Introduction

Smartphones are ubiquitous in people’s daily life. People use smartphones to communicate with others, search information and places, play games, listen to music, or watch videos nearly anywhere at any chosen time (Ariel et al., 2017; Stachl et al., 2017). These smartphone functions are convenient and offer many benefits for maintaining social connections, for getting work and other tasks done, as well as to entertain oneself. At the same time, (heavy) use of smartphones might also have disadvantages. For example, excessive social media use can have negative physical (e.g., sleep deficits), social (e.g., loneliness), and psychological (e.g., well-being) consequences (Griffiths et al., 2018). The current study examines both benefits and costs of smartphone use in multiple important life domains, comprehensively.

Using a broad theoretical approach to life consequences (Ozer & Benet-Martinez, 2006), the current study focuses on the individual domain (e.g., well-being and sleep), interpersonal domain (e.g., social contact), and the work domain (e.g., working hours and job satisfaction). These domains cover important aspects of humans’ daily life, often summarized as well-being, love, and work (Maslow, 1954). To match these domains, we focus on three aspects of smartphone use: general smartphone use; communicative smartphone use; and work-related smartphone use. Whereas the majority of previous work mostly focused on one area of either positive or negative consequences of smartphone use and mainly relied on convenience samples of adolescents and university students, the current study examines both benefits and costs in individual, interpersonal, and work-related outcomes in a representative sample from young to late adulthood. With additional machine learning analyses, we explore as yet undiscovered associations between smartphone use and the three domains of human functioning. Two research questions guide the current study:

  1. I.

    In which parts of their life do adults benefit from what functions of their smartphone and at what cost?

  2. II.

    Do the costs and benefits of smartphone use vary with adults’ age?

Literature Review and Derivation of Research Hypotheses

Most research on the positive or negative consequences of smartphone use has been guided, at least implicitly, by the enhancement or the displacement hypothesis (Hale & Guan, 2015; Liu et al., 2016): on the one hand, smartphones are thought to enhance daily life, to simplify communication, and to increase work productivity (Cavazotte et al., 2014). On the other hand, smartphones, and digital media in general, are assumed to affect the quality of life negatively because digital media use may displace more valuable activities like meeting friends or sleeping (Exelmans & Van den Bulck, 2016; Sbarra et al., 2019). Yet, when multiple life domains are considered simultaneously, it is likely that both the enhancement and the displacement hypothesis will find support: different types of smartphone use will likely show benefits in some domains, costs in others, and non-measurable effects in again other domains.

Individual Domain: Smartphone Use, Psychological Well-Being, and Sleep

Results regarding the associations between smartphone use and psychological well-being are often inconsistent, but overall point towards a quadratic relationship with both rare and heavy smartphone use being weakly associated with lower psychological well-being. Still, the discussion about digital media use mostly focuses on negative consequences on people’s lives. Accordingly, many studies have reported negative correlations between digital media use and psychological well-being (for a meta-analysis see Baker & Algorta, 2016), while some other studies found null or even positive effects (e.g., Granic et al., 2014; Przybylski & Weinstein, 2017). Several large-scale studies with adolescents thus observed non-linear associations with both very low and very high daily screen time being associated with multiple indicators of lowered psychological well-being and mental health (Orben & Przybylski, 2019; Przybylski & Weinstein, 2017; Twenge & Campbell, 2018). It has to be noted, though, that digital technology use explains only very little of the variation in adolescents’ well-being (Orben & Przybylski, 2019) and researchers are still debating the causal direction of these effects (Hartanto et al., 2021). Based on the findings of quadratic associations between general media use and well-being, we assume:

  • H1a. Adults who rarely or heavily use their smartphone report less psychological well-being than adults who moderately use their smartphone.

Reduced well-being in association with high levels of digital media use (i.e., screen time) might also (partly) be explained by reduced sleep duration and quality. Most adults who own a smartphone also report using it prior to bedtime (Exelmans & Van den Bulck, 2016). Importantly, smartphone use prior to bedtime is associated with delayed sleep onset, shorter sleep duration, and poorer sleep quality among children, adolescents, and adults (Exelmans & Van den Bulck, 2016; for a review among children/adolescents see Hale & Guan, 2015). Smartphone use might negatively influence sleep, because it is an unstructured, mostly pleasant leisure activity without predefined ending and thus difficult to terminate (Kubiszewski et al., 2013). In addition, exposure to bright screen light can suppress the secretion of melatonin, which regulates the wake–sleep cycle, thus delaying the sleep onset (Chellappa et al., 2013). In summary, smartphone use prior to bedtime is associated with reduced sleep duration and quality. As most adults who use their phone a lot during the day likely also use it before going to bed, we assume:

  • H2a. Adults who heavily use their smartphone sleep less and worse compared to adults with low to moderate smartphone use.

With regard to how smartphone use might relate to well-being and sleep in different age groups, one has to consider that, compared to older adults, younger adults use their smartphones more often and are thus more likely to develop problems because of their excessive smartphone use (Exelmans & Van den Bulck, 2016; Van Deursen et al., 2015). Accordingly, smartphone use seems to be more closely linked to well-being among adolescents and perhaps young adults: a meta-analysis including 127,714 participants demonstrated a significantly higher risk of depression with longer screen time among teenagers younger than 14 years, but not in those older than 14 years (Liu et al., 2016). Similarly, increased mobile phone use was related to more intense negative emotions among young adults compared to older adults (Chan, 2018).

With respect to sleep, media use often results in a shorter sleep duration among adolescents, also known as sleep displacement (King et al., 2014), whereas adults reported instances of time shifting, that is later bedtimes and rise times after media use, without an impact on total sleep duration (Custers & Van den Bulck, 2012). Accordingly, the effects of increased smartphone use on shorter sleep duration and lower sleep quality have been more pronounced in younger adults (King et al., 2014; but regarding sleep duration see Exelmans & Van den Bulck, 2016). We therefore expect:

  • H1b. The effect of smartphone use on psychological well-being is more pronounced for younger adults compared to older adults.

  • H2b. The effect of smartphone use on sleep is more pronounced for younger adults compared to older adults.

Interpersonal Domain: Smartphone Use and Social Outcomes

Research on social outcomes of smartphone use is mixed, but points towards small positive effects: some authors argue that smartphone use displaces face-to-face interaction and has negative implications for people’s social lives (for a review see Sbarra et al., 2019). Indeed, in a study with more than 2,000 US couples, 25% of couples between 18 and 75 years, and even 42% of couples between 18 and 29 years reported that their partners had been distracted by a mobile phone while they were spending time together (Lenhart & Duggan, 2014). Yet, other researchers found positive effects of communicative smartphone use on social outcomes. For example, college students who used online social networks a lot reported attending more social events and hanging out with friends more often (Kim et al., 2016). Online social network sites allow people to expand their circle of acquaintances to also include more distant people and may help to maintain long-distance relationships (Steijn & Schouten, 2013). Moreover, instructing people to post more status updates led to reduced feelings of loneliness (Deters & Mehl, 2013).

Together this suggests that some uses of the smartphone might be detrimental to people’s social life, but most of the studies show that there is, on average, a small positive effect of using communicative functions of the smartphone, such as social networks or messaging applications. This leads to the following hypothesis:

  • H3a. The more adults use the communicative functions of their smartphone, the bigger are their social networks and the less socially isolated they feel.

While smartphone use for communicative purposes may help sustain social connectedness, evidence from the life course literature suggests that the effects of smartphones on social outcomes may vary depending on people’s age. Younger adults often use social media for both maintaining existing offline networks and gaining information from tweets, newsfeeds, or videos (Ariel et al., 2017), whereas older adults use social media primarily for strengthening their existing offline networks, which relates to less loneliness (Deters & Mehl, 2013; Rosales & Fernández-Ardèvol, 2016). Besides, many older adults suffer from impaired mobility, which limits their ability to engage in social activities and thus leads to higher loneliness (Luhmann & Hawkley, 2016). Smartphones may help to buffer the effects of mobility limitations on loneliness because smartphones enable communication with low effort, even from home. Because of the differences in smartphone use and loneliness risk factors, older adults may benefit even more from using the communicative functions of the smartphone than younger adults.

  • H3b. The effect of communication-related smartphone use on social outcomes is more pronounced for older adults compared to younger adults.

Work Domain: Smartphone Use, Overtime Work, and Job Satisfaction

During recent decades, working life has changed profoundly as the Internet enabled the fast transmission of knowledge and communication across physical and temporal boundaries. As the physical and temporal distinction between work and leisure were blurred, positive (e.g., increased flexibility), yet also negative effects (e.g., work-life infiltrations) of technology use for work have been reported (Eurofound and the International Labour Office, 2017). Employees who use technology to work from home spend more hours working and are more likely to work in excess of their contracted hours (for a review see Eurofound and the International Labour Office, 2017). Expectations to be available and to use a smartphone for work outside official working hours can impede employees’ psychological detachment from work (Mellner, 2016) and a lack of psychological detachment is associated with increased strain and decreased life satisfaction (Sonnentag & Fritz, 2015). Since smartphone use for work purposes outside of regular work hours can lead to intrusions in one’s personal life and thus may lead to decreased job satisfaction (Wright et al., 2014), we tested the following hypotheses:

  • H4a. The more employed adults use work-related functions of their smartphone, the more hours they work overtime and during evenings or nights.

  • H5a. The more employed adults use work-related functions of their smartphone, the less satisfied they are with their job.

Most of the previous studies were conducted with young employees, which raises the question as to whether work-related smartphone use has equal effects on working hours and job satisfaction among middle-aged and older employed adults. Teleworking is more prevalent in younger employees and they work more often in (night) shifts and at weekends than employees 55 years and older (Rothe et al., 2017). Interestingly, expectations of work-related additional accessibility seem not to differ with age (Pangert et al., 2014). As younger employees have a stronger preference for keeping work and private life separate than older employees (Derks et al., 2016), they might be more influenced by smartphone use outside of official working hours. Thus, younger employed adults are more at risk of suffering from a work-life infiltration, and we assumed:

  • H4b. The effect of work-related smartphone use on overtime hours and on work during evenings or nights is more pronounced for younger employed adults compared to older employed adults.

  • H5b. The effect of work-related smartphone use on job satisfaction is more pronounced for younger employed adults compared to older employed adults.

Present Study

In summary, previous research has led to interesting, but somewhat inconclusive findings regarding the costs and benefits of smartphone use in different life domains. Inconsistencies in earlier research may partly arise from many studies relying on non-representative samples and focusing on a narrow range of outcome variables. In contrast, the current study examined multiple outcome variables from three important life domains. We tested preregistered hypotheses on associations between smartphone uses and multiple outcome variables, as well as possible age moderations of these associations, in a sample representative of the German adult population. Additionally, we used structural equation model tree analyses (SEM trees) to probe the robustness of the findings across the entire sample and to explore non-linear associations between different indicators of smartphone use and the three domains of human functioning.

Method

Sample and Procedures

Smartphone use was assessed in the 2015 wave of the German Socio-Economical Panel ‘Innovation Sample’ (SOEP-IS) study. In the SOEP-IS, a large sample representative of the German adult population is surveyed yearly in computer-assisted personal interviews (Richter & Schupp, 2015). The full data set from the 2015 wave contained data from 5,131 participants (53% female, 47% male). Participants were, on average, 52.26 years old, with a standard deviation of 18.37 years (range: 17–96 years). For the analyses containing social isolation and work-related variables, subsets of this data set were used: for the analyses with social isolation as an outcome, the data of 4,053 participants (53% female, 47% male) were used, who participated both in 2015 and in 2017, when social isolation was assessed. For the analyses of work-related variables, the data of 2,813 employed adults (50% female, 50% male) were used. Employed adults were, on average, 44.40 years old with a standard deviation of 13.32 years (range: 17–85 years).

Measures for Predictor Variables

Smartphone use

Participants were asked ‘How often do you use the following functions of your phone?’ regarding seven categories: ‘calls’; ‘SMS/WhatsApp’; ‘social media’; ‘email’; ‘taking pictures/videos’; ‘internet’; and ‘games’. For each category, participants indicated on a scale of -1 (phone does not have this function), 0 (never) to 8 (several times a day) how often they used that specific function of their phone (supplementary Fig. S1 provides labels of the other categories). Using the same scale, participants were further asked how often they use ‘health/physical fitness’; ‘finances/shopping’; ‘learning/mental fitness’; ‘news’; ‘office’; ‘travel/traffic/maps’; or ‘weather’ software on their phone. Participants’ answers to those questions were used to calculate the following composite scores:

Smartphone use composite scores

General smartphone use was calculated as the mean of the use frequency of all the aforementioned functions and apps of the smartphone; communicative smartphone use was calculated as the mean of the frequency of calls, SMS/WhatsApp and social media use; work-related smartphone use was calculated as the mean of the frequency of email-function and office software use; and other smartphone use was calculated as the mean of the use of all functions and apps of the smartphone, except those used to calculate communicative and work-related smartphone use.

Measures for Outcome Variables

Psychological well-being

At the end of the interview, participants answered the following question: ‘In conclusion, we would like to ask you about your satisfaction with your life in general. How satisfied are you with your life, all things considered?’ on a scale of 0 (completely dissatisfied) to 10 (completely satisfied).

Sleep time

Participants indicated ‘How many hours do you sleep on average on a normal day during the working week?’ and ‘How many hours on a normal weekend day?’ in whole hours.

Subjective sleep quality

Participants responded to ‘How satisfied are you with your sleep?’ on a scale of 0 (completely dissatisfied) to 10 (completely satisfied).

Social network size

Participants answered ‘How many close friends would you say that you have?’ in an open-answering format, that is, providing a whole number.

Social isolation

Social isolation was measured with a three-item version of the ULCA loneliness scale (Hawkley et al., 2015). Participants indicated how often they missed the company of other people, how often they felt left out, and how often they felt socially isolated on a scale of 0 (never) to 3 (very often). Cronbach’s alpha of the resulting composite score was α = 0.75. Social isolation was measured in the 2017 wave for the first time and thus we used the 2017 measurement for the current study.

Job satisfaction

Participants responded to ‘How satisfied are you with your job?’ on a scale of 0 (completely dissatisfied) to 10 (completely satisfied).

Overtime work

Participants answered: ‘How many hours per week are stipulated in your contract (excluding overtime)?’ and ‘And how many hours do you generally work, including any overtime?’. Overtime work was estimated by subtracting monthly working hours as stipulated in their contract from monthly working hours including any overtime.

Evening work and night work

Participants responded to ‘Do you sometimes have to work in the evenings (after 19:00) or at night (after 22:00)? If so, how often?’ on scale of 0 (No, never) to 3 (Yes, every day). Participants provided two answers, one for evening work and one for work at night. Participants working in weekly shifts (n = 377, 13% of employed adults) were excluded from analyses with these variables.

Measures for Exploratory Analyses

Marital status

Answers to the question ‘What is your marital status?’ were recoded into two categories: ‘married’ and ‘not married’. Married persons that were living (permanently) separated from there spouse were treated as ‘not married’.

Employment status

Answers to the question ‘Are you currently employed?’ were recoded into three categories: ‘not working’, ‘working part-time’, and ‘working full-time’. Completing in-service training, apprenticeship, and in-service retraining were treated as ‘working full-time’. Participants that worked in a sheltered workshop (n = 5) or voluntary social/ecological year/federal volunteer service (n = 8) received a missing value on this variable.

Statistical Approaches for Hypotheses Testing and Exploratory Analyses

While the single items to assess the use of smartphone apps and functions often showed skewed and bimodal distributions (online appendix, available at https://osf.io/nb89p, Fig. S1, Fig. S2), the variables used for the main analyses, that is, general smartphone use, communicative smartphone use, and work-related smartphone use, showed more even distributions (Fig. S3). Regarding the outcome variables, all satisfaction measures were skewed to the right, number of close friends, social isolation, as well as overtime and evening work were skewed to the left, and sleep durations were approximately normal distributed. Since the sample was very large and no range restriction occurred, we used ordinary least squares regression.

Statistical tests for all hypotheses were performed using the R software version 3.6.2 (R Core Team, 2019), with p < 0.05 as criterion for determining statistical significance. Whenever one hypothesis was tested using multiple regression models (H2, H3, H4), we adjusted the alpha level for multiple tests. Missing values occurred in all study variables but concerned at most 0.2% of the sample and were completely at random. Multiple regression analyses did not consider cases with missing values (i.e., listwise deletion per model). All analyses that contained variables with statistical outliers (M ± 3SD) were calculated twice: once with the original variables and again after all variables with statistical outliers were winsorized to M + 3SD or M—3SD, respectively. Unless otherwise stated, the results obtained from the winsorized data did not differ from the results obtained from the original variables. All variables were z-standardized to facilitate the interpretation of the results.

All hypotheses-testing analyses were preregistered and can be found at https://osf.io/skcwe. The conducted analyses differed slightly from the preregistration in three instances. First, other than assumed, social isolation was not measured in 2015 but in 2017 for the first time and, thus, this information was used. Second, participants answered separate questions on working in the evening or at nighttime. Since answers correlated only moderately (r = 0.38; 95% CI [0.35, 0.41]) and working at night outside a regular shift-work schedule is very rare (Eurofound and the International Labour Office, 2017), we decided against a composite score and based the analyses only on evening work. Third, after working through additional literature and before conducting any analyses, the direction of Hypothesis H5b was changed because it was theoretically more plausible to expect the association between work-related smartphone use and job satisfaction to be more pronounced for younger adults instead of for older adults, as stated in the preregistration. All other analyses were conducted as stated in the preregistration.

As a robustness check, we conducted SEM tree analyses using the semtree package (Brandmaier et al., 2020) to explore whether specific, non-linear combinations of smartphone uses and demographic variables could explain heterogeneity in the associations between smartphone use and the different life domain outcomes. SEM trees combine recursive partitioning (i.e., repeated dividing of the sample based on model-derived ‘splitting variables’ such as gender) with structural equation modelling to foster theory-guided exploration of empirical data. In the current study, simple regression models of the association between (specific) smartphone uses and functioning in the three life domains (i.e., without age as a moderator) were used as pre-specified models. The SEM tree algorithm automatically selects variables from a list of potential moderators (i.e., splitting candidates) and splits the data into subgroups with similar ‘focus parameters’ – in the current study into subgroups with similar regression slopes. Splitting candidates were the composite smartphone use variables (communicative smartphone use, work-related smartphone use and other smartphone use) and demographics (age, gender, marital status, and employment status).Footnote 1

To increase the ease of interpretation and to keep the processing times manageable, all trees were restricted to a depth of three consecutive splits (max.depth = 3). When we ran the first trees, the algorithm tended to select splits that resulted in very small samples for the terminal nodes (n around 20), often resulting in uninterpretable parameter estimates with large standard errors. We therefore set the minimum size of the subsamples identified in splits (i.e., nodes) to min.N = 200. Bonferroni correction was applied within each tree to control for multiple testing. This way, nine trees corresponding to the nine life domain outcomes were calculated.Footnote 2

Single SEM trees possess a high flexibility in fitting any data, which involves the risk of fitting (only) noise. Therefore, parameter estimates that SEM trees identify for subgroups need to be validated on a separate sample. For the current analyses, we randomly split the data set in a training (50%, n = 2565) and in a test sample (50%, n = 2566). The training and the test sample did not differ significantly in smartphone use, outcome variables, or demographic variables (all p > 0.05). For any tree in the training sample, we first determined in which subgroups of the terminal nodes (i.e., subsample of the overall sample) the estimated models had a regression slope that was significantly different from zero. For any significant regression slope, we tested whether the regression slopes could be replicated in the corresponding subsample of the test sample by applying the same splits (i.e., subsamples) that the algorithm suggested in the training sample. We considered a regression slope replicated if (a) it was also significantly different from zero in the test sample, and (b) the original point estimate of the regression slope from the training sample was within the 95% confidence interval of the regression slope from the test sample.

Results

Out of all participants, 615 (12%) reported that they did not own a mobile phone and some reported that their phone did not have certain functions or apps.Footnote 3 Since the predictor variables general, communicative, and work-related smartphone use were composite variables, which averaged across different apps, missing data for specific apps were unproblematic. The means and standard deviations of the composite measures of smartphone use and all outcome variables, as well as their bivariate correlations and their partial correlations controlled for participants’ age, are reported in Table 1.

Table 1 Means, Standard Deviations, and Correlations
Full size table

General Smartphone Use, Psychological Well-Being, and Sleep

To test Hypotheses 1a and 1b, general smartphone use (linear and squared effects), age, as well as the interaction between smartphone use and age were included in the regression model. The results showed both linear (β = 0.12, 95% CI [0.07,0.16]) and quadratic effects of general smartphone use (β = -0.05, 95% CI [-0.10,-0.01]) on well-being, linear associations between age and well-being (β = 0.10, 95% CI [0.06,0.13]), and also age significantly moderated the linear effect of smartphone use on well-being (β = -0.08, 95% CI [-0.12, -0.04]), but not the quadratic effect (β = -0.02, 95% CI [-0.06, 0.01]; overall model fit F(5, 5091) = 11.93, R2 = 0.01, Fig. 1).Footnote 4 To better understand the significant interaction of the linear effect, the region of significance was calculated with the Johnson–Neyman technique. For participants younger than 67.69 years, the linear association between general smartphone use and higher psychological well-being was significant (p < 0.05), and more pronounced with younger age. Nevertheless, the linear interaction term should be interpreted in conjunction with the other regression parameters. Thus, the predicted values from the regression model suggest that moderate smartphone use was associated with the highest psychological well-being in middle and late adulthood, and that more smartphone use was associated with somewhat higher psychological well-being in young adulthood, but with diminishing returns for heavy smartphone use (Fig. 1). Including sex, marital status and employment status as covariates slightly increased the size of the age moderation to β = -0.14, 95% CI [-0.18, -0.10] (see Table S2).

Fig. 1
figure 1

The Effects of Age and General Smartphone Use on Psychological Well-Being. Note. Predicted values from the regression with psychological well-being as the outcome. For middle-aged and older adults, the association is primarily quadratic. For younger adults, the association is positive, but with diminishing returns

Full size image

Three hierarchical regression analyses were performed to examine the effects of general smartphone use and age on sleep duration for weekends and for weekdays, and on sleep satisfaction, separately (Table 2). In general, the effects of general smartphone use and sleep duration varied with participants’ age (Fig. 2): contrary to the expectations, for sleep duration on weekdays, more smartphone use was associated with longer sleep among adults younger than 30.6 years, but with shorter sleep among adults older than 42.1 years (Fig. 2A and B). For adults between 30.6 and 42.1 years, general smartphone use was not significantly associated with sleep duration on weekdays. For sleep duration at weekends, more smartphone use was associated again with longer sleep among adults younger than 38.7 years, but with shorter sleep among adults older than 53.4 years (Fig. 2C and D). For adults between 38.7 and 53.4 years, general smartphone use was not significantly associated with sleep duration at weekends. In contrast, sleep satisfaction was lower with higher age, but neither general smartphone use nor the interaction with age significantly predicted sleep satisfaction (Table 2). In the sensitivity analyses with covariates, parameter estimates for general use and the age moderation were very similar (within ± 0.02 of original β, see Table S1).

Table 2 Model Parameters for Sleep Outcomes Predicted by General Smartphone Use and Age
Full size table
Fig. 2
figure 2

Interaction Effects of General Smartphone Use and Age on Sleep Duration. Note. A) Interaction of general smartphone use and age on sleep duration on weekdays; B) Johnson–Neyman plot for sleep duration on weekdays; C) Interaction of general smartphone use and age on sleep duration at weekends; D) Johnson–Neyman plot for sleep duration at weekends. Around the estimated slopes, 95% confidence intervals are shown

Full size image

Communicative Smartphone Use and Social Outcomes

The variable ‘number of close friends’ included some plausible, but extreme values (e.g., > 15; 1% of sample). To reduce the impact of outliers on the results, estimates obtained from the winsorized number of close friends are reported.Footnote 5 Consistent with H3a, with higher communicative smartphone use, adults reported slightly more close friends (Table 3). This association did not differ significantly with participants’ age. Inconsistent with H3a, communicative smartphone use did not significantly predict social isolation two years later, after correction for multiple testing; only with higher age, adults reported somewhat lower social isolation (Table 3). The effect of communicative smartphone use on number of close friends did not change after including additional covariates (see Table S2).

Table 3 Model Parameters for Social Outcomes predicted by Communicative Smartphone Use and Age
Full size table

Work-related Smartphone Use and Work Outcomes

In partial support for H4a, more frequent work-related smartphone use was significantly associated with longer overtime hours, working more often in the evening, yet not with job satisfaction (Table 4). After adding covariates, parameter estimates for work use were very similar (within ± 0.02 of original β, see Table S3). In contrast to hypotheses H4b and H5b, age and the interaction between work-related smartphone use and age did not significantly predict work outcomes (Table 4).

Table 4 Model Parameters for Work Outcomes predicted by Work-related Smartphone Use and Age
Full size table

Exploratory Analyses Using SEM Trees

The previous regression analyses tested preregistered hypotheses, and we next explored with SEM trees whether unpredicted associations and moderation effects existed among smartphone use, demographic characteristics, and life domain outcomes. In addition to exploration of unpredicted effects, SEM tree analyses can be seen as a robustness test of the results from the hypotheses testing. If the associations between smartphone use and life domain outcomes differed depending on demographics (i.e., gender, marital status and employment status) or other smartphone use variables (i.e., communicative smartphone use, work-related smartphone use and other smartphone use), the SEM tree analyses should find corresponding subgroups with differing regression slopes in the data.

Overall, nine SEM trees (corresponding to the nine outcomes, see Method section) were computed separately in the training and test data, which resulted in 51 pairs of regression slopes. Out of the 51 pairs of regression slopes, 23 (45%) were non-significant in both samples, 17 (33%) were significant only in the training data, 6 (12%) were significant only in the test data and 5 (10%) were significant in both samples. First, we exemplarily describe one of the SEM trees in detail. After that, we will report on the other two SEM trees that contained one or more subgroups with a significant regression slope that could be replicated in the test sample. Plots of all nine SEM trees are provided in the online appendix.

In the SEM tree displayed in Fig. 3, the underlying model was a simple ordinary least squares regression, with communicative smartphone use as predictor and number of close friends as the outcome. We explored whether the significant association between having more close friends and higher communicative smartphone use was less or more pronounced for people with certain demographic characteristics or further smartphone use habits. Splitting candidates were thus work-related smartphone use and other smartphone use, as well as the demographic variables (age, gender, marital status, employment status). In the terminal nodes of this tree, several regression slopes (b1 parameters) were significant in the training data (see Fig. 3, bolded b1 parameters). The association between more frequent communicative phone use and reporting more close friends was more pronounced for older adults (between 63.5 and 71.5 years, b = 0.26, 95% CI [0.09,0.43]) and even more pronounced for old participants (between 71.5 and 76.5 years, b = 0.53, 95% CI [0.29,0.76]) in the training sample. Because these effects did not replicate in the test sample (see Fig. 3), we will refrain from further discussion of these results.

Fig. 3
figure 3

SEM Tree: Close Friends Predicted by Communicative Smartphone Use with Demographics and Further Smartphone Uses as Splitting Candidates. Note. The winsorized variable of close friends was used. Significant regression slopes are bold. Other Use = other smartphone use. b1 = slope. b0 = intercept. VarY = variance of number of close friends. VarX = variance of communicative smartphone use. M(X) = mean of communicative smartphone use

Full size image

In one instance of this SEM tree, the slope parameter (i.e., b1) was replicated in the test data (Fig. 3, bottom left, marked with a box, married males younger than 63.5 years). For a proper interpretation, it is important to know whether this effect is just a replication of the overall sample effect in a certain subsample, or whether the effect in the subsample is significantly different from the overall sample effect. The slopes that were estimated for the subsamples of unmarried males younger than 63.5 years were larger (training data: b1 = 0.31, 95% CI [0.18,0.44]; test data: b1 = 0.24, 95% CI [0.12,0.35]) than the regression slopes estimated for the overall samples (training data: b1 = 0.12, 95% CI [0.08,0.16]; test data: b1 = 0.12, 95% CI [0.07,0.16]). In both cases, the point estimate of the regression slope in this specific subgroup was outside the confidence interval of the overall sample effect, and vice versa. Thus, the association between having more close friends and higher communicative smartphone use was more pronounced in the subgroup of unmarried males younger than 63.5 years than in the overall sample.

We did not find any replicable regression slopes in the SEM trees with psychological well-being, sleep duration at weekends, or subjective sleep quality as the individual domain outcome (see online appendix, figs. S4, S5, S6). Only in the SEM tree with sleep duration on weekdays as the outcome, among men older than 53.5 years, who used smartphone apps subsumed under ‘other smartphone use’ on average more than once a year (other smartphone use >  = 1.5), more frequent general smartphone use was related to shorter sleep duration on weekdays (training data: b1 = -0.38, 95% CI [-0.60,-0.16]; test data: b1 = -0.44, 95% CI [-0.68,-0.21], Fig. S7). These associations were different from the association in the overall sample (training data: b1 = -0.05, 95% CI [-0.08,-0.03]; test data: b1 = -0.04, 95% CI [-0.07,-0.02]). This result is in line with the finding from the regression analysis that the association between general smartphone use and less sleep on weekdays was more pronounced with higher age.

Regarding the interpersonal domain, there were no replicable regression slopes in the SEM tree with social isolation as the outcome (Fig. S8). We reported the results of the SEM tree with number of close friends as the outcome in detail in Fig. 3.

Regarding the work domain, the SEM tree with work-related smartphone use predicting overtime work showed significant associations in three out of four subgroups in the training data and in all four subgroups in the test data (Fig. S9). However, the confidence intervals of the regression slopes of all subgroups overlapped with the overall sample effect in the test data (see Fig. S9). Thus, the overall sample effect (i.e., that more work-related smartphone use was associated with more overtime work) could be replicated in the subgroups, but none of the subgroups showed a pattern that was clearly distinct from the overall effect. No replicable regression slopes in subgroups occurred in the SEM trees with job satisfaction or evening work as the outcome (Figs. S10 and S11).

Discussion

We took a broad theoretical perspective on the consequences of smartphone use and analysed both benefits and costs in the individual, interpersonal, and work domain within a sample representative of the German adult population. In the individual domain, middle-aged and older adults who moderately used their smartphone reported the highest psychological well-being (quadratic effect). The more often younger adults used their smartphone, the more they reported somewhat higher psychological well-being, but with diminishing returns for heavy smartphone use. These findings support previous research on adolescents, which also found quadratic effects (Orben & Przybylski, 2019; Przybylski & Weinstein, 2017; Twenge & Campbell, 2018). Overall, the effect sizes of smartphone use on psychological well-being were small, which is in line with the most recent literature (Orben & Przybylski, 2019).

With regard to sleep, younger adults slept slightly more and older adults slept less if they used their smartphones a lot, which partly supports Hypothesis 2a, but the effect sizes were small. As smartphone use prior to bedtime has been linked to a multitude of sleep problems (Exelmans & Van den Bulck, 2016), younger adults, who use their smartphone late at night, might rise later and thus sleep longer overall compared to young adults with less smartphone use (Custers & Van den Bulck, 2012). Contrary to H2b, no association existed between smartphone use and satisfaction with sleep, which might result from the single question not being sensitive enough to detect sleep problems.

In the interpersonal domain, we found partial support for H3a: Adults, who used communicative functions of their smartphone more often, reported having slightly more close friends, but did not differ in their social isolation two years later. These effects occurred irrespective of age, which contradicts the assumed age moderation specified in H3b. Thus, the data are coherent with the idea that, on average, people use communicative functions of their smartphone in a way that benefits their social lives (Kim et al., 2016). The reverse explanation is also plausible given the correlational nature of the findings: people might use the communicative functions of their smartphone more often because they have more close friends to communicate with. In addition, further factors—for example, the sociability of the smartphone-user—could lead to both a slightly higher number of close friends and increased use of the communicative smartphone functions.

Concerning the work domain, employed adults who used their smartphone a lot for work-related purposes reported working more overtime and more often in the evenings than employed adults, who used their phone less for work-related purposes, which supports H4a. Contrary to the prediction (H4b, H5a, H5b), age did not moderate these effects and no association between work-related smartphone use and job satisfaction existed. Job satisfaction is believed to result from a complex interplay of job characteristics and person characteristics (Judge & Klinger, 2008), which might explain why the sheer frequency of work-related smartphone use did not explain differences in job satisfaction. For example, the effects of work-related smartphone use on job satisfaction might depend on characteristics of the job (e.g., autonomy) as well as on characteristics of the person (e.g., whether they wish to separate work and private life – Derks et al., 2016).

To summarize, at the highest level of interpretation, only small associations of smartphone use with a diverse set of daily life consequences occurred. This pattern of findings contrasts two prominent positions that are often advocated in the mainstream media: first, we did not find strong evidence for the idea that more (use of) technology is generally desirable. Second, we also did not find strong evidence that (over)use of smartphones constitutes a broad societal problem among the German adult population; in other words, most people were not impaired from using their smartphone regularly—within the observed amount of use.

Nevertheless, larger effects of smartphone use could exist at more fine-grained levels of analysis (e.g., how daily changes in smartphone use relate to daily fluctuations in well-being within individuals), or for certain subpopulations. The associations between smartphone use and well-being as well as the associations between smartphone use and sleep were both moderated by age. The age differences indicate that the effects of smartphone use might be stronger for some people, weaker for others, and not really present for yet others. We explored further heterogeneity in the strength of the associations between smartphone use and the life domain outcomes using SEM tree analyses. These exploratory analyses suggested that the strength of the associations between smartphone use and some outcome variables differed for certain subpopulations of the current sample, yet many patterns did not replicate. As before, the small effect sizes might be attributable to smartphone use contributing only very little to overall life satisfaction, job satisfaction, or perceived loneliness compared to other personal and environmental factors (Diener et al., 2018; Luhmann & Hawkley, 2016). In addition, it is also possible that the frequency of smartphone use shows little effect because the purpose and content of the use as well as situational and cultural factors probably influence the impacts of smartphone use on important life domains.

Purpose and Context of Smartphone Usage Needs to be Considered

Up to now, most studies have focused on the frequency of (specific) smartphone use or on the duration of screen time exposure (Kim et al, 2016; Orben & Przybylski, 2019; Twenge & Campbell, 2018). Yet how and with what intentions the smartphone is used is likely more important for life domain outcomes than the frequency of use. For example, using the smartphone to maintain and deepen existing friendships was related to less loneliness and a more fulfilling social life, while passively scrolling through feeds on social network sites was associated with feeling more lonely (Nowland et al., 2018).

Moreover, people differ in how they want to integrate their smartphone into their lives and how well they regulate their smartphone use. For example, the effects of work-related smartphone use on job satisfaction differ depending on how much employees wish to separate work and private life (Derks et al., 2016; Mellner, 2016). Furthermore, compared to students who believe in owning their fate, students who believe strongly in fortune or bad luck had less control over their smartphone use – that is, they were more likely to use the phone at disadvantageous times such as bedtime, in class or while studying. At the same time, these students were more likely to report reduced quality of sleep, academic performance, or subjective well-being (Li et al., 2015). Similarly, while social media use can have some important benefits among its users, research over the last few years has consistently shown that a small minority of adolescents can experience negative consequences because of excessive and problematic use of social networking sites (Griffiths et al., 2018).

Strengths and Limitations

In this study, we used a large sample representative of the German adult population to test preregistered hypotheses on the costs and benefits of smartphone use in multiple life domains. In contrast to many previous studies, we tested whether these associations differed with people’s age and we applied SEM trees to search for additional factors that might explain heterogeneity in the associations of smartphone use and different life-domain outcomes. The focus on a broad range of outcomes allows for a comprehensive interpretation of the results that would not have been possible had we only examined one domain. Despite these strengths, some limitations need to be discussed.

Firstly, it was not possible to make any claims on the effects of very time-intensive smartphone use with the current data. The answer option for the highest smartphone use for each function or app was ‘multiple times a day’. Therefore, potentially adverse smartphone use of many hours per day could not be distinguished from moderate, innocuous smartphone use (Przybylski & Weinstein, 2017). We addressed this problem by combining the use of different apps and functions into a variable of general smartphone use to build an indicator of ‘heavy smartphone use’. However, people excessively using a small number of functions or apps could not be distinguished from people moderately using many different apps.

Secondly, relying on self-reported smartphone use potentially introduced some bias in the measurement; yet, we would argue that the bias introduced by generalizing from non-representative samples is likely higher than the small inaccuracies documented for self-reports of smartphone use (Hodes & Thomas, 2021). In addition, the indicators for life-domain outcomes were often single-item scales, which do not offer the breadth and reliability of more comprehensive questionnaires.

Thirdly, the reported effect sizes were small. Although small effect sizes are common and expectable for many complex psychological outcomes (Götz et al., 2022), little consensus exists on how to interpret small effect sizes, especially those derived from a single study with a large sample (Ferguson & Heene, 2021; Götz et al., 2022; Orben & Przybylski, 2019). Therefore, it is important to keep in mind that researchers should not draw strong theoretical conclusions from a single study and cannot derive causality from the cross-sectional data. In the introduction, we reviewed evidence that pointed more or less strongly towards causal relationships – depending on the outcome. For many outcomes, bi-directional explanations seem plausible, for example, increased smartphone use might reduce sleep duration and quality, while experiencing sleep disturbances might also lead to using the smartphone more often (Hartanto et al., 2021).

Regarding the explanatory analyses, it should be noted that SEM trees can only detect heterogeneity in the associations between smartphone use and life-domain outcomes that can be explained by the variables considered as splitting candidates. Given the necessity to restrict the number of splitting candidates and the depth of the trees for more stable parameter estimates, it is likely that heterogeneity remained in the data, which could be explained by other, untested variables. Although we replicated some significant associations from the initial SEM trees in a test sample, we are careful in drawing strong conclusions from any single SEM tree because of the large number of tests and the scarcity of replicated effects. The scarcity of robust effects offers further support for the idea that for the majority of the population, effects of smartphone use on life outcomes are on average rather small or non-existent at all.

Outlook and Further Considerations

A guiding question for future studies on smartphone usage could be ‘Who is using the smartphone when, where, and for what purpose?’ A promising approach would be to gather calling, texting and app use behaviours from smartphone system logs (e.g., Matz & Harari 2021). With those system logs, it is possible to continuously track with high precision how long which function of the smartphone was used. Such detailed data on smartphone use behaviour from system logs could be combined with data on personality, locations, and daily activities for even more insights (e.g., Matz & Harari, 2021). Tracking both smartphone use and other activities continuously and over longer periods of time would also allow observing how smartphone use and daily activities compete against or complement each other within days or weeks instead of only aggregated over time and persons.

In addition to tracking how much certain functions or apps are used, it might be useful to assess how and with which goals people use these functions and apps (Rosales & Fernández-Ardèvol, 2016). Furthermore, situational appropriateness (e.g., regarding daytime, conflicting activities) is likely an important moderator of the costs and benefits of smartphone use. Thus, the situational context, with both objective cues and subjective characteristics (i.e., the psychological meaning) should be assessed in future studies.

Conclusion

We showed that, on average, typical smartphone use had only minor associations with different life domain outcomes, that is, most German adults were not impaired from regularly using their smartphones. However, we would not take the small average effects to mean that the many hours people spend on smartphones do not have any effects on their lives. Rather, the literature suggests that smartphones are best viewed as tools that can be used in very different ways, some of which are connected to desirable outcomes, while others might lead to undesired consequences. Smartphones are now ubiquitous in our daily life. We can use them at unsuitable times, like at bedtime or in class, and we can overuse the smartphone for work purposes or media consumption, isolating ourselves and leaving not enough time for family and friends. But we can also use the smartphone to quickly retrieve information as needed or to maintain and deepen our existing friendships – it is up to us.