Technological Trust Perceptions in Wearable Fitness Technology: A Person-Centred Approach

Technological trust is a key element impacting the success of a technology. This study focuses on fitness tracker users and their perceptions of trust towards the device. The aim of the study is to identify distinct subgroups of technological trust perceptions of fitness tracker users (n = 150) adopting a person-centred approach. Furthermore, the role of age, experience with the device and effort expectancy in trust perception subgroup membership are examined. Three distinct subgroups could be identified. The first subgroup is characterised by neutral to low data privacy perceptions and moderate perceptions towards reliability, validity, system capability and system transparency perceptions. The second subgroup is characterised by moderate to high trust perceptions, and the third group shows the highest trust perceptions in comparison to the other subgroups. Age and experience with the device were no significant predictors of trust profile membership; effort expectancy, however, was a significant predictor. Users who perceive that using the device does not require high effort and that using the tracker is easy are more likely to be classified in the high trust subgroup and in the moderate to high trust subgroup than in the low to moderate trust subgroup. Furthermore, differences between two latent profiles regarding their usage of the wearable in their physical activities were found.


Introduction
Wearable fitness technologies are self-tracking devices that help people to self-monitor their physical activity.These include smart watches or smart wristband fitness trackers.Most of these technological devices are attached to the body and should be worn while being physically active to track various health-related parameters like heart rate, calories burned, GPS data, distances and steps (Lupton, 2013;Rupp et al., 2016Rupp et al., , 2018;;Shin et al., 2019).Reasons for using these technologies are the possibility to monitor one's activity, their usefulness for self-motivation, curiosity and health aspects (Wiesner et al., 2018).However, while the tracking of individuals' activity data, training progress, goals etc. is beneficial for the user (e.g.increasing one's physical activity) (Brickwood et al., 2019;Coughlin & Stewart, 2016), it is also linked to issues like privacy risks, data security (Huckvale et al., 2015;Kang & Jung, 2021;Wiesner et al., 2018) and trust in the technology issues (Rupp et al., 2018).The fast development of technology, the integration of technologies in our lives and our increased dependency on technology have led to research into the role of technological trust (e.g.privacy issues) in human-technology interaction (Ejdys, 2018).
Even though there is a growing interest in technological trust issues in different domains like health information technology (Xie et al., 2020), smart healthcare services (Liu & Tao, 2022), shared technology (Xu et al., 2014), e-voting technology (Lippert & Ojumu, 2008), information and communication technology (Ejdys, 2018) and autonomous vehicles (Choi & Ji, 2015;Hegner et al., 2019), few researchers (see systematic review using a topic modelling approach by Shin et al., 2019) have addressed technological trust issues in wearable fitness device or fitness app users (Beldad & Hegner, 2018;Kang & Jung, 2021;Rupp et al., 2016Rupp et al., , 2018;;Wiesner et al., 2018) although these technologies track sensitive health-related and locationbased data.Trust perceptions are significant predictors of individuals' intention to use a technology and a lack of trust could lead to low continuing use of a technology (Gu & Wei, 2021;Luarn & Juo, 2010).This makes it important to consider trust perceptions when studying users use and adoption of technologies.In current research of technology trust, variable-centred approaches are dominant, aiming to identify associations among variables (e.g. using structural equation modelling).In the present paper, a person-centred approach (Bergman & Magnusson, 1997;Masyn, 2013) will be used to identify distinct subgroups of users based on individual patterns of various technological trust perceptions of wearable fitness tracker users.
The aim of the present study is to get information about differences among individuals in how trust perceptions are related to each other.By adopting a more holistic perspective, the combinations of different trust perceptions within individuals are surveyed.Therefore, subgroups regarding trust perceptions should be identified.A further aim is to examine if the trust subgroups differ in their usage of the technology and whether effort expectancy, the experience with the wearable fitness device and age are predictors of technological trust profile membership.

Acceptance and Usage of Technology
Research dealing with different thematic focuses of wearable fitness/health technology, like the potential for indicating (health) behaviour change, technology acceptance, usefulness in medical settings and technological issues (e.g.reliability, validity of the devices), has been growing over the last years (for review see Shin et al., 2019).This research has identified factors that are related to technology use and acceptance.According to the Unified Theory of Acceptance and Usage of Technology (UTAT) (Venkatesh et al., 2003) and the extended UTAT 2 framework (Venkatesh et al., 2012), performance expectancy, effort expectancy, social influence, hedonic motivation, facilitating conditions, price value and habits are important factors that influence the behavioural intention to use a technological device and the acceptance of these devices.Age, gender and experience are assumed to be moderator variables in these relations (Venkatesh et al., 2012).Recent empirical research in the fitness and health technology context (e.g.smartwatches, wearable fitness trackers, wearable technology in healthcare) indicate a link between performance expectancy, effort expectancy, hedonic motivation, facilitating conditions and the behavioural intention to use a technological device (Beh et al., 2021;Cavdar Aksoy et al., 2020;Gao et al., 2015;Hoque & Sorwar, 2017;Reyes-Mercado, 2018).Using new technologies is also associated with different risks (loss of privacy, information quality etc.) (Huckvale et al., 2015;Lewis & Wyatt, 2014).Especially in situations (e.g. using a technological device the first time) which are characterised by risks, trust perceptions play an important role in the utilisation of these technologies.For example, trust perception towards the operating company of a location sharing application predicts the use of this location sharing application (Beldad & Citra Kusumadewi, 2015).

Technological Trust
According to Lee and See (2004, p. 54), trust can be defined as "the attitude that an agent will help achieve an individual's goals in a situation characterized by uncertainty and vulnerability".Technological trust perceptions are considered to predict behavioural intention to use a technology (Dhagarra et al., 2020;Gefen et al., 2003;Gu & Wei, 2021;Liu & Tao, 2022;Luarn & Juo, 2010).Users who have a lack of trust in devices will not use the device or will not use all functions of the technology (and will not exhaust the full potential of the technology), and this could lead to decrease of usage behaviour (Xu et al., 2014).
As the complexity of technological processes and devices increases, trust helps people to deal with this complexity (Lee & See, 2004).For example, users have to trust that the device provides reliable and valid information.Referring to the analytic process of trust formation (Lee & See, 2004), rational evaluations of the characteristics of the technological device may be used to build trust if the cognitive process there is less demanding.The analytic process does not take emotions and their impact on trust formation into account (Lee & See, 2004).When cognitive resources of the users are limited and/or the technological processes are complex, emotions are essential in building trust (Hoff & Bashir, 2014;Lee & See, 2004).Users' positive and negative experiences when using a technology, as well as their expectations as to how the technology will perform in future, shape their trust perception towards this technology (Lippert, 2007;Lippert & Ojumu, 2008).Furthermore, the transfer of previous experiences with similar or comparable technologies shapes individuals' trust perceptions when using a device for the first time.All these previous experiences are linked to positive or negative emotions.This underpins the role of previous experiences from similar situations or with similar technologies in initial trust building by transferring these experiences to the current technological device (swift trust) (Robert et al., 2009).Technological trust is also related to design characteristics of the device meaning (unique) functionality (Gu & Wei, 2021), perceived usefulness, ease of use (Corritore et al., 2003;Sarkar et al., 2020), organisational reputation of the producer of the device (Adebesin & Mwalugha, 2020), usability and motivational affordance (Rupp et al., 2018).Trust in the app developer (not in the technology itself) has an impact on the users' perception of the usefulness of fitness apps, while social norms and ease of use is linked to users' trust perceptions in the app developer (Beldad & Hegner, 2018).Research in the context of NFC mobile payment shows that trust in the firm has a positive effect on trust in the technology, which, in turn, is a predictor for behavioural intention to use this technology (Luarn & Juo, 2010).
Most research dealing with technological trust is using variable-centred approaches.Person-centred approaches could expand our knowledge about technological trust and its consequences and antecedents by adopting a more holistic perspective when studying the multidimensional construct of technological trust.The novelty of the present paper is to adopt a person-centred approach to understand if users differ in their trust perceptions towards a fitness device by analysing different trust indicators (privacy, reliability etc.) in combination.
We want to answer the question if there are different technology trust subgroups of users and if they differ in their frequency using the wearable.Furthermore, the aim is to identify antecedents of trust group membership.Is effort expectancy, age, and the experience of the user with the fitness wearable is linked to specific trust patterns?The use of different trust perceptions (e.g.towards privacy, reliability; see "Materials and Method" section) reflects the complexity of technology trust perceptions.
Therefore, different trust perceptions were used.In this paper, we refer to the technological trust conceptualisation of Rupp et al. (2018) including trust towards security of one's personal data (privacy), validity of the data displayed by the device, reliability of the data provided by the device, system capability (relevant features and functionality of the device) and system transparency (understand the limitations as well as the methods of data processing of the device).It is assumed that users who have more experience with a device are characterised by higher trust perceptions towards the device than users with less experience.Referring to research in different technological contexts (Ejdys, 2018;Rupp et al., 2018) showing a link between usability aspects and technological trust, it is assumed that low effort expectancy is linked with higher trust perceptions.Furthermore, Wiesner et al. (2018) results indicate that older healthy active individuals (runners) are more concerned about privacy issues than younger individuals.Thus, the effect of age on trust will be surveyed.

Research Question 1
Which distinct subgroups (latent profiles) of wearable fitness tracker users regarding their technological trust perceptions (privacy, validity, reliability, system capability and system transparency) can be identified?

Research Question 2
Are previous experiences with the device (in months), effort expectancy and age significant predictors of technological trust profile membership?

Research Question 3
Are there significant differences between the technology trust subgroups regarding the frequency of using of the device in their physical activities?

Sample and Data Collection
Data from 150 fitness tracker users (male: 43.9%; female; 56.1%; age: 19-69 years) M age = 29.79;SD age = 9.63) were collected in a cross-sectional design using an online questionnaire.Individuals in fitness courses were asked to participate in the study.Furthermore, the link for the online questionnaire was distributed over social media and among university student and networks.

Measures
Technological trust: Technological trust was measured using the Wearable Technology Trust Scale (Rupp et al., 2018).This scale consists of five subscales using a 5-point rating scale (1 = totally disagree to 5 = totally agree) namely: • Privacy (security of personal data; example item: "I feel this device will keep my data secure") (3 items; α = 0.88) • Validity (information displayed on the device is correct; example item: "I feel this device is measuring what it says it measures") (3 items; α = 0.8) • Reliability (the device information is consistent over time; example item: "I feel this device will give me consistent readings over time") (4 items; α = 0.92) • System capability (the wearable has relevant features for the user; example item: "I feel this device has the functionality I need") (3 items; α = 0.87) • System transparency (transparency of methods to calculate information displayed by the device and knowing the limitations of the device; example item: "It is easy to follow what this device does") (3 items; α = 0.85) Effort expectancy (4 items; example item: "I find the tracker to be easy to use" α = 0.88; 1 = totally disagree to 5 = totally agree) was measured using an adapted four-item scale (Lee et al., 2015;Venkatesh et al., 2003).Furthermore, participants were asked for their experience with their device (in months), the frequency of usage of the wearable in their physical activity (0 = never to 100 = always) and sociodemographic information.

Analytic Approach
Variable-centred approaches like regression analysis or structural equation modelling focus on relationships between variables (Laursen & Hoff, 2006;Masyn, 2013)."These analyses identify processes found to a similar degree in all members of a group" (Laursen & Hoff, 2006, p. 379).Person-centred approaches are useful to analyse individual differences and similarities in heterogeneous populations by categorising individuals with respect to their patterns of individual characteristics (in the present study: trust perceptions) and differences in these patterns (Bauer et al., 2018;Bergman & Magnusson, 1997;Masyn, 2013).Based on the assumption that the population is heterogeneous, personcentred approaches (e.g.latent profile analysis) aim to identify subgroups of individuals called latent profiles or latent classes (Laursen & Hoff, 2006;Masyn, 2013;Vermunt & Magidson, 2002), which have similar (in this study) trust perception patterns (privacy, validity, reliability, system capability and system transparency) within a subgroup.Latent profile analyses (LPA) were conducted using Mplus 8 (Muthén & Muthén, 1998-2017).The number of latent profiles (subgroups) was increased from two to eight, and fit statistics were examined to find the solution with the best model fit.Common fit statistics like Bayesian information criterion (BIC) (Schwarz, 1978), adjusted likelihood ratio test (LRT) (Lo et al., 2001) and entropy value (Celeux & Soromenho, 1996) were used to evaluate the model fit.Smaller values on the Bayesian information criterion (BIC) indicate better model fit (Nylund et al., 2007;Nylund-Gibson & Choi, 2018).Furthermore, a significant LRT indicates that the k-profile solution has a better model fit than the k-1 profile solution (Nylund-Gibson & Choi, 2018).Entropy values (ranging from 0 to 1) provide information on classification accuracy.Larger entropy values show better classification accuracy (Celeux & Soromenho, 1996).
In a further step, auxiliary variables (antecedents of profile membership) were included (3STEP command in Mplus) to identify predictors (effort expectancy, experience with the device, age) for profile membership of the final LPA solution using multinomial logistic regression analysis.
The measurement model shows excellent model fit (χ 2 (153) = 166.653,p = 0.213; RMSEA = 0.025, SRMR = 0.069, CFI = 0.99, TLI = 0.99).Standardised factor loadings are between 0.66 and 0.92.The reliabilities of the scales are good (α between 0.8 and 0.92; see "Measures" section).Mean, standard deviation and intercorrelations of the variables are shown in Table 1.All trust perceptions show significant positive correlations with effort expectancy, which is in line with our previous assumption regarding associations between trust and effort expectancy.Effort expectancy explains notable 10 to 29% of the variance of the trust perceptions (r 2 ranging from 0.10 to 0.29).Age shows a weak but significant positive correlation with perception of reliability (r = 0.188, r 2 = 0.035, p < 0.05) of the wearable fitness device.Experience with the device (in months) was significantly correlated with system transparency (r = 0.234, r 2 = 0.054, p < 0.05) and effort expectancy (r = 0.204, r 2 = 0.042, p < 0.05).The lack of significant associations, and low significant correlation coefficients, between age and trust perceptions and experience with the device and trust perceptions will be explored in more detail by analysing associations between age, experience with the device and trust profiles (incorporating all trust perceptions).The frequency with which individuals uses the wearable in their physical activities show significant associations with perceptions of privacy (r 2 = 0.037, p < 0.05), reliability (r 2 = 0.048, p < 0.05), system capability (r 2 = 0.008, p < 0.01), effort expectancy (r 2 = 0.11, p < 0.01) and age (r 2 = 0.010, p < 0.01).

Latent Profile Analyses
The adjusted Lo-Mendell-Rubin likelihood-ratio-test and the BIC indicate that the three-profile solution shows the best model fit (see Table 2).Furthermore, the entropy value of 0.86 shows adequate classification accuracy of the three-profile solution (see Table 2).Analyses of variance and Kruskal-Wallis tests indicate significant differences between all latent profiles regarding the different trust perceptions (see Table 3).

Latent Profile 1 (Neutral to Low Privacy, Neutral to Moderate Validity, Reliability, System Capability and System Transparency Perceptions; n = 54)
Individuals classified in profile 1 are characterised by neutral to relatively low trust perceptions concerning privacy issues and neutral to moderate trust perceptions concerning the validity, reliability, system capability and system transparency trust factors of the wearable (see Figs. 1, 2 and Table 3).Figure 2 displays the standardised means (z scores).Profile 1 shows in all trust perceptions the lowest standardised means.Individuals assigned to latent profile 2 report moderate trust perception towards the privacy of their data and relatively high trust perceptions regarding validity, reliability, positive appraisal towards system capability (relevant features of the device) and system transparency, but not as high as in latent profile 3 (see Figs. 1, 2 and Table 3).The standardised means (see Fig. 2) show that this is the average profile.

Latent Profile 3 (High Trust; n = 45)
Profile 3 can be labelled "high technological trust perceptions".Individuals assigned to this latent profile report the highest technology trust perceptions in all factors, namely privacy, validity, reliability, system capability and system transparency (see Figs. 1, 2 and Table 3).

Predictors of Latent Profile Membership
In the next step, predictors of trust profile membership were examined.Results of the multinomial logistic regression (see Table 4) show that individuals scoring high on effort expectancy (high values mean low effort expectancy) are also more likely to be classified to the profile 3 (p = 0.000; OR = 48.327)and profile 2 (p = 0.002; OR = 4.112) than to profile 1.Individuals who have low effort expectancy (high values on this scale) are more likely to be classified in the "high trust perceptions" profile 3 and the moderate to high trust profile 2 (average profile) than in the profile 1 with the lowest trust perceptions of all profiles.Age and experience with the device (in months) were no significant predictors for profile membership.

Differences Between Latent Profiles in the Frequency with Which the Device is Used
Individuals assigned to profile 1 use the wearable less frequent when they are physical active than individuals assigned to profile 3 (see Table 5).

Discussion
Technological trust issues are key elements impacting the success of a technology.Especially in situations characterised by different risks (e.g.data privacy issues) (Beldad & Citra Kusumadewi, 2015;Pavlou, 2003) and in the case of complex technologies when users do not know or understand the data processing of the technological device (Lee & See, 2004), trust perceptions play an important role when using a technology.Previous work has shown that trust perceptions in a technology are associated with behavioural intentions to use the technology (Gu & Wei, 2021;Liu & Tao, 2022;Rupp et al., 2018).This leads to the question of antecedents of trust in technologies like m-commerce (Sarkar et al., 2020), ICT applications (Ejdys, 2018) or wearable fitness devices (Rupp et al., 2016(Rupp et al., , 2018)).The aim of this study was to identify distinct subgroups (latent profiles) of wearable fitness device users, based on a wide array of trust perceptions towards the device, by adopting a person-centred approach.Referring to technological trust as a multidimensional construct, various technological trust perceptions, namely privacy, validity, reliability, system capability and system transparency (Rupp et al., 2018), were used.Latent profile analyses demonstrate the existence of three distinct latent profiles: users who are characterised by neutral to low privacy perceptions (negative perception of data security), moderate validity, reliability, system capability and system transparency perceptions (latent profile 1); a user subgroup with moderate privacy perceptions, relatively high trust perception concerning reliability and validity issues and positive appraisal of system capability and transparency (latent profile 2); and users showing high trust perceptions in all scales (latent profile 3).In all latent profiles, perceptions of privacy have the lowest value compared to perceptions of validity, reliability, system capability and system transparency.Even though the trust perception of privacy was the lowest compared to the other trust perceptions within a subgroup, there was no extreme profile which was characterised by very low privacy perceptions and very high trust perceptions of reliability, validity, system capability and system transparency.It seems to be somewhat harder to build trust concerning privacy issues than to build trust concerning the other trust perceptions, but the privacy perceptions also increase between the profiles like the other perceptions.
The existence of these trust profiles raises the question of which factors are associated with these trust profiles.Recent variable-centred research in the context of different technologies (e.g.application at a university (see Ejdys, 2018); wearable technology (see Gu & Wei, 2021)) shows associations between technological trust and usability or functionality of the technology.In the present study, users' effort expectancy (high values mean low effort expectancy; example item: "I find the tracker to be easy to use"1 ) is a significant predictor of trust profile membership.Users who perceive that using the device does not require high effort and that using the tracker is easy are more likely to be classified in the high trust profile (profile 3) and the moderate to high trust profile (profile 2) than in the low to moderate trust profile (profile 1).These results are mostly in line with recent research (Ejdys, 2018;Gu & Wei, 2021;Lee & See, 2004;Liu & Tao, 2022) adopting a variable-centred approach and different trust conceptualisations.Our results provide further evidence of the connection between trust perceptions and effort expectancy.Furthermore, users which are characterized by high trust perceptions are using the wearable more frequently than users which are characterized by moderate perceptions towards validity, reliability, system capability and system transparency and neutral to low privacy perceptions (latent profile 1).This indicated the link between trust perceptions and usage behaviour.
It was also assumed that age and experience with the device (in months) shape trust perceptions.When using a device for a longer time, people collect more and more positive and negative experiences with the device, which should shape their trust perceptions (Lippert, 2007;Lippert & Ojumu, 2008).In the present study, age and experience using the device were not related to profile membership which was quite surprising.Most participants in the study (25th percentile = 23 years; Mdn = 26,5 years, 75th percentile = 34.2years) were relatively young and can be described as digital natives, which may be an explanation of missing age effects.One explanation for the missing effect of experience with the device on trust perception patterns could be that prior experience with a similar technology could also influence trust perceptions towards the technology when people have recently started using their current wearable fitness device (Robert et al., 2009).Furthermore, emotions play an important role in trust formation, especially when the technological processes of the device are complex (Lee & See, 2004).It could be assumed that experiencing positive emotions while using the device for a short period of time could also lead to high trust perceptions.
Trust perceptions towards the developer of the device could also influence trust perception towards the specific device (Luarn & Juo, 2010) without users having much experience using it.The social environment may also play an important role in building trust perception.For example, Beldad and Hegner (2018) show that social norms are predictors for users' trust in the app developer.Social interactions and media influences (recommendations of friends, social media or professional coaches in fitness clubs, influencer marketing activities, online etc.) may foster or hinder trust building without users having much experience with the device.Further research is needed to clarify these issues.Therefore, different stages of trust building as well as changes in trust perceptions over time of usage should be surveyed in longitudinal studies, especially when individuals start using a device.Influences from the social environment, which provide users with information regarding the device, should be incorporated into future studies of trust formation.Furthermore, future research could focus on more diverse age groups and participants with different education level to get more insights about the acceptance and use of wearable fitness technologies in different populations.

Limitations
The present study is not free from limitations.It focuses on the link between effort expectancy, age and experience with the device and technological trust in a cross-sectional design.Based on the data, we can just provide potential explanations for the reasons why experience with the device and age has no effect on trust.Using longitudinal designs with users who start using a device (as mentioned above) would be beneficial for studying trust formation.The present study does not incorporate technological skills and knowledge measures which could also be considered in future studies.

Table 2
Fit statistics of the LPA (n = 150) AIC Akaike information criterion, BIC Bayesian information criterion, LRT Likelihood ratio test, the threeprofile solution has the best model fit

Table 3
Means, standard deviations of the technology trust scales; ANOVA, Kruskal-Wallis tests and pairwise comparisons of the three latent profiles M estimated mean, SD estimated standard deviation; all scales use 5-point rating scale (high values display high trust perceptions) *p < .05;**p < .01;***p <.001; pairwise comparisons: Bonferroni or Mann-Whitney-U-Test a Kruskal-Wallis Test H(df); n=150

Table 4
Multinomial logistic  regression (n = 150)Profile 1: neutral to low privacy, moderate validity, reliability, system capability and system transparency; Profile 2: moderate privacy trust, relatively high trust perception in reliability and validity issues and positive appraisal of system capability and transparency; Profile 3: high trust profile in all trust perceptions; effort expectancy: 5-point rating scale expectancy (high values display low effort expectancy) Parameterization using reference profile 1; Estimate regression coefficient B, OR odds ratio, SE standard error

Table 5
Mean, standard deviation of the frequency with which the device is used; Kruskal-Wallis Tests and pairwise comparisons of the three latent profiles M estimated mean, SD estimated standard deviation *p < 0.05; **p < 0.01; ***p < 0.001 pairwise comparisons: Mann-Whitney U test; Kruskal-Wallis test were performed due to violations in normal distribution; n = 150 a Individuals were asked to rate the frequency with which they use the wearable in their physical activity ranging from never = 0 to always = 100; participants rated the frequency for different activities and the mean of all physical activities was calculated