Sage Journals: Discover world-class research

Abstract

Smartphone applications routinely collect precise location data from users by offering free services and this information is often monetized for advertising and marketing purposes. While many companies limit sales to aggregate behaviors, data aggregators and data brokers (DA&DB) provide access to individual-level location data. Some DA&DB have implemented privacy-preserving rules and the Federal Trade Commission (FTC) has also intensified its regulations on location-data practices. This paper presents an in-depth exploration of U.S. privacy perceptions regarding specific location features that can be derived from data made available by DA&DB. These findings can provide policy-relevant insights that could assist organizations such as the FTC in defining clearer access rules. Using a factorial vignette survey, we collected responses from 1,405 participants to assess comfort levels with sharing various types of location features, including individual trajectories and visits to points of interest, which are currently available for purchase from DA&DB worldwide. Our results show that trajectory-related features elicit higher privacy concerns, that certain data broker based obfuscation practices increase comfort, and that race, ethnicity and education have an effect on data sharing privacy perceptions.

Keywords

Location data privacy perceptions data brokers online survey ordinal regression analysis

Introduction

Location (GPS) data collection has become widespread, with smartphones routinely collecting precise location data from users in exchange for free services. App developers use this data to understand app usage, improve functionalities and most importantly, to deliver personalized advertisements that drives monetization.

Monetization is primarily driven by two distinct models. Large technology companies like Google,¹ Apple, and Meta collect precise location data for internal purposes such as improving app functionality and sell only aggregate behaviors insights (e.g. geographical areas visited) for advertising, without sharing individual-level location data. In contrast, data aggregators and data brokers (DA&DB) collect, repackage, and sell individual level location data to third parties, including private organizations and academic researchers (Chen and Poorthuis, 2021; Hunter et al., 2021; Sila-Nowicka et al., 2016; Wang and Hu, 2024). DA&DB either develop their own applications, or distribute software development kits (SDKs) to app developers to collect location data from specific apps. Major technology companies have faced lawsuits for tracking user’ location without consent.^2,³ The Federal Trade Commission (FTC) has also penalized two DA&DB in 2024 for selling location data to advertisers without adequately informing consumers or obtaining consent (Record, 2024a). Unlike tech companies which do not share individual level location data with third parties, DA&DB do and this has enabled several instances of inappropriate uses, including: the US Defense Department’s access to a prayer app’s location data to monitor Muslim communities (Center, 2024), local law enforcement agencies tracking racial-justice protesters using location apps (Center, 2024), identifying gay priests (Morrison, 2021) and individuals who visit abortion clinics (Cox, 2022).

In response to these concerns, some DA&DB have introduced internal policies to limit disclosure of sensitive information about individuals. For example, DA&DBs such as Cuebiq prohibit the identification of visits to healthcare facilities on their platforms, or only limit identification of home locations at the census tract level.⁴ However, these internal guidelines are solely based on DA&DB’s decisions rather than on the perceptions of users whose location data is being monetized. We put forward that true consumer privacy rules need to come directly from consumer privacy perceptions.

In this paper, we evaluate user comfort with the use of individual level location data—acquired from DA&DB—to identify and characterize personal trips and visits to places. Previous studies have used survey approaches to explore people’s perceptions of the collection and use of individual level location data, primarily focused on points of interest (POI) visited (Gamarra et al., 2019; Gilbert et al., 2023; Martin and Nissenbaum, 2019). For example, previous research has shown that users are more comfortable in sharing their work or home location than their location when they attend a rally or visit a medical facility (Gilbert et al., 2023).

However, points of interest are only one of the category of features available from DA&DB. In fact, travel history of millions of individual devices are available from data aggregators.⁵ These data enables derivation of trajectory-based features such as origin-destination trips (e.g. trips from home to work) or modes of transport inference (e.g. identifying driving vs. use of public transit vs. walks). We extend the state of the art in individual level location data sharing perceptions by evaluating user perceptions of both trajectory-based features and points of interest (POIs) visits, using a U.S.-representative survey.

Factorial vignette approach. Past work (Martin and Nissenbaum, 2019; Vitak et al., 2022) shows that the level of comfort with location data sharing varies depending on who accesses the data and for what purpose. We draw on Nissembaum’s (Nissenbaum, 2004) contextual integrity (CI) framework to design survey questions that ask participants to evaluate their comfort with specific feature being analyzed by a particular actors and for particular purposes. Following Martin and Nissenbaum (2019), we use a factorial-vignette survey design to randomly generate plausible combinations of actors, purposes and features, exposing survey participants to different ways in which their location data could be used.

Understanding these nuanced privacy perceptions can inform the creation of policies governing the types of location features that can be extracted from location data and the features that should be restricted due to low comfort levels, and which obfuscation practices effectively increase users’ level of comfort.

The FTC has started to regulate consumer privacy providing “bright-line rules” for companies to clarify what can and cannot be done with location data (Record, 2024b). This paper contributes to the effort by offering empirical guidance for potential future FTC rulemaking. The main contributions and findings of this paper are:

An evaluation of user comfort with respect to the use of trajectory data and visits to points of interest taking into account actors and purposes. Participants low comfort with movement (trajectory-based) features such as frequent travel paths. Comfort increased when the data was used for social-benefit purposes but decreased when data was accessed by actors such as government agencies or employers.

An analysis on how obfuscation approaches affect levels of comfort. Obfuscation significantly increased levels of comfort, especially for detailed trajectory data but also for points of interests, such as home or work locations.

An analysis of how levels of comfort vary with educational background, race and ethnicity. Our analysis show that there are significant differences in comfort between racial and ethnic groups, with Hispanics having the highest levels of comfort (even after controlling for education levels). Highly educated individuals report significantly higher comfort with obfuscated features, while those with lower education levels show no significant effect. To the best of our knowledge, this is the first study to evaluate user privacy perceptions across racial and educational groups with respect to trajectory data, taking into account current data aggregator obfuscation practices.

Literature review

Location data tracking

Location data has been widely used for pandemic modeling (Akinbi et al., 2021; Organizers et al., 2019), inferring socio-economic indicators (Hong et al., 2016; Yang et al., 2023a), assessing health outcomes (Garcia-Bulle et al., 2022), supporting urban development and disaster risk (Fraser et al., 2024; Frias-Martinez et al., 2012; Wang and Hu, 2024; Wu et al., 2022), enabling targeted marketing and advertising (Bauer and Strauss, 2016), understanding migration patterns (Hong et al., 2019), or improving public transit systems (Ma and Wang, 2014).

Obfuscation techniques

The archetypes for location privacy-preserving mechanisms fall under Anonymization, Dummy data, or a combination of both. Anonymization-based methods report coarser information to reduce geo-distinguishability (Andrés et al., 2013; Gao et al., 2022; Gramaglia et al., 2021; Shen et al., 2023), whereas dummy methods attempt to simulate realistic location data (Du et al., 2019; Shankar et al., 2009).

In this paper, we focus on anonymization techniques commonly used by DA&DB. For example, instead of computing an exact home location, some providers report home area as a census tract (SafeGraph SafeGraph, 2025); or rather than sharing exact route trajectories, data points are changed to obfuscate movement patterns towards sensitive locations and home/work location (Spectus, 2025).

Privacy perceptions of location data

Our research builds on previous work examining privacy perceptions related to Gamarra et al. (2019); Martin and Nissenbaum (2019); Vitak et al. (2022): (1) Actors, defined as entities accessing the location data, for example, commercial entities or federal agencies (Gilbert et al., 2023; Martin and Nissenbaum, 2019; Vitak et al., 2022; Vitak and Zimmer, 2023) (2) Purposes, referring to how the location data will be used, such as advertising or public health applications (Gilbert et al., 2023; Gorra, 2007; Martin and Nissenbaum, 2019) (3) Time, or duration of the access to location data (Gorra, 2007; Martin and Nissenbaum, 2019) (4) Age (Chakraborty et al., 2013; Gamarra et al., 2019; Haffner et al., 2018) (5) Privacy attitudes, including general disposition towards data sharing (Bansal et al., 2016; Junglas and Spitzmuller, 2006), and (6) Socio-cultural aspects, such as norms and contextual expectations (Gorra, 2007; Ioannou and Tussyadiah, 2021; Wang and Lin, 2017; Zhang et al., 2020).

Survey design

We used a factorial vignette survey approach following Nissembaum’s contextual integrity framework. Each vignette in the survey follows the format: Actor X wants to do Purpose Y and for that, they need to use Feature Z. We systematically varied actors, features and purposes across participants to test their influence on levels of comfort. For example, a vignette might read ”A doctor wants to monitor your personal wellness. For that purpose they want to access your detailed walking activity. How comfortable would you feel with this use of your personal location data?”. Levels of comfort were measured using a 5-point Likert scale, ranging from Very Uncomfortable to Very Comfortable. Participants were also asked to fill out a free-form text box explaining their answer (see Figure 1 and Appendix Figure 5 for two survey question examples). Next, we describe location features, actors, and purposes in detail.

Location features. We advance the state-of-the-art in the evaluation of location-data privacy perceptions along two main dimensions. First, we consider location data features beyond commonly focused ones such as points of interest (POI) (visiting a hospital or a liquor store). Specifically, we evaluate trajectory-based features that are extracted from spatio-temporal trajectory datasets. For example, participants were showed specific routes on a map and asked about their level of comfort with a company using their detailed driving trips or their detailed walking activity.

Figure 1.

Example of vignette question (Actor: Commercial entity, Purpose: Personal wellness and physical activity, Feature: Walking activity). The feature is obfuscating and only shows general statistics.

Second, we evaluate trajectory-based and POI visit features in two forms: detailed and obfuscated. Detailed features use all available GPS points whereas obfuscated features characterize location data using current state-of-the-art data aggregator practices. For example, an obfuscated driving trip only specifies the origin and destination census tract of the trip rather than the full GPS trajectory. Similarly, an obfuscated home location would only report home location at the county level instead of the exact location.

Table 1 lists all features used to construct the survey vignettes. These features can be categorized into six groups: Home+Work, Places Visited, Transportation, Movement, Walking Activity, and International Trips. For each feature in each category, we define both its Detailed (D) and its Obfuscated (Ob) version. Both the features and obfuscation approaches are inspired by real-world uses of location data by academia, data aggregators, and location-intelligence companies. Further information is provided in the Appendix, section Features Definitions and Obfuscation. Next to each feature in the Table, we provide links to papers that compute these features using location data acquired from DA&DB, showing that these are in fact state-of-the-art features when working with datasets from DA&DB.

Table 1.

Location features and abbreviations.

Feature Cluster	Feature	Abbreviation	Description
Home + Work (Chen and Poorthuis, 2021;	Your inferred home location	Home location (Detailed)	Home location as a point on map (e.g. Appendix Figure 7)
Sila-Nowicka et al., 2016)	Your inferred home location represented as a census tract	Home location (Obfuscated)	Home’s county location (e.g. Appendix Figure 8)
	Your inferred work location	Work location (Detailed)	Work location as a point on map (e.g. Appendix Figure 9)
	Your inferred work location represented as a census tract	Work location (Obfuscated)	Work’s county location (e.g. Appendix Figure 10)
Places Visited (Yang et al., 2023b; Sila-Nowicka et al., 2016; Coleman et al., 2023)	The places you visit	Places you visit (Detailed)	Chart indicating types and frequency of places visited. Includes a map depicting the detailed locations of the places visited (e.g. Appendix Figures 11, 12)
	The types of places you visit	Places you visit (Obfuscated)	Chart indicating types and frequency of places visited (e.g. Appendix Figure 12)
	The geographical area where you spend most of your time	Area you spent most of your time (Obfuscated)	Map displaying radius of gyration (e.g. AppendixFigure 13)
Transportation (Adler et al., 2017; Chuang et al., 2023; Sila-Nowicka et al., 2016)	The modes of transportation you use, with what frequency and their corresponding routes	Modes of transportation (Detailed)	Chart indicating frequency and types of mode of transport used. It also includes a map with lines indicating detailed routes for each frequent mode of transport (e.g. AppendixFigure 14,15)
	The modes of transportation you use and with what frequency	Modes of transportation (Obfuscated)	Chart indicating frequency and types of mode of transport used (e.g. Appendix Figure 15)
Movement (Adler et al., 2017; Coleman et al., 2023; Li et al., 2020)	Your most frequent trips	Most frequent trips (Detailed)	Map with frequently taken routes. Routes are detailed showing start and end locations as well as the in-between GPS points visited (e.g. Figure 2)
	Your least frequent trips	Least frequent trips (Detailed)	Map with infrequently taken routes. Routes show start and end locations as well as all the GPS points visited (e.g. Appendix Figure 17)
	Your most frequent types of trips	Most frequent type of trips (Detailed)	Map with the different frequent trips taken and the inferred trip purpose by type of destination (e.g. Appendix Figure 18)
	Your most frequent trips represented by their origin and destination census tracts and connected by a line	Most frequent trips between counties (Obfuscated)	Map with frequently taken routes. It protects privacy by showing the start and end points as counties, and the frequently taken routes as straight lines between counties instead of detailed GPS (e.g. Appendix Figure 19)
	Your most frequent trips represented by their origin and destination census tracts and connected by an approximate route	Most frequent trips between counties (‘‘Google’’) (Obfuscated)	Map with frequently taken routes. It protects privacy by showing the start points, end points as counties, and the frequently taken routes as suggested by Google Maps, instead of the actual GPS route (e.g. Appendix Figure 20)
Walking Activity (Chuang et al., 2023; Chuang and Chen, 2024; Hunter et al., 2021; Kim et al., 2022)	Your walking activity and the corresponding routes	Frequent walking activity (Detailed)	Chart indicating frequency and duration of walks. It also includes a map with detailed GPS trajectories indicating routes for each frequent walking path (e.g. Appendix Figure 21,22 )
	Your walking activity	Frequent walking activity (Ob)	Chart indicating frequency and duration of walks with no detailed trajectories (e.g. Appendix Figure 22)
International Trips (Cuebiq et al., 2025)	The foreign countries you have visited and the duration of the visit, including all the locations where you have stayed	International visits (Detailed)	Chart indicating frequency,duration and location of international trips. It includes a map with detailed GPS points indicating regions visited in the foreign county (e.g. Appendix Figure 23)
	The foreign countries you have visited, and for how long	International visits (Ob)	Chart indicating frequency, duration and location of international trips (e.g. Appendix Figure 24)

We cite examples of recent studies where such features are used in the ‘‘feature cluster’’ column. HTML visualizations for detailed features and image format charts available on OSF Awasthi et al., 2025. Additional details provided to survey participants on each feature are in the appendix section feature background information.

While features such as home, work and places visited have already been explored in previous works, comfort perceptions for obfuscating approaches and for other trajectory-based features have not been explored.

Given the complexity of these features, each vignette includes background information explaining the feature and its typical uses, along with an interactive visualization. See Figure 3 in the Appendix for an example vignette with background information, visualizations and the corresponding question. See Description column in Table 1 for links to sample visualizations shown to survey participants. All individual visualizations are also available in our anonymized Open Science Foundation link visualization folder. More details about survey design choices, including visualizations, are explained at the end of this section.

Actors Table 2 lists the nine actors used to construct the vignette questions.

Table 2.

Actors and the abbreviations.

Actor	Abbreviation
Your employer	Employer
A federal government agency—like the FBI or CIA	Federal government agency
A law enforcement agency—like a city police department or a county sheriff’s office	Law enforcement agency
A commercial entity	Commercial Entity
A local government agency	Local government agency
Your doctor	Doctor
Your family	Family
Academic researchers	Academic researchers
An emergency service—like emergency medical services or fire and rescue services	Emergency services

Actors taken from previous studies like (Gorra, 2007; Martin and Nissenbaum, 2019; Vitak et al., 2022).

These actors have been examined in previous studies related to location-data privacy (Martin and Nissenbaum, 2019; Vitak et al., 2022) and represent the major entities that may access an individual’s location information. Our work’s novel contribution lies in examining how these actors influence privacy perceptions in conjunction with novel trajectory-based features, while controlling for demographic and educational factors.

Purposes Table 3 lists the ten different purposes used to construct the vignette questions. These purposes are informed by previous work on privacy perceptions related to location data (Martin and Nissenbaum, 2019; Vitak et al., 2022). They cover wide range of applications, including public-service uses (e.g. identifying where to build new hospitals), public-health uses (e.g. monitoring population mobility during a pandemic), law-enforcement uses (e.g. identifying criminal activity), and economic or commercial uses (e.g. showing targeted ads or optimizing productivity).

Table 3.

Purposes and abbreviations.

Purpose	Abbreviation
Understand your commute to work so as to optimize work productivity	Optimize work productivity
Monitor your mobility patterns, e.g. the places you visit or the trips you make	Monitor mobility patterns
Analyze terrorist attacks by looking into people’s movements and locations visited	Analysis of terrorist attacks
Understand criminal activity by looking into the relationship between crime, people’s movements and locations visited	Analysis of criminal activity
Monitor your personal wellness and physical activity	Personal wellness
Show you targeted ads or personalized announcements	Show Ads
Monitor how people move (or don’t) to control the spread of a disease, e.g.COVID-19	Control spread of diseases
Analyze traveling experiences and public transit services	Analysis of Public transit services
Understand how people move in a city so as to inform the design of new walking and cycling infrastructure	Design new walking/cycling infrastructure
Understand how people move so as to identify optimal locations for hospitals, libraries or parks	Identify locations for infrastructure

Purposes taken from previous studies like (Gorra, 2007; Martin and Nissenbaum, 2019; Vitak et al., 2022).

Some combinations of actor, feature and purpose are implausible. For example, a doctor will not be interested in getting access to mobility patterns at city scale. Therefore, we manually eliminated such implausible combinations from the pool of possible vignette questions. After this process, a total of 445 valid combinations of actor, purpose and feature remain and were randomly shown to participants. Additional details are provided in Appendix section Approach to Selecting Plausible Survey Questions for details. All vignette questions are available in OSF (Awasthi et al., 2025) file allQuestionsForViewing.csv column humanreadableQuestions.

Design choices

Prior work in privacy-risk perception studies has shown three important insights that we build on when creating the survey. First, Gerber et al. (2019) showed that abstract scenarios are often perceived as less risky than personal scenarios. Hence, before starting the survey questions, we ask participants to situate themselves in the vignettes as if this was their own personal data. The questions are also phrased to put the participant at the center of the vignette using the ”you” pronoun (see Appendix section ‘‘Consent and briefing’’ for more details on this design).

Second, previous work has shown that visualizations can improve understanding of privacy risks, and that when privacy-related questions are asked with visualizations, privacy risk concerns with data sharing tend to decrease (Farke et al., 2021). Hence, each of the vignette includes an interactive visualization of the location feature and a brief explanation of the visualization to make sure that the user understand the feature extracted from the location data. Figure 2 illustrates the Detailed Frequent Trips features, where actual trips and destinations are identified. Additional examples of feature visualizations can be found in the Appendix, including transportation-mode visualization(Appendix Figure 3), visits to points of interest (Appendix Figure 11), and obfuscated frequent trips using synthetic trajectories generated with Google Maps (Appendix Figure 20).

Figure 2.

Snapshot of interactive map with most frequent trips feature (FreqT (D)) as visible to participants. For our hypothetical data contributor, the pins denote the start and end points and the line joining the pins denote one of the most frequent routes. For example red pins and line segment denote the most frequent trip between home and work.

Third, previous work has shown that having technical knowledge or specific demographic or personality traits might affect privacy perceptions (Martin and Nissenbaum, 2019). Hence, we also ask participants to fill out questions covering both privacy attitudes, technical knowledge, and demographic data. These questions were divided into two blocks, Demographic and general technology knowledge questions appeared before the vignettes and the Privacy-attitudes questionnaire appeared afterwards.

Following Martin and Nissenbaum (2019), the privacy-attitudes questionnaire asked participants to rate their agreement with different privacy attitude statements on a 5-point Likert scale (from “strongly disagree” to “strongly agree”). These statements captured attitudes such as trust in business or government agencies. Figures 5 and 6 in the Appendix list all the questionnaire items, and Appendix section Privacy Attitudes provides further details on the rationale behind the privacy questionnaire design.

We conducted a qualitative study with five respondents recruited from Craigslist to evaluate vignette comprehension and survey usability. Details of the study procedures and findings are provided in the Appendix section Qualitative Survey Evaluation. Using feedback from participants of the qualitative survey and methodological considerations of effective sample sizes, we set the number of vignette questions to five. Additional details of effective sample size are provided in the Appendix section Ordinal Regression, Design Effect.

Platform

We advertised our online survey on Cint⁶ from March 2023 to March 2024. Interested participants clicked on a link on Cint that redirected them to an external webpage where the survey was hosted. We did not use Cint built-in survey tools because of the elaborate nature of the factorial vignettes that were randomly sampled and the custom map visualizations needed to explain the survey questions. Each participant was paid $7 for completing the survey, which consisted of five vignette questions, seven demographic and computer knowledge questions, and ten privacy-attitude questions. All participants were protected under IRB protocol 1768475-4 (see Appendix section Ethical considerations for further details).

Analytical approach

We analyze users’ levels of comfort with the use of individual-level trajectory and POI visits, accounting for the actors and purposes involved in the data analysis, as well as the presence or absence of obfuscation approaches. We also examine how these perceptions vary with respect the participants’ educational background and racial or ethnic identities.

To conduct this analysis, we use a mixed-effects ordinal logistic regression model (the clmm function in R Christensen, 2023) appropriate for ordinal 5-point Likert scale responses (ranging from very uncomfortable to very comfortable). We follow best practices for the vignette data analysis using ordinal regressions (Baguley et al., 2022). Participant ID (randomly assigned) and vignette ID (vignette number) are included as two random effects in the analysis. The dependent variable is the level of comfort (five ordered categories). The independent variables are the variables whose effect we want to evaluate: actors, purposes, features, education, and race or ethnicity. We also include responses to the privacy attitudes, demographic and technical knowledge questionnaires as control variables in the regression, given their documented associated with privacy comfort levels in prior work (Martin and Nissenbaum, 2019).

Analyzing regression coefficients allows us to identify which variables are statistically significant and thus influence the level of comfort that users have with different types of location data being shared.

A detailed description of the ordinal-regression model and the model analysis using AIC (see Appendix Table 4) is provided in the Appendix section Ordinal Regression. We also verify that the assumptions for the ordinal-regression model are satisfied.

In addition, we are also interested in pairwise comparisons between independent variables. For example, while the ordinal regression may show that a trajectory-based feature has a significant negative effect on the level of comfort, we are also interested in quantifying significant changes in levels of comfort for that feature across different actors or purposes. For these analyses, we first transform the 5-point Likert scale responses into a numerical scale from -2 to +2, (-2 $=$ very uncomfortable, 0 $=$ neutral, +2 $=$ very comfortable). We then perform Kruskal–Wallis (KW) (Kruskal and Wallis, 1952) test for relevant pairwise comparisons (e.g. features and actors), followed by post hoc Dunn tests (Dunn, 1961) with the Benjamini–Hochberg correction (Benjamini and Hochberg, 1995) to identify statistically significant differences in comfort levels across specific pairs (e.g. specific features and actors). Further details for the KW and post hoc Dunn tests are provided in the Appendix section KW and Dunn post hoc test. We do not include the interaction terms in the ordinal-regression model because that would restrict comparisons to specific regression baselines rather than across all feature values.

In the next section, we provide general statistics about the survey participants and the distribution of levels of comfort across features, actors, and purposes. We then address our three main research questions: (1) RQ1: What are the effect of individual trajectory features and POI visits on user levels of comfort?, (2) RQ2: What are the effect of obfuscation approaches on user levels of comfort?, and (3) RQ3: What are the effect of educational background or race and ethnicity on privacy perceptions?

Survey response analysis

We analyze answers from 1,405 participants. On average, each participant spent 18 minutes completing the survey, which included five factorial vignettes as well as the demographic, technical-knowledge and privacy-attitudes questionnaires.

Overall, 7,930 participants viewed the survey, and, 5,577 agreeing to participate in the survey. From the completed surveys, we removed responses whose required free-form text-box explanations did not align with the user comfort level selected on the Likert scale, including both lack of quality explanations or low quality text. Two researchers independently reviewed all responses and agreed upon its quality. Additional details are in the Appendix section Survey Response Quality Control.

We aimed for a U.S. representative sample, with demographic proportions similar to those reported by the U.S. Census American Community Survey (ACS) (U.S. Census Bureau, 2022). Table 4 summarizes the participants’ education levels and racial or ethnic identities, which are the focus of our analysis. Additionally, statistics for age and gender can be found in the Appendix Table 1 and Appendix Table 2, respectively.

Table 4.

U.S. census and survey population distribution across four major races/ethnicities and education levels.

Education	Under High School	High school to Bachelors	Bachelors and above	U.S. Census
Race/Eth	Under High School	High school to Bachelors	Bachelors and above	U.S. Census
White	6.2%	37.4%	24.5%	68.2%
Black	1.2%	7.2%	2.9%	11.3%
Asian	1.2%	2.0%	3.2%	5.7%
Hispanic	3.6%	8.4%	2.9%	14.9%
Census	12.1%	55.0%	33.5%

Education	Under High school	High school to Bachelors	Bachelors and above	Survey Population
Race/Eth	Under High school	High school to Bachelors	Bachelors and above	Survey Population
White	3.60%	48.80%	21.40%	73.80%
Black	1.00%	8.80%	3.30%	13.10%
Asian	0.20%	2.80%	2.60%	5.60%
Hispanic	0.40%	5.50%	1.70%	7.60%
Survey	5.20%	65.90%	29.00%

The bold values indicate total for each row and column.

The sample is approximately representative of the U.S. population for White and Black groups. The percentages for Asian, Black and White groups closely match ACS estimates, while the values for Hispanic are lower. To account for this imbalance, we apply weights in our ordinal-regression analysis, giving more importance to groups that are less represented (further details are provided in the Appendix section Ordinal Regression).

Response analyses. Figure 3 shows the distribution of participants’ comfort levels for each location feature. Trajectory features, including detailed most-frequent trips, least-frequent trips, frequent walking trips, and places visited were often rated as Uncomfortable or Very Uncomfortable. In contrast, several obfuscated features elicited considerably higher levels of comfort when compared to their detailed counterpart, including frequent trips, frequent walking activity and work location. These results indicate that efforts to obfuscate individual trajectory and POI-visit features are viewed favorably by participants, increasing their comfort with these analytical approaches. Table 5 shows all mean comfort levels per location feature when levels are transformed into a numeric variable.

Figure 3.

Percentage of responses per feature type and level of comfort (ordered left to right: Very uncomfortable, Uncomfortable, Neutral, Comfortable, Very comfortable).

Table 5.

Feature and comfortableness (means).

Feature	Answer
Feature: Home location (Detailed)	0.155
Feature: Home location (Obfuscated)	0.210
Feature: Work location (Detailed)	0.268
Feature: Work location (Obfuscated)	0.263
Feature: Places you visit (Detailed)	0.092
Feature: Places you visit (Obfuscated)	0.224
Feature: Area you spent most of your time (Obfuscated)	0.200
Feature: Modes of transportation (Detailed)	0.175
Feature: Modes of transportation (Obfuscated)	0.220
Feature: Most frequent trips (Detailed)	$- 0.058$
Feature: Least frequent trips (Detailed)	0.026
Feature: Most frequent type of trips (Detailed)	0.145
Feature: Most frequent trips between counties (Obfuscated)	0.129
Feature: Most frequent trips between counties(’Google’) (Obfuscated)	0.152
Feature: Frequent walking activity (Detailed)	0.054
Feature: Frequent walking activity (Obfuscated)	0.356
Feature: International visits (Detailed)	0.286
Feature: International visits (Obfuscated)	0.247

The privacy-attitudes questionnaire at the end of the survey asked participants to rate their agreement with statements related to trust in businesses or in government agencies, as well as attitude towards authority. Interestingly, trust in institutions and a general predisposition towards compliance associates with increase comfort levels with sharing location data (detailed or obfuscated), while skepticism towards authority is associated with reduced willingness to share location information. Detailed results can be explored in the Appendix Table 3. These results confirm the importance of adding participant responses to privacy attitude questionnaires in the regression model as control variables, since they appear to play a role in the perception of comfort.

The distribution of comfort levels across purposes and actors aligns with prior work like (Akinbi et al., 2021; Apthorpe et al., 2018; Martin and Nissenbaum, 2019; Vitak et al., 2022) (see Figures 4 and 5 and Tables 6 and 7 for detailed results). Participants were more likely to respond with ”Very Uncomfortable” or ”Uncomfortable” for purposes related to optimizing work productivity and monitoring mobility patterns. In contrast, Purposes with perceived societal benefits, such as identifying locations for infrastructure development, designing walking/cycling infrastructure, and analyzing public transit services were more frequently rated as comfortable or very comfortable. Consistent with prior work, participants expressed greater discomfort sharing data with Employers and Federal Government Agencies whereas actors such as Academic Researchers, Family, Emergency Services, and Doctors were rated higher on the comfort scale (see Figure 5 and Table 7 for further details).

Figure 4.

Percentage of responses per actor and level of comfort (ordered left to right: Very uncomfortable, Uncomfortable, Neutral, Comfortable, Very comfortable).

Figure 5.

Percentage of responses per purpose and level (ordered left to right: Very uncomfortable, Uncomfortable, Neutral, Comfortable, Very comfortable). Infra stands for Infrastructure.

Table 6.

Actor and comfortableness (mean).

Actor	Answer
Actor: Employer	$- 0.424$
Actor: Federal government agency	$- 0.073$
Actor: Law enforcement agency	0.081
Actor: Commercial Entity	0.162
Actor: Local government agency	0.189
Actor: Doctor	0.256
Actor: Family	0.383
Actor: Academic researchers	0.387
Actor: Emergency services	0.525

Table 7.

Purpose and comfortableness (means).

Purpose	Answer
Purpose: Optimize work productivity	$- 0.054$
Purpose: Monitor mobility patterns	$- 0.022$
Purpose: Analysis of terrorist attacks	0.126
Purpose: Analysis of criminal activity	0.147
Purpose: Personal wellness	0.157
Purpose: Show ads	0.185
Purpose: Control spread of diseases	0.233
Purpose: Analysis of Public transit services	0.276
Purpose: Design new walking/cycling infrastructure	0.425
Purpose: Identify locations for infrastructure	0.446

RQ1: Effect of location features, actors and purposes on privacy perceptions

Table 8 reports the ordinal-regression coefficients and odds ratios for actors, purposes and location features. Statistically significant coefficients indicate variables that have a significant effect on participants’ levels of comfort. The odds ratio quantify how much more or less likely the participants are to report higher comfort with respect to the baseline, holding all other variables constant. First, we analyze the effect of location features (individual trajectory and POI visit data) on privacy perceptions, including pairwise comparisons between location features and actors or purposes. Next, we report our main findings for actors and purposes, confirming significant insights found in prior work.

Table 8.

Regression coefficients for actors, purposes, features, ethnicity and education.

	Coefficient	Odds Ratio	Std Err	Statistic	p-Value
Actor (Baseline: Commercial Entity)
Employer	$- {0.626}^{* *}$	0.535	0.149	$- 4.205$	0
Federal government agency	$- {0.249}^{* *}$	0.78	0.11	$- 2.266$	0.023
Law enforcement agency	0.14	1.15	0.11	1.272	0.203
Local government agency	0.105	1.111	0.097	1.081	0.28
Doctor	${1.078}^{* *}$	2.939	0.169	6.393	0
Family	${1.409}^{* *}$	4.092	0.166	8.493	0
Academic researchers	${0.539}^{* *}$	1.714	0.087	6.199	0
Emergency services	${1.166}^{* *}$	3.209	0.185	6.297	0
Purpose (Baseline: Show ads)
Optimize work productivity	0.152	1.164	0.224	0.679	0.497
Monitor mobility patterns	$- {0.654}^{* *}$	0.52	0.178	$- 3.666$	0
Analysis of terrorist attacks	0.09	1.094	0.183	0.488	0.626
Analysis of criminal activity	$- 0.075$	0.928	0.179	$- 0.416$	0.677
Personal wellness	$- 0.338$	0.713	0.212	$- 1.594$	0.111
Control spread of diseases	0.138	1.148	0.188	0.731	0.465
Analysis of Public transit services	$- 0.047$	0.954	0.185	$- 0.255$	0.799
Design new walking/cycling infrastructure	${0.583}^{* *}$	1.791	0.195	2.986	0.003
Identify locations for infrastructure	0.296	1.344	0.19	1.56	0.119
Feature (Baseline: Places you Visit (Detailed))
Home location (Detailed)	$- 0.101$	0.904	0.21	$- 0.48$	0.631
Home location (Obfuscated)	${0.441}^{* *}$	1.554	0.213	2.068	0.039
Work location (Detailed)	$- 0.2$	0.819	0.191	$- 1.043$	0.297
Work location (Obfuscated)	${0.376}^{* *}$	1.456	0.19	1.973	0.049
Places you visit (Obfuscated)	0.166	1.181	0.139	1.191	0.234
Area you spent most of your time (Obfuscated)	0.012	1.012	0.148	0.081	0.936
Modes of transportation (Detailed)	0.01	1.01	0.149	0.066	0.948
Modes of transportation (Obfuscated)	0.192	1.212	0.144	1.327	0.185
Most frequent trips (Detailed)	$- {0.297}^{* *}$	0.743	0.146	$- 2.039$	0.041
Least frequent trips (Detailed)	$- 0.198$	0.82	0.153	$- 1.292$	0.196
Most frequent type of trips (Detailed)	$- 0.063$	0.939	0.147	$- 0.431$	0.667
Most frequent trips between counties (Obfuscated)	0.168	1.183	0.148	1.134	0.257
Most frequent trips between counties (’Google’) (Obfuscated)	0.23	1.259	0.147	1.562	0.118
Frequent walking activity (Detailed)	$- 0.191$	0.826	0.143	$- 1.335$	0.182
Frequent walking activity (Obfuscated)	${0.458}^{* *}$	1.581	0.145	3.166	0.002
International visits (Detailed)	$- 0.061$	0.941	0.209	$- 0.29$	0.772
International visits (Obfuscated)	${0.369}^{*}$	1.446	0.21	1.753	0.08
Ethnicity/Race (Baseline: White)
Asian	$- 0.017$	0.983	0.233	$- 0.072$	0.942
Black	0.221	1.247	0.162	1.361	0.174
Hispanic	${0.504}^{* *}$	1.655	0.196	2.574	0.01
Education (Baseline: Highschool to Bachelors)
Under Highschool	$- 0.017$	0.983	0.227	$- 0.075$	0.94
Bachelors and above	$- 0.206$	0.814	0.126	$- 1.631$	0.103

Odds ratio column is the likelihood of being more comfortable compared to the baseline class. Coefficients for the control variables are shown in the appendix, in Table 5. Significance ** $p - value < 0.05$ , * $p - value < 0.1$

Location features

Table 8 shows that participants reported statistically significant lower levels of comfort when sharing detailed trajectory data compared to detailed visits to POIs (the regression baseline). For example, the odds of being comfortable sharing Most Frequent Trips (Detailed) were 0.74 times that of sharing detailed visits to POIs. In other words, participants were significantly less comfortable with the use of detailed trajectory data than with the use of visits to specific types of POIs (e.g. restaurants, libraries, schools, etc.). As one participant stated: ”There is no good reason for them to monitor my trips on the regular at all.”(Participant-8a6a).

In contrast, participants expressed higher comfort levels sharing obfuscated trajectory data compared to detailed visits to POIs. For example, walking (Frequent Walking Activity) was 1.5 times more likely to be rated as comfortable and international trips (International Visits) were 1.44 times more likely to be rated as comfortable relative to the baseline.

Participants were also significantly more comfortable sharing their home and work locations (obfuscated to the census tract) than sharing detailed visits to specific POIs, although no significant difference was observed when home and work locations were not obfuscated (we discuss this further in RQ2).

These results reflect that if appropriately obfuscated, users are willing to share some forms of trajectory data. As Participant-D303 noted, ”This is general information and I would be interested in this as well”.

Pairwise analysis. Table 9 summarizes the (feature, actor) and (feature, purpose) pairs that differ statistically significantly in median level of comfort. Each row in the table compares two pairs (feature, actor1) and (feature, actor2) whose median level of comfort (M1 and M2, respectively) are significantly different, obtained after applying the KW and Dunn tests. Table 6 in the Appendix summarizes the complete tests. We focus on two interesting findings from Table 9. First, the median level of comfort with sharing obfuscated trajectory data decreases depending on the actor accessing the data. For researchers, family members, and doctors, the median level of comfort is ”Comfortable” (median $=$ 1, p-value<0.05) for Federal Government agency or Employer actors the level of comfort is Neutral (median=0, p-value<0.05)). Second, the median levels of comfort with respect to visits to POIs (Ob) is significantly higher for socially beneficial purposes such as ”Identifying locations for infrastructure or cycling” (median $=$ 1, p-value<0.05) than for more generic purposes such as ”monitoring mobility patterns” (median $=$ 0, p-value<0.05).

Table 9.

Subset of significant differences in (feature,actor) and (feature,purpose) pairs after applying Kruskal–Wallis analysis with post hoc Dunn test.

Feature	Actor1	Actor2	M1	M2	M2-M1
Freq Walks(Ob)	Employer	Researchers	0	1	1
		Doc	0	1	1
		Family	0	1	1
Freq Walks(Ob)	Fed	Researchers	0	1	1
		Doc	0	1	1
		Family	0	1	1
Freq Walks(Ob)	Local gov	Doc	0	1	1
Feature	Purpose1	Purpose2	M1	M2	M2-M1
Visits(Ob)	Monitor mobility	Public transit	0	1	1
		Infrastructure (Walk/Cycling)	0	1	1

Complete results can be found in Table 6 in the appendix. M”i” represents median level of comfort value for distribution ”i.” Significance at p-value <0.05).

Actors and purposes

Our ordinal-regression results align with prior findings on the effect of actors and purposes on privacy perceptions (Martin and Nissenbaum, 2019; Vitak et al., 2022). Federal Government Agencies and Employers were significantly negatively related to comfort when accessing location data (Table 8). In fact, participants had 0.78 and 0.53 times the odds (p<0.01), respectively, of rating their level of comfort higher for these actors than for the baseline actor (Commercial Entities).

On the other hand, Academic Researchers, Emergency Services, Family members, and Doctors were positively related to levels of comfort, with participants having 1.7, 3.2, 4 and 2.9 times the odds (p<0.01) of rating their level of comfort higher for these actors than for the baseline actor (Commercial Entities).

For purposes, we observe that socially beneficial purposes such as Designing public infrastructure were perceived more favorably than marketing purposes (regression baseline), with the odds ratio being 1.79 times higher.

In contrast, generic purposes (e.g. Monitor mobility patterns) were associated with significantly lower comfort levels compared to the baseline, that is, people were approximately half as likely to be comfortable with sharing data (p-value<0.05).

Pairwise analysis.

Appendix Table 7 summarizes the KW and Dunn test results for the pairwise analysis between (actor, feature) and (purpose, feature) pairs. Here, we would like to highlight two significant findings. First, sharing walking activity (obfuscated) was associated with higher median levels of comfort (median $= 1$ , Comfortable) than other features such as most frequent type of trips (obfuscated) (median $= 0$ , Neutral) or most frequent trips (detailed) (median $= - 0.5$ , leaning uncomfortable). In contrast, when the actor is an Employer, giving access to the most frequent places of visit (obfuscated) was perceived as more uncomfortable (median $= - 1$ , Uncomfortable) than giving access to walking activity (obfuscated) (median $= 0$ , Neutral).

Finally, no significant differences were identified between purposes and location features, indicating similar levels of comfort independently of the trajectory feature or POI visit feature.

RQ2: Effect of obfuscating approaches on privacy perceptions

The Features column in Table 8 shows that several obfuscated trajectory and POI visit features are associated with statistically significant higher levels of comfort than their detailed counterparts (Places you visit (detailed) as baseline). Specifically, obfuscated frequent walking activity, obfuscated international trips, obfuscated home and work locations had 1.58, 1.4, 1.55 and 1.45 times the odds (p<0.05) of being rated higher in level of comfort than detailed places visited, indicating that individuals were more prone to sharing certain types of trajectory and visits data if protected by obfuscation approaches. Beyond statistical significance, it is important to highlight that all means (see Table 5) were higher for obfuscation features than for their detailed counterpart, for example, Home (Detailed $-$ Obfuscated) $= 0.055$ , Modes of Transport (Detailed $-$ Obfuscated) $= 0.045$ . In other words, even when not significant in the regression model, obfuscated location features were associated with more positive levels of comfort towards data sharing.

Pairwise analysis. We performed KW and Dunn post hoc tests to identify (actor, obfuscated/detailed feature) and (purpose, obfuscated/detailed feature) pairs that statistically significantly differ in median levels of comfort. We aim to evaluate whether the use of obfuscated or detailed features changes participants’ level of comfort for a given actor or purpose. Detailed results are shown in the Appendix Table 8. Here, we discuss two findings. First, we observe that when the actor accessing the location feature is a family member (actor=Family), the users’ level of comfort increases from neutral to comfortable if the location feature is obfuscated rather than detailed. Second, we found that obfuscated features improved the level of comfort when compared to detailed features for certain purposes such as “control the spread of diseases”. This finding aligns with previous studies showing privacy-preserving location sharing apps during COVID were perceived more positively than apps requiring detailed location sharing app (Ioannou and Tussyadiah, 2021; Kim and Kwan, 2021; Zhang et al., 2020).

RQ3: Effect of race/ethnicity and education on privacy perceptions

Race and ethnicity

The results in Table 8 show that Hispanic participants reported significantly higher levels of comfort that White participants (regression baseline), having 1.6 times the odds (OR $= 1.6$ , p<0.01) of rating their level of comfort with location data higher than White participants. Free-form text boxes align with these observations. For example, one participant who self-identified as Hispanic noted: ”I have nothing to hide so it doesn’t really make a difference to me.” (Participant-050e). Another participant noted

”(In) this particular scenario I feel very comfortable with my data being accessed & used for this purpose. Hospital locations are important and their location within a community can severely impact the quality of life for that community too.” (Participant-2ae1). No significant difference in level of comfort was observed between any other racial or ethnic groups.

Pairwise analysis. To examine whether the level of comfort vary for race or ethnicity across location features, actors or purposes, we perform KW and Dunn post hoc statistical tests to identify pairs of (race/ethnicity, location feature), (race/ethnicity, obfuscated/detailed location feature), (race/ethnicity, actor), and (race/ethnicity, purpose) that significantly differ in their median levels of comfort. See Appendix Tables 9 and 10 for the results. Here, we discuss the most relevant findings. For Hispanic participants, the levels of comfort were the lowest for marketing purposes (i.e. showing ads) with median $= - 1$ (Uncomfortable, p-value<0.05).

In contrast, Black participants perceived marketing purposes more positively (median $=$ 1 Comfortable, p-value< 0.05) when compared to other purposes like analysis of terrorist attacks (median $=$ 0), public transit (median $=$ 0) or mobility monitoring (median $=$ 0). Nevertheless, no significant differences were observed between Hispanic or Black participants and actor or feature types. This suggests that, for Hispanic and Black participants, data comfort revolves around purpose of data access, and not around actor who access the data or the specific type of data feature shared.

In contrast, for Asian and White respondents showed significant change in level of comfort between detailed and obfuscated features. For example, ”Most frequent trips (Detailed)” had lower median comfort (median $=$ 0, p-value<0.05) than ”Frequent walking activity (Obfuscated)” (median $=$ 1) for White individuals. In addition, Asian and White respondents were also significantly less comfortable with the Federal Government, Law enforcement, Local government (median $=$ 0) and Employers (median $= - 1$ ) than with Researchers, Family, Doctors or Emergency services (median $=$ 1). White participants were also more comfortable in sharing data for infrastructure-related purposes such as designing cycling infrastructure and designing built infrastructure (median $=$ 1, p-value<0.05) than for all the other purposes (median $=$ 0). Thus, for Asian and White participants we find evidence that privacy perceptions depend on the interactions of actors, purposes and features.

Education

We do not observe any statistically significant difference in comfort levels for different education groups when compared to participants with education high school to bachelors (baseline).

Pairwise analysis. We perform KW and post hoc Dunn tests to identify (education, location feature), (education, obfuscated/detailed location feature), (education, actor), and (education, purpose) pairs with significant differences in median levels of comfort. Table 11 in the Appendix has all the test results, here we discuss a few important findings. Participants with a ”Bachelors and above” felt more comfortable in sharing obfuscated features (median $=$ 1, p-value<0.05) compared to detailed (median $=$ 0, p-value<0.05).

No statistical difference between detailed and obfuscated location data were observed for the other two educational groups (”High School to Bachelors” and ”Under High School”). Participants with a ”Bachelors and above” degree also showed significantly lower levels of comfort with purposes related to monitoring mobility (median $=$ 0, p-value<0.05) and with sharing location data with employers (median $= - 1$ , p-value<0.05). These findings suggest that decisions for this educational group depend on the interaction of the full spectrum of actor-purpose-feature (along with privacy attitudes).

Participants in the “High School to Bachelors” group were significantly less comfortable sharing their location data with the Federal Government, Law Enforcement agencies (median $= 0$ , p-value<0.05) and employers (median $= - 1$ , p-value<0.05) than with Researchers and Emergency services (median $=$ 1, p-value <0.05). These results reveal that the privacy perceptions for this educational group are mostly shaped around actors and purposes, and not types of features or obfuscation techniques. Finally, we did not observe any significant difference between pairs of actors, purposes or features for participants with “Under High School” education, signaling that for these participants their decisions are mostly driven by their individual privacy attitudes.

Implications for policy and practice

Location features. Our regression analysis demonstrates that participants have higher levels of comfort in sharing detailed visits to POIs than detailed trajectory data. However, obfuscating some of the trajectory variables resulted in higher levels of comfort even above sharing detailed visits to POIs. Our analyses also showed that the levels of comfort increased when the actors accessing the data are researchers or family, or when the purpose is involved on social benefit which is consistent with previous research like (Martin and Nissenbaum, 2019; Vitak et al., 2022). Conversely, our study empirically demonstrates a significant level of discomfort regarding government agencies’ access to location data, aligning with the work by legal scholars and practitioners, such as Rahbar (Rahbar, 2022) and Conner (Media, 2025), who have argued against government agencies purchasing location data from DA&DB, citing potential violations of the Fourth Amendment.

These findings suggest the need for implementation of more granular control mechanisms that allow users to specify which types of location features can be derived from the data collected. As DA&DB collect data from apps via SDKs, they could allow users to select their preferences of sharing feature clusters (defined in Table 1) during the app installation or permissions page. Users should also have a mechanism to control which actors may access their location features and for which purposes.

Obfuscation approaches. Obfuscated location features (both trajectory and visits to POIs) elicited higher levels of comfort when compared to detailed visits to POIs (baseline) in the regression model (see Table 8). Levels of comfort increased when the actor accessing the data was a family member. Although not all obfuscated approaches were significantly associated with higher levels of comfort in the regression, the mean values of obfuscated features were higher. Previous studies on people’s willingness to share location data in obfuscated vs detailed settings such as (Ackermann et al., 2022; Bilogrevic et al., 2015; Brush et al., 2010) have found similar results for home, work and places of visit data. In our study, we observed the same trend in trajectories, mode of transport and international trip data. We posit that there might be no need for more granular control over obfuscated versus detailed features. The location features could be obfuscated, and during the app installation users can be given the choice to opt-in to share more detailed features, together with an explanation of why this would be valuable for analytical purposes (actor, purpose), given that vague purposes such as “monitoring mobility patterns” was particularly rated as uncomfortable by participants in our survey. In future research, we will also explore why obfuscation techniques appear to not always alleviate privacy concerns.

Race, ethnicity, and education effects. Our regression analysis revealed significant differences in comfort levels for the Hispanic participants when compared to White participants, with Hispanic participants being more comfortable sharing trajectory and visit data. Our pairwise analysis has also revealed a complex interplay between cultural factors, educational experiences, and privacy perceptions while controlling for privacy attitudes and computer knowledge. Hispanic and Black participants as well as those with “High School to Bachelors” appear to base comfort primarily on the purpose of the data access or the actors accessing the data. In contrast, White participants and those an educational level “Bachelors and above” shape their data sharing comfort around the combined effects of actors, purposes and by the types of location features being shared. We did not observe any significant association between education and levels of comfort which is consistent with prior research on general data sharing preferences (Longin et al., 2025), but opposed to findings specific to the health domain (Trinidad et al., 2020).

These findings could be used to pre-populate app settings for location data sharing. Settings could be detailed with actors or purposes, and include, the educational background of the person installing the app, thus reducing the burden of having to select among many different options from scratch.

However, offering granular controls can limit access to detailed variables valuable in specific circumstances, for example, monitoring evacuation during a disaster. Hence, companies using location data for societal benefit, could consider explaining well the benefits of sharing location data, encouraging users to support data for good initiatives; or could use economic incentives that would encourage participants to share more of their data.

Policy suggestions. Building on our empirical findings, the FTC could adopt several policies to strengthen consumer control and privacy protections in the location intelligence ecosystem. First, the agency could require data aggregators to implement fine-grained, purpose-specific consent mechanisms that clearly distinguish actors, data features, and intended uses. Standardized consent interfaces such as “privacy nutrition labels” would help ensure that users can easily understand and manage the permissions they grant. Complementing these controls, the FTC could mandate that individuals have simple, unified tools to revise or revoke consent at any time, supported by auditable logs that show when and how their data have been accessed. Such initiatives have already been adopted in state governments such as California⁷, Massachusetts⁸ and Illinois.⁹ These measures would not only enhance transparency but also reduce asymmetries between consumers and DA&DB, who currently operate with limited oversight.

Second, the FTC could promote meaningful data minimization by adopting an obfuscation-by-default rule, requiring companies to use privacy-preserving transformations unless users explicitly opt-in to granular data sharing. Importantly, the FTC could also incorporate equity-aware protections informed by our findings. Because comfort with specific data uses varies substantially across racial and ethnic identities, the agency could permit companies to offer empirically grounded, pre-populated privacy profiles that reduce user burden while preserving full control for individuals to modify defaults. Alternatively, the policies could default to the most cautious user. Further, to prevent disparate privacy risks, the FTC could require periodic assessments of racial and socioeconomic disparities in data exposure and mandate disclosures to users when certain groups face higher risks due to a company’s data practices. Together, these measures would ensure that location data governance is not only transparent and user-centered but also attentive to equity considerations evident in public preferences.

Limitations and future work

Our list of actors, purposes and features is not exhaustive, mostly due to a need to narrow down the number of potential actor-purpose-feature combinations to collect statistically significant information in the survey. Our research focuses on only 18 features derived from location data and examines how six actors may use these features for ten purposes. However, most actors do not publicly disclose the full range of purposes for which they use collected location data, except in limited cases such as academic publications or investigative reporting that uncovers significant privacy breaches by government or commercial entities. Important actors such as foreign entities purchasing U.S. data or large language model (LLM) developers using personal data to generate behavioral profiles are not captured in our framework. Similarly, additional purposes, such as selecting billboard locations or identifying optimal sites for new retail stores, could be considered in future work.

Future vignette designs could incorporate additional factors that influence comfort with data sharing, such as the time span of data collection, the presence of monetary incentives and the granularity of obfuscation (e.g. census tract vs counties, different radii of obfuscation). The manual assessment of free-form text responses could be replaced in future work by an LLM-based few-shot learning system, especially now that we have identified several examples of both high-quality and low-quality responses.

Conclusion

This study advances our understanding of the nuanced factors shaping privacy perceptions in location data sharing. By highlighting the importance of the type of location feature (trajectory or visits to POIs), the presence of an obfuscation approach, contextual factors such as actors and purpose, and privacy attitudes and demographic variations our findings provide a foundation for user-centric, feature focused approach to location privacy. Our results show that trajectory-related features are associated with higher privacy concerns, current obfuscation approaches by DA&DBs are sometimes successful, and that race, ethnicity and education have an effect on privacy perceptions with Hispanic participants and those with High School to Bachelor’s education reporting higher levels of comfort in location data sharing. As technology continues to evolve, ongoing research in this area will be crucial for developing privacy practices and policies that effectively balance the benefits of location-based services with individual privacy rights.

Supplemental Material

sj-pdf-1-bds-10.1177_20539517261429203 - Supplemental material for From “I have nothing to hide” to “Its stalking”: Americans’ comfort sharing individual mobility features

Supplemental material, sj-pdf-1-bds-10.1177_20539517261429203 for From “I have nothing to hide” to “Its stalking”: Americans’ comfort sharing individual mobility features by Naman Awasthi, Saad Mohammad Abrar, Daniel Smolyak and Vanessa Frias-Martinez in Big Data & Society

Footnotes

ORCID iDs

Naman Awasthi

Daniel Smolyak

Vanessa Frias-Martinez

Saad Mohammad Abrar

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Supplemental material

Supplemental material for this article is available online.

Notes

References

Ackermann

Burkhalter

Mildenberger

, et al. (2022) Willingness to share data: Contextual determinants of consumers’ decisions to share private data with companies. Journal of Consumer Behaviour 21(2): 375–386.

Adler

Bernardin

Dumont

, et al. (2017) The promise and limitations of locational app data for origin-destination analysis: A case study. Technical Report FHWA-HEP-20-022, United States. Federal Highway Administration. https://rosap.ntl.bts.gov/view/dot/55798.

Akinbi

Forshaw

Blinkhorn

(2021) Contact tracing apps for the covid-19 pandemic: A systematic literature review of challenges and future directions for neo-liberal societies. Health Information Science and Systems 9(1): Article 18. DOI: 10.1007/s13755-021-00147-7.

Andrés

Bordenabe

Chatzikokolakis

, et al. (2013) Geo-indistinguishability: Differential privacy for location-based systems. In: Proceedings of the 2013 ACM SIGSAC conference on computer & communications security, CCS ’13, pp.901–914. New York, NY: Association for Computing Machinery. ISBN 9781450324779. DOI: 10.1145/2508859.2516735.

Apthorpe

Shvartzshnaider

Mathur

, et al. (2018) Discovering smart home internet of things privacy norms using contextual integrity. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 2(2): Article 59. DOI: 10.1145/3214262.

Awasthi

(2025) From “I have nothing to hide” to “It’s stalking”: Americans’ comfort sharing individual mobility features [Project]. OSF. DOI: https://doi.org/10.17605/OSF.IO/HY9M3.

Baguley

Dunham

Steer

(2022) Statistical modelling of vignette data in psychology. British Journal of Psychology 113(4): 1143–1163.

Bansal

Zahedi

Gefen

(2016) Do context and personality matter? trust and privacy concerns in disclosing private information online. Information & Management 53(1): 1–21.

Bauer

Strauss

(2016) Location-based advertising on mobile devices. Management Review Quarterly 66(3): 159–194.

10.

Benjamini

Hochberg

(1995) Controlling the false discovery rate: A practical and powerful approach to multiple testing. Journal of the Royal Statistical Society. Series B (Methodological) 57(1): 289–300.

11.

Bilogrevic

Huguenin

Mihaila

, et al. (2015) Predicting users’ motivations behind location check-ins and utility implications of privacy protection mechanisms. In: 22nd network and distributed system security symposium (NDSS).

12.

Brush

Krumm

Scott

(2010) Exploring end user preferences for location obfuscation, location-based services, and the value of location. In: Proceedings of the 12th ACM international conference on ubiquitous computing, UbiComp ’10, p.95–104. New York, NY, USA: Association for Computing Machinery. ISBN 9781605588438. DOI: 10.1145/1864349.1864381.

13.

Center

(2024) Closing the data broker loophole — brennan center for justice. https://www.brennancenter.org/our-work/research-reports/closing-data-broker-loophole.

14.

Chakraborty

Vishik

Rao

(2013) Privacy preserving actions of older adults on social media: Exploring the behavior of opting out of information sharing. Decision Support Systems 55(4): 948–956.

15.

Chen

Poorthuis

(2021) Identifying home locations in human mobility data: An open-source r package for comparison and reproducibility. International Journal of Geographical Information Science 35(7): 1425–1448.

16.

Christensen

RHB

(2023) ordinal—Regression Models for Ordinal Data. https://CRAN.R-project.org/package=ordinal. R package version 2023.12-4.1.

17.

Chuang

Beattie

Feng

(2023) Analysing the relationship between proximity to transit stations and local living patterns: A study of human mobility within a 15 min walking distance through mobile location data. Urban Science 7(4): Article 105. DOI: 10.3390/urbansci7040105.

18.

Chuang

Chen

(2024) Urban street dynamics: Assessing the relationship of sidewalk width and pedestrian activity in auckland, new zealand, based on mobile phone data. Urban Studies 0(0): 00420980241293659.

19.

Coleman

Liu

Zhao

, et al. (2023) Lifestyle pattern analysis unveils recovery trajectories of communities impacted by disasters. Humanities and Social Sciences Communications 10(1): Article 803 .

20.

Cox

(2022) Data broker is selling location data of people who visit abortion clinics. https://www.vice.com/en/article/m7vzjb/location-data-abortion-clinics-safegraph-planned-parenthood.

21.

Cuebiq, Marketing and Team (2025) Come Back Soon! Measuring the Effectiveness of Tourism Campaigns with — cuebiq.com. https://cuebiq.com/come-back-soon-tourism/ (accessed 06 February 2025).

22.

Cai

Zhang

, et al. (2019) An efficient dummy-based location privacy-preserving scheme for internet of things services. Information 10(9): Article 278. DOI: 10.3390/info10090278.

23.

Dunn

(1961) Multiple comparisons among means. Journal of the American Statistical Association 56(293): 52–64.

24.

Farke

Balash

Golla

, et al. (2021) Are privacy dashboards good for end users? evaluating user perceptions and reactions to google’s my activity. In: 30th USENIX security symposium (USENIX Security 21). USENIX Association. ISBN 978-1-939133-24-3, pp.483–500. https://www.usenix.org/conference/usenixsecurity21/presentation/farke.

25.

Fraser

Yabe

Aldrich

, et al. (2024) The great equalizer? mixed effects of social infrastructure on diverse encounters in cities. Computers, Environment and Urban Systems 113: 102173.

26.

Frias-Martinez

Soto

Virseda

, et al. (2012) Computing cost-effective census maps from cell phone traces. In: Workshop on pervasive urban applications.

27.

Gamarra

Meriño Fuentes

Calabria Sarmiento

, et al. (2019) Privacy perception in location-based services for mobile devices in the university community of the north coast of colombia. Ingenieria y Universidad.

28.

Gao

Huang

Zheng

, et al. (2022) Protecting location privacy of users based on trajectory obfuscation in mobile crowdsensing. IEEE Transactions on Industrial Informatics 18(9): 6290–6299.

29.

Garcia-Bulle

Horn

Bell

, et al. (2022) You are where you eat: Effect of mobile food environments on fast food visits. medRxiv DOI: 10.1101/2022.09.20.22280128.

30.

Gerber

Reinheimer

Volkamer

(2019) Investigating people’s privacy risk perception. Proceedings on Privacy Enhancing Technologies 2019(3): 267–288.

31.

Gilbert

Shilton

Vitak

(2023) When research is the context: Cross-platform user expectations for social media data reuse. Big Data & Society 10(1): 20539517231164108.

32.

Gorra

(2007) An analysis of the relationship between individuals? perceptions of privacy and mobile phone location data—A grounded theory study. https://eprints.leedsbeckett.ac.uk/id/eprint/1554/.

33.

Gramaglia

Fiore

Furno

, et al. (2021) Glove: Towards privacy-preserving publishing of record-level-truthful mobile phone trajectories. ACM/IMS Transactions on Data Science 2(3). DOI: 10.1145/3451178

34.

Haffner

Mathews

Finchum

(2018) Location-based social media behavior and perception: Views of university students. Geographical Review 108(2): 203–224.

35.

Hong

Frias-Martinez

(2016) Topic models to infer socio-economic maps. In: Proceedings of the AAAI conference on artificial intelligence, Vol. 30.

36.

Hong

Frias-Martinez

, et al. (2019) Characterization of internal migrant behavior in the immediate post-migration period using cell phone traces. In: Proceedings of the tenth international conference on information and communication technologies and development. pp.1–12.

37.

Hunter

Garcia

de Sa

, et al. (2021) Effect of COVID-19 response policies on walking behavior in US cities. Nature Communications 12(1): 3652.

38.

Ioannou

Tussyadiah

(2021) Privacy and surveillance attitudes during health crises: Acceptance of surveillance and privacy protection behaviours. Technology in Society 67: 101774.

39.

Junglas

Spitzmuller

(2006) Personality traits and privacy perceptions: An empirical study in the context of location-based services. In: 2006 International conference on mobile business. pp.36–36. DOI: 10.1109/ICMB.2006.40.

40.

Kim

Lee

(2022) Gps data-based mobility mode inference model using long-term recurrent convolutional networks. Transportation Research Part C: Emerging Technologies 135: 103523.

41.

Kim

Kwan

(2021) An examination of people’s privacy concerns, perceptions of social benefits, and acceptance of covid-19 mitigation measures that harness location information: A comparative study of the us and south korea. ISPRS International Journal of Geo-Information 10(1): 25.

42.

Kruskal

Wallis

(1952) Use of ranks in one-criterion variance analysis. Journal of the American Statistical Association 47(260): 583–621.

43.

Wang

Zhang

, et al. (2020) Understanding intra-urban human mobility through an exploratory spatiotemporal analysis of bike-sharing trajectories. International Journal of Geographical Information Science 34(12): 2451–2474.

44.

Longin

Briceno

Poquet

(2025) Data sharing in learning analytics: How context and group discussion influence the individual willingness to share. Humanities and Social Sciences Communications 12(1): Article 849.

45.

Wang

(2014) Development of a data-driven platform for transit performance measures using smart card and gps data. Journal of Transportation Engineering 140(12): 04014063.

46.

Martin

Nissenbaum

(2019) What is it about location? Berkeley technology law journal / Boalt Hall School of Law, University of California, Berkeley DOI: 10.2139/ssrn.3360409.

47.

Media

(2025) Data broker sales and the fourth amendment. https://www.lawfaremedia.org/article/data-broker-sales-and-the-fourth-amendment (accessed: 27 Auguset 2024).

48.

Morrison

(2021) This outed priest’s story is a warning for everyone about the need for data privacy laws. https://www.vox.com/recode/22587248/grindr-app-location-data-outed-priest-jeffrey-burrill-pillar-data-harvesting.

49.

Nissenbaum

(2004) Privacy as contextual integrity. Washington Law Review 79: 119.

50.

Organizers, Kang

Oshan

Wolf

, et al. (2019) A roundtable discussion: Defining urban data science. Environment and Planning B: Urban Analytics and City Science 46(9): 1756–1768.

51.

Rahbar

(2022) How the government’s purchase of commercial location data violates carpenter and evades the fourth amendment. Columbia Law Review 122(3): 713–754.

52.

Record (2024a) FTC settles unprecedented case against geolocation data broker — therecord.media. https://therecord.media/ftc-settles-case-geolocation-data-broker-xmode-outlogic (accessed 27 Auguset 2024).

53.

Record (2024b) FTC’s Khan warns tech industry that agency will strictly enforce AI data privacy — therecord.media. https://therecord.media/ftc-warns-tech-industry-ai-data-privacy (accessed 27 Auguset 2024).

54.

SafeGraph (2025) Places data curated for accurate geospatial analytics. https://www.safegraph.com/.

55.

Shankar

Ganapathy

Iftode

(2009) Privately querying location-based services with sybilquery. In: Proceedings of the 11th international conference on ubiquitous computing, UbiComp ’09, p.31–40. New York, NY, USA: Association for Computing Machinery. ISBN 9781605584317. DOI: 10.1145/1620545.1620550.

56.

Shen

Wang

Zhang

(2023) A privacy-preserving trajectory publishing method based on multi-dimensional sub-trajectory similarities. Sensors 23(24): Article 9652. DOI: 10.3390/s23249652.

57.

Sila-Nowicka

Vandrol

Oshan

, et al. (2016) Analysis of human mobility patterns from gps trajectories and contextual information. International Journal of Geographical Information Science 30(5): 881–906.

58.

Spectus (2025) Device Recurring Areas and Sensitive LocationsSpectus Documentation Portal—docs.spectus.ai. https://docs.spectus.ai/Getting%20Started/User_Guides/Data_Assets/Device_Recurring_Areas_and_Sensitive_Locations/#how-we-expose-the-devices-recurring-areas (accessed 05 February 2025).

59.

Trinidad

Platt

Kardia

SLR

(2020) The public’s comfort with sharing health data with third-party commercial companies. Humanities and Social Sciences Communications 7(1).

60.

US Census Bureau (2022) Age and sex. American Community Survey 5-Year Estimates Subject Tables, Table S0101. Available at: https://data.census.gov/table/ACSST1Y2023.S1501?q=education%20attainment.

61.

Vitak

Liao

Mols

, et al. (2022) When do data collection and use become a matter of concern? a cross-cultural comparison of U.S. and Dutch privacy attitudes. https://api.semanticscholar.org/CorpusID:261211421.

62.

Vitak

Zimmer

(2023) Surveillance and the future of work: Exploring employees’ attitudes toward monitoring in a post-covid workplace. https://api.semanticscholar.org/CorpusID:259666183 .

63.

Wang

EST

Lin

(2017) Perceived quality factors of location-based apps on trust, perceived privacy risk, and continuous usage intention. Behaviour & Information Technology 36(1): 2–10.

64.

Wang

(2024) Unraveling hurricane ian’s impact: A multiscale analysis of mobility networks in florida. Transportation Research Part D: Transport and Environment 136: 104482.

65.

Abrar

Awasthi

, et al. (2022) Enhancing short-term crime prediction with human mobility flows and deep learning architectures. EPJ Data Science 11(1): 53.

66.

Yang

Pentland

Moro

(2023a) Identifying latent activity behaviors and lifestyles using mobility data to describe urban dynamics. EPJ Data Science 12(1): 15.

67.

Yang

Pentland

Moro

(2023b) Identifying latent activity behaviors and lifestyles using mobility data to describe urban dynamics. EPJ Data Science 12(1): 15.

68.

Zhang

Kreps

McMurry

, et al. (2020) Americans’ perceptions of privacy and surveillance in the COVID-19 pandemic. PLoS One 15(12): e0242652.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

13.16 MB