In the initial development of the MPAS, factors that may be associated with mobile phone use (and more broadly technology use) were identified by conducting a search of the relevant literature using PubMed. Identified constructs included social connectivity, dependence, addiction, mood, and continuous use [21]. To better assess positive functions of mobile phone use, the same search procedures were used in this study to expand the MPAS to include constructs related to empowerment, safety, and usefulness in the domains of social and personal use, and work and school-related use. Individuals from our research team independently wrote a series of items for each of these three additional constructs in English at a sixth grade reading level. The entire research team reviewed the items to determine face validity. Any confusing or ambiguous items were edited and duplicates were deleted. Instructions and response format were also reviewed for clarity. The resulting instrument contained 57 items measuring 7 constructs, with 6-to-9 items per construct, and used a Likert-type scale response format. This initial draft of the revised MPAS was then pretested with eight adults to confirm item clarity and comprehension before administering the measure to a larger sample.

To test the MPAS, we conducted a national cross-sectional survey between March 25 and April 1, 2016, to assess respondents’ mobile phone use, attitudes toward their mobile phone, current mood, demographic information, and other characteristics. Participants were registered users (workers) of Amazon Mechanical Turk (MTurk). We used MTurk to recruit for this survey since research has shown it to be a fast, inexpensive, reliable, and useful approach to collect data from a large and ethnically diverse sample [22-26].

To participate in the study, workers were redirected through the MTurk website to our project survey website, which presented detailed information about the study and an informed consent document. Interested workers who indicated their consent were then linked to the screening questions to assess eligibility. Workers were eligible to participate in the study if they were 18 years of age or older, lived in the United States, could read fluently in English, and owned a mobile phone. If eligible, participants were then directed to the full survey, which asked questions about their mobile phone use, attitudes toward their mobile phone, their current mood, and demographic information.

The online survey was managed using a secure Web-based application, Research Electronic Data Capture [27], hosted in our institution’s Information Services department. Participants were compensated US $1 upon completing the survey. No identifying information was asked of participants, thus keeping their responses anonymous. The informed consent process, assessment of eligibility, and completion of the surveys took an average of 17 minutes to complete. This study was approved by the Institutional Review Board at The Miriam Hospital.

First, we used summary statistics (means, standard deviations, and frequencies) to describe the sample characteristics and measures for the entire sample. Second, a preliminary analysis was conducted to examine the MPAS item characteristics, and psychometric analyses were then conducted using confirmatory factor analysis (CFA) in Mplus Version 7.3 [32]. Model fit was evaluated based on the minimum fit function chi-square statistic, the comparative fit index (CFI; [33]), the nonnormed or Tucker-Lewis index (TLI; [34]), the root means square error of approximation (RMSEA; [35]), the standardized root mean square residual (SRMR; [36]), and the model chi-square. Respecifications to the model were guided by theory, in combination with modification indices that are part of the statistical output, and designed to produce a brief measure of mobile phone attachment that is suitable for use in field research. Third, Cronbach alpha [37], a measure of internal consistency reliability, was estimated for each subscale. Finally, to test for concurrent validity of the final measure, we examined the association between MPAS subscales and measures of anxiety, depression, and impulsivity.

Of the 1241 MTurk workers who were redirected to our survey website, 1.05% (13/1241) never completed the informed consent process, 5.32% (66/1241) of the potential participants completed the informed consent process but were deemed to be ineligible for the study, and 8.30% (103/1241) of the participants did not complete the survey. The data for one additional participant was corrupted and removed from the analyses.

Our analyses were restricted to the 1058 participants who agreed to participate in the study and completed all aspects of the survey. Participants were predominately white (877/1058; 82.89%). Men accounted for 49.91% (528/1058) of the participants, and women comprised 50.09% (530/1058) of the sample. Participants were between 18 and 87 years of age (mean 32.5, standard deviation 10.3). Table 1 provides the complete demographic information for the sample.

Preliminary item level analyses were conducted to examine individual item means, standard deviations, skew, and kurtosis in each of the 57 items, and the results were judged adequate to proceed with the dimensional analysis. An initial CFA model was fit using the full 57-item variable set and specified 7 correlated factors, with each of the items only allowed to load and be freely estimated on its hypothesized factor. This initial model fit poorly (χ²_1,518=9375.0, RMSEA=.07, CFA=.79, TLI=.78, SRMR=.071). The model fit was improved by removing items with poor loadings (<.4), very high cross-factor error correlations, or potentially high cross-factor loading, which indicated a complex item. One further adjustment included collapsing the initially posited separate factors of safety and empowerment into a single factor, based on the modification indices and conceptual similarity of the item content.

The final model (χ²₂₃₇=1100.9, RMSEA=.059, CFA=.941, TLI=.931, SRMR=.042) represented a parsimonious and balanced solution with 6 correlated factors, each measured by 4 items, creating a final measure consisting of 24 items (Multimedia Appendix 1). This 24-variable solution fit very well based on alternative criteria suggested by Hu and Bentler [38] and did not use any superfluous adjustments, such as freeing error covariance parameters or allowing variables to load on additional factors, to achieve the final improved model fit. Individual item loadings were high for all items on their respective factors (range=0.57-0.87, mean 0.75; see Table 2), and internal consistency reliability was also very good for each factor, as measured by Cronbach alpha (range=0.76-0.88, mean 0.83; see Table 2). The disattenuated correlations for the latent constructs are presented in Table 3.

Concurrent validity analyses examined correlations between each MPAS subscale and measures of depressive symptoms (CESD-10), anxiety (STAI), resilience (BRS), and impulsiveness (BIS). Depressive symptoms, anxiety, and impulsivity were significantly correlated with Anxious Attachment, Addiction, and Continuous Use subscales, but not with the Connectedness, Productivity, or Empowerment subscales. Resilience was significantly negatively correlated with scores on the Addiction and Anxious Attachment subscales and positively correlated with the Productivity subscale (Table 4).

Ultimately, we anticipate the value of the MPAS will be to provide predictive value for mHealth interventions, necessitating the assessment of whether scores on the MPAS (or any of its subscales) relate significantly to outcomes of interventions. However, this instrument was not developed in the context of an intervention study, and it is a limitation of this study that we did not assess other factors (eg, intentions to change health behaviors or outcome expectations) that may have served as a proxy for intervention outcomes. A second important limitation relates to the nature of the study sample. Although MTurk workers tend to be racially and ethnically diverse, they are likely to be more technologically savvy than most mobile phone users, and may not be representative of the overall US population in that regard. Thus, it is possible that the mean value of any individual MPAS subscale may differ between the MTurk study sample and other US population samples, but the high internal consistency reliability of each subscale, and the associated high item loadings on each subscale, indicate that the instrument has strong internal validity. We expect that this instrument will prove valuable for its intended purposes when used in other adult samples.

Additional work is needed to examine whether scores on the MPAS or its subscales are predictive of uptake, maintenance, and successful outcomes among individuals who are interacting with a behavioral health intervention delivered (in whole or in part) through mobile technology. It is our hope that the MPAS developed in this study may prove to be a useful indicator of the quality of the individual’s relationship with their mobile phone, and may comprise an important element in understanding the efficacy of mHealth interventions and programs.

There is tremendous growth in technologies that can provide health care and behavioral health interventions through mobile channels, such as apps and text messaging [16-18]. While the qualities of both the intervention and the delivery technology are important, in our quest to understand the effects that these interventions have on behaviors, it is also important to understand the relationship that the individual has with their mobile phone. mHealth involves not only health behaviors, but also those behaviors and attitudes relevant to interacting with mobile technology, and interacting with other people through mobile devices. Given the immediate and reciprocal nature of both health behavior and the behavior of interacting with a mobile device, thought leaders have suggested that our current behavioral health theories may be inadequate, particularly as mHealth interventions become increasingly interactive and adaptive [45]. Several papers have called for an expansion of our understanding of how interacting with a mobile device impacts health behavior [46,47]. Research is needed that can contribute to new theories regarding the interaction between mobile technology use, mHealth interventions, and behavior change.