Individuals' willingness to provide geospatial global positioning system (GPS) data from their smartphone during the COVID-19 pandemic

The COVID-19 pandemic underscored the critical need for digital epidemiology leveraging public health data, creating a tension between public health and individual privacy. A major challenge was acquiring sufficient public digital data, particularly concerning mobile phone location data. This study focuses on understanding the motivational mechanisms driving data donation, specifically comparing the influence of monetary versus altruistic motivations. Geospatial GPS data from smartphones offers valuable insights into human mobility patterns, crucial for understanding COVID-19 transmission and evaluating public health interventions. However, significant privacy concerns hinder data acquisition. Data altruism—donating data for the common good—could be a key motivator, but its effectiveness may be limited without additional incentives. Behavioral economic framing suggests that the message format influences data donation decisions. While monetary rewards can incentivize data provision, the "crowding-out effect" suggests that extrinsic incentives might undermine intrinsic and altruistic motivations. This study aims to investigate the willingness of smartphone users to provide GPS data under different incentive and framing conditions, exploring the existence of a crowding-out effect and the impact of positive versus negative framing.

Existing research highlights the challenges of acquiring personal data due to privacy concerns (Acquisti et al., 2016; Posner, 1981). The study draws on self-determination theory and behavioral economics, referencing prior work on crowding-out effects of extrinsic incentives on prosocial behavior (Deci, 1972; Frey et al., 1997; Gneezy and Rustichini, 2000; Mellström and Johannesson, 2008). The literature also demonstrates the impact of framing on data donation decisions (Oullier et al., 2010) and the effectiveness of monetary rewards in influencing data sharing (Gefen et al., 2020). The use of a Vickrey auction (sealed-bid second-price auction) as a truth-revealing mechanism for data valuation is also supported by previous research (Gefen et al., 2020).

Participants (N=1055) were recruited via Amazon Mechanical Turk (MTurk) from 41 countries. Inclusion criteria were age 18+, smartphone ownership, and specification of operating system (Android or iOS). Participants received a standard MTurk compensation ($0.05) plus additional payments depending on their experimental condition. Participants were randomly assigned to one of six experimental groups based on three motivational frames (self-interest, pro-social, monetary) and two valence types (positive, negative). The messages were as follows: * **Arm 1, self-interest (+valence):** Safe navigation guidance during COVID-19. * **Arm 2, self-interest (-valence):** Notification of potential COVID-19 contact. * **Arm 3, pro-social (+valence):** Data helps identify safe community reopening strategies. * **Arm 4, pro-social (-valence):** Data helps identify and isolate COVID-19 hotspots. * **Arm 5, monetary (+valence):** $5 bonus for providing GPS data. * **Arm 6, monetary (-valence):** Loss of $5 bonus if GPS data not provided. The monetary incentive arm offered a $5 payment. Those unwilling to provide data in the initial phase were offered a Vickrey auction, allowing them to bid for compensation. Data analysis used logistic regression and a probit model to examine the influence of different motivational messages, framing, and demographics on willingness to provide GPS data and monetary bid values in the Vickrey auction.

The study involved 1055 participants from 41 countries, with a mean age of 34. Participants from India and Brazil were more willing to share data than those from the US. There was no significant difference in data provision based on positive versus negative valence framing in the initial message. However, a $5 monetary incentive significantly increased willingness to provide data (approximately 64% acceptance rate). In the Vickrey auction, the average accepted bid was $17 for the self-interest condition and $21 for the pro-social condition. Those initially exposed to the monetary condition were less likely to participate in the auction. The results indicated no crowding-out effect of monetary incentives; instead, monetary incentives appeared to complement intrinsic motivations. Significant associations were found between the type of mobile operating system and the decision to provide GPS data. Participants using an IOS operating system were significantly less willing to provide data. Additionally, COVID-19 testing status significantly influenced participation. Participants who had been tested for COVID-19, regardless of result, were significantly more willing to provide data compared to those who had not been tested, with those testing positive showing the strongest willingness.

The findings challenge the assumption of a crowding-out effect in this context. Monetary incentives significantly increased data provision without diminishing intrinsic motivations (self-interest or pro-social). The lack of significant impact from positive versus negative framing suggests that simply emphasizing positive or negative consequences is not sufficient to overcome privacy concerns. The Vickrey auction proved effective in eliciting further data provision, especially among those initially unwilling to participate. The higher average bid for pro-social framing versus self-interest framing suggests a willingness to value data more highly when it benefits the collective good. The results also suggest that individual experiences with COVID-19 and contextual factors (country of origin, operating system) significantly impact data sharing behavior.

This study demonstrates that monetary incentives, particularly when combined with effective framing, can significantly enhance the acquisition of private smartphone GPS data for public health purposes. The absence of a crowding-out effect suggests that combining intrinsic and extrinsic motivations is a viable strategy. Future research could explore the long-term effects of such incentives and examine the role of cultural and technological factors in shaping data donation decisions.

The study's reliance on MTurk participants may introduce sampling bias, as this population may be more inclined to share data than the general public. The limited exploration of loss aversion and the lack of a true endowment effect in the framing could be further investigated. The study's findings may not be generalizable to all populations.