Institutes such as public health offices, statistical offices, ministries, and municipalities measure their population’s level of physical activity (PA) and sedentary behavior to investigate public health and develop policy accordingly. PA is defined as any bodily movement produced by the skeletal muscle that results in energy expenditure (Prince et al. 2008). Traditionally, research of PA is based on survey data. However, survey data are known to suffer from errors related to representativeness and measurement error (Ferrari, Friedenreich, and Matthews 2007; Fruin and Rankin 2004; Helmerhorst et al. 2012; Sallis and Saelens 2000; Shephard 2003; Welk et al. 2007; Wijndaele et al. 2015). Response rates for surveys are typically low and are declining over the past couple of decades (Luiten, Hox, and de Leeuw 2020). Response rates for objective measures such as accelerometer data are probably not going to be better due to high respondent burden. In addition, PA in surveys is often overestimated due to socially desirable responding (Adams et al. 2005; Helmerhorst et al. 2012; Prince et al. 2008; Warnecke et al. 1997; Welk et al. 2007).
In addition to surveys, PA can also be measured with accelerometer data generated by activity trackers (Ward et al. 2005). Activity trackers measure PA objectively by, for instance, counting steps, heart rate, etc. Accelerometer approximations can be applied to general daily activities by calculating the metabolic equivalent of tasks (METs value). The MET value is calculated by dividing the oxygen uptake by the mass of a person in kilogram times 3.5 (Mortazavi et al. 2013). Some people have their own activity trackers (e.g., Fitbit or smartwatches). The quality, algorithms, and type of parameters available differ per device. There are also high quality activity trackers where one can get the raw data. These activity trackers can be expensive. For research purposes, these trackers have to be lent out and sent back by respondents. In general, it is relatively unclear to what extent respondents are willing to wear activity trackers for research purposes or to what extent activity trackers can replace surveys from a data quality perspective. Furthermore, it is unclear to what extent the data of personal activity trackers could be useful in measuring PA.
In this paper, we use the probability-based I&O Research Panel to investigate to what extent the Dutch general population is in possession of personal activity trackers and monitors PA on a daily basis. We compare PA of people with and without personal trackers to find out if there are differences in their PA. In addition, we investigate the willingness to wear a professional high quality tracker (ActivPAL) for a week. In addition to these panel results, we present results from a follow-up study of panel members who actually wore the ActivPAL for a week.
This study should enable researchers to make more considered choices regarding methods for studies on PA as well as the use of sensor data for research purposes.
The survey was conducted in the I&O Research Panel. This is a probability-based online panel of people in the Netherlands. Internet penetration in the Netherlands is about 98% (Central Bureau of Statistics 2018). The general questionnaire consisted of a questionnaire on PA (in Dutch) and was administered June 2-8, 2020. See Appendix A for the full questionnaire. We randomly divided respondents into two versions of a health survey: the SQUASH questionnaire and I&O’s own survey on PA.
In addition, we asked questions about possession of activity trackers (smartwatch and activity tracker, such as Fitbit or app on phone); questions about willingness to share data of one’s own device by copying daily information in the survey or by uploading a data file to a research portal; questions about the willingness to wear a professional loaned device and questions about conditions under which people would be willing to participate in research wearing a professional loaned device. A total of 5,000 I&O panel members were invited to complete ‘a questionnaire on physical activity’; 2,276 panel members completed the survey (see Appendix B for descriptives of the gross and net sample for gender, age, and education). In a follow-up on the survey, 80 respondents who indicated their willingness to wear a professional device were invited to wear an ActivPAL accelerometer, which is worn on the thigh. We chose a sample of 80 due to availability of the ActivPAL accelerometers. The follow-up study served as a pilot for investigating feasibility and logistics of lending a professional device to measure physical activity. The follow-up study was fielded from June 22 to July 12, 2020. Selection for the follow-up study was based on respondent’s possession of a personal device. Half of the sample (40) were in possession of a personal device to measure PA, the other half (40) did not own a personal device to monitor PA. In addition, we made sure we had a heterogenous sample based on gender, age, and education. Consenting respondents in the follow-up study were sent the accelerometer plus instructions on how to position the device on the body. Respondents were asked to wear the accelerometer for one week and send the accelerometer back afterwards. An incentive of 20 euros was used in the follow-up study. (Respondents in the general survey received an incentive through the standard point system used by the panel.) No reminders were sent. In the follow-up study, we compare ActivPAL measures of respondents who own a personal device to monitor PA and respondents who do not own a device. Respondents’ adherence to the Dutch activity guidelines in terms of minutes per week moderate and vigorous activity, number of days of bone and muscle strengthening activity, and number of days of balance inducing activity were calculated and compared.
3.1 Participation rates
As shown in Table 1, 49.2% of the respondents in the survey use a smartwatch, an activity tracker (e.g., Fitbit), or an app on their phone to measure PA. A total of 58.0% of the respondents who use such a device are willing to copy personal PA data in a questionnaire while 40.6% of activity tracker users are willing to upload their data to a research portal. In total, 48.7% of the respondents are willing to wear a loaned accelerometer for a week.
From the respondents who indicated their willingness to wear a loaned device for a week (n=1,109), 80 respondents were invited to participate in the follow-up study to wear the loaned accelerometer for a week. Once we approached them and requested their actual participation in the follow-up study, 48 (60% of the 80 invited) were willing to actually participate in the follow-up study and wear the loaned accelerometer. In the end, 45 (56.3%) of the 80 invited respondents actually wore the accelerometer for at least 4 days.
3.2 Reasons to reconsider participation
Respondents noted multiple reasons to reconsider participation in the follow-up study. The saliency of the research topic is most prevalent; 159 respondents (29.1%) would reconsider participation if they were totally convinced of the purpose of the study. In addition, 134 respondents (24.5%) would reconsider participation if they were convinced the study to be totally anonymous. Other reasons were if given a higher incentive (n=69; 12.6%, minimal amount mentioned is 40 euro), if given advice on physical activity (n=68; 12.5%), if given feedback on physical activity (n=51; 9.3%), if they would move like they normally do (n=29; 5.3%), if they were younger, slimmer, or not ill (n=27; 4.9%), and various other reasons (n=145; 19.8% of all answers given), ranging from ‘if it doesn’t give me a skin rash’ to ‘if I can combine it with work’, to ‘I already move enough’ to ‘I can’t be bothered.’
3.3 Willingness to wear the accelerometer
Respondents who own a smartwatch or activity tracker such as Fitbit were more willing to wear the accelerometer than respondents who use their phone or do not monitor their PA (61.8%, 50.4%, and 42.3%, respectively; means differ significantly according to LSD post-hoc tests, F(2,2275) = 27.4 p < .001).
3.4 Willingness to copy or upload data
Respondents who own a smartwatch or activity tracker to measure their PA are more willing to copy their data into a questionnaire than to upload the data to a research portal. In addition, respondents who own a smartwatch or activity tracker are more willing to copy or upload their data than respondents who use their phone to monitor their PA (see Table 2).
Chi2 analyses showed that in both willingness to copy data and upload data, smartwatch and activity tracker owners were significantly more willing than respondents who only used their phone to monitor their PA (Chi2 (1) = 14.40, Cramèr’s V = .128, p<.001 and Chi2 (1) = 22.29, Cramèr’s V = .165, p < .001), respectively).
As shown in Table 3, respondents who own a smartwatch or activity tracker are more willing to copy their data into a questionnaire than respondents who monitor their PA on their phone. In addition, respondents who monitor their PA daily are more willing to copy their data into a questionnaire than respondents who monitor their PA less frequently. Respondents with an activity tracker are more willing to upload their PA data than respondents with a smartwatch or phone. Finally, respondents with an activity tracker are more inclined to wear a loaned meter than respondents with a smartwatch or phone.
3.5 PA of respondents with or without own device
In total, 52.9% of respondents complies with the Dutch PA guidelines, a slightly higher percentage than in the general population: 51.9% (RIVM 2020). Monitoring one’s PA with own devices is significantly related to this compliance, as is shown in Table 4.
With the exception of adhering to the guidelines for balance, people who monitor their activity comply to all aspects of the guidelines to a higher degree than people who do not monitor their PA. People with a smartwatch or other activity tracker do so to a larger extent than people who use their smartphone, but this difference only reaches significance in the adherence to the complete guidelines.
This pattern is confirmed by the data of the follow-up study with the 45 respondents who wore the ActivPAL, half of whom did not have a personal device. Table 5 shows that respondents with a personal device take significantly more steps, walk longer times with high intensity, and expend more energy (MET) than respondents with no personal device. However, sitting, standing and lying time do not significantly differ between respondents with and without their own personal device.
4. Discussion and Conclusion
In this paper, we investigated the willingness of respondents to wear an accelerometer, the willingness to share data from their own PA meter, and the difference in PA between respondents with an own device and respondents with no device. The stated willingness of respondents to wear an accelerometer was high (48.7%). However, the number of respondents who actually wore an accelerometer for an acceptable number of days once invited was fairly low (n=45 out of 80 invites). The sample used in the follow-up study is small, however. The participation rate could be improved by providing more and better information and better procedures, e.g., to emphasize the saliency of the study or privacy. Furthermore, an important reason not to participate in wearing a loaned accelerometer was “I do not move enough (now).” It is important to take this reason into account in further research, because people who are willing to participate in accelerometer research do not seem to reflect the general population in PA. We note that the study was conducted during the Covid-19 pandemic, which could have affected levels of physical activity. Although there was no lockdown, people were advised to work from home.
This research shows that there is a clear difference in self-reported level of activity between respondents who own a device to measure their PA (smartwatch and activity tracker such as a Fitbit or phone) and respondents who do not own such a device. Data of the follow-up study show a clear difference in level of activity measured with the ActivPAL between respondents who own a device to measure their PA and respondents who do not own such a device. We found no difference in sedentary behavior such as sitting, lying and standing, however. Note that we report bivariate analyses in this article, which do not control for any covariates. We checked for age effects but did not find any.
The willingness of respondents to share their personal data by copying device data into a questionnaire is high. The willingness to upload the data in a spread sheet to a research portal is slightly lower. Respondents who use their own device more frequently (daily) are more willing to upload their data than respondents who use their own device less frequently (sometimes). Further research is necessary to investigate the difference between data of a professional accelerometer (such as ActivPAL) and the data of respondents’ personal devices (smartwatches and activity trackers), in order to decide what the optimal method is for measuring PA in the 21st century, considering aspects of data quality, costs, reliability, privacy, and usability: via survey, a respondent’s own activity device or a professional (loaned) activity tracker.
In addition, future research should investigate how to improve representativeness of samples in PA studies. The low participation rates in the follow-up study might be related to the unfamiliar device (ActivPAL). People might be more willing to participate in a study of this kind if they were asked to wear a regular device, rather than a professional one (which they are likely to be unfamiliar with). With consumer-level activity trackers, respondents are more in control, and monitoring is integrated more effectively into respondents’ lifestyle, which may increase willingness to participate in studies making use of these devices compared with professional devices (Loh et al. 2017). Scientific research often prefers research-based devices over commercial devices, however, since consumer companies develop their own equations and algorithms and work to improve them, making consumer level data less transparent. Another method that could increase response rates is the use of interviewers that can persuade respondents to participate, provide additional information, pick up the devices, etc.