Table 1.

Overview of included studies.

Authors	Design and Sample	Aims	Method	App used/developed Objective measurement of behaviour	Findings	Strengths/limitations
Choi et al. (2017)	41, 683 logs of 48 smartphone users collected from March 8, 2015 – January 8, 2018. For each participant, log data were collected for an average of 15.8 days. 48 participants from South Korea recruited by polling company Hankook Research, Inc. (aged between 20 and 39 years, mean age and SD not reported; 60.42% male) 25 participants were in their 20 s (control group = 11 and addiction group = 14), 23 participants were in their 30 s (control group = 12 and addiction group = 11). There were 29 males (control group = 17 and addiction group = 12) and 19 females (control group = 6 and addiction group = 13).	To derive usage patterns that were directly correlated with smartphone dependence from usage data, including apps and timeslots. Also to predict smartphone dependence through data-driven prediction algorithm.	Analysis procedure consisted of: 1. Collection of smartphone usage log data 2. Derivation of smartphone usage patterns via tensor factorisation (a reduction method to derive meaningful concepts from high dimensional data) 3. Prediction of smartphone dependence based on the patterns Data collected over period of ten months (March 2015- January 2016). Korean Smartphone Addiction Proneness Scale for Adults (S-Scale) and interview with psychiatrist and psychologist (using Korean version of Mini International Neuropsychiatric Interview (MINI) also implemented- used to classify control group and addicted group.	“Smartphone Overdependence Management System” (Developed) Supports only Android phones. Monitoring achieved through 4 main ‘sessions’. – For collection of mobile device usage data this was done in the ‘Sensoring & Monitoring Session’; Mobile data usage collected included general phone usage, e.g., when phone is turned on/off and general app data (internet, SNS and game monitoring) - exact usage time and period logs monitored through background app.	Usage patterns and membership vectors are effective tools for the assessment and prediction of smartphone dependence.	Limitations: The 6 indicators that were developed and used to assess smartphone overdependence were only developed for internet dependence. App used was only available for Android phones. Strengths: Tensor factorisation can obtain meaningful patterns from large-scale data.
Felisoni and Godoi (2018)	43undergraduate students from Business Administration of Fundaҫão Getúlio Vargas in São Paulo, Brazil (mean age and SD not reported; 46.5% male)	To investigate whether increasing smartphone usage among college students has a significant impact on their academic performance.	Both survey (personal information, self-efficacy while learning and usage perception) and objective data (through apps) collected. Questionnaire also included Self-Efficacy for Self-Regulated Learning (SE:SRL scale) to assess student ability to self-regulate in learning related activities. Academic performance for the college entry exam for each student was obtained through Undergraduate’s Office used as a predictor for academic performance in college Objective data collected across 14 days.	“Moment” (iPhone) and “App Usage Tracker” (Android) Usage time is only computed when cell phones are unlocked, therefore does not include time checking time/notifications. Data collected contained total minutes on phone each day for two week period. Average usage time subsequently calculated.	Significant negative relationship between total time spent using smartphones on academic performance. Average usage time for men = 217.7 min per day Average usage time for women = 240.7 min per day	Direct measurement of usage as opposed to relying on self-report data; allows observation of students’ natural behaviour during the day and to collect data unrelated to own bias. Allows automatic extraction of information from students’ regular routine with least intervention possible.
Lee et al. (2017)	35 college students enrolled at a public University in the Metropolitan region of northeast Asia (mean age 22.3, SD = 2.4, 68.57% male)	To examine the similarity and variance in smartphone usage patterns between measured and self-reported data.	Both survey (demographic information, smartphone addiction scale short version (SAS-SV) and smartphone usage patterns) and objective measure implemented. Objective data collected over 6 weeks.	‘Smartphone Addiction Management System’ (SAMS). (Custom app) Android phones. SAMS software runs in the background and measures which application, website or document is used. (Usage time, pattern and most used application types.)	Unconscious users underestimate their usage time. Findings show that there are significant cognitive biases in actual usage patterns in self-report of smartphone addictions.	Limitations: IT usage trends change rapidly, therefore continual and successive studies should be taken in a systematic way regularly. Ambiguity in time periods measured; definition of time period, e.g., evening/night can depend on personal life and culture. This should be clarified when considering life patterns of participants Strength: The app implemented demonstrated that there is significant cognitive bias in actual usage pattern and self-report of smartphone addictions; participants reported favourite app and usage time did not match with their most used ones. (It is proposed by the authors that underestimate of real usage time may suggest the development of tolerance)
Lee et al. (2018)	125 students; most of which attended computer classes (49% male) Age (mean and SD) and location not reported. 64 participants agreed to participate in objective data collection. These data were combined with results of addiction on Smartphone Addiction Scale (SAS) with the final dataset.	To analyse smartphone addiction by considering the differences between smartphone usage patterns as well as cognition.	A standardised smartphone addiction self-diagnosis scale was used as the smartphone addiction self-diagnosis scale (based on SAS). Objective data collected over a period of a month, twice a week.	‘How often do you use’ Android system. Data was collected from the following items: total usage time, usage time by day, data usage, number of screen turns, usage time by app, number of executions by app and frequently used apps.	Average smartphone usage based on results is more than 6hr a day. There is significant cognitive bias between self-reports and behavioural data. The higher the ‘recurrence’ item, the higher the addiction. The number of times screen was turned on and/cognitive time use had the greatest influence in higher risk users.	Combination of self-report and smartphone data can improve the accuracy of data and ensuring data reliability from respondents. Smartphone usage data is beneficial to be mined for useful correlations. Only 64 participants agreed to participate in objective measures from the original 125 respondents.
Lin, Lin, Lin, et al. (2017)	79 young adults recruited form the Department of Electrical Engineering and Department of Computer and Communication Engineering of two Universities in northern Taiwan (mean age = 22.4 years, SD = 2.3; 72.15% male).	1. To develop parameter needed to assess use/non-use reciprocity (i.e., screen off to screen on, which indicates impaired control for smartphone use). 2. To examine the predictive ability of smartphone use, non-use and use/non-use parameters when making a problematic smartphone diagnosis.	Predominantly based on App developed. Data recorded across at least 3 weeks. Psychiatrists also determined whether individual participants were smartphone addicts or non-addicts using criteria consisting of three parts: Criterion A: eight characteristic symptoms of smartphone use Criterion B: functional impairment caused by smartphone use, or that causes distress Criterion C: excluded addictive behaviours that accounted for obsessive compulsive disorder or bipolar I disorders.	App developed by authors, to support data collection on Android phones. Smartphone use parameters: screen on to successive screen off was defined as one epoch of use; the app calculated the average daily epoch count for one month as the use frequency parameter. Smartphone non-use parameters: event from screen off to screen on was defined as one epoch of non-use. Defined as maximal non-use epoch between 21:00 h and 12:00 h. Use/non-use reciprocity: two parameters introduced to assess the reciprocity between use and non-use patters- Roots Mean Square of the Successive Differences (RMSSD) and Similarity Index.	App-generated parameters were more associated with the App-assisted diagnosis than with psychiatric interviews alone. Frequency of use and non-use demonstrated identical prediction in relation to problematic smartphone use diagnosis.	Strengths: The high predictive natures of RMSSD and the Similarity Index imply that use/non-use reciprocity is validated with respect to the compulsive symptoms of problematic smartphone use. Limitations: Any smartphone use epoch is recorded as screen on to screen off by the app- however the app is unable to distinguish between proactive and reactive use, which may have resulted in non-use parameters being more accurate when predicting problematic smartphone use that the use parameters. Smartphone use and non-use were defined as screen on and screen off; this cannot wholly represent the status of smartphone use
Lin, Chang et al. (2015)	79 young adults recruited form the Department of Electrical Engineering and Department of Computer and Communication Engineering of two Universities in northern Taiwan (mean age = 22.4, SD = 2.3; 72.15% male).	To develop and validate proposed diagnostic criteria for smartphone addiction based on interviews with psychiatrists. To examine the relationship between smartphone addiction and the parameters generated by the app using EMD and criteria to excessive use, tolerance and time estimation. To test the differences between actual and self-aware usage time.	App recorded phone data across three weeks. Psychiatrist interviews also undertaken with participants, based on criteria of the Diagnostic Criteria of Internet Addiction for College Students (DC-IA-C) and Internet gaming disorder in DSM-5.	App developed by authors to support data collection on Android phones. App operates in background to record smartphone behaviours e.g. power on, program on, calls in/out, alarm clock, screen on/off, notifications.	Daily use count and frequency are associated with smartphone addiction (rather than duration). Self-reported time use was significantly lower than the recorded use via the app. Frequent short-period smartphone use may result in subjective distress or functional impairment. Excessive use- both frequency and duration - are part of smartphone addiction.	Limitations: Further information such as how many and what kinds of apps are used were not looked at. Data collection of one month may not be enough to allow for the detection of trends in some significant app-generated parameters. Strengths: Empirical mode decomposition (EMD) analysis was used, allowing for the ability of decomposing a complex series of smartphone use into a set of intrinsic mode functions.
Montag et al. (2015)	58 participants recruited through psychology and computer sciences classes (mean age = 24.22, SD = 5.02; 56.9% males)	To further support studies that indicate actual smartphone behaviour constitutes a better predictor for addictive tendencies that self-reported variables. To also investigate excessive mobile phone and smartphone behaviour.	App recorded phone data across five weeks. Participant also provided self-report of average mobile phone behaviour for a week, prior installing app, for comparison against actual usage and self-report – Mobile Phone Problem Use Scale (MPPUS)	Self-developed app ‘Menthal’-(non-private version, which presents no feedback to the user). Records behaviour such as incoming/outgoing calls, screen lock/unlocked and length of app use. Also recorded call and SMS related variables and computed means for average use of variable on a weekly basis. All events were associated with a user-ID and timestamp.	Weekly phone usage in hours was overestimated, while call and text message variables were underestimate. Associations between actual usage and addiction to mobile phones could be derived from recorded behaviour, but not through self-report variables.	Strengths: Overall patterns and correlations between recorded and self-reported variable and mobile phone addiction scores demonstrate recorded behaviour is more strongly associated with addictive tendencies-potential benefits in diagnostic process by direct tracking of behaviours. Limitations: Did not monitor activities in social networking sites that may have been more strongly associated with smartphone addiction compared to the present variables. First week of data discarded due to possibility results may have been influenced by being observed
Pan et al. (2019)	33 adult participants (mean age = 29.48, SD = 10.44; 84.84% male).	To illustrate the time periods or span of weeks required to reliably infer patterns of long term smartphone use. To investigate how long could a smartphone use cycle perpetuate by assessing maximum time intervals (i.e., weeks) between two smartphone use periods. To validate smartphone use and use/non-use reciprocity parameters.	Self-report; The 5 item Smartphone Addiction Inventory (SPAI = 5). To assess smartphone addiction. Objective measure data collection across two months.	The ‘Know Addiction’ database. (Custom app) Measured smartphone use patterns; predominantly frequency and use duration. Parameters developed; root mean square of successive differences (RMSSD), control index (CI) and similarity index (SI) to indicate impaired control and compulsive behaviours. Episode of smartphone use defined as a time period from screen on to the successive screen off. App calculated daily episode count as total use frequency, and total daily episode lengths were calculated as total use duration. Proactive use defined as one use episode without any notification within one minute before the screen on. Subsequently the proactive use frequency and proactive use duration were calculated.	Two week (bi-weekly) smartphone use is an adequate fundamental time unit to infer a two-month period of use. Significant correlation found between proactive use duration in two months and smartphone addiction; suggesting that self-reported smartphone addiction may correlate with long term duration than with short term use as indicated in previous studies. However this may be due to the adaptation of more app generated parameters within the current study.	Smartphone use episode was recorded as screen-on to screen off, providing an opportunity to distinguish between proactive and reactive use. The use/non-use parameters (RMSSD, SI and CI) allow the assessment of reciprocal patterns of smartphone use and may represent control ability of individuals; CI demonstrated better temporal stability than SI and RMSSD. Smartphone uses were defined as screen on/off, which cannot completely represent the status of smartphone use. Used a selective sample with excessive smartphone use; limited generalisability. Different smartphone use patterns may generate identical values on the temporal stability on use/non-use parameters. E.g. frequent, .long use periods spread out in short intervals may generate similar CI with sparse use period with sporadic checking.
Prasad et al. (2018)	140 undergraduate and postgraduate students from a tertiary care hospital were recruited in India (mean age = 22.89, SD = 2.79; 50% male)	To evaluate psychological correlates and predictors of excessive smartphone use with a telemetric (objective) approach.	Both psychometric tests (including the Smartphone Addiction Scale) and objective measures (three apps). Objective data collected across seven days.	‘Callistics’; ‘App Usage Tracker’; ‘Instant’ Android phone only. Callistics’; tracks number and duration of calls made and received from device. ‘App Usage Tracker’; tracks duration of minutes spent on all apps by the user- recorded in minutes and seconds. ‘Instant’; keeps track of duration in minutes spent on all apps by the user-recorded in min. It also provides the number lock/unlock cycle an individual has performed on the phone over a certain time-frame.	SAS score significantly predicted time spent on a smartphone in a seven day period. Psychological factors predict overall smartphone usage as well as usage on individual apps. Predictors for time spent on social networking sites were ego resiliency, conscientiousness, neuroticism and openness.	Limitations: Unwillingness of participants to install apps to track usage and reset WhatsApp usage statistics. Exclusion of iOS/Windows users.
Rozgonjuk et al. (2018)	101 college students recruited from a Midwestern, U.S. public university. (mean age = 19.53, SD = 4.31, 76.2% female)	To investigate how self-reported levels of PSU, depression, anxiety and daily depressive mood relate to objectively measured smartphone use over one week	Implementation of both psychometric test (SAS) and objective measure. Objective data collected over period of one week.	‘Moment’; Support iOS system only. Tracks usage of screen time (time phone screen is active and unlocked) and number of screen unlocks (unlocking phone).	Self-reported PSU was positively associated with the average minutes of screen time over a week, and that it positively predicted the minutes of screen time over a week in growth curve analysis. Phone screen locks could not be predicted from PSU scores. Self-reported PSU was not significantly related to the number of phone screen unlocks over a week.	Different types of smartphone usage measures e.g. screen time and screen unlocks could provide insight into PSU and negative mood from different perspectives. Time lag between web survey completion and participating in the week-long phone observation study, which may have influenced findings. Participants aware of smartphone usage being monitored, which may have increased self-criticism and self-monitoring in those with depression/anxiety monitoring, potentially influencing them to adjust smartphone usage downwardly over the study period.
Shin and Lee (2017)	195 undergraduate and graduate students from a university in Korea (age range 18–30 years, mean age and SD not reported; 63.59% males).	To discover the relationship between smartphone addiction diagnostic scale and smartphone usage patterns. To characterise smartphone addiction in terms of categorial usage patterns of smartphone, and to discriminate smartphone addicts from non-addicts.	Participant to install app and send average smartphone usage patterns to research. Also filled out modified version of the smartphone addiction self-diagnosis scale (S scale).	‘Smartphone Usage Tracker’ Android system only. Collects usage patterns; monitors the usage time of each individuals app and averages them to get the total usage time per day	Smartphone addiction is highly correlated with communication but not entertainment. Solely measuring total usage time is not enough to predict whether a smartphone user is addicted.	While smartphone usage is more accurate, it is limited in representing the multifaced nature of smartphone addiction. Usage time does not capture psychopathological symptoms, such as compulsive smartphone usage and interpersonal conflict, implying that measuring smartphone usage alone is not sufficient enough to predict smartphone addiction.
Tossell et al. (2015)	34 students from both a community college and university in Houston Texas. (mean age and SD not reported; 55.88% male)	To examine smartphone user behaviours and their relation to self-reported smartphone addiction through the use of both survey and telemetric data.	Quasi-experimental approach. Use of both survey (Smartphone Addiction Measurement Instrument (SAMI) and Internet Addiction Test and objective measure. Objective data collected over one year.	‘LiveLAb’ (Custom developed) Data captured every night. Data that was collected included all application launches, the duration of application launches, and when the application launches occurred (i.e. date/time stamps). Further information such as how many texts were sent/received and URL’s visited on Safari, was also collected.	Addicted users demonstrated differentiated smartphone use as compared to users who did not indicate addiction. Addicted used spent twice as much time on their phone and launched application almost twice as often compared to the non-addicted user; mail, messaging, Facebook and the Web drove this use. Addictive users showed significantly lower time-per-interaction than non-addicts for the above apps.	The telemetric use data provides more depth and precision than typical survey = based research and helps to mitigate small sample sizes.
Wilcockson et al. (2018)	27 students and staff from the University of Lincoln (mean age = 22.52, SD not reported; 62.96% female)	To examine how much time should be spent measuring mobile phone operation to reliably infer general patterns of usage and repetitive checking behaviours, and whether self-report measures of problematic smartphone use is associated with real-time patterns of use.	Both psychometric test (Mobile Phone Problem Use Scale; MPPUS) and objective measure implemented. Objective data collected across 14 days.	Custom developed app through Funf in a Box framework. Android only. Provided timestamp when the phone became active, and a second when the activity stopped and phone became inactive- primarily that involved screen use, but also included processor intensive activities, e.g., calls and playing music. Two behavioural measures were generated by the end of the day: total hours of usage and the frequency of use. (Total hours of usage determined by the amount of time the phone was active, whilst frequency of use was measured by the number of smartphone checks.)	Smartphone usage collected for a minimum of five days will reflect typical weekly usage in hours, but habitual checking behaviours can be reliably inferred within two days. Objective measures did not reliably correlate with self-reported measure.	Relatively little data is required to quantify typical usage for longer periods of time. The first day of data collection was removed due to participant time differences when the app was installed, which may have implicated the inference of typical behaviour.
Ellis et al. (2019)	238 participants recruited from Lancaster, Bath and Lincoln universities and via Prolific Academic (mean age = 31.88, SD = 11.19, 52.10% female).	To compare the accuracy of ten smartphone usage scales and single estimates against objective measures of smartphone behaviour.	Self-report estimate on number of hours/minutes spent on smartphone daily, in addition to number of notification received daily and how many times they pick up their device each day. Psychometric tests (Mobile Phone Problem Use Scale; MPPUS, Nomophobia Questionnaire; NMP-Q, Possession Incorporation in the Extended Self, Attachment Scale, Smartphone Addiction Scale; SAS, Smartphone Application-Based Addiction Scale; SABAS, Problematic Mobile Phone Use Questionnaire; PMPUQ, Media and Technology Usage and Attitudes Scale; MTUAS, Smartphone Use Questionnaires (SUQ-G&A). Objective measure implemented; data collected from a period of one week.	Apple’s Screen Time App. iOS system only. Measure of number of hours and minutes spent on phone, number of notifications received and number of times device picked up.	Correlations between psychometric scales and objective behaviour are generally poor. Single estimates and measures that attempt to frame technology use as habitual as opposed to addictive correlate more favourably with subsequent smartphone behaviour.	Behavioural measures utilised were limited; use of daily tracking as opposed to finer temporal measurements based on hourly patterns of usage. System used allows participants to view their own data in real, which may have implicated correlation between self-report data and objective measure.
Elhai et al. (2018)	68 college students from a Midwestern, U.S. university (mean age = 19.75, SD = 2.03, 64.70% female).	To examine smartphone use over the course of one week by employing a repeated measures design that allowed for direct tests of associations between depression severity and emotion regulation, in addition to the correlates involved in increased and problematic smartphone use.	Both objective measure and survey implemented: (self-report on frequency of smartphone features, Smartphone Addiction Scale-Short Version; SAS-SV, Patient Health Questionnaire-9; PHQ-9, Emotion Regulation Questionnaire; ERQ). Objective measure collected across one week.	‘Moment’; Support iOS system only. Measures screen time actively used daily (time that phone is locked is not included)	Lower depression severity predicted increased smartphone use over a period of one week. Greater use of expressive suppression as an emotion regulation strategy predicted more baseline smartphone use, but less smartphone use during the week.	Strengths: Moment app ran in the background, therefore it is possible that that participants did not think/forgot that their smartphone use was being monitored, subsequently maintaining their regular use over the course of the week without bias or influence. Limitations: Similarly, participants were aware that their smartphone usage was being monitored, which may have implicated their smartphone use behaviour. Due to limitations on Moment app, data on specific types of smartphone features used over the week were not acquired.
Giunchiglia et al. (2018)	72 undergraduate students from the University of Trento, Italy (mean age and SD not reported; 61.1% male).	To define new metrics in representing social media use and using smartphones to both track app usage and to administer time diaries. To employ both time diaries and smartphone data to establish the correlation between social media usage and academic performance.	Objective data collection and time diaries through application used. Academic performance assessed with two measures: Grade Point Average (GPA)- the average grade of point student obtained during the semester. Represents qualitative dimension of academic performance. Credito Formativo Universitario (CFU) - course credits obtained by students for each exam taken. Represents quantitative dimension of academic performance. Data collection across two weeks.	‘iLog’ (Custom developed) Both data collection from multiple sensors (hardware- GPS, accelerometer, gyroscope) and software (in/out calls, apps running on device) and time diaries, consisting of three-sub questions on activities, location and social relations of students every 30 mins. Data included social media app usage (most used), screen status information (collection of data of apps that are running at the time at which they are running) and academic performance.	Social media app usage during academic activities (in terms of session and duration) is negatively associated with student academic performance.	Limitations: small time frame of two weeks. However in regards to time diaries this is more than usual (one week) allowing a bigger window to extract patterns from through this data Strengths: three different parameters defined (social media, usage and academic performance) - distinction allows capturing different types of usage patterns.
Lee, Lee, Ko, Lee, Kim, Yang et al. (2014)	95 college students from university in Korea (mean age = 20.6, SD = 1.7, 70.5%)	To identify the usage patterns related to smartphone overuse and to provide several guidelines to facilitate the design of intervention software.	Survey (Smartphone Addiction Proneness Scale for Adults) and interview implemented in addition to objective measure. Data collected across an average of 27 days.	‘SmartLogger’ Custom app. Android only. Logs active/inactive apps, touch and text input events, web browsing URLs and notifications, power on/off, screen on/off, calls and SMS.	Compared to non-risk group, risk group has longer usage time per day and different diurnal usage patterns. Risk group more susceptible to push notification and tend to consume more online content. Usage time and frequency correlated to smartphone overuse.	Fine-grained usage features such as session time distribution exhibited consistent patterns across datasets. Allowed for unobtrusive monitoring that has minimal impacts on user behaviour.
Shin and Dey (2013)	48 participants recruited through local university community and Android market place (mean age = 26.7, SD not reported; 70.83% male).	To explore and automated, objective and repeatable approach for assessing problematic smartphone usage.	Psychometric assessments of addiction based on Mobile Phone Problematic Use Scale (MPPUS). Individual interviews. Objective measure. Objective data collected over a period of 25.1 days	Custom app. Android only. Collected sensory data, including apps that were installed and in use, battery usage, events and notifications and screen status data. Also extracted usage features of smartphones such as battery usage, network data usage, session usage (interval between screen turning on/off (a session indicates a unit of usage that involves app and event usage), app usage, touch inputs and push event usage (events sent form apps, e.g., new incoming SMS or email, upcoming events from calendar).	The number of apps used per day, ratio of SMSs to calls, event-initiated sessions, number of apps used event initiated session and length of non-event initiated sessions are useful in detecting problematic smartphone usage.	Strengths: Since the detection approach for problematic usage implemented is objected and automated, it can be repeated as frequently as desired. Also low inconvenience for the user and can detect problematic use after behaviour is exhibited. Limitations: Limited to Android users. Observation deemed relatively short (3.5 weeks average) - long term data may be more insightful in terms of changes in usage depending on context.