1 Introduction

Several studies have revealed that fear of missing out (FOMO), boredom, and loneliness predict phubbing, which is the act of looking at the smartphone while one is talking to another in person (Al-Saggaf et al. 2019; Franchina et al. 2018; Yaseen et al. 2021b). This means experiencing these emotional states can trigger smartphone users to phub others during face-to-face conversations. At the same time, a few other studies have shown that phubbing increased levels of loneliness, and boredom (Zhao et al. 2021; Wang et al. 2021). This shows that the relationship between phubbing and emotional states is complex. Moreover, a recent study has revealed that FOMO, phubbing, and time spent on social media are related (Al-Saggaf 2021). Phubbing fully mediated the effect of FOMO on time spent on social media meaning that users coped with FOMO by ignoring others in social situations in favour of the smartphone and this resulted in users spending more time on social media (Al-Saggaf 2021). It is clear from this research that phubbing links social media and smartphone use with emotional states. In addition, phubbing is associated with a plethora of other effects (Al-Saggaf and O’Donnell 2019a). Partner phubbing ignited the phubbed partner’s feeling of anxiety leading them to engage in spying on their phubbing partner's digital activities (Schokkenbroek et al. 2022). In another recent study, being phubbed by a partner was found to be associated with lower relationship satisfaction, anger, frustration, and retaliation (by the phubbed partner) (Thomas et al. 2022). In the case of parents phubbing, a study from China revealed that parents phubbing lowered children's self-esteem to which children responded by showing signs of suicidal ideation (Wang and Qiao 2022). These harmful effects show that phubbing is not a harmless act, rather a harmful behaviour (Hunter-Brown 2021). While efforts to investigate the relationships among social media and smartphones use, phubbing, and emotional states using validated scales and employing the scientific method of hypothesis testing, should continue, as these investigations offer robust findings, another way to learn more about the relationships among social media and smartphone use, phubbing, and emotional states is to use data mining (Al-Saggaf 1152). Unlike traditional investigations (validated measurements), data mining facilitates the analysis of numerous factors at the same time, while overcoming issues of multicollinearity and non-normality violations.

In addition, because data mining can select attributes on which to split data (Al-Saggaf and Walsh Oct. 2021), it can accurately estimate factor importance while avoiding Type I errors (i.e., rejecting a true null hypothesis) (Mullainathan and Spiess 2017; Obermeyer and Emanuel 2016; Spielberger 1989). This paper uses data mining techniques to find out more about the relationships among social media and smartphone use, phubbing, and emotional states. For example, a decision tree algorithm such as J48 produces a set of logic rules (i.e., root to leaf path) that are used to understand the reason why a particular incident has happened (Quinlan 1996, 1993; Rahman et al. 2020a, 2020b, 2022). This study uses data mining techniques to discover new patterns related to social media and smartphone use, phubbing, loneliness, state boredom, state FOMO and trait boredom using two datasets whereas a previous study (Rahman et al. 2022) discovered patterns related to phubbing using one dataset. Also, the emotional states covered in the study (Rahman et al. 2022) are different from the emotional states covered in this study. The contributions of the paper are:

  • Advancing the field of social media and smartphone use and psychology by using data mining techniques to reveal previously unknown patterns about the phubbing phenomenon based on two existing datasets.

  • Novel patterns related to phubbing, loneliness, state boredom, state FOMO, and trait boredom are discovered and explored in detail.

  • Providing a different perspective to social media and smartphone use and emotional states derived through an alternative data-driven research approach, thereby facilitating the comparison between this approach and the traditional investigations that use validated measurements and the scientific method of hypothesis testing.

The rest of the paper is organized as follows: Sect. 2 discusses related work, Sect. 3 presents a discussion on datasets and the model-building process, Sect. 4 presents the results and patterns of the models, Sect. 5 discusses the models, and Sect. 6 presents conclusions and future plans.

2 Related work

Social media and smartphone use and emotional states are interlinked. Phubbing occurs when someone momentarily shifts their attention from a face-to-face conversation with another person to the smartphone (Al-Saggaf 2022). Numerous factors that can trigger phubbing have been suggested in the literature, including FOMO (Franchina et al. 2319), boredom (Yam and Kumcagiz 2020), both state and trait (Al-Saggaf 1152), and loneliness (Yaseen et al. 2021a). FOMO is the feeling of anxiety or restlessness that arises from the belief that others might be experiencing more enjoyable activities or events than oneself (Franchina et al. 2319). State FOMO (Balta et al. 2020) is a temporary form of FOMO that is triggered by a particular situation or event, such as scrolling through social media and seeing friends' posts about a fun activity that you were not invited to Przybylski et al. (2013).

State boredom refers to a temporary and situational experience of boredom, such as waiting in a long line or attending a tedious meeting, while trait boredom reflects a more stable and enduring tendency to experience boredom across a variety of situations and contexts, which may be related to personality traits or other individual factors. Boredom is the state often arising from a lack of stimulation or interest in one's surroundings or activities (Vodanovich et al. 2005). Scholars differentiate several types of boredom, but the main two are state and trait boredom (Al-Saggaf et al. 2019). Trait boredom is the recurring tendency or chronic disposition of individuals to experience boredom (Sharp and Hemmings 2013). This is different from state boredom, which is the fleeting feeling of boredom in a given moment (Damon and Louis 2010). Loneliness is experienced when social connections are considered to be lacking (Maclean et al. 2022). Like FOMO and boredom, loneliness can also be experienced temporarily (state) or chronically (trait) (Al-Saggaf and O’Donnell 2019b).

An existing study (Al-Saggaf et al. 2019) showed that boredom has a relationship with phubbing. The authors in study (Al-Saggaf et al. 2019) used a regression model to show that trait boredom predicts the phubbing frequency. Another study (Al-Saggaf and O’Donnell 2019b) also used a regression model to show state FOMO has a strong correlation with state phubbing and state boredom. The datasets used in our paper are also used in previous studies (Al-Saggaf et al. 2019; Al-Saggaf and O’Donnell 2019b). However, the studies (Al-Saggaf et al. 2019; Al-Saggaf and O’Donnell 2019b) did not apply data mining techniques to discover patterns related to phubbing. An existing data mining-based study (Rahman et al. 2022) is used to discover patterns related to phubbing. However, the study (Rahman et al. 2022) discovers patterns only related to phubbing whereas the present study discovers patterns related to phubbing, loneliness, state boredom, state FOMO, and trait boredom. Moreover, the predictors (i.e., attributes) used in Rahman et al. (2022) are also different than the predictors used in the present study.

3 Materials and methods

3.1 Datasets

Data used in the generation of data mining-based models were sourced from two datasets collected as part of two different but related studies on phubbing behaviour (Al-Saggaf et al. 2019; Al-Saggaf and O’Donnell 2019b). Both studies received approval from the first two authors’ University’s Human Research Ethics Committee (HREC)—equivalent to Institutional Review Board approval—Protocol Number H200201770.

The first study (Study 1) aimed at understanding phubbing behaviour more broadly, addressing questions, such as how often people phub, who do people phub more often, in which situations people phub others more frequently, and which apps do people use the phubbing of others, as well as phubbing association with trait boredom, which is the chronic form of boredom. Data for Study 1 was collected in 2018 using an online survey. Participation in the survey was solicited through social media platforms, such as Twitter, Facebook, and LinkedIn, and through traditional media channels, such as newspaper articles, and radio and television appearances. A total of 385 participants completed the SurveyMonkey questionnaire. Respondents were mostly Australian residents (96.64%) and predominantly females (71.07%). They aged between 18 and 71 years and while 47.47% of the respondents lived in metropolitan areas, 52.53% lived in regional areas. See (Al-Saggaf et al. 2019) for more information on Study 1.

The second study (Study 2) aimed at understanding phubbing as a fleeting reaction and investigated its association with other fleeting behaviours, such as state loneliness (the trainset form of loneliness), state boredom, and state FOMO. This second study focussed mainly on the predictors of phubbing as a momentary behaviour. Data for Study 2 was collected in 2019 also using an online survey and in addition to recruiting participants through social media platforms, participation in the survey also occurred through several websites and discussion mediums, such as Reddit.com. A total of 325 participants completed the Google Forms questionnaire. Respondents came from several Western countries, such as the United States, the United Kingdom, and Australia but a good number of respondents also came from Asian countries. Interestingly, respondents were again predominantly females (73.87%). They aged between 18 and 65 years and while 50.2% of the respondents lived in metropolitan areas, 49.8% lived in regional areas. See (Al-Saggaf and O’Donnell 2019b) for more information on Study 2.

3.2 Methods

Our method is a data mining-based method which has four basic steps that we explain below:

  • Step 1: Data collection

  • Step 2: Data pre-processing

  • Step 3: Model development

  • Step 4: Insights of the model

Step 1 of our method is data collection. We used two datasets which were used in two previous studies on phubbing. A dataset has a number of rows (i.e., records) and a number of columns (i.e., attributes). The attribute of a dataset can be either numerical and/or categorical.

In Step 2, we eliminate records with missing values, if there are any. For model-building purposes, the attributes of a dataset are grouped into two categories namely class attribute and non-class attribute. If a dataset has more than one class attribute, then we prepare a sub dataset for each class attribute separately. In each sub-dataset, there is only one class attribute and all non-class attributes of the original dataset. The number of records and the number of non-class attributes in each sub-dataset is the same except for the class attribute.

In Step 3, for each sub dataset, we build models using data mining techniques namely J48 Decision Tree (Quinlan 1996, 1993), Random Tree (RT) (Rahman 2014), Random Forest (RF) (Frank et al. 2005), and Naive Bayes (NB) (Breiman 2000). J48, NB, RT and RF algorithms are available in Weka (Rahman 2014). The main reason for using four interpretable algorithms is to compare their performance to select the best-performing algorithm and to enable the results to be human interpretable. The J48 algorithm demonstrated the best performance (see Tables 1, 2, 3, 4) and was selected for the subsequent data mining tasks. The J48 decision tree algorithm is generally used for pattern analysis (Rahman et al. 2020a, 2020b, 2022). A decision tree contains nodes, leaves and branches as shown in Fig. 1. The J48 decision tree algorithm divides a dataset into mutually exclusive horizontal segments based on the values of the attribute tested on a node. For the J48 decision tree, we used tenfold cross-validation, the minimum number of records per leaf is two and the value of the confidence factor is 0.99. We aimed to perform limited pruning of the decision tree. Hence, we used a higher confidence factor value (i.e., 0.99) instead of the default confidence factor value (i.e., 0.25). We observed that for a confidence factor of 0.25, we get a small decision tree that does not provide useful patterns. For the other algorithms, we used their default parameters. A branch of the decision tree is presented in Fig. 1 considering state FOMO as the class attribute where the leaves are shown by ovals, the nodes are shown by rectangles, the branches are shown by lines, and the dotted line indicates the remaining part of the tree. The root to a leaf path of a decision tree is called a logic rule (i.e., pattern). A leaf of a decision tree can be homogeneous and/or heterogeneous. If the class values of the records in a leaf are the same, then the leaf is called a homogeneous leaf, otherwise, it is a heterogeneous leaf.

Table 1 Accuracy of the techniques
Full size table
Table 2 Recall of the techniques
Full size table
Table 3 Precision of the techniques
Full size table
Table 4 F1-score of the techniques
Full size table
Fig. 1
figure 1

A branch of the decision tree considering State FOMO as the class attribute

Full size image

In Step 4, we obtain the insights of each model. For the J48 decision tree (as an example), we obtain insights by examining the patterns of each decision tree. The patterns from each decision tree model are written in a.csv file by using a Java program that takes the output of the decision tree as input. The number of patterns in a decision tree is the same as the number of leaves in that decision tree. The support (i.e., number /of records) in a leaf, the number of records in a leaf with the majority class value, the number of miss-classified records, and the confidence of the majority class value in the leaf are also stored in the.csv file. The confidence for each pattern is calculated using the following formula:

$$Confidence = \frac{{{\text{TR}} - {\text{MR}}}}{{{\text{TR}}}}$$
(1)

where TR denotes the number of records in a leaf and MR denotes the number of misclassified records in the leaf. For all the algorithms, we obtain the classification accuracy, recall, precision, and F1-score values to provide a broad set of metrics to best handle possible dataset challenges such as class imbalance.

4 Results

As mentioned in Sect. 3.1, we used two datasets from two previous studies on phubbing. After removing the records with missing values (Step 2) there were 352 records in the Study 1 dataset and 325 records in the Study 2 dataset. In the Study 1 dataset, there were three independent variables and two dependent variables. Age, gender, and geographical area were independent variables whereas trait boredom and phubbing were dependent variables. The domain values of trait boredom were {bored, not-bored} and for phubbing, they were {phubber, not-phubber}.

There were six independent variables and four dependent variables in the Study 2 dataset. Age, gender, geographical area, the number of people the participants follow on social media (i.e., followees), the number of people who follow the participants on social media (i.e., followers), and daily time spent on social media (in minutes) were independent variables, whereas lonely, state FOMO, state boredom, and phubbing were dependent variables. For each dependent variable, a separate analysis was performed. The domain values of lonely were {lonely and not-lonely}, the domain values of state FOMO were {slight to extreme and never}, the domain values of state boredom were {agree, disagree, and neutral}, and the domain values of phubbing were {phubber and not-phubber}.

4.1 Models

We compared the performance of the techniques in terms of accuracy, recall, precision, and F1 score (Rahman 2014; Frank et al. 2005). The performance of the models is presented in Tables 1, 2, 3, and 4, respectively. The accuracy is defined as the total number of correctly classified records divided by the total number of records in a dataset (Quinlan 1996, 1993; Frank et al. 2005). If TP is the number of true positive records, FP is the number of false positive records, FN is the number of false negative records and FP is the number of false positive records then accuracy is calculated using the following formula:

$$Accuracy = \frac{TP + TN}{{TP + TN + FP + FN}}$$
(2)

The F1 score is calculated based on precision and recall. The precision is defined as the number of positive class predictions that belong to the positive class, whereas the recall is defined as the number of positive class predictions of all positive records in the dataset (Frank et al. 2005). The formulas for precision, recall, and F1-score are given below:

$$Precision = \frac{{{\text{TP}}}}{{{\text{TP}} + {\text{FP}}}}$$
(3)
$$Recall = \frac{{{\text{TP}}}}{{{\text{TP}} + {\text{FN}}}}$$
(4)
$$F1{ - }score = \frac{{2{*}Precision{*}Recall}}{Precision + Recall}$$
(5)

We have compared the performance of J48 with three baseline classification techniques namely Random Tree (RT) (Rahman 2014), Random Forest (RF) (Frank et al. xxxx), and Naive Bayes (NB) (Breiman 2000). From Tables 1, 2, 3, and 4, we can see that in terms of accuracy, recall, precision, and F1-score the overall performance of J48 is better than RT, RF and NB. Therefore, we use J48 for model-building purposes. Moreover, from J48 we can easily extract the logic rules (i.e., patterns) that are used to understand why a particular incident has occurred. The two models using J48 for the Study 1 dataset are labelled Trait Boredom and Phubbing-Study 1 whereas Loneliness, State Boredom, State FOMO, and Phubbing-Study 2 are the four models from the Study 2 dataset. Some interesting patterns that emerged from these models are presented in the following sections.

4.1.1 Loneliness model

We present some interesting patterns of the loneliness model in Table 5. It appears from Table 4 that age, gender, geographical area and, to some extent, the number of people the participants follow on social media and daily time spent on social media are important factors in the experience of loneliness among the research participants. Specifically, those under the age of 44 years and who live in a regional area are likely to report feeling lonely. Males living in a metropolitan area whose number of followees is less than or equal to 382 are also likely to report feeling lonely. In the case of females, those who are less than or equal to 24 years of age are likely to report feeling lonely, if they were living in a metropolitan area and their number of followees were less than or equal to 382.

Table 5 Patterns related to loneliness
Full size table

However, if females with these characteristics were above the age of 24, then they were less likely to report feeling lonely. On the other hand, participants whose age was above 44 years and who spent less than or equal to 70 min a day on social media are less likely to report feeling lonely.

4.1.2 State Boredom model

In Table 6, we present some interesting patterns of the state boredom model. In a similar vein to patterns from the previous model, age, gender geographical area and, to some extent, the number of followers on social media and daily time spent on social media seem to influence the research participants' temporary experience of feeling bored.

Table 6 Patterns related to state boredom
Full size table

With strong confidence but weak support, participants above 47 years of age are likely to momentarily feel bored. Females under the age of 47 who live in a regional area and spend less than an hour and a half on social media and whose number of followers is greater than 22 are likely to feel momentarily bored. The same can be said about females under the age of 47 who live in a metropolitan area and spend less than an hour and a half on social media. The only difference in their case is that their number of followers was not included in the rule. In the case of males, if they are under the age of 47, live in a metropolitan area and their number of followers is under 164 but spend five minutes or less a day on social media then are likely to feel momentarily bored. Likewise, if these men living in a metropolitan area, were under the age of 47 but older than 17 years, and whose daily use of social media is under an hour and 30 min but more than five minutes, then they are likely to report feeling momentarily bored.

4.1.3 State FOMO model

In Table 7, we present some interesting patterns of the state FOMO model. With regards to FOMO, daily time spent on social media and age were the crucial factors. Participants who spend more than 20 min a day on social media are likely to experience some degree of FOMO. Participants under the age of 30 who spend less than 20 min a day on social media are likely to experience some form of FOMO. However, participants above the age of 30 who spend less than 2 min a day on social media are not likely to experience any FOMO.

Table 7 Patterns related to state FOMO
Full size table

4.1.4 Phubbing (Study 2 dataset) model

We present some interesting patterns of the Phubbing (Study 2 dataset) model in Table 8. Several individual characteristics provide clues on who is likely to phub others including daily spend on social media, gender, geographical area and number of followers. Participants who spend more than three minutes on social media are likely to engage in phubbing behaviour. If the participant is female above the age of 27 and living in a metropolitan area, then they are likely to phub others if their number of followees is less than 10 even if the time spent on special media is just one minute. The same can’t be said about males who share these characteristics with females even if the time spent on special media is under three minutes.

Table 8 Patterns related to phubbing (Study 2 dataset)
Full size table

4.1.5 Trait Boredom model

We present some interesting patterns of the trait boredom model in Table 9. Contrary to patterns from the previous models, male participants under the age of 31 living in a regional area are likely to be bored if they occasionally or sometimes phub others. Participants under the age of 22 are likely to be bored if they rarely or occasionally phub others. Interestingly, participants under the age of 55 are likely to be bored if they most frequently phub others. However, female participants under the age of 55 are not likely to be bored if they only occasionally or sometimes phub others. Likewise, female participants between the ages of 22 and 55 are not likely to be bored if they only rarely or occasionally phub others.

Table 9 Patterns related to trait boredom
Full size table

4.1.6 Phubbing (Study 1 dataset) model

We present some interesting patterns of the Phubbing (Study 1 dataset) model in Table 10. Phubbing and boredom are interrelated. Participants who are not feeling bored and whose age is under 44 years, then they are likely to engage in phubbing behaviour. Similarly, if participants are not feeling bored and whose age is above 44 years but who live in a metropolitan area then they are likely to engage in phubbing behaviour. However, if participants are males who live in a metropolitan area, then they are likely to engage in phubbing behaviour if they feel bored if age is not included in the equation. In the case of females, if they are feeling bored, they are likely to engage in phubbing behaviour.

Table 10 Patterns related to phubbing (Study 1 dataset)
Full size table

5 Discussion

This section discusses the key results from this research to relate these results to previous research findings that primarily employed the scientific method of hypothesis testing.

Loneliness and FOMO appeared to be strongly influenced by age and time spent on social media. Participants under 44 who lived in a regional area are likely to feel lonely. This is consistent with the literature. In Australia, for example, where one in three feel lonely,Footnote 1 young people and middle-aged people are lonelier than other age groups and those in regional areas are more likely than those in metropolitan areas to feel socially isolated. On the other hand, those over 44 are not likely to feel lonely if they spend less than or equal to 70 min per day on social media. Indeed, Australians who reported feeling lonely, are more likely to have a social media addiction. A recent study (Al-Saggaf 2021) found that respondents in regional areas spent less time on social media compared to their metropolitan counterparts. It may be that the reason that participants under 44 feel lonely is not because they spend less time on social media, but because they live in a regional area as the first rule suggested. Likewise, the reason participants in metropolitan areas, especially those over 44, are not likely to feel lonely is not because they spend more time on social media, but because they spend less time on social media. Both males and females under the age of 24, are likely to feel lonely if they live in a metropolitan area and followed less than or equal to 382 social media contacts. This finding is not surprising. A study (Al-Saggaf et al. 2016), which adopted the scientific method, has revealed that social media users who expressed feelings of loneliness had smaller network sizes (M = 307).

With FOMO, if social media users spend just over 20 min a day on social media, they are likely to experience fear of missing out. If they are under 30 and spend less than 20 min a day, they are likely to experience FOMO. A recent study (Al-Saggaf 2021) has revealed that state FOMO predicted time spent on social media. An increase of one unit of change in state FOMO resulted in an increase of six minutes spent on social media. The current study, however, suggests that the more time one spends on social media, the more likely they are to experience fear of missing out. The variation in the direction of the relationship between time spent on social media and FOMO does not mean the results of these studies are inconsistent. On the contrary, it means these two influence each other. The more time spent on social media, the more likely one is to experience FOMO and the more one experiences FOMO, the more likely they will spend more time on social media. This is especially true if the individual is younger. On the other hand, those over 30, will not experience FOMO if they spend less than 2 min a day, which suggests the less time spent on social media, the less likely one is to experience FOMO.

State boredom and trait boredom were both influenced by age, gender, and geographic location. In this study, participants over 47 are likely to feel fleetingly bored. This is not in line with a recent study on the relationship between phubbing, fear of missing out and boredom (Al-Saggaf Jun. 2021), which found that as age increased, state boredom decreased. This discrepancy is worth investigating in a future study. Males under 31 who live in a regional area and tend to phub others are likely to feel chronically bored. This is consistent with the findings of two recent studies, (Al-Saggaf et al. 2019; Boylan et al. 2021), that found that as age increased, trait boredom decreased. However, the pattern in this study applies only to those who tend to phub others more, it does not apply to the association between boredom proneness and age in general. Females under the age of 47 who live in a regional area, spend less than an hour and a half on social media, and who have more than 22 followers on social media are likely to feel temporarily bored. Their male counterparts who live in a metropolitan area, spend under five minutes on social media, and have less than 164 may feel bored from time to time. In Al-Saggaf and O’Donnell’s (Al-Saggaf and O’Donnell 2019b) study, the associations between gender and state boredom and geographical location and state boredom were not statistically significant. The trend in this study must have been influenced more by time spent on social media and social media network size. On the other hand, Al-Saggaf (Al-Saggaf 2021) found that participants in regional areas spent less time on social media but this research offered additional clues about the characteristics of those participants that were females under the age of 47. Participants under the age of 55 who frequently phub others are susceptible to long-term boredom.

Age, gender, and geographic location also affected the predisposition to phubbing. Female participants above 27 who live in a metropolitan area, whose number of followees is under 10, and who spend at least a minute daily on social media are likely to phub others. Male participants, who live in a metropolitan area but who report feeling bored are likely to phub others. Likewise, participants over the age of 44 who live in a metropolitan area and who are not bored are likely to phub others. These observations are all novel and to the best of our knowledge, they have not been reported previously in the literature.

6 Conclusion

This paper presented a data-driven approach to discover previously unknown patterns about the use of the smartphone and social media, phubbing, state FOMO, loneliness, state boredom and trait boredom using two datasets. We built several models using two datasets to discover patterns relating to these constructs. These patterns are novel as they have not been reported in previous research. Studies that employed the scientific method of hypothesis testing, tested a single dependent variable. This limitation does not apply to the data mining approach. Thus, more than one dependent variable was tested in this research.

The patterns of the models showed that phubbing, state FOMO, loneliness, state boredom and trait boredom are not dependent on a single attribute. The patterns about loneliness and FOMO showed that they are strongly influenced by age and time spent on social media. The patterns about FOMO showed that social media users who spent more than 20 min daily on social media are likely to experience FOMO. Age, gender, and geographic location are common factors affecting both state and trait boredom. The participants under the age of 55 who frequently phub others are vulnerable to long-term boredom. Age, gender, and geographic location also affected the predisposition to phubbing.

Three limitations in our study should be outlined. First, the number of participants in each dataset is relatively small. Second, while each dataset encompassed a wide range of demographics and geographies, the inclusion of an even wider range of demographics and geographies would have no doubt significantly improved the generalizability and relevance of the findings to a broader population with diverse cultural and socio-economic backgrounds. That said the literature in this area (see (Al-Saggaf 2022) for more information), has indicated that not all demographic variables (attributes from a data mining perspective) are determinantal to how emotional states are experienced. While age has consistently been found to be a significant factor in the relationship between social media and smartphone use and emotional states, gender and regionality, i.e., metropolitan vs regional, have not been influential. Third, is that both surveys, from which our datasets originated, used cross-sectional data. It is true that adopting longitudinal data to discover patterns around the use of social media and smartphones and how they impact individuals’ emotional states would have provided deeper insights into the cause-and-effect relationship between technology use and emotional states, but the aim of this research was not to establish a cause-and-effect relationship. The aim was to discover patterns around the use of technology and emotional states. Future research should adopt a longitudinal design to investigate the cause-and-effect relationship between these variables thereby facilitating the creation of a more comprehensive understanding of these complex relationships.

As part of future work, we plan to collect a new dataset with more attributes and more participants. We also plan to include a more balanced set of participants from different countries with culturally and linguistically diverse backgrounds to explore what effects this might have on the results.