Predicting Subjective Well-Being by Smartphone Usage Behaviors

Yusong Gao

, He Li

and Tingshao Zhu

Institute of Psychology, Chinese Academy of Sciences, 16 Lincui Road, Beijing 100101, China

The 6th Research Institute of China Electronics Corporation, Beijing 100083, China

Keywords: Smartphone Usage Behaviours, Subjective Well-Being, Mental Health.

Abstract: Subjective Well-Being (SWB) refers to how people experience the quality of their lives, thus to acquire

people’s SWB levels timely and effectively is very important. Self-report and interviewing are mostly used

techniques for assessing SWB, but cannot be done in time. This study aims to predict one’s SWB levels by

smartphone usage behaviours. We collect users’ smartphone usage and self-reported subjective well-being,

and found that several usage behaviours correlate with SWB, especially for females. For example,

smartphone users with higher SWB scores tend to use more communicating apps, play more games and read

more, but take fewer photos. Based on these findings, we trained a predicting model of user’s SWB based

on smartphone usage behaviours, and the accuracy is up to 62%. The result indicates that SWB can be

identified based on smartphone usage fairly well.

1 INTRODUCTION

Subjective well-being (SWB) means how people

experience the quality of their lives, comprising

longer-term levels of pleasant affect, lack of

unpleasant affect, and life satisfaction (Angner, 2010;

Diener, 1994). SWB focus on how a person

evaluates his/her own life, including emotional

experiences of pleasure versus pain in response to

specific events and cognitive evaluations of what a

person considers a good life (Diener, 2000).

Positive psychologists have done much research

on SWB (

Lyubomirsky, 2001), and they claimed that

the pursuit of happiness is regarded as one of the

most valued goals in almost every culture. Some

studies found that health and SWB influence each

other, as good health leads to greater happiness

(Okun, 1984), and a number of studies also found

that positive emotions and optimism have a

beneficial influence on health (Frey, 2011).

As SWB plays an important role in both mental

and physical health of people, it is critical to acquire

either individual or population SWB level. Currently,

SWB is generally measured by using self-reported

questionnaires, in which participants fill in the

related scales to get the SWB levels. However, this

method has three disadvantages.

1. It’s not convenient to collect SWB on large

population. It may take a lot of manpower

and material resource, and it is prone to

human errors in the process of questionnaires.

2. It is not easy to identify the driving force of

SWB level, if only SWB is presented. We

cannot identify any detailed personal

conditions or daily behaviours that can

explain his or her mental state.

3. Sometimes, it is difficult to acquire SWB in

time. Self-reported method is always time-

consuming, and conditions may have some

changes after the data process (i.e.,

questionnaire results) has been totally

completed.

In order to avoid these disadvantages, we

propose to predict SWB based on users’ phone

usage behaviours. Since a smartphone is built on a

mobile operating system with advanced computing

capability, it becomes an indispensable part of our

daily lives as an attractive tool for communication

and interpersonal interaction. IDC (the Internet Data

Center) statistics show that in the first quarter of

2013, total global mobile phone sales to 418.6

million, in which smartphones to 216.2 million,

accounting for 51.6% of the total mobile phone sales

volume.

There are various definitions of phone usage

behaviour, some researchers take it as part of the

functions of mobile phones, such as telephone,

calendar, SMS using frequency, etc. (Coen et al.,

2003). Suh et al. took some specific phone features

317

Gao Y., Li H. and Zhu T..

Predicting Subjective Well-Being by Smartphone Usage Behaviors.

DOI: 10.5220/0004800203170322

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 317-322

ISBN: 978-989-758-010-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

as phone usage behaviour, such as mobile e-

commerce usage (Suh et al., 2003). Falaki et al.

described it as users’ interactions with the device

and the applications used (Soikkeli, Karikoski and

Hammainen, 2011). In this paper, we investigate the

interactions between users and their devices to

explore individual’s phone usage behaviour, such as

frequency of telephone, SMS, application use, phone

switch habits, wallpaper changing, GPS using, etc.

Since smartphone is now a very powerful

platform, it is possible to record various interactions

between users and devices, ranging from the most

basic communication functions to a variety of third-

party applications use, such as the frequency of

making calls, sending text messages, App using,

GPS using. This data intensive platform provides a

new opportunity, as it allows us to study users’

various phone usage behaviours.

In order to record user's usage behaviour

automatically, we have developed an Android

application named MobileSens (Li et al., 2013; Guo

et al., 2011). This application can record smart

phone usage and upload data to the server. At the

same time, the application also allows the user to fill

a questionnaire and upload the answers to the server.

In this paper, we recruited 106 participants using

smartphones to record phone usage and SWB scores.

We analysed relations between user’s phone usage

and SWB, and found that it is possible to predict

SWB by phone usage behaviours with an accuracy

of 62%.

The rest of the paper is organized as following:

Section 2 introduces some related works conducted

by other researchers. We describe the details of our

data in Section 3. Section 4 mainly describes the

process of feature extraction, and the results will be

thoroughly discussed in Section 5. Section 6

concludes our work in this paper and gives a brief

discussion on future work.

2 RELATED WORKS

Much research has been conducted to investigate the

users’ psychological characteristics and their usage

of Internet and social media, such as Youtube, blogs,

Facebook. It is reported that Internet and social

media usage can reflect psychological characteristics

(Back et al., 2010; Biel et al., 2011; Counts and

Stecher, 2009; Stecher, 2008; Yeo, 2010).

Since smartphones become one kind of

ubiquitous communication and interpersonal

interaction tools nowadays, smartphone usage may

reflect our mood and mental health. Traditional

researches mostly used questionnaires to acquire

users’ phone usage preference, and then analysed

these data with the psychological scale scores. Butt

pointed out that mobile phone supports interpersonal

interaction, and concluded that psychological theory

can explain patterns of mobile phone use (

Butt and

Phillips, 2008). Ehrenberg analysed mobile IM

(instant messaging, IM) application use, and found

agreeable and low self-esteem have a negative

relationship with IM-related applications usage

among teenagers (Ehrenberg and Juckes, 2008).

With a self-designed questionnaire assessing

problematic mobile phone use, Billieux found

impulsivity played a specific role in mobile phones

usage (Billieux and Van der Linden, 2008).

Chittaranjan found some aggregated features

obtained from smartphone usage data can indict the

Big-Five traits (Chittaranjan and Blom, 2011). Gross

et al. reported that time spent on-line was not

associated with dispositional or daily well-being.

However, the closeness of instant message

communication partners was associated with daily

social anxiety and loneliness in school, above and

beyond the contribution of dispositional measures

(Gross et al., 2002).

Some research have been done to investigate

mobile phone usage and psychological

characteristics. Chittaranjan installed a data

collection procedure into Nokia N95, collected 117

participants’ phone use logs during 17 months,

including text messaging, applications, phone card

use logging (Chittaranjan et al., 2011). They found

several aggregated features obtained from

smartphone usage can be indicators of the Big-Five

traits. LiKamWa developed an iPhone software

system to record phone use data and remained user

of noting down their mood regularly (LiKamWa,

2013). He found smartphone use patterns fluctuate

as mood changes, and he built a smartphone

software system which can infer user’s mood by

how the smartphone is used.

In this study, we explore relationship between

smartphone usage and SWB using data collected

automatically. Differing from studies mentioned

above, we intend to predict users’ SWB based on

smartphone usage, thus to acquire the user’s SWB

timely and accurately.

3 METHODS

After MobileSens has been implemented, this study

was carried out by three steps. Firstly, we recruited

106 participants who owned Android smartphones

HEALTHINF2014-InternationalConferenceonHealthInformatics

318

and also use phones in daily life. During the

experimental time (one month), all the participants

were required to fill out the questionnaire at least

once, and should keep the phone network

connectivity when phone use records were

uploading. Secondly, we extract 48 phone use

behavioural features, including SMS, telephone use

and wallpaper changed frequency, GPS use

frequency, etc. We analysed the relationships

between these features and user’s well-being by

calculating the correlation coefficients. Finally, we

extract features and build a prediction model using

data mining method.

3.1 Participants

We recruited participants by advertising on social

networks. Finally, 98 participants between 18 and 32

years (mean 23.5, SD=2.48) were finally included in

the data analysis; 61.22% were male, and 38.78%

were female. Most of the participants are college

students and postgraduates, and Fig. 1 shows

detailed education levels of the participants.

Figure 1: Participants Education level distribution.

3.2 Data Collection

MobileSens consists of two modules, one is for

collecting phone usage, and the other is for filling

the questionnaire, which can get usage data and the

corresponding questionnaire results at the same time.

Its framework is shown in Fig. 2.

Figure 2: MobileSens Modules.

After one user installed MobileSens into their

Android devices, most of their interactions with the

device were recorded in the Android Sqlite, and

these data would upload to our server later.

MobileSens recorded 14 categories of information,

as listed in Tab. 1.

3.3 Psychological Scale

In our study, SWB is assessed by Satisfaction With

Life Scale (SWLS), which was developed to assess

subjective feelings of well-being (Diener et al.,

1985). SWLS is a single scale that has been used by

UNESCO (The United Nations Educational,

Scientific and Cultural Organization), the CIA (the

Table 1: Categories of Phone usage information and details.

Type Record Content

activity application log creating, starting, resuming, stopping, and exiting of activity application

application package log adding, changing, and removing package

calling log state, number, contact, and direction of calling

configuration log configuration change information (e.g., font, screen size, and keyboard type)

contact log adding, changing, and deleting of contacts

date changed log changing of system date and time

GPS log user’s locale, altitude, latitude, longitude and direction of movement

headset log plugging in headset or not

power connected log connecting or disconnecting the power

power log powering on smartphone or not

screen log state of the screen

service application log creating, starting, and deleting service application

sms log state, content, and contacts of SMS

wallpaper log changing wallpaper

PredictingSubjectiveWell-BeingbySmartphoneUsageBehaviors

319

Central Intelligence Agency), the New Economics

Foundation, and the WHO (the World Health

Organization) to measure how one views his or her

self-esteem, well-being and overall happiness with

life (

Diener, 1999

). Among the various components

of subjective well-being, SWLS focuses on assess

global life satisfaction and does not tap related

constructs such as positive affect or loneliness.

SWLS items are rated on a 7-point scale: 0 =

strongly disagree, 1 = disagree, 2 = slightly disagree,

3 = neither agree nor disagree, 4 =slightly agree, 5 =

agree, 6 = strongly agree. Five items are shown as

follows.

1. In most ways my life is close to my ideal.

2. The conditions of my life are excellent.

3. I am satisfied with my life.

4. So far I have gotten the important things I

want in life.

5. If I could live my life over, I would change

almost nothing.

The distribution of participants’ SWB scores in

this study is shown in Tab. 2.

Table 2: The mean, standard deviation and range of SWLS

scores in this study.

Mean StdDev Range

Male 14.600 6.239 4~30

Female 15.395 5.299 7~28

Total 14.908 5.877 3~30

4 FEATURES EXTRACTION

After data collection, we then extract several

features from smartphone usage. There are three

types of features extracted as shown in Fig. 3, such

as the frequency of some basic smartphone functions

usage. We classify Apps according to Wandoujia

Android market (http://www.wandoujia.com/apps),

which is very popular and comprehensive in China.

Figure 3: The extracted feature classification and details.

48 features are extracted from phone usage data

by MobileSens running as a background service.

Therefore, these features were objective and

captured various aspects of communication and

applications use on the phone.

5 DATA ANALYSIS

5.1 Correlation Analysis

In this section, we run statistical analysis to examine

the correlation between smartphone usage and SWB

across genders.

In psychology literature, Pearson’s correlation

coefficient is commonly used as a bounded measure

of correlation, or linear dependence between two

variables. For two random variables X and Y, it is

given by



X,



√

 ∙ 

(1)

where cov(X,Y) is the covariance between the

random variables X and Y. r=1 denotes a positive

sloped linear relationship, and r=-1 denotes a

negative one. Values in-between -1 and 1 indicate

sub-linear relationships between the variables.

We compute the Pearson’s correlation coefficient

between the Well-being and the smartphone

features. We enlist parts of the correlation analysis

results where p<0.05 in Table 3.

Table 3: Correlations between Well-being and parts of

Smartphone use features.

Female Male Total

Communication

App using

0.371* 0.234* 0.251*

Strategy Game 0.320* 0.004 -0.002

Reading App using 0.299* -0.099 0.063

Camera using -0.311* -0.242* -0.251*

Competitive Game -0.050 0.265* 0.197*

Browser using 0.368* -0.027 0.124

*p < 0.05

Table 3 shows that users with high SWB tend to

use communication Apps (MSN, weixin, feixin)

more frequently, but less camera Apps usage,

especially for female participants. Besides, all the

users with higher SWB are more likely to play

games, with females prefer strategy games while

males prefer competitive games. In addition, females

with higher Well-being may use reading Apps and

browsers more frequently, which is not significant

on male participants.

HEALTHINF2014-InternationalConferenceonHealthInformatics

320

5.2 Prediction Model

To identify SWB based on smartphone usage

behaviours, we train prediction models by machine

learning.

To achieve better performance, we need to get

more meaningful features at first. In this study, we

use StepWise regression as a feature extraction

method. StepWise Regression algorithm is a

regression-based algorithm for automatic filtering

features, and this selection method is applicable to a

variety of other models (Kwak and

Choi, 2002). It is

a forward selection algorithm, which involves

starting with no variables in the model, testing the

addition of each variable using a chosen model

comparison criterion, adding the variable (if any)

that improves the model the most, and repeating this

process until none improves the model. In this study,

we have adopted StepWise regression algorithm for

feature selection. The selection result shows that

communicate apps usage, certain types of game

(strategy games are for female and competitive

games are for male), and photo-taken frequency are

the most important variables for predicting SWB.

After feature extraction, we built prediction

models on WEKA. WEKA contains a collection of

visualization tools and algorithms for data analysis

and prediction model, together with graphical user

interfaces (Witten and

Frank, 2005).

In this study, we choose pace regression to build

the prediction model in WEKA. Pace regression is

one kind of linear regression algorithm, to create

parametric linear relationship between independent

variables and dependent variable (Wang, 1999).

Compared with other algorithms, pace regression

performs much better in this study. To make the

trained model more general, we use four criterions to

measure the quality of prediction model.

1. Correlation Coefficient (CORR)

(X )(Y )

cov( X, Y)

XY XY

EXY



 













(2)

2. Mean Absolute Error (MAE):

(x ) y

MAE f







(3)

3. Root Mean Squared Error (RMSE):

((x) y)

RMSE f







(4)

The results of CORR, MAE and RMSE are

shown in Table 4.

Table 4: Results of the prediction model.

CORR MAE RMAE

Female 0.62 3.49 4.23

Male 0.34 4.82 5.89

Total 0.39 4.76 5.92

We can see that the prediction model for females

obviously get a much higher accuracy than models

for males or total users. The reason for the result

hasn’t been found up to now, but we have supposed

that females interact with their smartphones more by

using more kinds of Apps and spending more time

on phones than male users. Therefore, we can

predict females’ SWB more easily based on females’

rich datasets.

6 CONCLUSION

AND FUTURE WORK

This paper mainly investigates how smartphone

usage correlates with SWB, and further develops a

SWB prediction model. We intend to provide an

alternative method to measure subjective well-being

based on smartphone usage behaviours.

The result demonstrates that users with different

SWB level behave differently on their smartphones.

Users with higher SWB would like to use more

communication Apps but less camera Apps,

especially for females users. Besides, all the users

with higher well-being are more likely to play

games, female users prefer to strategy games and

males play competitive games more. In addition,

females with higher Well-being use reading Apps

and browsers more frequently, which does not

appear on male users.

In the near future, we plan to tune the

performance of prediction model by developing new

behavioural features, and take into account short

message and phone voice. We also intend to

implement SWB predicting App, to get more

feedback from real users.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the generous

support from National High-tech R&D Program of

China (2013AA01A606), National Basic Research

Program of China (973 Program， 2014CB744600),

PredictingSubjectiveWell-BeingbySmartphoneUsageBehaviors

321

Key Research Program of CAS (KJZD-EW-L04),

Strategic Priority Research Program

(XDA06030800) and 100-Talent Project

(Y2CX093006) from Chinese Academy of Sciences.

REFERENCES

Angner, E., 2010. Subjective well-being. The Journal of

Socio-Economics, 39(3), 361-368.

Diener, E., 1994. Assessing subjective well-being:

Progress and opportunities. Social indicators research,

31, 103--157.

Diener, E., 2000. Subjective well-being: The science of

happiness and a proposal for a national index.

American psychologist, 55(1), 34.

Lyubomirsky, S., 2001. Why are some people happier than

others? The role of cognitive and motivational

processes in well-being. American Psychologist,56(3),

239.

Billieux, J., Van der Linden, M., & Rochat, L., 2008. The

role of impulsivity in actual and problematic use of the

mobile phone. Applied Cognitive Psychology, 22(9),

1195-1210.

Chittaranjan, G., Blom, J., & Gatica-Perez, D., 2011.

Mining large-scale smartphone data for personality

studies. Personal and Ubiquitous Computing, 17(3),

433-450.

Ehrenberg, A., Juckes, S., White, K. M., & Walsh, S. P.,

2008. Personality and self-esteem as predictors of

young people's technology use. Cyberpsychol Behav,

11(6), 739-741.

Soikkeli, T., Karikoski, J., & Hammainen, H., 2011.

Diversity and End User Context in Smartphone Usage

Sessions. 7-12.

Frey, B. S., 2011. Happy people live longer. Science,

331(6017), 542-543.

Coen A, Dai L, Herzig S, et al., 2002. The Analysis:

Investigation of the cellular phone industry [J].

Retrieved October, 2002, 21: 2006.

Suh, B., & Han, I., 2003. Effect of trust on customer

acceptance of Internet banking. Electronic Commerce

research and applications, 1(3), 247-263.

Back MD, Stopfer JM, Vazire S, Gaddis S, Schmukle SC,

Egloff B, Gosling SD., 2010. Facebook profiles reflect

actual personality, not self-idealization. Psychol Sci

21:372–374

Biel, J. I., Aran, O., & Gatica-Perez, D., 2011. You Are

Known by How You Vlog: Personality Impressions

and Nonverbal Behavior in YouTube. In ICWSM.

Counts, S., & Stecher, K. B., 2009. Self-Presentation of

Personality During Online Profile Creation. In ICWSM.

Stecher, K. B., & Counts, S., 2008. Spontaneous Inference

of Personality Traits and Effects on Memory for

Online Profiles. In ICWSM.

Yeo, T. D., 2010. Modeling Personality Influences on

YouTube Usage. In ICWSM.

Butt, S., & Phillips, J. G., 2008. Personality and self

reported mobile phone use. Computers in Human

Behavior, 24(2), 346-360.

Gross, E. F., Juvonen, J., & Gable, S. L., 2002. Internet

use and well-being in adolescence. Journal of Social

Issues, 58(1), 75-90.

LiKamWa R., 2013. MoodScope: Building a Mood Sensor

from Smartphone Use Patterns. In ACM.

Diener, E. D., Emmons, R. A., Larsen, R. J., & Griffin, S.,

1985. The satisfaction with life scale. Journal of

personality assessment, 49(1), 71-75.

Diener, E., Suh, E. M., Lucas, R. E., & Smith, H. L., 1999.

Subjective well-being: Three decades of

progress. Psychological bulletin, 125(2), 276.

Kwak, N., & Choi, C. H., 2002. Input feature selection for

classification problems. Neural Networks, IEEE

Transactions on, 13(1), 143-159.

Witten, I. H., & Frank, E., 2005. Data Mining: Practical

machine learning tools and techniques. Morgan

Kaufmann.

Wang, Y., & Witten, I. H., 1999. Pace regression.

Li, A., Li, H., Guo, R., & Zhu, T., 2013. MobileSens: A

Ubiquitous Psychological Laboratory based on Mobile

Device. International Journal of Cyber Behavior,

Psychology and Learning (IJCBPL), 3(2), 47-55.

Guo, R., Zhu, T., Wang, Y., & Xu, X., 2011. MobileSens:

A framework of behavior logger on Andriod mobile

device. In Pervasive Computing and Applications

(ICPCA), 2011 6th International Conference on. IEEE,

2011: 281-286.

Workshop on New Computationally-Enabled Theoretical

Models to Support Health Behavior Change and

Maintenance. (2012). Realizing effective behavioural

management of health. Available:

http://www.behaviorchange.be/

HEALTHINF2014-InternationalConferenceonHealthInformatics

322