The Association Between Problematic Smartphone Use, Depression and T Anxiety Symptom Severity, and Objectively Measured Smartphone Use Over One Week

Computers in Human Behavior 87 (2018) 10–17

Contents lists available at ScienceDirect

Computers in Human Behavior

journal homepage: www.elsevier.com/locate/comphumbeh

Full length article The association between problematic smartphone use, depression and T anxiety symptom severity, and objectively measured smartphone use over one week

∗ ∗∗ Dmitri Rozgonjuka,b, , Jason C. Levineb,c, Brian J. Halld,e, , Jon D. Elhaib,c a Institute of Psychology, University of Tartu, Tartu, Estonia b Department of Psychology, University of Toledo, Toledo, OH, USA c Department of Psychiatry, University of Toledo, Toledo, OH, USA d Department of Psychology, University of Macau, Taipa, Macau (SAR), People's Republic of China e Department of Health, Behavior and Society, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA

ARTICLE INFO ABSTRACT

Keywords: Problematic smartphone use (PSU) is associated with several types of psychopathology, such as depression and Psychopathology anxiety severity. However, the majority of studies reporting these associations have not used objective smart- Information technology phone use data and/or repeated-measures study design. Our aim was to investigate how self-reported levels of Mobile phones PSU, depression, anxiety, and daily depressive mood relate to objectively measured smartphone use over one Addictions week. We assessed objective smartphone use by daily minutes of screen time and number of phone screen Smartphone addiction unlocks. One hundred and one undergraduate university students participated. Bivariate correlations and latent Problematic smartphone use growth curve analyses showed that PSU severity related to screen time minutes, but not to phone screen unlocking. Depression and anxiety severity were not related to screen time minutes, but negatively correlated with frequency of phone screen unlocking. Additionally, daily depressive mood items did not, in general, predict objective smartphone use for the corresponding day. However, average daily depressive mood over one week positively correlated with PSU levels. These ﬁndings and their implications are discussed.

1. Introduction psychopathology (Contractor, Weiss, Tull, & Elhai, 2017; Elhai, Dvorak, Levine, & Hall, 2017), detrimental consequences in the workplace Problematic smartphone use (PSU) is characterized by excessive (Derks & Bakker, 2014) and in academic settings (Rozgonjuk, Saal, & smartphone usage with negative functional consequences. PSU man- Täht, 2018; Samaha & Hawi, 2016). Additionally, PSU is related to poor ifests in several symptoms resembling behavioral addiction (Billieux, sleep quality (Demirci, Akgonul, & Akpinar, 2015) and physical health Maurage, Lopez-Fernandez, Kuss, & Griffiths, 2015). In fact, earlier (Kim, Min, Kim, & Min, 2017). Studies have consistently found that PSU research used the term “smartphone addiction” to describe this phe- is related to specific types of psychopathology – in particular, depres- nomenon. It is debatable if smartphone “addiction” should be regarded sion and anxiety (Elhai, Dvorak, et al., 2017). However, the majority of as a genuine psychopathology construct in the (behavioral) addictions relevant studies measured smartphone use frequency either by im- framework, due to several debatable issues, such as inconsistencies in plementing a self-report methodology and/or cross-sectional study de- the theoretical framework and the conceptualization of excessive be- sign. haviors as behavioral addictions (Griffiths, 2017; Kardefelt-Winther Compensatory internet use theory (CIUT) could shed light on the et al., 2017; Sussman, Rozgonjuk, & van den Eijnden, 2017). Yet a explanation of these findings (Kardefelt-Winther, 2014). Specifically, considerable body of evidence demonstrates relationships between ex- this theory posits that different motivations drive excessive, or pro- cessive smartphone use and adverse daily life outcomes - suggesting blematic technology use. Among these motivations, people experien- that PSU as a construct could be evident and requires further research. cing stress and/or negative affect often engage in technology use to As mentioned, different studies have confirmed that PSU is asso- relieve emotional discomfort. Therefore, people experiencing affective ciated with several negative outcomes, such as various types of psychopathology symptoms, such as depression and anxiety, may

∗∗ Corresponding author. Humanities and Social Sciences Building, E21-3040, University of Macau, Av. da Universidade, Taipa, Macau, Taiwan, ROC. ∗ Corresponding author. Institute of Psychology, University of Tartu, Näituse 2, Tartu 50409, Estonia. E-mail addresses: [email protected] (D. Rozgonjuk), [email protected] (B.J. Hall). https://doi.org/10.1016/j.chb.2018.05.019 Received 2 January 2018; Accepted 14 May 2018 Available online 16 May 2018 0747-5632/ © 2018 Elsevier Ltd. All rights reserved. D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17 increase technology use to relieve their stress and regulate negative emotion; in fact, studies have tested this theoretical framework and Initial sample have found support for it with PSU (Elhai, Tiamiyu, & Weeks, 2017; N = 301 Wang, Wang, Gaskin, & Wang, 2015; Wolniewicz, Tiamiyu, Weeks, & Elhai, 2017; Zhitomirsky-Geffet & Blau, 2016). However, the majority of studies investigating PSU in relation to affective symptomatology have used participant self-reported smartphone usage and/or cross-sectional study design. This is an important Total sample with at limitation, as studies show that self-reported smartphone use does not least 50% scale items correlate robustly with objectively measured smartphone use (Boase & N = 300 Ling, 2013; Kobayashi & Boase, 2012). Elhai, Tiamiyu, Weeks, et al. (2017) measured daily minutes of smartphone screen time use over one week to correlate this measurement with psychopathology measures - depression and emotion regulation scale scores. They found that baseline depression severity and emotion regulation predicted objective Consent for smartphone usage over the week. The findings fit within the framework contacting of CIUT, as those who experience more distress may regulate such N = 256 distress by using their smartphones. However, the psychopathology measures in Elhai, Tiamiyu, Weeks, et al. (2017) were assesed only at baseline, and thus they could not assess the effect of daily changes in psychopathology on daily measured smartphone use. In addition, while iPhone users they investigated objective smartphone use, they did not examine the N = 213 relationships with PSU.

1.1. Aims of the study

There is a considerable amount of scientiﬁc literature reporting Newer iPhone users relationships between PSU and various mental health outcomes, based N = 209 on cross-sectional and self-report designs. In the current study, we focus on relationships between self-reported baseline PSU, depression and anxiety symptomatology, objective smartphone use, and self-reported daily depressive mood over one week. Our objective smartphone usage measures included daily minutes of phone screen time and number of Responded to phone screen unlocks over one week. We aimed to gain insight into follow-up e-mail three research questions: N = 127 1. How is self-reported PSU as a baseline measure related to objectively measured smartphone usage over one week? 2. How are baseline depression and anxiety symptom severity related Agreed to participate in to objectively measured smartphone usage over one week? phone use tracking study 3. How are daily depressive mood ratings related to objectively mea- N = 101 sured smartphone usage over a week?

These research questions are accompanied by hypotheses derived Fig. 1. The recruitment procedure. from the literature: to objective smartphone use during each corresponding day, as a means H1. Levels of PSU should be positively correlated with objectively measured of regulating mood. This hypothesis is novel, as it is tested by smartphone use over a week. Increased, habitual smartphone use is implementing an ecological momentary assessment methodology. associated with PSU (Oulasvirta, Rattenbury, Ma, & Raita, 2011), and Elhai, Tiamiyu, Weeks, et al. (2017) have shown that PSU is related to objectively measured smartphone use. The next hypotheses are derived from CIUT, positing that more 2. Method frequent technology use may be motivated by maladaptive coping with emotional distress: 2.1. Sample H2. Baseline depression and anxiety severity should correlate with The recruitment procedure is depicted in Fig. 1. 301 college stu- objectively measured smartphone use over a week. According to CIUT, dents were recruited from a large Midwestern, United States public emotional distress can drive individuals to frequent smartphone use in university for a web survey. One participant was excluded for only order to cope with their distress (Kardefelt-Winther, 2014). Using a responding to demographic items. Of the remaining 300 participants, smartphone may work as a stress-relieving activity. Indeed, Elhai, 256 students (85.3%) provided consent to be contacted for a follow-up Tiamiyu, Weeks, et al. (2017) found some evidence that baseline study for additional academic credit. depression scores are related to changes in objective smartphone The smartphone application (called Moment; Holesh (2017)) was usage over a week. used to objectively track smartphone usage. It was designed only for H3. Daily depressive mood ratings should correlate with objectively Apple's mobile operating system (iOS), so the sample was limited to measured smartphone usage over one week. This hypothesis, too, is iPhone users. Of 256 potential participants interested in the follow-up based on CIUT. It is expected that daily (depressive) mood is related study, 213 were current iPhone users. Four participants used an older

11 D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17 generation iPhone than the iPhone 6, and were excluded due to app 2.3.2. Demographics performance issues. Information about age, gender, race/ethnicity, year in college, and Of the remaining 209 participants, 127 responded to our follow-up employment status were asked. invitation. Of the 127 individuals, 101 (48.3% of the 209 invited) agreed to participate, while others declined for various reasons (e.g., 2.3.3. The Smartphone Addiction Scale (SAS) not interested in the follow-up, meeting could not fit with their sche- To measure PSU, we used the SAS developed by Kwon et al. (2013). dule, etc.). All 101 students (the effective sample) who agreed to par- The SAS consists of 33 Likert-type items (1 = not at all to 6 = very ticipate, completed the study. much), and can be used as a summed score to measure PSU severity. The effective sample of 101 students had an average age of 19.53 Internal reliability is very good, α = 0.97 (α = 0.93 for the current (SD = 4.31). Most were freshmen (n = 64, 63.4%) or sophomores sample), with convergent validity relative to other instruments mea- (n = 25, 24.8%). The majority were women (n = 77, 76.2%). 85 suring internet and smartphone use. We reworded several items into a (84.2%) were Caucasian, 12 (11.9%) African-Americans, 7 (6.9%) first-person voice for greater accessibility and consistency for partici- Hispanic/Latinos, and 3 (3.0%) Asian-Americans (responses were not pants, such as rewording “Missing planned work due to smartphone mutually exclusive). 49 participants (48.5%) were working part-time use” to “I missed planned work due to smartphone use” (Duke & (in addition to being in school), 8 (7.9%) were working full-time, and Montag, 2017). others were unemployed (n = 44, 43.6%). The sample size of 101 people is sufficient for latent growth curve modelling, as at least 100 2.3.4. Depression Anxiety Stress Scales (DASS-21) people is preferred (Curran, Obeidat, & Losardo, 2010). We used the shorter, 21-item version of the original 42-item DASS The effective sample (101 participants) was compared with web (Lovibond & Lovibond, 1995). DASS-21 items use a Likert-type scale survey participants not included in the follow-up (200 participants). (0 = did not apply to me at all to 3 = applied to me very much or most of Analyses of variance (ANOVA) for continuous outcomes (age, year in the time). We used summed depression and anxiety scores. The cut-off college) and chi-square analyses for categorical outcomes (gender, criteria for summed scores are the following: Normal (0–4 for depres- race/ethnicity, employment status) were used. There were no sig- sion, 0–3 for anxiety), Mild (5–6 for depression, 4–5 for anxiety), nificant differences between those who did and did not participate in Moderate (7–10 for depression, 6–7 for anxiety), Severe (11–13 for de- the follow-up (all ps > .05). pression, 8–9 for anxiety), and Extremely Severe (14+ for depression and 10+ for anxiety). Coefficient alpha is 0.97 for depression and 0.87 2.2. Procedure for anxiety (αs of 0.88 and 0.80, respectively, for the current sample), with convergent validity relative to other depression and anxiety scales Study participants were recruited in fall 2017 from the university's (Antony, Bieling, Cox, Enns, & Swinson, 1998; Brown, Chorpita, psychology department research pool (hosted on Sona Systems' web- Korotitsch, & Barlow, 1997). site), approved by the university's institutional review board. Those enrolling were routed to an online consent statement hosted on 2.4. Daily depressive mood items psychdata.com. No monetary compensation was provided, but partici- pation in the 20- to 30-min web survey earned required course research When inquiring about daily depressive mood, we first queried a points. After completing the survey, participants were asked to take unique identification number chosen by participants to match their part in a follow-up study (for additional research points), and if willing, daily depressive mood and other data. Two items measured daily de- their contact information was solicited. pressive mood, assessing depressive symptoms from the previous day. For the 210 participants with a newer generation iPhone, we con- We used the following instruction and rating scale: To what extent did tacted them for the follow-up study. Those agreeable to participate you feel this way yesterday, using the following scale (1 = very slightly or were scheduled for in-person, ∼10-min appointments. Participants not at all,2=a little,3=moderately,4=quite a bit,5=extremely), were instructed beforehand to bring their iPhone, updated to iOS 10 or followed by: (a) Little interest or pleasure in doing things, and (b) Feeling higher, with at least 70 MB of free storage for app installation. down, depressed, or hopeless. Mood items are from the Patient Health The number of days between participants' web survey and ap- Questionnaire-9, or PHQ-9 (Kroenke, Spitzer, & Williams, 2001) – pointment date ranged from 0 to 19 (M = 8.74, SD = 4.57). During specifically, representing the abbreviated 2-item version (PHQ-2; scheduled meetings, participants were asked several non-meaningful Kroenke, Spitzer, Williams, and Lowe (2010)), a valid depression se- questions (e.g., shoe size) to match their new data and web survey data. verity scale. However, as the scale assesses symptom frequency over We assisted with Moment app installation, and demonstration (e.g., the two weeks, we used the symptom intensity rating scale from the Posi- app runs in the background and should not be force quit; location tive and Negative Affect Scale's, or PANAS (Watson, Clark, & Tellegen, services should not be disabled; airplane mode should not be enabled). 1988). The two items were summed for each day. Moment's notifications were turned off, except for the notification alerting that the app has been force quit. In addition, participants were 2.5. Data analysis asked to set phone auto-locking for the minimum 15-s duration, and screen to lock at night before sleep (if using an alarm or sleep tracking There was a minimal amount of missing item-level data, with app, so active usage time would not be counted). We sent text messages roughly 5–10% of participants typically missing 1–2 items per scale. (through clicksend.com's service) each morning (around 9:45AM) for Missing items were estimated with maximum likelihood estimation the next seven days querying two daily depressive mood items. After procedures, summed to form scale scores. Descriptive statistics and one week, we contacted participants to obtain their Moment data. correlations were computed in RStudio (version 3.2.3). Because some of the items violated assumptions for the traditional benchmarks of 2.3. Measures [−1;1] in skewness and kurtosis (Field, Miles, & Field, 2012), we used robust methods in data analysis, as detailed below. 2.3.1. The Moment app Next, we conducted latent growth curve analysis (also known as The Moment app tracks usage of one's iOS device (Holesh, 2017). repeated-measures multilevel modelling) (Raudenbush & Bryk, 2002) Specifically, the number of minutes of screen time (time the phone's in Mplus version 8 to investigate our research questions. PSU, depres- screen is active and unlocked) and number of phone screen unlocks sion, and anxiety severity were predictors of baseline smartphone use (unlocking the phone) are tracked. The app is valid in measuring minutes and changes in use over a week. In addition, daily depressive smartphone usage (Elhai, Tiamiyu, Weeks, et al., 2017). mood ratings were added as time-varying covariates for each day. Time

12 D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17 scores were specified as equidistant across days. Dependent variables correlated to phone screen unlocks for each day (with moderate ef- were treated as continuous. Latent intercept (baseline estimate) and fects), daily depressive mood ratings were not, in general, associated latent slope (growth over time) parameters were estimated. with objectively measured smartphone use. There were two outcome variables for measuring objective smartphone use over a week: daily minutes of screen time, and daily number 3.2. Latent growth curve analysis of phone screen unlocks. We followed the same modelling procedure for both variables. First, we tested the unconditional model with the out- First, unconditional growth models without covariates were tested come variable (either minutes of screen time or number of phone screen to assess changes in objective smartphone usage over the week in unlocks). Then, the time-invariant covariates (age, gender, PSU, de- minutes of screen time and the number of phone screen unlocks. Next, pression, and anxiety measures) were added, followed by inclusion of we included time-varying covariants (daily depressive mood items). time-varying covariates (daily depressive mood items for each mea- Finally, we also tested these models without the PSU measure. surement day). We included age and gender as covariates because women and younger people use their smartphones more (Elhai, Dvorak, 3.2.1. Minutes of screen time as outcome variable et al., 2017; Rozgonjuk, Rosenvald, Janno, & Täht, 2016; van Deursen, The time-invariant covariates were age, gender, PSU, depression, Bolle, Hegner, & Kommers, 2015). The estimation method was max- and anxiety measures. The model had a good fit, with CFI = 0.955, imum likelihood with robust standard errors (MLR), which corrects for TLI = 0.948, RMSEA = 0.056. non-normality (Yuan & Bentler, 2000). Both bivariate and multivariate coefficients for covariates pre- To evaluate model fit, we used traditional measures and bench- dicting the intercept (average baseline) and slope (average growth) of marks of adequate fit: Comparative Fit Index (CFI) ≥ 0.90, Tucker- screen time minutes over a week are presented in Table 3. Among the Lewis Index (TLI) ≥ 0.90, and root mean square error of approximation baseline psychopathology measures, the only signi ficant predictor was (RMSEA) ≤ 0.08 (Hu & Bentler, 1999). PSU – higher levels of PSU indicate more baseline screen time minutes. This relationship holds both in bivariate and multivariate analyses. 3. Results However, changes in smartphone usage (slope) over a week in terms of screen time minutes could not be predicted from the model's predictor 3.1. Descriptive statistics and correlations variables. Next, time-varying covariates (daily depressive mood ratings) were According to the DASS depression and anxiety cut-off scores dis- introduced to the model. The results are presented in Table 4. Similar to cussed above (Lovibond & Lovibond, 1995), the majority out of 101 the model with time-invariant covariates, PSU scores were the only fi participants scored in the normal range (67 people for depression and signi cant predictor of baseline smartphone usage in terms of screen 62 people for anxiety), with some scoring in the mild range (10 people time minutes. Age and gender did not predict screen time minutes. for depression and people 16 for anxiety), the moderate range (17 Finally, we re-conducted our models by excluding PSU as a pre- fi people for depression, 11 people for anxiety), the severe range (1 dictor to assess if ndings changed as a result. In both models, results person for depression and 3 people for anxiety), and extremely severe did not substantially change from the original models. As PSU was the fi range (6 people for depression and 9 people for anxiety). only signi cant predictor in the models described above, models fi Below, descriptive statistics and correlations between variables are without PSU did not have any signi cant predictors of minutes of presented (Table 1). screen time. PSU was significantly positively correlated to depression and anxiety measures, to average screen time minutes over a week, and 3.2.2. Number of phone screen unlocks as outcome variable average daily depressive mood ratings. However, PSU was not sig- The time-invariant covariates were age, gender, PSU, depression, nificantly related to averaged phone screen unlocks. and anxiety measures. The model showed a nearly adequate fit, Averaged phone screen unlocks over a week were significantly ne- CFI = 0.880, TLI = 0.860, and RMSEA = 0.130. Results are presented gatively related to anxiety and depression scores, whereas averaged in Table 3. While PSU did not predict baseline phone screen unlocks, screen time minutes were not related to these psychopathology mea- interestingly, both age and depression negatively related to baseline sures. phone screen unlocks in bivariate and multivariate analyses. Similar to Finally, in addition to being related to PSU, averaged daily de- results with screen time minutes, the slope or growth of phone screen pressive mood measures were also positively associated with depression unlocks over a week could not be predicted from these covariates. and anxiety measures, but not related to averaged objectively measured Controlling for daily depressive mood ratings as time-varying cov- smartphone use over a week. ariates, similar effects were present (see Table 5). Both age and de- In Table 2, correlations of objectively measured smartphone use and pression scores negatively predicted baseline phone screen unlocks. daily depressive mood items are presented. However, daily depressive mood also negatively predicted phone screen Results show that though screen time minutes were positively unlocks during Day 3 and Day 5 (the baseline was centered at Day 1).

Table 1 Descriptive statistics of measures, and Spearman correlations between the variables.

Descriptive statistics Correlations

min max M SD 1 2 3 4 5

1. PSU 41 146 93.998 24.075 ∗ 2. DASS-21 Depression 0 19 4.069 4.441 .250 ∗∗∗ ∗∗∗ 3. DASS-21 Anxiety 0 20 3.517 3.742 .346 .634 ∗ 4. Average minutes of screen time over a week 46.571 608.143 243.008 103.519 .231 .053 −.043 ∗∗ ∗ ∗∗ 5. Average phone screen unlocks over a week 16.857 332.000 88.197 46.671 .041 −.257 −.227 .318 ∗∗ ∗∗∗ ∗∗ 6. Average daily depressive mood over a week 1.571 6.857 3.388 1.175 .272 .400 .307 .027 −.048

∗ Notes. PSU = problematic smartphone use measured with the Smartphone Addiction Scale; DASS-21 = the 21-item Depression Anxiety Stress Scale. p < .05, ∗∗ ∗∗∗ p < .01, p < .001.

13 D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17

In this case, too, we re-computed our models but excluding PSU as a covariate. As in the case of our revised screen time minutes analyses, excluding PSU as a covariate minimally changed the results for screen unlocks. This similarity manifested in the signs and magnitudes of coeﬃcients and p-values. time time time time time time Day 5 min of screen time 4. Discussion

Numerous studies have shown that higher PSU levels are associated with several negative outcomes, and aﬀective disorder symptomatology, such as depression and anxiety (Contractor et al., 2017; Demirci et al., 2015; Elhai, Levine, Dvorak, & Hall, 2016; Elhai, Tiamiyu, et al., 2017; Kim et al., 2017; Samaha & Hawi, 2016). However, a majority of this work has been either cross-sectional or did not incorporate information about objective smartphone use. Our aim was to overcome these limitations and upgrade the literature in that regard. We found that self-reported PSU was positively associated with the average minutes of screen time over a week in bivariate correlation

∗ analysis, and self-reported PSU positively predicted the minutes of .049 Day 2 min of screen Correlations between daily depressiveminutes mood of and screen time .077− .042.103.197 .097.007 Day 1 min of screen Day 3 min of screen Day 4 min of screen screen Day 6 min of screen time Day 7 min of screen over a week in growth curve analysis. However, phone screen unlocks could not be predicted from PSU scores. Speciﬁcally, self-reported PSU was not signiﬁcantly related to the number of phone screen unlocks over a week in bivariate correlation analysis, nor did

Mood Mood Mood Mood Mood Mood Mood PSU predict the number of phone screen unlocks in latent growth curve models. These results partially confirm our first hypothesis. The findings, however, are interesting, suggesting that time spent using one's smartphone could give better insight into PSU than frequency of checking one's phone. Screen time minutes and phone screen unlocks are essentially two variations of measuring smartphone use. One could spend a considerable amount of time using a smartphone without a high number of phone screen unlocks – such as video/movie watching. Alternatively, one may have a high number of phone screen unlocks without having much phone screen time – such as constant phone checking, e.g. to see if there are any new e-mail or (social media) notifications. Also, one may have a high amount of phone screen unlocks if the main function of their smartphone is for communication, e.g., phone calls and text-messaging. Indeed, studies have suggested that the .021.089.130.051 Day 1 Day 3 Day 6 Day 7 Correlations between phone screendaily unlocks depressive and mood − .042− .121.006− − Day 2 Day 4 Day 5 relationships between types of smartphone use for social or non-social purposes could be differently related to both PSU and psychopathology symptoms (Elhai, Hall, Levine, & Dvorak, 2017; Elhai, Levine, et al., 2017). Our results indicate the possibility that objective smartphone use measures might reflect these differences in types of smartphone use and that there may be differential associations with PSU between these

p < .001. measurements of smartphone use. ∗∗∗ Day 1 phone screen unlocks Day 2 phone screen unlocks Day 3 phone screen unlocks Day 4 phone screen unlocks Day 5 phone screen unlocks Day 6 phone screen unlocks Day 7 phone screen unlocks Interestingly, depression and anxiety severity were not associated with average screen time minutes over a week in bivariate analyses, nor baseline or changes in objectively measured screen time duration over a

p < .01, week in latent growth curve analysis. Elhai, Tiamiyu, Weeks, et al. ∗∗ (2017) found similar results, although they found that depression symptoms predicted the growth of smartphone use over a week; in addition, that study reported only results with regards to minutes of p < .05, ∗ screen time but not the number of phone screen unlocks. We found that the average number of phone screen unlocks over a week was negatively correlated with depression and anxiety scores in bivariate analyses. In addition, depression scores predicted baseline phone screen unlocks, where higher depression scores correlated with less frequent phone screen unlocks in latent growth curve analysis. This finding ∗∗ ∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗ ∗∗∗ sheds light on the idea that baseline frequency of checking one's phone Correlations between minutes of screenphone time and screen unlocks .286 .359 .262 .398 .392 .335 .317 could be negatively predicted by proneness to negative affect – depression, specifically. These results are not in accordance with the compensatory internet use theory (CIUT; Kardefelt-Winther (2014)) that would relate more negative affect to more technology use. Alter- native explanation to this finding could be that people who are experiencing more depressive tendencies could attempt to relieve their time time time time time time time distress by not using their smartphone, instead engaging in behavioral Day 1 min of screen Day 3 min of screen Day 2 min of screen Day 4 min of screen Day 5 min of screen Day 7 min of screen Day 6 min of screen

Table 2 Spearman correlations between minutes of screen time, number of phone screen unlocks, and daily depressive mood items. Notes. Mood = daily depressive mood rating. avoidance or social isolation (De Silva, McKenzie, Harpham, & Huttly,

14 D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17

Table 3 Results of latent growth curve analysis with time-invariant covariates predicting objective smartphone use measures.

Covariate Intercept Slope

Bivariate B (SE) t Multivariate B (SE) t Bivariate B (SE) t Multivariate B (SE) t

Dependent variable: minutes of screen time over a week Age −3.730 (2.469) −1.511 −.153 (.109) −1.400 .129 (.281) .461 .056 (.123) .459 Gender −12.358 (29.762) −.415 −.051 (.121) −.420 3.709 (4.425) .838 .162 (.187) .865 PSU 1.030 (.488) 2.109a .235 (.112) 2.100b −.021 (.067) −.313 −.051 (.163) −.313 Depression 3.139 (3.707) .847 .132 (.157) .845 −.057 (.478) −.120 −.026 (.215) −.120 Anxiety −7.670 (4.238) −1.810 −.272 (.151) −1.801 .249 (.557) .446 .094 (.215) .436 Dependent variable: number of phone screen unlocks over a week Age −2.671 (.757) −3.527c −.258 (.099) −2.613d .053 (.071) .737 .127 (.205) .620 Gender 13.223 (9.705) 1.363 .129 (.091) 1.411 1.070 (1.452) .737 .260 (.382) .681 PSU −.133 (.242) −.552 −.072 (.127) −.567 −.014 (.026) −.561 −.195 (.337) −.579 Depression −2.633 (.907) −2.902e −.262 (.091) −2.882e .073 (.191) .382 .182 (.469) .387 Anxiety −1.007 (1.330) −.757 −.085 (.114) −.740 .179 (.256) .701 .377 (.632) .597

Notes. PSU = problematic smartphone use measured with the Smartphone Addiction Scale. ap = .035, bp = .036, cp < .001, dp = .009, ep = .004. Parameter estimation method in both models: robust maximum likelihood.

2005). These findings partially confirm our second hypothesis. mood did not relate to objective smartphone usage, suggesting that our Depression was related to screen time minutes, but not phone screen third hypothesis was not supported. This result may be explained by unlocks in latent growth curve models. Moreover, anxiety was not study limitations (discussed below). correlated with screen time minutes. However, averaged phone screen Slope (or the growth over the week) was not predicted by other unlocks over a week were negatively related both to depression and covariates. A potential explanation is that study participants were using anxiety in bivariate analyses – even though anxiety was not a sig- their smartphones so that they might have reached their typical nificant predictor in latent growth curve models predicting phone smartphone usage pattern in terms of screen time and phone-checking, screen unlocks. As our study participants were aware of their smart- and, therefore, expectation in systematic smartphone usage growth phone usage being monitored, monitoring itself could have increased over a week was not justified. Of course, it should also be noted that self-criticism and self-monitoring in those with more depression and/or tracking participants' smartphone usage and daily depressive mood anxiety severity (Dunkley, Sanislow, Grilo, & McGlashan, 2009), po- varied in terms of tracking starting day (e.g., some started on a tentially influencing them to adjust their smartphone usage down- Wednesday, and others on other days, such as Friday or Saturday) wardly over the study period. In addition, as mentioned earlier, it could which could also potentially explain these findings. be that those with higher depression and/or anxiety could have en- One of the main contributions of this paper was introducing re- gaged in social isolation or behavioral avoidance (De Silva et al., 2005), peated-measures design for objectively measuring smartphone usage in and, therefore, they had less motivation to check their phones for in- combination with self-reported daily depressive mood measures to coming e-mails, status updates from others, and notifications from so- previously collected web survey data. We found that different types of cial media. smartphone usage measures – screen time minutes and phone screen As a novel approach in studying PSU, we included daily depressive unlocks – could provide insights into PSU and negative mood from mood monitoring and objective smartphone usage data. Although different perspectives. The results suggest that the field of (problematic) average daily depressive mood ratings positively related to cross-sec- smartphone use could benefit by distinguishing between how much time tionally measured PSU measure, they did not predict objective smart- vs. how frequently does one use one's smartphone, as the former was phone usage. Even though the phone screen unlocks could be predicted related to PSU in this study, but the latter was not. We also found that by daily depressive mood during two days, in general, daily depressive daily depressive mood ratings are not necessarily related to objective

Table 4 Results of latent growth curve analysis with time-invariant and time-varying covariates (daily depressive mood measures) for predicting minutes of screen time.

Covariate Intercept Slope

Bivariate B (SE) t Multivariate B (SE) t Bivariate B (SE) t Multivariate B (SE) t

Unconditional model with covariates, dependent variable: minutes of screen time over a week Age −3.644 (2.482) −1.468 −.149 (.109) −1.367 .137 (.287) .477 .059 (.124) .475 Gender −12.924 (29.776) −.434 −.053 (.121) −.439 3.754 (4.378) .857 .162 (.183) .887 PSU 1.026 (.491) 2.091a .234 (.112) 2.083b −.023 (.068) −.345 −.056 (.163) −.344 Depression 3.148 (3.720) .846 .132 (.157) .844 −.121 (.489) −.247 −.054 (.217) −.247 Anxiety −7.709 (4.239) −1.818 −.273 (.151) −1.808 .228 (.584) .391 .085 (.222) .383

Bivariate B (SE) t Multivariate B (SE) t

Coeﬃcients of PHQ measures of corresponding days of measuring minutes of screen time Day 1 Mood regressed on Day 1 Minutes of Screen Time 2.161 (3.552) .608 .027 (.045) .604 Day 2 Mood regressed on Day 2 Minutes of Screen Time 4.041 (3.236) 1.249 .053 (.043) 1.226 Day 3 Mood regressed on Day 3 Minutes of Screen Time −2.114 (3.395) −.623 −.024 (.039) −.613 Day 4 Mood regressed on Day 4 Minutes of Screen Time 4.993 (3.620) 1.379 .059 (.042) 1.396 Day 5 Mood regressed on Day 5 Minutes of Screen Time 3.098 (2.932) 1.057 .043 (.041) 1.030 Day 6 Mood regressed on Day 6 Minutes of Screen Time 4.294 (4.170) 1.030 .061 (.059) 1.029 Day 7 Mood regressed on Day 7 Minutes of Screen Time 5.584 (4.631) 1.206 .081 (.067) 1.222 Notes. PSU = problematic smartphone use measured with the Smartphone Addiction Scale; Mood = daily depressive mood rating. ap = .037, bp = .038.

15 D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17

Table 5 Results of latent growth curve analysis with time-invariant and time-varying covariates (daily depressive mood measures) for predicting phone screen unlocks.

Covariate Intercept Slope

Bivariate B (SE) t Multivariate B (SE) t Bivariate B (SE) t Multivariate B (SE) t

Unconditional model with covariates, dependent variable: number of phone screen unlocks over a week Age −2.671 (.754) −3.541a −.259 (.068) −3.782a .053 (.070) .750 .110 (.162) .678 Gender 13.312 (9.643) 1.381 .130 (.091) 1.429 1.118 (1.429) .782 .235 (.306) .769 PSU −.127 (.241) −.528 −.069 (.127) −.540 −.015 (.026) −.571 −.176 (.300) −.586 Depression −2.568 (.903) −2.844b −.256 (.089) −2.890b .073 (.195) .374 .157 (.411) .382 Anxiety −1.012 (1.323) −.765 −.085 (.114) −.749 −1.012 (1.323) .712 .335 (.516) .650

Bivariate B (SE) t Multivariate B (SE) t

Coefficients of PHQ measures of corresponding days of measuring phone screen unlocks Day 1 Mood regressed on Day 1 Phone Screen Unlocks .060 (1.579) .038 .002 (.047) .038 Day 2 Mood regressed on Day 2 Phone Screen Unlocks .427 (1.205) .355 .013 (.037) .359 Day 3 Mood regressed on Day 3 Phone Screen Unlocks −3.034 (1.268) −2.392c −.091 (.035) −2.560d Day 4 Mood regressed on Day 4 Phone Screen Unlocks .535 (1.391) .384 .015 (.038) .402 Day 5 Mood regressed on Day 5 Phone Screen Unlocks −2.226 (.988) −2.252e −.078 (.036) −2.200f Day 6 Mood regressed on Day 6 Phone Screen Unlocks −.180 (1.053) −.171 −.006 (.038 −.170 Day 7 Mood regressed on Day 7 Phone Screen Unlocks .370 (1.410) .262 .012 (.047) .263 Notes. PSU = problematic smartphone use measured with the Smartphone Addiction Scale; Mood = daily depressive mood rating. ap < .001, bp = .004, cp = .017, dp = .010, ep = .024, fp = .028. smartphone usage. The clinical implication of this study could be the Curran, P. J., Obeidat, K., & Losardo, D. (2010). Twelve frequently asked questions about consideration that while excessive technology use is considered to be growth curve modeling. Journal of Cognition and Development, 11(2), 121–136. http:// dx.doi.org/10.1080/15248371003699969. related to psychopathology, it could be that those who actively check De Silva, M. J., McKenzie, K., Harpham, T., & Huttly, S. R. (2005). Social capital and their smartphones might actually exhibit better psychological well- mental illness: A systematic review. Journal of Epidemiology & Community Health, being that could be caused by (expected) social interaction. On the 59(8), 619–627. http://dx.doi.org/10.1136/jech.2004.029678. Demirci, K., Akgonul, M., & Akpinar, A. (2015). Relationship of smartphone use severity other hand, constantly using smartphone that adds up to more minutes with sleep quality, depression, and anxiety in university students. Journal of of screen time might not necessarily be associated with psycho- Behavioral Addictions, 4(2), 85–92. http://dx.doi.org/10.1556/2006.4.2015.010. pathology but could be an important factor in PSU measures. Derks, D., & Bakker, A. B. (2014). Smartphone use, work-home interference, and burnout: – The following limitations should also be mentioned to provide some A diary study on the role of recovery. Applied Psychology, 63(3), 411 440. http://dx. doi.org/10.1111/j.1464-0597.2012.00530.x. caution to interpretation of results. First, we asked about the mood for Duke, É., & Montag, C. (2017). Smartphone addiction, daily interruptions and self-re- the previous day, and text messages were delivered during the morning. ported productivity. Addictive Behaviors Reports, 6,90–95. http://dx.doi.org/10. It could be that our measures were not adequate due to recall bias. 1016/j.abrep.2017.07.002. Dunkley, D. M., Sanislow, C. A., Grilo, C. M., & McGlashan, T. H. (2009). Self-criticism Second, participants were instructed to respond to daily depressive versus neuroticism in predicting depression and psychosocial impairment for 4 years mood items as soon as they could; yet, participants responded to daily in a clinical sample. Comprehensive Psychiatry, 50(4), 335–346. http://dx.doi.org/10. depressive mood items during different times of the day, challenging 1016/j.comppsych.2008.09.004. Elhai, J. D., Dvorak, R. D., Levine, J. C., & Hall, B. J. (2017). Problematic smartphone use: the validity of results. Third, the participants were college students and A conceptual overview and systematic review of relations with anxiety and depres- the age variability was restricted due to the college-based age range, sion psychopathology. Journal of Affective Disorders, 207, 251–259. http://dx.doi.org/ which may add constraints to the generalizability of the findings to 10.1016/j.jad.2016.08.030. Elhai, J. D., Hall, B. J., Levine, J. C., & Dvorak, R. D. (2017). Types of smartphone usage other samples. Fourth, there was a slight time-lag between web survey and relations with problematic smartphone behaviors: The role of content con- completion and participating in the week-long phone observation sumption vs. social smartphone use. Cyberpsychology: Journal of Psychosocial Research study, and this could influence findings. Finally, although we used re- on Cyberspace, 11(2), http://dx.doi.org/10.5817/cp2017-2-3. Elhai, J. D., Levine, J. C., Dvorak, R. D., & Hall, B. J. (2016). Fear of missing out, need for peated-measures study design, it was only over the course of one week, touch, anxiety and depression are related to problematic smartphone use. Computers not longer (e.g., over months or years). Nonetheless, these results add to in Human Behavior, 63, 509–516. http://dx.doi.org/10.1016/j.chb.2016.05.079. the growing literature on psychopathology related to PSU, and con- Elhai, J. D., Levine, J. C., Dvorak, R. D., & Hall, B. J. (2017). Non-social features of tribute by including objectively measured smartphone use estimates smartphone use are most related to depression, anxiety and problematic smartphone use. Computers in Human Behavior, 69,75–82. http://dx.doi.org/10.1016/j.chb.2016. and daily depressive mood ratings. 12.023. Elhai, J. D., Tiamiyu, M. F., & Weeks, J. W. (2017a). Depression and social anxiety in References relation to problematic smartphone use: The prominent role of rumination. Internet Research, 28(2), 315–332. http://dx.doi.org/10.1108/IntR-01-2017-0019. Elhai, J. D., Tiamiyu, M. F., Weeks, J. W., Levine, J. C., Picard, K. J., & Hall, B. J. (2017b). Antony, M. M., Bieling, P. J., Cox, B. J., Enns, M. W., & Swinson, R. P. (1998). Depression and emotion regulation predict objective smartphone use measured over Psychometric properties of the 42-item and 21-item versions of the Depression one week. Personality and Individual Differences. http://dx.doi.org/10.1016/j.paid. Anxiety Stress Scales in clinical groups and a community sample. Psychological 2017.04.051. Assessment, 10(2), 176–181. http://dx.doi.org/10.1037/1040-3590.10.2.176. Field, A., Miles, J., & Field, Z. (2012). Discovering statistics using R. SAGE Publications Ltd. Billieux, J., Maurage, P., Lopez-Fernandez, O., Kuss, D. J., & Griffiths, M. D. (2015). Can Griffiths, M. D. (2017). Behavioural addiction and substance addiction should be defined disordered mobile phone use be considered a behavioral addiction? An update on by their similarities not their dissimilarities. Addiction, 112(10), 1718–1720. current evidence and a comprehensive model for future research. Current Addiction Holesh, K. (2017). Moment – screen time tracker. Pittsburgh, Pennsylvania. Retrieved from Reports, 2(2), 156–162. https://inthemoment.io. Boase, J., & Ling, R. (2013). Measuring mobile phone use: Self-report versus log data. Hu, L. t., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure Journal of Computer-mediated Communication, 18(4), 508–519. http://dx.doi.org/10. analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: 1111/jcc4.12021. A Multidisciplinary Journal, 6(1), 1–55. http://dx.doi.org/10.1080/ Brown, T. A., Chorpita, B. F., Korotitsch, W., & Barlow, D. H. (1997). Psychometric 10705519909540118. properties of the depression anxiety stress scales (DASS) in clinical samples. Behaviour Kardefelt-Winther, D. (2014). A conceptual and methodological critique of internet ad- Research and Therapy, 35(1), 79–89. http://dx.doi.org/10.1016/s0005-7967(96) diction research: Towards a model of compensatory internet use. Computers in Human 00068-x. Behavior, 31, 351–354. http://dx.doi.org/10.1016/j.chb.2013.10.059. Contractor, A. A., Weiss, N. H., Tull, M. T., & Elhai, J. D. (2017). PTSD's relation with Kardefelt-Winther, D., Heeren, A., Schimmenti, A., van Rooij, A., Maurage, P., Carras, M., problematic smartphone use: Mediating role of impulsivity. Computers in Human & Billieux, J. (2017). How can we conceptualize behavioural addiction without pa- Behavior, 75, 177–183. http://dx.doi.org/10.1016/j.chb.2017.05.018. thologizing common behaviours? Addiction, 112(10), 1709–1715.

16 D. Rozgonjuk et al. Computers in Human Behavior 87 (2018) 10–17

Kim, H. J., Min, J. Y., Kim, H. J., & Min, K. B. (2017). Association between psychological approaches to learning, and social media use in lectures. International Journal of and self-assessed health status and smartphone overuse among Korean college stu- Environmental Research and Public Health, 15(1), http://dx.doi.org/10.3390/ dents. Journal of Mental Health, 1–6. http://dx.doi.org/10.1080/09638237.2017. ijerph15010092. 1370641. Samaha, M., & Hawi, N. S. (2016). Relationships among smartphone addiction, stress, Kobayashi, T., & Boase, J. (2012). No such Effect? The implications of measurement error academic performance, and satisfaction with life. Computers in Human Behavior, 57, in self-report measures of mobile communication use. Communication Methods and 321–325. Measures, 6(2), 126–143. http://dx.doi.org/10.1080/19312458.2012.679243. Sussman, S., Rozgonjuk, D., & van den Eijnden, R. (2017). Substance and behavioral Kroenke, K., Spitzer, R. L., & Williams, J. B. W. (2001). The PHQ-9. Journal of General addictions may share a similar underlying process of dysregulation. Addiction, Internal Medicine, 16(9), 606–613. http://dx.doi.org/10.1046/j.1525-1497.2001. 112(10), 1717–1718. http://dx.doi.org/10.1111/add.13825. 016009606.x. van Deursen, A. J. A. M., Bolle, C. L., Hegner, S. M., & Kommers, P. A. M. (2015). Kroenke, K., Spitzer, R. L., Williams, J. B., & Lowe, B. (2010). The patient health Modeling habitual and addictive smartphone behavior. Computers in Human Behavior, Questionnaire somatic, anxiety, and depressive symptom scales: A systematic review. 45, 411–420. http://dx.doi.org/10.1016/j.chb.2014.12.039. General Hospital Psychiatry, 32(4), 345–359. http://dx.doi.org/10.1016/j. Wang, J.-L., Wang, H.-Z., Gaskin, J., & Wang, L.-H. (2015). The role of stress and moti- genhosppsych.2010.03.006. vation in problematic smartphone use among college students. Computers in Human Kwon, M., Lee, J. Y., Won, W. Y., Park, J. W., Min, J. A., Hahn, C., & Kim, D. J. (2013). Behavior, 53, 181–188. http://dx.doi.org/10.1016/j.chb.2015.07.005. Development and validation of a smartphone addiction scale (SAS). PLoS One, 8(2), Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief e56936. http://dx.doi.org/10.1371/journal.pone.0056936. measures of positive and negative affect: The PANAS scales. Journal of Personality and Lovibond, P. F., & Lovibond, S. H. (1995). The structure of negative emotional states: Social Psychology, 54(6), 1063–1070. http://dx.doi.org/10.1037/0022-3514.54.6. Comparison of the depression anxiety stress scales (DASS) with the beck depression 1063. and anxiety inventories. Behaviour Research and Therapy, 33(3), 335–343. Wolniewicz, C. A., Tiamiyu, M. F., Weeks, J. W., & Elhai, J. D. (2017). Problematic Oulasvirta, A., Rattenbury, T., Ma, L., & Raita, E. (2011). Habits make smartphone use smartphone use and relations with negative affect, fear of missing out, and fear of more pervasive. Personal and Ubiquitous Computing, 16(1), 105–114. http://dx.doi. negative and positive evaluation. Psychiatry Research. http://dx.doi.org/10.1016/j. org/10.1007/s00779-011-0412-2. psychres.2017.09.058. Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data Yuan, K. H., & Bentler, P. M. (2000). Three likelihood-based methods for mean and analysis methods (2nd ed.). Thousand Oaks, CA: Sage Publications. covariance structure analysis with nonnormal missing data. In M. E. Sobel, & M. P. Rozgonjuk, D., Rosenvald, V., Janno, S., & Täht, K. (2016). Developing a shorter version Becker (Eds.). Sociological methodology (pp. 165–200). Washington, D.C: Wiley- of the Estonian smartphone addiction proneness scale (E-SAPS18). Cyberpsychology: Blackwell. Journal of Psychosocial Research on Cyberspace, 10(4), http://dx.doi.org/10.5817/ Zhitomirsky-Geffet, M., & Blau, M. (2016). Cross-generational analysis of predictive cp2016-4-4. factors of addictive behavior in smartphone usage. Computers in Human Behavior, 64, Rozgonjuk, D., Saal, K., & Täht, K. (2018). Problematic smartphone use, deep and surface 682–693. http://dx.doi.org/10.1016/j.chb.2016.07.061.