconception of a crowding-out effect, at least for the specific domain of data donation behavior. On the contrary, the financial motivation appears to be “crowding-in”, since the arms (5 and 6), with a monetary reward, are tested much more fruitful in terms of average adhesion to personal data access. We rather validate previous research on monetary incentives for the adoption of contact-tracing apps (Munzert et al., 2021) and the general statement that crowding theory can be either “in” or “out”, depending on the precise context (Frey et al., 2012; Gneezy et al., 2011). Our results build on these previous researches as we demonstrate that there is no crowding-out effect between intrinsic and extrinsic motivations for the provision of private smartphone data. Methods of framing and of incentivization can be used in conjunction to influence the user’s willingness to provide their private human mobility data from their smartphone. No loss for the public authority was found when a mixed strategy was selected. On the contrary, this dual strategy had the best cost-effectiveness ratio to incentivize data provision. Thus the study’s findings have the ability to contribute to lowering the social cost of epidemic control by informing evidence-based policymaking efficiently (Vuong, 2018).

Lastly, our results indicate that participants who knew their COVID-19 serology status are more willing to share their GPS data than those who have not been tested. The same was seen for participants who knew someone who had COVID-19. This may indicate that an “empathic” response was taking place whereby if participants have a personal health experience with the disease of study, they are more likely to exhibit altruistic or pro-social behaviors which have also been seen in previous data donation studies (Gefen et al., 2020; Vuong and Napier, 2015; Papachristou and Whitcomb, 2004). A theoretical explanation could come from the “mindsponge mechanism” which elucidates how and why an individual observes and ejects cultural values depending on the external setting (Vuong and Napier, 2015); in our study, this is shown through having been tested for COVID-19 or knowing someone positive. Our results also suggest that participants living in Brazil or India are more willing to provide their mobile phone GPS data compare to those living in the US. The latter result may reflect a local context behavior (Eggo et al., 2021). Indeed, during the period of study, those two countries (Brazil and India) were some of the countries with the highest prevalence rate of COVID-19, again perhaps stimulating a pro-social mindset (Vuong and Napier, 2015) and, then, eliciting an empathetic response. This may imply that greater experience with COVID-19 either through testing or through vicarious (indirect) community experiences moderated participants’ fears of privacy and increased their altruistic behavior toward public safety (Pomery et al. 2009). Finally, we found that the type of operating system of the user’s smartphone may play a role in the willingness to provide location data. Users of the IOS (Apple) operating system are less willing to share their mobile phone GPS data compare to users of the Android (Google) operating system which may indicate a selection bias that those who opt to use an IOS system have greater privacy concerns than the others.

This paper has some limitations that should be recognized. One of the first is the conceivable sensitivity of our results to the population surveyed. Recruitment on Amazon Mechanical Turk permits a large and costless sample but may create biases. Tests on the reliability, of behavioral research, of data collected through the Amazon-Turk platform, are reassuring (Crump et al., 2013; Arechar et al., 2018). However, they may be insufficient to be sure that in part the response we have in our experiment was not biased by the profile of the Amazon-Turk users, who may be more likely to accept data transfers (in other words, the sample members of this platform are self-selected and might be in general more willing to let access to their data than the general population). Note that, during the COVID-19 pandemic, this method of data collection was a necessary measure, as lockdowns prohibited in-person data collection. The RCT design was also assuring that the observed differences between arms were not biased in relative terms (measurement of the effect), although the absolute rates could be -as mentioned, due to the specific population recruited on platforms. Another limitation may lie in the way we introduced and studied the loss-aversion effect. Generally, this has to be associated with a first “endowment”, which, in turn, allows to put people in a lost context. This was not possible for arms 1–4 (self-interest/pro-social), so for a reason of parallelism, we did not do it for the monetary arms (5 and 6).

## Conclusion

Country of origin and COVID-19 testing status influences the behavioral response to sharing private GPS smartphone data. Self-interest and pro-social motivations for data donation are sufficient to encourage donation if the target is around a 50% acceptance rate. However, supplementing prior intrinsic motivations with monetary incentives regarding the provision of private human mobility mobile data has an additive effect on influencing users’ decisions, which could help to reach the 60% acceptance rate level needed to fight the epidemic efficiently. Communications that promote altruistic donation with the addition of financial compensation will encourage the most active participation of users to provide private location data.