Internet use and psychological well-being among college students: A latent profile approach

(1)

Internet use and psychological well-being among college students: A latent

proﬁle approach

Ssu-Kuang Chen

⇑

Institute of Education, National Chiao Tung University, 1001 Ta-Hsueh Rd., Hsinchu 300, Taiwan, ROC

a r t i c l e

i n f o

Article history:

Available online 19 July 2012 Keywords:

Internet use

Problematic Internet use Psychological well-being Latent proﬁle analysis

a b s t r a c t

Studies of correlations between Internet use and psychological well-being (PW) have produced mixed results. The present study used a latent profile analysis to distinguish among populations in terms of PW profiles, and then used a multinomial logistic regression to determine how online entertainment, social use, problematic Internet use (PIU), and gender predicted each latent PW profile. The initial sample consisted of 757 Taiwanese college freshmen. Four ordered latent groups were established as follows: good PW, normative, minor-disadvantageous, and severe-disadvantageous. No latent PW profile stability was found in subsequent years: three groups (normative, minor-disadvantageous and severe-disadvan-tageous) emerged in Year 2, and only one group emerged in Year 3. The results indicate (a) no relation-ship between PW and online entertainment or gender, (b) greater PIU increased the likelihood of disadvantageous PW and decreased the probability of good PW, and (c) greater use of online resources for social purposes was related to increased probability of a participant being in the good PW group, but not associated with fewer PW problems.

1. Introduction

Due to their unlimited access to the Internet via a range of wire-less tools, today’s college students tend to spend large amounts of time online, and therefore have signiﬁcant potential to experience symptoms of problematic Internet use (PIU), for example, remain-ing online longer than intended. The impacts of Internet use on psychological well-being (PW) are receiving greater research attention, but with mixed results and conclusions (Huang, 2010). Some researchers have found positive effects in terms of social relationships (McKenna, Green, & Gleason, 2002), while others re-port that prolonged Internet usage increases the potential for depression, loneliness, and other psychological and emotional problems (Ceyhan & Ceyhan, 2008). Still others have gathered evi-dence indicating concurrent positive and negative effects of Inter-net usage (Campbell, Cumming, & Hughes, 2006).

Huang (2010)points out that most researchers in this area have

relied on correlation analyses when reporting statistically signiﬁ-cant relationships between PW and Internet usage, however, many have reported small effect sizes. This implies that PW may need to be treated as categorical in order to capture more complex and per-haps non-linear relationships (see alsoPasta, 2009). In addition, most Internet use researchers have employed individual PW variables such as observed or latent depression and loneliness as

outcomes for their statistical analyses, with few considering the specific PW profiles of individual Internet users. When performing latent profile analyses, Internet users are classified so that all individuals in a group have similar PW profiles. Identifying latent profiles and classifying students according to their individual PW profiles is likely to provide insights to individual differences in psychological development (Magnusson & Cairns, 1996) and to support investigations of relationships between groups and Inter-net-use covariates.

The primary objectives for the present study were to identify la-tent PW proﬁles for a group of Taiwanese college students; to uti-lize the resulting information to evaluate connections between those proﬁles and frequency of online entertainment, frequency of social Internet use, and level of Internet dependence; and to examine the effect of gender. A cross-sectional approach was used to analyze data collected during the participants’ freshman, soph-omore, and junior years.

1.1. Positive impact of Internet usage on PW

Today, a large number of individuals use the Internet for signif-icant proportions of their social needs, including support and afﬁr-mation (Sproull & Faraj, 1997). According toCaplan and Turner

(2007), the anonymity and absence of non-verbal and demographic

cues provide Internet users relief from emotional distress. As a result, adolescents feel more comfortable to communicate with people online. Wei (2000) asserts that the Internet facilitates

http://dx.doi.org/10.1016/j.chb.2012.06.029

⇑ Tel.: +886 3 5712121x58065; fax: +886 3 5738083. E-mail address:Schen75025@gmail.com

Contents lists available atSciVerse ScienceDirect

Computers in Human Behavior

(2)

adolescent development by deemphasizing family ties and expanding ranges of interpersonal relationships. Cotten (2008)

suggests that Internet access helps students make the transition from high school to college by decreasing uncertainty, improving communication, and fostering online social interaction.

When emphasizing the positive impacts of Internet social usage on PW,Kang (2007)andShaw and Gant (2002)claim that Internet chatting decreases loneliness and depression, increases feelings of happiness, and signiﬁcantly enhances perceptions of social support and self-esteem. Kraut et al. (2002) assert that greater Internet usage leads to better communication and greater social involve-ment, resulting in an enhanced sense of well-being.

Amichai-Ham-burger and Furnham (2007)also provide evidence supporting the

argument that when used appropriately, the Internet can greatly improve user quality of life and PW.

1.2. Negative impacts of Internet usage on PW

Another group of researchers considers the lack of non-verbal cues and physical contact in Internet communication potentially problematic. These researchers argue that online interactions fail to provide sufﬁcient depth or feelings of support associated with in-person relationships, resulting in an overall decrease in well-being (Green et al., 2005).Kraut et al. (1998)assert that excessive Internet usage exerts a negative effect on face-to-face interactions by reducing time spent with friends and family members, thereby decreasing PW (see also Valkenburg & Peter, 2007). van den

Eijnden, Meerkerk, Vermulst, Spijkerman, and Engels (2008)report

a link between excessive use of instant messaging and depression in adolescents. In studies involving undergraduates, students using the Internet at excessively high levels have been described as lone-lier (Engelberg & Sjöberg, 2004; Moody, 2001) and showing weaker development in terms of identity and intimacy (Huang, 2006).Lin and Tsai (2002)indicate that Internet usage disturbs interpersonal relationships as well as work and study functions, resulting in re-duced adolescent happiness. According toWhang, Lee, and Chang (2003), Internet-dependent students report higher degrees of lone-liness, depression, and compulsivity compared to non-dependent students. Evidence indicating associations between prolonged Internet use, compulsive behaviors, and psychological problems among college students has been gathered by Ko (2006) and

Hwang and Yang (2004).van der Aa et al. (2009)also suggest that

compulsive Internet use impacts PW.

1.3. Latent proﬁle analysis

The development of extended algorithms supports complex computer analyses of data containing multiple variables, resulting in greater interest in latent models. In contrast to older models with parameters that only describe relationships among observed variables, latent proﬁle models can be used to identify latent and homogeneous populations within large heterogeneous popula-tions, with meaningful groups of individuals proﬁled in terms of observable variables. (For a detailed overview, seeMagidson &

Ver-munt, 2004andMuthén, 2001.) In educational psychology studies,

latent proﬁle analyses of achievement goal orientations have re-sulted in better differentiation and more accurate predictions of achievement-related outcomes (Pastor, Barron, Miller, & Davis,

2007; Tuominen-Soini, Salmela-Aro, & Niemivirta, 2008, 2011).

Furthermore, in studies of teenagers, latent proﬁle analyses have helped in the identiﬁcation of depressed adolescent subtypes, thus providing usable data in support of optimum treatment and pre-vention decisions (Herman, Ostrander, Walkup, Silva, & March, 2007).

1.4. The present study 1.4.1. Deﬁnition of PW

According to the literature, associations exist between PW and Internet usage, yet there is no clear agreement on a deﬁnition of

PW.van der Aa et al. (2009)describe depression, loneliness, and

self-esteem as indicators of PW; others such asvan den Eijnden

et al. (2008)andWhang et al. (2003)exclude self-esteem.Niemz,

Grifﬁths, and Banyard (2005)provide evidence indicating that prob-lematic Internet users are more likely to have low self-esteem. In contrast,Shaw and Gant (2002)believe that PIU affects self-esteem, but not as quickly as it triggers feelings of depression and loneliness. The present study will use a deﬁnition that includes self-esteem. 1.4.2. Type of Internet use

Today’s college students use the Internet for social purposes, leisure activities, information searches, and online shopping ( Mor-gan & Cotten, 2003). There is no signiﬁcant relationship between time spent using the Internet in general and life satisfaction (Green et al., 2005) or psychological adjustment (Gross, Juvonen, & Gable,

2002).Gordon, Juang, and Syed (2007) believe it is necessary to

precisely determine what online activities individuals participate in (as opposed to simply measuring total time spent online). Sup-port for this idea comes fromKim, LaRose, and Peng (2009), who note that compared to social usage, entertainment usage is more likely to result in psychological problems. In addition,Chou,

Con-dron, and Belland (2005)advocate speciﬁc online activities such

as chat rooms or online games are more important in determining PIU than other Internet usages. Therefore, type of Internet use will be considered in the present study.

1.4.3. Gender

Gender differences have been reported in terms of the relation-ships between PW and Internet usage (Amichai-Hamburger &

Ben-Artzi, 2003; Kang, 2007). According to Amichai-Hamburger

and Ben-Artzi (2003), the correlation between social use and lone-liness is stronger for females than for males, while for males the relationship between leisure use and loneliness is stronger.Kang

(2007)observes that female chat users tend to be happier than

male chat users. More research is needed in order to determine whether gender differences exist in terms of correlations between Internet usage and PW.

1.4.4. Study purpose

The present study aimed to determine the most appropriate la-tent classiﬁcations for a group of Taiwanese college students based on their self-reported levels of depression, loneliness, and self-es-teem. After identifying the most appropriate groups, the next step was to determine whether gender, Internet use type, and PIU are adequate predictors for the probability of an individual belonging to a speciﬁc PW category. An important motivation was to test an approach to inspecting the relationship between Internet usage and PW, one that differs from correlational analyses involving con-tinuously distributed variables. This investigation followed the same group of undergraduates from their freshman to junior years. Although the present study examined group changes and the pre-dictive power of covariates across 3 years, it was not the intention of the study to analyze trajectory changes longitudinally.

2. Method

2.1. Participants and instrument

A survey was used to collect data for 757 freshmen (51.1% male) entering a college in southern Taiwan in 2008. Participants were

(3)

asked to sign informed consent documents and received a small gift for completing the survey. The same sample was surveyed two more times, during the students’ sophomore and junior years. The instrument was designed to elicit data on demographics (age, gender, major), Internet habits, and four scales: Self-Esteem Scale (Rosenberg, 1965), Loneliness Scale (De Jong Gierveld & Kamphuis, 1985), Beck’s Depression Inventory II (Beck, Steer, & Brown, 1996), and short form PIU (Lin & Tsai, 2002). The survey items focus on two aspects of Internet usage: entertainment (e.g., watching vid-eos, gaming) and social usages (e.g., email, instant messaging). Fre-quencies were coded on a scale of 0 (never) to 8 (7 days per week). 2.1.1. Self-Esteem Scale

Five of the 10 items in Rosenberg’s (1965)Self-Esteem Scale were translated into Chinese. Global self-esteem was measured according to responses to ﬁve statements (e.g., ‘‘I feel I’m a person of worth’’) using a 5-point scale (1 = never true; 5 = always true). All items were reverse-coded, therefore higher scores indicate low-er self-esteem. Cronbach’s

a

coefﬁcients for this scale ranged from .88 to .93 over the 3-year study period, indicating good internal consistency.

2.1.2. Loneliness Scale

A Chinese version ofDe Jong Gierveld and Kamphuis’s (1985)

Loneliness Scale was employed. The scale consists of five positive and six negative dichotomously scored items (1 = yes; 0 = no). An example of a positive item is ‘‘There are many people that I can trust completely.’’ An example of a negative item is ‘‘I experience a general sense of emptiness.’’ KR20 coefficients for this scale were .78 (freshman), .85 (sophomore), and .74 (junior). According to the scale’s authors, the loneliness construct includes social (e.g., ‘‘There are plenty of people that I can rely on when I have problems’’) and emotional factors (e.g., ‘‘I often feel rejected’’). Negatively worded items were reverse-coded, thus higher scores reflect higher loneli-ness in either social or emotional.

2.1.3. Beck Depression Inventory II

Participants were asked to report their experiences with 21 depression symptoms in Beck Depression Inventory II (BDI-II) (Beck et al., 1996) using a 1–4 severity scale. Total scores ranged from 0 to 63. Cronbach’s

a

coefficients ranged from .88 to .91 over the 3-year study period.Byrne, Stewart, and Lee (2004)used BDI-II to measure the following three factors in a sample of Hong Kong adolescents: negative attitude, performance difficulty, and somatic elements. An item on change in sexual interest that was deleted from the Byrne et al. project was included in the present study. According toWard (2006), the sexual interest item should load onto the somatic factor. Negative attitude symptoms include sad-ness, pessimism, and self-dislike; performance difficulty symp-toms include loss of pleasure, loss of interest, agitation, indecisiveness, irritability, and concentration difficulty; and so-matic elements include loss of energy and changes in sleep pat-terns, appetite, and sexual interest. Higher scores indicate a higher level of depression in negative attitude, performance diffi-culty, or somatic elements.

2.1.4. PIU scale

Parts ofLin and Tsai’s (2002)Internet Addiction Scale were used to measure PIU behaviors such as tolerance and compulsivity. Re-sponses to the ﬁve items were measured along a 5-point scale ranging from 1 (‘‘never’’) to 5 (‘‘always’’). Example items are ‘‘Though I plan to use the Internet for just a while, I stay online longer than I originally intended’’ and ‘‘When I try to cut down or stop my Internet use, I feel anxious.’’ Cronbach’s

a

coefﬁcients for this scale ranged from .78 to .85 over the 3 years, indicating moderate-to-high internal consistency.

2.2. Data analysis

When using multiwave panel data, it is necessary to determine whether the lost data in the subsequent data collection is ‘‘missing at random’’ (Graham, 2009). Independent t-tests were performed to examine whether the participants who completed the same sur-vey and those who withdrew from the sursur-vey in Year 2 differed on Year 1 PW indicators (i.e, negative attitude, performance difﬁculty, somatic elements, social loneliness, emotional loneliness, and self-esteem). The test results showed that the former group exhibited slightly higher self-esteem than the latter group (M = 7.50, SD = 4.03, N = 643 vs. M = 6.67, SD = 3.52, N = 109; t[750] = 2.23, p < .05). No differences were noted for any of the other PW indica-tors. The results from a comparison of Year 1 self-esteem levels be-tween the students with full data and the students with incomplete data in Year 3 indicated a small difference favoring the group with full data (M = 7.55, SD = 3.52, N = 590 vs. M = 6.77, SD = 3.82, N = 162; t[750] = 2.46, p < .05). Again, no differences were noted for any of the other PW indicators. Cohen’s d values for the self-esteem t-tests were small (.23 and .22), therefore self-esteem was included in the analysis.

The use of different scales in the latent profile analysis required the calculation of standardized scores for all indicators. Latent PW groups were analyzed using Mplus 6 (Muthén & Muthén, 2010) with an exploratory approach. Multiple latent profile models (1-group to 5-(1-group) were fit to determine the best-fitting model and the most appropriate number of latent groups. The statistics that are most often used to select the best-fitting model are

Akaike’s (1987) Information Criterion (AIC), Schwartz’s (1978)

Bayesian Information Criterion (BIC), andSclove’s (1987)Adjusted BIC (ABIC). In all cases, lower values indicate better model-data fit and identify a simpler model given two models with similar fit to the same data. The Vuong–Lo–Mendell–Rubin likelihood ratio test (VLMR) (Lo, Mendell, & Rubin, 2001) for comparing models of k and k 1 groups produces p values that indicate whether the fit of k-group model is significantly better than that of (k 1)-group mod-el. Mplus also calculates the probabilities of membership in each group for each individual and assigns persons to the group associ-ated with the largest probability. Precision-of-classification indica-tors (e.g., entropy statistics) are used in assessing the classification utility of the model. The final model was determined by consulting the above statistics. To determine stability, PW profile analyses were conducted for data collected from the same group of students during their sophomore and junior years. After determining the most appropriate number of groups, the study investigated gender effects, and the relationships between latent groups and covariates using multinomial logistic regressions (SPSS 18.0).

3. Results

3.1. Latent constructs

Confirmatory factor analyses (CFAs) were used to examine the latent constructs for the Loneliness Scale and BDI-II. The model fit statistics shown inTable 1were compared with cutoff criteria of a comparative fit index (CFI) > 0.90 and a root mean square error of approximation (RMSEA) < 0.10 (Browne & Cudeck, 1993; Hu & Bentler, 1999; McDonald & Ho, 2002); the scales expressed moder-ate model-data fit. The same was true for data collected during the second and third years. Self-esteem was examined using explor-atory factor analyses because the model fit statistics were less sat-isfactory for CFA when error correlations were disallowed. In the factor analysis for Year 1, one factor was extracted according the rule of eigen value >1.0 and judgment based on a scree plot; this factor accounted for 67.1% of all item variance. Accordingly, it

(4)

was deemed appropriate to use scale or subscale scores (standard-ized) as PW indicators during the next analytical stage.

3.2. Latent proﬁles

Model fit statistics for the proposed latent profile models are presented in Table 2. As shown, AIC, BIC, and ABIC decreased slightly after the fifth group was specified, indicating little improvement. According to the p-value from the VLMR likelihood ratio test (p-VLMR), no statistically significant difference was found between the 5-group model and the 4-group model; that is, the 5-group model did not improve model specification com-pared to the 4-group model. Further, the high entropy value for the 4-group model (.813) suggested a high level of precision in assigning individuals to their appropriate groups. Therefore, the 4-group model was considered as the best-fitting model.

According to their proﬁles (Fig. 1a), Group 1 students were nor-mative (47%), Group 2 students were slightly PW disadvantageous (27%), Group 3 had good PW (18%), and Group 4 students were se-verely PW disadvantageous (7%). Therefore, the four groups were labeled as normative, minor-disadvantageous, good PW, and se-vere-disadvantageous. The normative and good PW groups repre-sented the majority of freshmen and exhibited below average negative PW indicator scores. In contrast, the minor- and severe-disadvantageous groups exhibited higher negative PW indicator scores than average. The general linear model analyses and post-hoc comparisons showed statistically signiﬁcant differences in average standardized scores among the four groups for all six indi-cators, the exception being for emotional loneliness scores be-tween the minor-disadvantageous and severe-disadvantageous groups (f[3, 750] = 286.65, p < .001, partial

g

2_{= .53; Games-Howell} post-hoc test p = .82). Note that the average emotional loneliness

score of the good PW group was lower than those of the other groups (seeFig. 1a). Overall, the good PW and severe-disadvanta-geous groups were only different quantitatively in terms of all six indicators. Students in the severe-disadvantageous group scored particularly high on negative attitude, whereas those in the good PW group scored relatively low on emotional loneliness. Latent profile analyses were also conducted for the Year 2 data; the model fit statistics are shown inTable 2. According to the p-VLMR data, the 4-group model did not improve model specification compared to the 3-group model, thus making the latter the most

Table 1

Model ﬁt statistics for CFA (N = 757).

Scale v2 df CFI RMSEA SRMR/WRMR Loneliness Scale, 2-factor 191.60*** ₄₃ _0.96 _0.068 _1.716 (WRMR) BDI-II, 3-factor 625.24*** ₁₈₆ _0.90 _0.056 _{0.043 (SRMR)} Note. CFI = robust comparative ﬁt index; RMSEA = robust root mean square error of approximation; SRMR = standard root mean squared residual; WRMR = weighted root mean squared residual.

***_{p < .001.}

Table 2

Latent proﬁle model ﬁt statistics for Years 1, 2 and 3.

Number of groups 1 2 3 4 5 Number of parameters 12 19 26 33 40 Year 1 (N = 757) BIC 12889.403 11982.399 11674.560 11495.684 11472.972 ABIC 12851.298 11922.066 11591.999 11390.896 11345.955 AIC 12833.851 11894.441 11554.196 11342.916 11287.797 p-VLMR 0.000 0.016 0.000 0.300 Entropy 0.827 0.821 0.813 0.793 Year 2 (N = 647) BIC 11059.925 10249.624 9881.453 9790.639 9724.937 ABIC 11021.826 10189.300 9798.904 9685.865 9597.938 AIC 11006.257 10164.649 9765.172 9643.052 9546.043 p-VLMR 0.016 0.001 0.178 0.697 Entropy 0.869 0.852 0.793 0.840 Year 3 (N = 592) BIC 10031.544 9335.231 9065.144 8928.481 8839.774 ABIC 9993.448 9274.912 8982.602 8823.717 8712.786 AIC 9978.942 9251.945 8951.173 8783.825 8664.433 p-VLMR 0.223 0.363 0.354 0.526 Entropy 0.864 0.889 0.840 0.835

Fig. 1. Latent proﬁles based on negative attitude (NA), performance difﬁculty (PD), and somatic elements (SE), social loneliness (SL), emotional loneliness (EL), and self-esteem (SELF) for individual groups (Year 1 vs. Year 2).

(5)

appropriate latent profile model. The majority of sophomores were in Group 1 (58%), followed by Groups 2 (35%) and 3 (7%). According to their profiles, these groups were labeled as normative, minor-disadvantageous, and severe-disadvantageous (Fig. 1b). The results from general linear model analyses and post-hoc comparisons indicated statistically significant differences among the three groups for all six indicators. The minor-disadvantageous and se-vere-disadvantageous group profiles of sophomores were similar to those of freshmen. The good PW group and normative group identified in the freshman year became homogeneous in latent profiles in the second year.

The Year 3 results indicated that the 1-group model exhibited the best model-data fit; that is, the PW profile of the junior stu-dents was homogeneous. The model fit statistics are shown in Ta-ble 2.

3.3. Covariates

Given that the 4-group model exhibited a better model-data fit, providing meaningful PW classifications for the freshmen students, the next task was to examine the connection of covariates with the four profiles. The groups were regressed on gender (binary), fre-quency of social and entertainment Internet usage (continuous), and PIU (continuous). The interaction of gender and frequency of social Internet usage was also examined, based on previous find-ings of a gender difference in the correlation between social Inter-net usage and PW (Amichai-Hamburger & Ben-Artzi, 2003; Morgan & Cotten, 2003). The multinomial logistic regression results are presented inTable 3. The normative group served as the reference group. According to the non-significant logistic regression coeffi-cient results, females and males were equally likely to be in any of the four groups. No significant relationship of entertainment usage with PW groups was found.

For each point of increase in social use frequency, the odds of being in the good PW group increased from 1.0 to 1.160 when the other covariates were held constant (Table 3). Students at the high end of social use frequency were more likely to be in the good PW group than the normative group. This result was consistent for both males and females, as the interaction of gender and social use was not signiﬁcant. As shown inFig. 2, the probabilities of being in

the good PW group changed when social use frequency increased, regardless of gender.

Students with higher PIU values were more likely to be in the minor- or severe-disadvantageous PW groups than the normative group. As shown inTable 3, with all other covariates held constant, for each point increase in PIU score, the odds of being in the minor-disadvantageous group increased from 1.0 to 1.498, the odds of being in the good PW group decreased from 1.0 to 0.499, and the odds of being in the severe-disadvantageous group increased from 1.0 to 2.325.

Next, the relationships between each of the three PW groups and the covariates were examined for the sophomore year partic-ipants. The multinomial logistic regression results indicated that PIU was associated with the likelihood of being in the minor- or se-vere-disadvantageous group compared to the normative group. None of the other covariates were related to the probability of being in any of the non-normative PW groups. Because there was only one latent proﬁle group among the Year 3 students, associa-tions with covariates were not analyzed.

4. Discussion 4.1. Latent PW proﬁles

There was a clear order of PW groups according to the 4-class model for the freshman year students as follows: good PW, norma-tive, minor-disadvantageous, and severe-disadvantageous. Individ-uals in the severe-disadvantageous group displayed relatively high scores on nearly all of the negative PW indicators, particularly depression. Members of the good PW group exhibited the lowest negative psychological indicator scores. The largest difference be-tween the good PW and severe-disadvantageous groups was in negative attitude. Students with severe PW disadvantages were more pessimistic than their good PW counterparts. The data also indicate that students with good PW had relatively low emotional loneliness scores compared to all other students, implying that col-lege freshmen with good PW enjoy quality relationships with friends and family despite entering a new environment.

However, the data also suggested that important changes in latent proﬁles occurred during the second year, with the four groups being reshufﬂed into three as follows: normative,

Table 3

Logistic regression coefﬁcients and odds ratios for 4-group model with gender, entertainment/social use frequency, and PIU as covariates for Year 1 (normative group = reference group).

Group Effect b SE Wald’sv2

Odds ratio Minor-disadvantageous Intercept 1.188 .427 7.753 Gender (female) .498 .560 .791 .607 Entertainment use .034 .040 .695 1.034 Social use .071 .052 1.859 .931 Gender social .126 .086 2.160 1.134 PIU .404*** _.140 _8.378 _1.498 Good PW Intercept 1.173 .568 4.259 Gender (female) .189 .747 .064 1.208 Entertainment use .070 .047 2.181 1.072 Social use .149** _.069 _4.602 _1.160 Gender social .012 .110 .013 .988 PIU .695*** _.196 _12.527 _.499 Severe-disadvantageous Intercept 2.935 .728 16.238 Gender (female) .053 .898 .003 .949 Entertainment use .076 .067 1.295 .927 Social use .081 .074 .740 .923 Gender social .088 .140 .400 1.092 PIU .844*** _.214 _15.566 _2.325

Note. Cox and Snell R2

= 0.083; Nagelkerke R2

= 0.091; McFadden R2

= 0.036.

**_{p < .01.} ***_{p < .001.}

(6)

minor-disadvantageous, and severe-disadvantageous. Only one la-tent group was identiﬁed for the third-year students. Students with either complete or missing data during the successive data collection period differed only slightly in terms of self-esteem, and ‘‘missing at random’’ was ensured in the present study. There-fore, structural changes in latent proﬁles over time may have been due to their adaptability to life in college, resulting in homoge-neous PW states. More research is required to identify the reasons for these changes.

4.2. Gender, entertainment, social usage, and PIU

According to the present study, high-PIU students were more likely to be in the severe- and minor-disadvantageous PW groups, and low-PIU students were more likely to be in the good PW group. Similar results were found for the second-year students, although the nature of the latent groups differed. This result conﬁrmsvan der Aa et al.’s (2009)ﬁnding that PIU exerts the strongest and most direct effect on depression, self-esteem, and loneliness among their study participants. The present study also provides support for

Whang et al.’s (2003)results that individuals who are strongly ad-dicted to the Internet are more likely to be lonely and depressed than those who are less strongly addicted.

According to the ﬁrst-year data, the logit of online social use fre-quency for the good PW group indicated that the students who used the Internet primarily for social purposes were more likely to be in the good PW group than the normative group. This result was not observed for the severe- or minor-disadvantageous PW groups; that is, online social use did not decrease the probability of being in the disadvantageous PW groups. This result partially supports claims made byKang (2007)andShaw and Gant (2002)

that social use mitigates negative feelings such as loneliness and

depression. The good PW group had relatively low emotional lone-liness compared to all other PW indicators, implying that social usage of the Internet is associated with decreased emotional lone-liness. Since a good PW group did not emerge during the second year, the signiﬁcant association with social usage disappeared.

AsAmichai-Hamburger and Ben-Artzi (2003)note, correlations

between loneliness and social Internet usage have been found for females, but not for males. In contrast,Morgan and Cotten (2003)

argue that email usage decreases depression in males but not in fe-males. They suggest that because females are more active in com-munication, the effects of e-mail on depression are less obvious for them. Findings from the present study indicate that female and male students were equally likely to be in any PW group, and that the relationship between social Internet usage and PW did not dif-fer by gender. This result supportsSubrahmanyam and Lin’s (2007)

ﬁnding of no gender difference in the relationship between loneli-ness and Internet usage, andvan den Eijnden et al.’s (2008)ﬁnding of no moderating effect of gender in predicting PW from Internet communication usage.

The findings also indicate that greater use of the Internet for entertainment purposes was not associated with the higher prob-ability of being in any non-normative PW group during Years 1 and 2. However, the entertainment category was not broken down into specific types such as gaming and non-gaming, which is important in light of previous research findings of a negative rela-tionship between gaming and PW (Anderson & Bushman, 2001; Mannell, Zuzanek, & Aronson, 2005).

5. Conclusion, implications and future research

Researchers of human behavior search for theoretical and sta-tistically meaningful patterns among multiple co-occurring variables. Person-centered approaches (e.g., latent profile analyses) use configurations of multiple variables to identify quantitative differences in profile level, qualitative differences in profile shape, or both (Marsh, Lüdtke, Trautwein, & Morin, 2009). This allows researchers to conceptualize PW as a holistic latent construct with complex relationships among depression, loneliness, and self-es-teem. The present study differs in that detailed profiles of depres-sion and loneliness were created and broken down into negative attitude, performance difficulty, and somatic elements measures (depression) as well as social loneliness and emotional loneliness measures (loneliness). Therefore, the latent profile results pre-sented in this paper represent a unique contribution to the understanding of PW among Taiwanese undergraduates, facilitat-ing risk detection among students and practical implications for treatment and prevention. PW profile identification may also sup-port adjustments in interventions to best address the needs of spe-cific groups. In the present study, negative attitude was identified as the most discriminated negative indicator of PW. According to

Beck (1993), the way individuals think about a problem can cause

stronger negative psychological states than the problem itself. Negative attitudes and strong pessimistic beliefs about the ability to complete tasks are strong obstacles to academic success; indi-viduals with negative attitudes are more inclined to dislike them-selves and to lack the ability to enjoy normal day-to-day activities (Morrison & O’Connor, 2004). It is hoped that the ﬁndings of the present study will support efforts to identify individuals suffering from negative attitudes so that useful interventions can be enacted.

This study also contributes to the literature by providing a com-plex analysis of the relationship between Internet usage and PW beyond a linear correlational relationship. Classifying students using latent proﬁle analyses plays a critical role in such an analysis for two reasons. First, latent proﬁles provide a more complete

Fig. 2. Estimated probability of PW group membership in light of increased social use (PIU held at 0) for Year 1 study participants (male vs. female).

(7)

picture by connecting all individual indicators rather than consid-ering them separately. Second, by treating PW as categorical in-stead of continuous, the non-linear relationships can be detected. The study found that the relationship between social use of the Internet and PW differed across PW groups. Online social usage did not decrease the probability of study participants having psy-chological problems, however, students who used the Internet for social purposes were likely to be psychologically healthy. Therefore, the findings suggest that online social usage should be encouraged (or at least not discouraged) if an individual does not exhibit high PIU. However, professional intervention is called for when a student is clearly in a severe-disadvantageous PW cate-gory. The study results also confirm that PIU is significantly associ-ated with negative PW, as reported byKo (2006)andvan der Aa et al. (2009). These results suggest that interventions in this popu-lation should focus on reducing PIU.Young (2007)has discussed several promising long-term approaches to reducing and managing PIU, especially cognitive behavioral interventions.

The study purpose was not to track changes in PW group membership over time, but to identify developmental changes in the PW construct. Since an exploratory approach was used, the-oretical hypotheses were neither proposed nor tested. Future researchers may be interested in constructing a latent transit model for students across all four undergraduate years to produce a complete picture of how college students develop in terms of PW and the effects of Internet usage over longer time periods. In addition, a larger and more representative sample would sup-port an effort to generalize the current ﬁndings to all Taiwanese college students.

Acknowledgements

This research was funded by the ROC National Science Council through three projects: ‘‘The Net and Taiwan Adolescents’ Physical and Mental Development’’ (NSC 97-2631-S-009-001), ‘‘Internet Use and Well-being of Taiwan Students: Longitudinal Analysis and the Extension of Fact Bank’’ (NSC-100-2631-S-009-001), and ‘‘Internet Use and Its Socio-cultural Impacts on Taiwan Adolescent Students’’ (NSC 97-2631-S-415-003).

References

Akaike, H. (1987). Factor analysis and AIC. Psychometrika, 52, 317–332.http:// dx.doi.org/10.1007/BF02294359.

Amichai-Hamburger, Y., & Ben-Artzi, E. (2003). Loneliness and Internet use. Computers in Human Behavior, 19, 71–80. http://dx.doi.org/10.1016/S0747-5632(02)00014-6.

Amichai-Hamburger, Y., & Furnham, A. (2007). The positive net. Computer in Human Behavior, 23, 1033–1045.http://dx.doi.org/10.1016/j.chb.2005.08.008.

Anderson, C. A., & Bushman, B. J. (2001). Effects of violent videogames on aggressive behavior, aggressive cognition, aggressive affect, physiological arousal and prosocial behavior: A meta-analytic review of the scientiﬁc literature. Psychological Science, 12, 353–359.http://dx.doi.org/10.1111/1467-9280.00366.

Beck, A. T. (1993). Cognitive therapy: Past present and future. Journal Consulting and Clinical Psychology, 61, 194–198. http://dx.doi.org/10.1037//0022-006X.61.2.194.

Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck Depression Inventory-II. San Antonio, TX: Psychological Corporation.

Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model ﬁt. In K. A. Bollen & J. S. Long (Eds.), Testing structural equation models (pp. 136–162). Beverly Hills, CA: Sage.

Byrne, B. M., Stewart, S. M., & Lee, P. W. H. (2004). Validating the Beck Depression Inventory-II for Hong Kong community adolescents. International Journal of Testing, 4, 199–216.http://dx.doi.org/10.1207/s15327574ijt0403_1.

Campbell, A. J., Cumming, S. R., & Hughes, I. (2006). Internet use by the socially fearful: Addiction or therapy? Cyberpsychology and Behavior, 9, 69–81.http:// dx.doi.org/10.1089/cpb.2006.9.69.

Caplan, S. E., & Turner, J. S. (2007). Bringing theory to research on computer-mediated comforting communication. Computers in Human Behavior, 23, 985–998.http://dx.doi.org/10.1016/j.chb.2005.08.003.

Ceyhan, A. A., & Ceyhan, E. (2008). Loneliness, depression, and computer self-efﬁcacy as predictors of problematic Internet use. Cyberpsychology and Behavior, 11, 699–701.http://dx.doi.org/10.1089/cpb.2007.0255.

Chou, C., Condron, L., & Belland, J. C. (2005). A review of the research on Internet addiction. Educational Psychology Review, 17, 363–388. http://dx.doi.org/ 10.1007/s10648-005-8138-1.

Cotten, S. R. (2008). Students’ technology use and the impacts on well-being. New Directions for Student Services, 124, 55–70.

De Jong Gierveld, J., & Kamphuis, F. (1985). The development of a Rasch-type loneliness scale. Applied Psychological Measurement, 9, 289–299.

Engelberg, E., & Sjöberg, L. (2004). Internet use, social skills, and adjustment. Cyberpsychology and Behavior, 7, 41–47. http://dx.doi.org/10.1089/ 109493104322820101.

Gordon, C. F., Juang, L. P., & Syed, M. (2007). Internet use and well-being among college students: Beyond frequency of use. Journal of College Student Development, 48, 674–688.http://dx.doi.org/10.1353/csd.2007.0065.

Graham, J. W. (2009). Missing data analysis: Making it work in the real world. Annual Review of Psychology, 60, 549–576. http://dx.doi.org/10.1146/ annurev.psych.58.110405.085530.

Green, M. C., Hilken, J., Friedman, H., Grossman, K., Gasiewskj, J., Adler, R., et al. (2005). Communication via instant messenger: Short- and long-term effects. Journal of Applied Social Psychology, 35, 445–462. http://dx.doi.org/10.1111/ j.1559-1816.2005.tb02130.x.

Gross, E. F., Juvonen, J., & Gable, S. L. (2002). Internet use and well-being in adolescence. Journal of Social Issues, 58, 75–90. http://dx.doi.org/10.1111/1540-4560.00249.

Herman, K. C., Ostrander, R., Walkup, J. T., Silva, S. G., & March, J. S. (2007). Empirically derived subtypes of adolescent depression: Latent proﬁle analysis of co-occurring symptoms in the treatment for adolescents with depression study (TADS). Journal of Consulting and Clinical Psychology, 75, 716–728.http:// dx.doi.org/10.1037/0022-006X.75.5.716.

Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for ﬁt indices in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6, 1–55. http://dx.doi.org/10.1080/ 10705519909540118.

Huang, C. (2010). Internet use and psychological well-being: A meta-analysis. Cyberpsychology, Behavior, and Social Networking, 13, 241–249.http://dx.doi.org/ 10.1089/cyber.2009.0217.

Huang, Y.-R. (2006). Identity and intimacy crises and their relationship to Internet dependence among college students. Cyberpsychology and Behavior, 9, 571–576.

http://dx.doi.org/10.1089/cpb.2006.9.571.

Hwang, F. M., & Yang, K. B. (2004). Internet dating and female middle-school sexual abuse victims. Journal of National Taiwan Normal University, 49, 21–40. Kang, S. (2007). Disembodiment in online social interaction: Impact of online chat

on social support and psychosocial well-being. Cyberpsychology and Behavior, 10, 475–477.http://dx.doi.org/10.1089/cpb.2006.9929.

Kim, J., LaRose, R., & Peng, W. (2009). Loneliness as the cause and the effect of problematic Internet use: The relationship between Internet use and psychological well-being. Cyberpsychology and Behavior, 12, 451–455.http:// dx.doi.org/10.1089/cpb.2008.0327.

Ko, H. J. (2006). Internet usage pattern and relations with learning and physical/mental adjustment. Taiwan, ROC: Executive Yuan.

Kraut, R., Kiesler, S., Boneva, B., Cummings, J., Helgeson, V., & Crawford, A. (2002). Internet paradox revisited. Journal of Social Issues, 58, 49–74.http://dx.doi.org/ 10.1111/1540-4560.00248.

Kraut, R., Patterson, M., Lundmark, V., Kiesler, S., Mukopadhyay, T., & Scherlis, W. (1998). Internet paradox: A social technology that reduces social involvement and psychological well-being? American Psychologist, 53, 1017–1031.http:// dx.doi.org/10.1037//0003-066X.53.9.1017.

Lin, S. S. J., & Tsai, C. C. (2002). Sensation seeking and Internet dependence of Taiwanese high school adolescents. Computers in Human Behavior, 18, 411–426.

http://dx.doi.org/10.1016/S0747-5632(01)00056-5.

Lo, Y., Mendell, N., & Rubin, D. (2001). Testing the number of components in a normal mixture. Biometrika, 88, 767–778. http://dx.doi.org/10.1093/biomet/ 88.3.767.

Magidson, J., & Vermunt, J. K. (2004). Latent class models. In D. Kaplan (Ed.), Handbook of quantitative methodology for the social sciences (pp. 345–368). Newbury Park, CA: Sage Publications. http://dx.doi.org/10.4135/ 9781412986311.n10.

Magnusson, D., & Cairns, R. B. (1996). Developmental science: Toward a uniﬁed framework. In R. B. Cairns, & G. H. Elder Jr. (Eds.), Developmental science. Cambridge studies in social and emotional development (pp. 7–30). New York: Cambridge University Press.

Mannell, R. C., Zuzanek, J., & Aronson, R. (2005). Internet/computer use and adolescent leisure behavior, ﬂow experiences and psychological well-being: The displacement hypothesis. In Papers presented at the eleventh Canadian Congress on Leisure Research, Nanaimo, British Columbia, Canada.

Marsh, H. W., Lüdtke, O., Trautwein, U., & Morin, A. J. S. (2009). Classical latent proﬁle analysis of academic self-concept dimensions: Synergy of person-and variable-centered approaches to theoretical models of self-concept. Structural Equation Modeling, 16, 191–225.http://dx.doi.org/10.1080/10705510902751010.

McDonald, R. P., & Ho, M.-H. (2002). Principles and practice in reporting structural equation analyses. Psychological Methods, 7, 64–82.http://dx.doi.org/10.1037// 1082-989X.7.1.64.

McKenna, K. Y. A., Green, A. S., & Gleason, M. E. J. (2002). Relationship formation on the Internet: What’s the big attraction? Journal of Social Issues, 58, 9–31.http:// dx.doi.org/10.1111/1540-4560.00246.

Moody, E. (2001). Internet use and its relationship to loneliness. Cyberpsychology and Behavior, 4, 393–401.http://dx.doi.org/10.1089/109493101300210303.

(8)

Morgan, C., & Cotten, S. R. (2003). The relationship between Internet activities and depressive symptoms in a sample of college freshmen. Cyberpsychology and Behavior, 6, 133–142.http://dx.doi.org/10.1089/109493103321640329.

Morrison, R., & O’Connor, R. C. (2004). Predicting psychological distress in college students: The role of rumination and stress. Journal of Clinical Psychology, 61, 447–460.http://dx.doi.org/10.1002/jclp. 20021.

Muthén, B. O. (2001). Latent variable mixture modeling. In G. A. Marcoulides & R. E. Schumacker (Eds.), New developments and techniques in structural equation modeling (pp. 1–33). Hillsdale, NJ: Lawrence Erlbaum.

Muthén, L. K., & Muthén, B. O. (2010). Mplus user’s guide (6th ed.). Los Angeles, CA: Muthén & Muthén.

Niemz, K., Grifﬁths, M., & Banyard, P. (2005). Prevalence of pathological Internet use among university students and correlations with self-esteem, the General Health Questionnaire (GHQ), and disinhibition. Cyberpsychology and Behavior, 6, 562–570.http://dx.doi.org/10.1089/cpb.2005.8.562.

Pasta, D. J. (2009). Learning when to be discrete: Continuous vs. categorical predictors. In Paper presented at SAS Global Forum, Washington, DC.

Pastor, D. A., Barron, K. E., Miller, B. J., & Davis, S. L. (2007). A latent proﬁle analysis of college students’ achievement goal orientation. Contemporary Educational Psychology, 32, 8–47. http://dx.doi.org/10.1016/j.cedpsych.2006. 10.003.

Rosenberg, M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press.

Schwartz, G. (1978). Estimating the dimension of a model. The Annals of Statistics, 6, 461–464.

Sclove, S. L. (1987). Application of model-selection criteria to some problems in multivariate analysis. Psychometrika, 52, 333–343.

Shaw, L. H., & Gant, L. M. (2002). In defense of the Internet: The relationship between Internet communication and depression, loneliness, self-esteem, and perceived social support. Cyberpsychology and Behavior, 5, 157–171.http:// dx.doi.org/10.1089/109493102753770552.

Sproull, L., & Faraj, S. (1997). Atheism, sex, and databases: The Net as a social technology. In B. Kahin & J. Keller (Eds.), Public access to the Internet (pp. 62–81). Cambridge, MA: MIT Press.

Subrahmanyam, K., & Lin, G. (2007). Adolescents of the net: Internet use and well-being. Adolescence, 42, 659–677.

Tuominen-Soini, H., Salmela-Aro, K., & Niemivirta, M. (2008). Achievement goal orientations and subjective well-being: A person-centred analysis. Learning and Instruction, 18, 251–266.http://dx.doi.org/10.1016/j.learninstruc.2007.05.003.

Tuominen-Soini, H., Salmela-Aro, K., & Niemivirta, M. (2011). Stability and change in achievement goal orientations: A person-centered approach. Contemporary Educational Psychology, 36, 82–100.http://dx.doi.org/10.1016/j.cedpsych.2010. 08.002.

Valkenburg, P. M., & Peter, J. (2007). Online communication and adolescent well-being testing the stimulation versus the displacement hypothesis. Journal of Computer-Mediated Communication, 12, 1169–1182.http://dx.doi.org/10.1111/ j.1083-6101.2007.00368.x.

van der Aa, N., Overbeek, G., Engels, R. C. M. E., Scholte, R. H. J., Meerkerk, G.-J., & van den Eijnden, R. J. J. M. (2009). Daily and compulsive Internet use and well-being in adolescence. A diathesis-stress model based on big ﬁve personality traits. Journal of Youth Adolescence, 38, 765–776. http://dx.doi.org/10.1007/s10964-008-9298-3.

van den Eijnden, R. J. J. M., Meerkerk, G.-J., Vermulst, A. A., Spijkerman, R., & Engels, R. C. M. E. (2008). Online communication, compulsive Internet use, and psychosocial well-being among adolescents: A longitudinal study. Developmental Psychology, 44, 655–665.http://dx.doi.org/10.1037/0012-1649. 44.3.655.

Ward, L. C. (2006). Comparison of factor structure models for the Beck Depression Inventory-II. Psychological Assessment, 18, 81–88. http://dx.doi.org/10.1037/ 1040-3590.18.1.81.

Wei, H. I. (2000). Internet and psychological well-being, academic achievement. Unpublished master’s thesis, National Taitung University, Taitung, Taiwan, ROC. Whang, L. S.-M., Lee, S., & Chang, G. (2003). Internet over-users’ psychological proﬁles: A behavior sampling analysis on Internet addiction. Cyberpsychology and Behavior, 6, 143–150.http://dx.doi.org/10.1089/109493103321640338.

Young, K. S. (2007). Cognitive behavior therapy with Internet addicts: Treatment outcomes and implications. Cyberpsychology and Behavior, 10, 671–679.http:// dx.doi.org/10.1089/cpb.2007.9971.