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

The study addresses the tension between public health needs and personal privacy during the COVID-19 pandemic, focusing on the donation of smartphone GPS data for digital epidemiology (e.g., mobility analysis, contact tracing). It investigates how different motivations—self-interest, pro-social benefits, and monetary incentives—affect individuals’ willingness to share private location data, and whether message valence (positive vs negative framing) matters. Grounded in self-determination theory and behavioral economics, the authors specifically test for a potential crowding-out effect where extrinsic (financial) incentives might undermine intrinsic (altruistic) motivations. The research question is whether framing and incentivization strategies can effectively increase voluntary GPS data sharing without deterring altruistic motives, thereby informing public health data collection strategies during pandemics.

The paper situates its work within literature on digital epidemiology and the utility of smartphone-derived mobility data for tracking disease transmission and evaluating interventions. It reviews economic and behavioral research on privacy concerns and the conditions under which individuals disclose personal data (Acquisti et al., Posner), as well as framing effects on pro-social behavior. Central is the crowding-out literature from self-determination theory and behavioral economics (Deci; Frey; Gneezy & Rustichini), with evidence from health contexts like blood donation where monetary rewards sometimes reduce altruistic contributions (Mellström & Johannesson). Contrasting findings show contexts where incentives can increase desired behaviors (e.g., adoption of COVID-19 contact-tracing apps). The authors also reference loss aversion and the potential differential impact of positive versus negative framing, and introduce the Vickrey auction as a truth-revealing mechanism for valuing personal data (Gefen et al.).

Design: Randomized experimental study conducted on Amazon Mechanical Turk (MTurk). Inclusion criteria: age ≥18, ownership of a smartphone (Android or iOS). After consent and baseline questions (demographics, COVID-19 testing status, acquaintance with COVID-19 cases, phone OS), participants were randomized in equal proportions to one of six message frames combining three motives and two valences: self-interest (+/−), pro-social (+/−), monetary (+/−). Sample: 1055 participants from 41 countries (mean age 34), with reported distributions across the US, India, Europe, Brazil, and other countries. Smartphone ownership 96.4%; Android 73.0%, iOS 27.0%. Arms and messages:

• Self-interest (+): feedback on safe navigation of daily schedule with COVID-19.
• Self-interest (−): feedback on contact with someone who tested positive.
• Pro-social (+): help identify how to re-open the community safely.
• Pro-social (−): help identify community hotspots needing shelter-in-place.
• Monetary (+): receive a $5 bonus if GPS data is given.
• Monetary (−): will not receive a $5 bonus if renouncing giving GPS data. Participants in the monetary arms were offered a $5 bonus for sharing GPS data in addition to the base $0.05 MTurk payment. Randomization balance was verified (chi-squared test p=0.4857). Follow-up Vickrey auction: Participants who initially refused were offered a sealed-bid second-price auction to state a monetary bid they would accept for their data. Instructions noted payment to the lowest 100 bids under a hidden threshold; a $5 anchor was mentioned for all. Bids above $10 were not accepted. This mechanism aimed to elicit truthful valuation of GPS data. Outcomes and analysis: Primary outcome was willingness to share GPS data after the initial frame; secondary outcomes included acceptance of the Vickrey auction and the monetary valuation (accepted bids). Probit/logistic models estimated determinants of willingness, including arm, valence, and controls (gender, age, phone OS, knowing someone with COVID-19, testing status, location). A two-step selection model tested for bias in the auction analysis (Inverse Mills ratio nonsignificant). Analyses were performed in R 4.1.1.

• Randomization: Arms were balanced (X^2 test p=0.4857).
• Overall willingness: 55.95% (566/1017) accepted sharing GPS data after the initial message.
• By condition: Approximately half accepted under self-interest and pro-social frames; nearly two-thirds were willing under the $5 monetary incentive (as summarized in the abstract). Aggregated acceptance shares among the total sample: self-interest 16.42% (7.17% arm 1; 9.24% arm 2), pro-social 17.60% (8.45% arm 3; 9.14% arm 4), monetary 21.93% (10.22% arm 5; 11.70% arm 6).
• Valence effect: No significant difference between positive and negative framing on willingness to share (Table 2; both aggregated and arm-by-arm tests nonsignificant).
• Monetary incentive effect: The monetary condition (arms 5+6) was significantly associated with increased willingness to provide GPS data (Table 2; coefficients for monetary vs self-interest significant at p<0.01 in models 2 and 4).
• Demographic/contextual determinants (Table 2): iOS users were significantly less willing than Android users (coef ≈ −0.24, p<0.05). Participants in India and Brazil were more willing than those in the US (Brazil and India coefficients positive and significant; Europe/Other not significant). Having been tested for COVID-19 (positive or negative) increased willingness; being COVID-19 positive showed a strong positive association relative to negative/untested.
• Vickrey auction: Among initial refusers, 16.96% (76) accepted to participate in the bidding process. Exposure to a monetary first request (arms 5+6) significantly reduced subsequent acceptance of the auction (Table 3: coef ≈ −0.57 to −0.65, p<0.01), suggesting those who declined the initial monetary offer were less likely to engage in follow-up compensation-based sharing.
• Monetary valuation (Table 4): Average accepted bids among follow-up participants were approximately $17 (self-interest), $21 (pro-social), and $19 (monetary), with variability; the monetary-arm estimate was not significantly different from zero given high uncertainty. Distribution of bids shown in Fig. 3.
• Cost-effectiveness: Estimated average cost per accepted sharing was ~$3 (self-interest framing), ~$4 (pro-social framing), and ~$6 (monetary incentive framing), with a significant difference between these averages (p=0.0104).

Findings indicate that simple positive versus negative message valence does not meaningfully alter willingness to share private GPS data, suggesting that privacy concerns driven by incomplete/asymmetric information are not easily shifted by valence alone. In contrast, offering a modest ($5) monetary incentive significantly increases willingness, without evidence of crowding-out intrinsic altruistic motivations; rather, incentives appear to have a crowding-in effect in this context. A sequential strategy—starting with intrinsic frames (self-interest or pro-social) and following up with monetary incentives for those still reluctant—improves acceptance and yields favorable cost-effectiveness. Behavioral responses also varied by personal experience and context: participants tested for COVID-19 or knowing someone with COVID-19 showed higher willingness, consistent with empathic or prosocial activation mechanisms (e.g., mindsponge), and participants from Brazil and India (high-prevalence contexts during the study period) were more willing than those in the US. iOS users were less willing than Android users, potentially reflecting differing privacy preferences. Collectively, the results address the initial research questions by showing that targeted framing combined with moderate financial incentives can increase data provision for public health without diminishing altruistic motives.

Country of origin and COVID-19 testing status influence behavioral responses to sharing private GPS data. Self-interest and pro-social motivations can achieve about a 50% acceptance rate, while supplementing with monetary incentives has an additive effect that can help reach the ~60% acceptance level considered useful for epidemic control. Communications that promote altruistic donation with added financial compensation encourage greater participation in providing private location data. The study finds no evidence of a crowding-out effect between intrinsic and extrinsic motivations in this data donation context, supporting the use of combined framing and incentives to enhance public health data acquisition.

• Sample and generalizability: Recruitment via MTurk enables rapid, large-scale data collection but may introduce selection bias; MTurk users could be more willing to share data than the general population. Pandemic constraints precluded in-person recruitment.
• Relative vs absolute effects: The RCT design supports unbiased comparisons between arms (relative effects), but absolute acceptance rates may be influenced by the platform’s user profile.
• Loss-aversion framing: A classical loss-aversion setup typically requires an initial endowment; this was not feasible for arms 1–4, and for parallelism was not used in arms 5–6, potentially limiting the assessment of loss aversion.
• Auction participation: Only refusers were offered the Vickrey auction; although a two-step selection model found no significant selection bias (lambda p=0.21), the follow-up sample size for valuation estimates was modest.
• Measurement constraints: Willingness measures and bids were stated preferences within an online experiment; actual behavior in real-world data-sharing contexts may differ.