Journal of Online Behavior

Reformulating the Internet Paradox:
Social Cognitive Explanations of Internet Use and Depression

by Robert LaRose, Ph.D.; Matthew S. Eastin, and Jennifer Gregg

LaRose, R., Eastin, M. S., Gregg, J. (2001). Reformulating the Internet paradox: Social cognitive explanations of Internet use and depression. Journal of Online Behavior, 1 (2). Retrieved <date> from the World Wide Web:


The Internet Paradox study (Kraut et al., 1998) found evidence of a causal link between Internet use and depression, but it may have been specific to novice Internet users. The relationship between Internet use, social support and depression was reformulated drawing on social cognitive theory (Bandura, 1997) to account for the possible influence of self-efficacy, Internet-related stress, and perceived social support. A path analysis revealed a link between Internet use and depression, but one mediated by self-efficacy and the expectation of encountering stressful situations on the Internet. A path also was found linking Internet use to decreased depression through the use of e-mail exchanges with known associates to obtain social support.

The Paradoxical Internet Paradox

The Internet paradox study (Kraut et al., 1998), part of the HomeNet project at Carnegie Mellon University, provided important preliminary evidence of the possible harmful effects of Internet use. The paradox was how a "social technology" used primarily for interpersonal interaction could increase social isolation and thereby decrease psychological well-being among its users. Internet use was associated with increases in loneliness and depression and tended to increase stress in a sample of 169 persons who received free computers and Internet access over a period of one to two years. These results seemed paradoxical indeed to those--the researchers and their sponsors among them--who viewed the Internet as a vibrant new means of social interaction through the use of e-mail, newsgroups, and chatrooms. To explain the paradox, the researchers reasoned that superficial relationships (weak ties) formed online displaced meaningful (strong tie) relationships in the real world.

The results were also paradoxical in the face of competing, if inconsistent, evidence of the positive social impacts of Internet use. Wynn and Katz (1997) emphasized the inherent "situatedness" of Internet use in a broader social context that makes it impossible to completely separate the virtual world online from the real world off-line. Ethnographic research suggests that online communication supplements existing real world relationships rather than displaces them (Hamman, 1999). In a review of ethnographic and anecdotal evidence about Internet communities, Wellman and Gulia (1999) concluded that online relationships can be strong and intimate and may strengthen real world relationships as much as diminish them. They attributed concerns about negative effects to an overly idealized view of real world social interaction. Superficial relationships are found there, too.

Surveys (Katz & Aspden, 1997; Parks & Floyd, 1996; Parks & Roberts, 1998) indicated that the Internet spawned highly developed online relationships, many of which led to real world social contacts, suggesting that social isolation might decrease with greater Internet use. Online relationships equal off-line ones that fulfill similar roles in terms of their breadth, depth, and development of private communication codes, despite the fact that online relationships have fewer weekly contact hours and shorter histories than offline relationships (Parks & Roberts, 1998). Scherer (1997) found no differences in self-perceptions of sociability between college students who were dependent on the Internet (i.e., exhibited 3 or more symptoms of excessive Internet use paralleling those of substance abuse) and those who were not, even though the dependent users utilized newsgroups, chat, and multi-user environments more, and socialized less face-to-face. Katz and Aspden (1997) concluded from a national survey of 1500 respondents that Internet use had no impact on off-line social participation.NOTE 1 In a Pew Research Center poll (Pew Research Center, 2000) most Internet users said that e-mail had improved their connections to family and friends, and those perceptions increased the longer users had been on the Internet and the more they used it. There were also fewer socially isolated individuals among Internet users than non-users, and Internet users were more likely to have recent social contacts and sources of social support.NOTE 2 Computer mediated communication research has demonstrated that even media lacking in nonverbal cues, including text-based e-mail on the Internet, may foster supportive relationships over time (see for review Walther, 1996).

In support of the Internet paradox hypothesis, other scholars have warned about the potential harmful effects of online interpersonal communication, blaming online technology for disrupting real world networks (Heim, 1993; Stoll, 1995) and creating a "lonely crowd" in cyberspace (Kroker & Weinstein, 1994). Turkle (1995, p. 235) pointed out the absurdity of the notion that community can arise from among people sitting alone, typing messages to virtual friends. Nie and Erbring (2000) found that as Internet use increased, users were more likely to report a decrease in time spent talking to family and friends and attending social events. Online relationships may develop less interdependence, understanding, and commitment than comparable off-line ones do (Parks & Roberts, 1998).

The latter studies bolster the post hoc explanation that Kraut et al. (1998) applied to their findings, that superficial online relationships diminish close real-world ties, reducing social support and increasing depression (although no significant effect on social support was actually found; see Walther & Reid, 2000). However, all these studies make the possibly mistaken assumption that face-to-face relationships are inherently superior to online relationships (Hamman, 1999; Parks & Roberts, 1998) and neglect the possibility of hyperpersonal online interactions that may be more intimate than their offline counterparts (Walther, 1996).

Aside from Turkle's ethnographic case studies (which are contradicted by Hamman's, 1999), the hypothesis that online relationships diminish real world relationships has sparse empirical support. Riphagen and Kanfer (1997) found that e-mail users had more distance relationships than non-users and that the total number of relationships was about equal, suggesting that local (presumably strong) ties suffered as a result of having e-mail. However, their survey methodology could not rule out the competing explanation that people who had strong long distance ties to maintain were more likely to adopt e-mail. Nie and Erbring (2000) did not account for the possibility that users may have substituted e-mail contacts for face-to-face or telephone communication, and are contradicted by another survey (Pew Research Center, 2000) in which Internet users were more likely to report recent social contacts and the availability of social support than non-users. Parks and Floyd (1996) found that online communication frequently covered issues that went beyond the stated boundaries of the Internet communities in which it originated, a key distinguishing characteristic of strong social ties. And, a survey of German Internet users found a positive relation between Internet use and the number of friends one had (D÷ring, 1996).NOTE 3

Historically, the introduction of new social technology was not linked to increased social isolation. Kraut et al. (1998) viewed the telephone as a means of providing real world support when e-mail failed for their subjects (p. 1030). However, the telephone is a social technology itself, and one of its central functions is to provide an enjoyable source of social interaction (LaRose, 1999). In contemporary society, Wellman (1996) concluded that the phone was used more to maintain local relationships than to supplant them with distant ones.

The Role of Experience

The amount of experience with the Internet may be a pivotal factor in interpreting the competing findings. The respondents in the Internet Paradox study were all novice users, introduced to the Internet by the researchers’ treatment, and all therefore had less than two years’ experience on line. In other research, veteran users with over three years on line were more likely to observe improvements in social interactions as a result of Internet use than were novice users with less than six months’ experience (Pew Research Center, 2000). They also were more likely to state that they had someone to turn to when they needed help. Parks and Floyd’s (1996) respondents tended to be long-term users with over two years of experience with online communication, and typically formed their online relationships a year or more after joining a community. New users are less comfortable using the Internet and less satisfied with their Internet skills (GVU, 1999, question 101, question 102). Over time, users of computer mediated communication are able to compensate for the relative lack of social cues available in e-mail (Walther, 1996). Thus, novices may be simply less competent at using the medium to obtain social support.

Novice Internet users may also experience new sources of stress from technical problems encountered when using the Internet (Charney & Greenberg, in press; GVU, 1999, question 11). That stress may contribute to depression and negate the benefits of any social support received on line. Populations that include experienced users may therefore yield differing results from Kraut et al. (1998).

The subjects in Kraut et al.’s research may also have had better access to social support from face-to-face sources than is the norm in a highly mobile society. Part of the sample was recruited from members of community groups, a population that might be well integrated into local community life and to have large numbers of geographically proximate associates (Shapiro, 1999). Respondents who moved or went away to college during the period of the study were dropped from the sample and as much as a third of the original panel was apparently lost for these reasons (Kraut et al., 1998, p. 1021). Thus, there is still the possibly that individuals who are mobile, and who must rely on social technologies to maintain relationships, may use the Internet to obtain social support and relieve depression.

Finally, the HomeNet participants had low levels of depression overall (Rierdan, 1999) and so may not have been in any great need of social support. Other, more mobile populations that are cut off from stable face-to-face relationships, and with higher levels of depression and stress, may derive more benefit from online interactions than those with stable local community ties and normal levels of depression. According to the buffering hypothesis (Cobb, 1976; Cohen & Wills, 1985), social support protects psychological well-being primarily under conditions of high stress.

In searching for new explanations of the relationship between Internet use and psychological well-being an overarching theoretical framework may be desirable. Kraut et al. (1998) combined disparate constructs from sociology (e.g. strong vs weak social ties), various schools of psychology (e.g. personality constructs such as extraversion, and social psychological variables such as loneliness, social support and depression) and media studies (for the relationship between media exposure and social involvement). The current research reformulates the Internet paradox in terms of a comprehensive theory of human behavior that better accounts for users’ experience levels and for the possibility of obtaining social support from distant associates, among populations with a greater need for such support.

Social Cognitive Explanations of the Internet Paradox

Social cognitive theory provides a comprehensive theoretical framework for understanding human behavior, social interaction and psychological well-being (Bandura, 1986; 1989; 1997) with which we propose to reformulate the relationship between Internet use and depression. The theory recognizes a variety of mechanisms that govern human behavior, including enactive learning (learning through one’s own experience), vicarious learning (learning by observing others), self-regulation (the practice of self control) and self-efficacy (or the belief in one's ability to perform a task successfully). The self-efficacy mechanism (Bandura, 1977; 1982; 1997) pertains since it describes the cognitive processes that relate the acquisition to the performance of new behaviors. This concept may explain the implications of the transition from novice to veteran Internet user for psychological well-being.

Kraut et al. (1998) raised the self-efficacy issue in mentioning the possible impact of Internet use on self-esteem. But they dismissed it on the grounds that they were engaged in a study of social behavior while self-esteem was deemed a separate issue. Although self-esteem (the judgment of one's own self-worth) is distinct from self-efficacy (the judgment of one's own personal capacities), the two terms are often used interchangeably (Bandura, 1997), and indeed Kraut et al. were evidently being dismissive of self-efficacy ("self esteem related to computer skill learning," p. 1029). However, within social cognitive theory, self-efficacy is an important mediating factor between social behavior and depression. Thus, from the perspective of social cognitive theory, self-efficacy is a pivotal variable that implies a different causal mechanism, and was overlooked.

Whereas Kraut et al. found that Internet use caused depression, which was also directly linked to stress, the sociocognitive view differentiates the relationships among these effects. According to Bandura (1997, p. 153), depression results from "the inability to influence events and social conditions that significantly affect one's life," while stress is an emotional state generated by threats and taxing demands (p. 262). Adversity leads to depression when people create a depressing social environment for themselves, provoking social rejection through their own alienating behavior. Self-efficacy may mediate the effect of both stress and social support on depression. Cutrona and Troutman (1986) presented a path analysis of the relationship among these variables, in which stress reduced self-efficacy while social support increased it and in which self-efficacy then directly reduced depression.  Kraut et al. (1998) did include social support in their model, but as a general controlling variable, while social cognitive theory assigns it a more direct role, acting through self-efficacy, in the genesis of depression.

In the Internet paradox study, general life stress was treated as an external control factor. The Internet itself, however, is a source of stressful stimuli, and perhaps a very relevant one when investigating the link between its use and psychological well-being, particularly among novice users. For instance, most Internet users in the GVU surveys reported problems with slow downloads and unwanted e-mail (GVU, 1999, question 11). For those who depend upon the Internet to complete important life activities, the stress resulting from such problems could be a significant source of depression. Indeed, if the HomeNet subjects felt compelled to persist in Internet use as part of their arrangement for the free equipment and Internet service they received, a new competing explanation for the link between Internet use and depression in the Kraut et al. (1998) study emerges: As Internet use increased among these novice users, Internet stress also increased, leading in turn to depression. Perhaps the novice users in the HomeNet study never achieved the levels of self-efficacy required to control Internet-related stress.


From this perspective we may reformulate the relationship between Internet usage and depression, adding the intervening variable of self-efficacy. We propose two separate, but interrelated, mechanisms describing the relationship between Internet use and depression. One focuses on stress-inducing interactions with the Internet that contribute to depression while the other emphasizes the use of the Internet to obtain social support that reduces depression.

Novice users experience stressful interactions with the Internet that may trigger depression when they feel unable to control important events that depend upon successful use of the Internet. This is especially likely in cases where the stressful Internet events are beyond volitional control (such as encountering a busy signal when establishing a network connection, or encountering 404 errors or slow downloads on the Web). However, users gradually gain confidence in their ability to control the sources of Internet stress as they learn to dial alternate access numbers, set their dialers to autodial, or avoid the times of day when busy signals are most common, for example. So, the effect of Internet stress on depression should be mediated by Internet self-efficacy, the belief in one’s ability to use the Internet successfully. Following Cutrona and Troutman (1986), we hypothesize that stress reduces self-efficacy, leading to depression, while social support increases self-efficacy. An important antecedent of self-efficacy is previous experience (Bandura, 1997), so the amount of prior Internet experience should act on depression through self-efficacy.

H1: Internet usage is positively related to depression as an inverse function of Internet self efficacy.
H1a: Internet self-efficacy reduces the effect of Internet stress on depression.
H1b: Self-efficacy is positively related to social support and prior Internet experience.

However, Cutrona and Troutman (1986) found evidence (mirrored by the Internet Paradox study) of a direct link between stress and depression that was not mediated through self-efficacy. General life stress may also be related to situation-specific forms of stress (Kanner, Coyne, Schaefer, & Lazarus, 1981; Lu, 1994), although there is no indication in the literature of the direction of the relationship. It was initially assumed that general life stress would intensify Internet stress by creating a general feeling of "being hassled" and so precede it causally. Social support may also have a direct, inverse relationship to depression as well as a buffering effect (Cohen & Wills, 1985; Hashimoto et al., 1999). A direct effect from social support on depression, not found by Kraut et al. (1998), may be expected in populations with higher levels of depression than that of the Paradox study, since depressed people may be more likely to need social support.

H1c: General life stress is positively related to depression both directly and as a function of Internet stress and self-efficacy.
H1d: Social support is negatively related to depression

A second mechanism may decrease depression: Profligate Internet users might obtain social support from distant associates, and thereby either directly relieve depression, or buffer the effect of stress on depression through self-efficacy. Electronic mail would seem to be the crucial Internet application in this regard. E-mail was the single most frequent Internet activity in Kraut et al. (1999), a finding confirmed in national surveys (Katz & Aspden, 1997; Pew Research Center, 2000). Kraut et al. (1998) conducted (unreported) analyses that showed a positive relationship between e-mail use and depression (p.1029). However, their approach to measuring e-mail use may have obscured the  relationship. They used computer logs to count the actual number of e-mail messages sent and received, and excluded only those messages in which the respondent was not explicitly named, as these were presumably from mass distribution lists (i.e. listservs) that provide information rather than social support. They thus may have counted a great deal of unwanted e-mail either from unsolicited commercial "spammers" or from individuals with whom users might not wish to communicate (e.g., complete strangers or bothersome acquaintances). Since unwanted e-mail is a potential source of Internet stress and the receipt of such mail is likely to increase with use--especially among novice users who haven't learned to control it--spam emerges as a competing explanation for the Internet paradox effect.

However, electronic communication with people we know should enhance social support. Kraut et al. (1998) noted that socially isolated individuals might become less depressed as the result of social contacts made on the Internet. College students are one such lonely and depressed population (Rich & Scovel, 1987) for which social support buffers the effects of stress on depression (Cohen et al., 1986) and for which the Internet paradox might be stood on its head. Indeed, the situation of college students exposes the questionable assumption of equating distant ties with weak ones. For the lonely student, the most meaningful sources of social support may be available only by using social technologies to maintain distant ties with family and former high school classmates.

H2: Internet use is negatively related to depression among college students as a function of e-mailing known associates and social support.

However, the ability to obtain social support may itself be an acquired skill that takes some years of Internet experience to master (Pew Research Center, 2000), therefore:

H2a: Prior Internet experience is positively related to social support.

The hypothesized relationships among these variables are summarized in a path model shown in Figure 1.

Figure 1: Hypothesized Path Model

figure1 (12980 bytes)

Research Methods


Respondents were 171 students enrolled in an introductory telecommunication class at a large midwestern university in the USA. The sample was 59 percent male and 39 percent female. Thirty-five percent of the respondents were freshmen; 22 percent, sophomores;18 percent, juniors; and the rest, seniors. The mean age was 21 years old,   (SD = 5.00) . Respondents were offered extra credit for their participation in the study,while alternate extra credit assignments involving participation in other research projects with comparable time commitments were available at the students' options.


Questionnaires were distributed over two successive weeks so that students who were not present in class during the first week would have an opportunity to participate in the second week. Respondents completed questionnaires only once. Respondents picked up the questionnaires and returned them two days later. They also kept a diary of their Internet use during that time (results not reported here).


Measures of social involvement and psychological well-being previously used by Kraut et al. (1998) were included in the present study and reliability indices (Cronbach a) were computed. In each case, mean values were substituted for missing data on individual scale items. The depression measure included all 20 of the items from the Center for Epidemiologic Studies Depression (CES-D) scale (Radloff, 1977, a = .91), scored as 0=Rarely/None, 1=Some/Little, 2=Occasionally/Moderate, 3=Most/All.  Positively-worded items (e.g., "I enjoyed life") were reflected.  A 57-item scale was employed including 54 of 156 items from Kanner et al.'s (1981) Hassles Scale plus three (non-Internet specific) computer hassles items (lost computer files, caught a computer virus, computer hardware failure; a = .93). The same 16 (out of 40) items from the Interpersonal Support Evaluation List (ISEL; Cohen, et al, 1985, a = .81) used in the previous study were also included.  See Appendix for notes on these and other scales.

Eastin and LaRose's (2000) Internet Self-efficacy scale was used. The eight-item measure (e.g., "I feel confident using the Internet to gather data") was highly reliable (a = .93). The subjects rated their efficacy beliefs on a seven-point scale ranging from 7 for strongly agree to 1 for strongly disagree.

A four-item measure of Internet stress (a = .61 ) was developed from previous work on Internet frustrations (Charney & Greenberg, in press) and from GVU research on problems using the Web (GVU, 1999). Respondents were asked to rate their likelihood of experiencing each type of stressful Internet behavior (e.g. have trouble getting on the Internet, have trouble finding what I am looking for, have my computer freeze up, and get blocked by password protection) on a seven-point scale that ranged from 7 for very likely to 1 for very unlikely.

Internet usage was an additive index of four self-reported items (a = .82).

E-mail use (a = .67) was the sum of two items measuring the number of e-mail messages sent (M = 2.60, SD = .86) and the number received (M = 3.20, SD = .94) from people known to the respondent in the preceding two days. They were coded 1 for no messages, 2 for one to five messages, 3 for six to ten, 4 for eleven to twenty-five, and 5 for twenty-six messages or more.


Path analytic techniques were used to analyze the data (McClendon, 1994). Path analysis allows the researcher to specify and test the pattern and direction of causal relationships among the variables where mediated effects are predicted. Kraut et al. (1998) used path analysis to analyze the results of a time series study in which the dependent variable was observed at two points in time, allowing them to make interpretations about the direction of causation (i.e., that Internet use causes depression rather than depressed people use the Internet more). Path analysis may also be used when observations are performed at a single point in time. Here, path analysis allowed us to test all proposed relationships within the theoretical model. The results of Kraut et al. (1998) gave us confidence about the direction of causation from Internet usage to depression, but we tested competing models of the relationships among intervening variables.

The present study was a cross-sectional survey so we could not replicate the longitudinal controls for depression performed by Kraut et al. (1998). Consistent with our theoretical approach, we did not use demographic variables (e.g. gender) as controls. Within social cognitive theory, the explanatory power of such variables is subsumed by social cognitive constructs.


A matrix showing the Pearson product-moment correlation coefficients between variables is presented in Table 1, with means and standard deviations for each variable. All significant correlations reported are based on two-tailed tests. Overall, the present sample was significantly more depressed than both Kraut et al.’s (1998), t = 4.37, p < .001 (two-tailed), and the general population sample used to validate the CES-D originally (Radloff, 1977), t = 10.27, p < .001 (two-tailed), but less so than members of the general population who believed they "need help" in the validation study, t = -4.21, p < .001. The scores on the CES-D ranged between 0 and 55, against a theoretical range of 0 to 60. Thirty-seven percent of the respondents scored at or above the arbitrary cut-off point of 16 that distinguishes moderately from clinically depressed people. Internet usage scores ranged from the minimum possible score of 3 to the maximum possible 24. The Internet self-efficacy scores also covered the entire possible range of 8 to 56 as did Internet stress (4 to 28) and e-mail usage scores (2 to 10). Internet experience ranged from 2 to 5 (no respondent reported less than two months’ experience with the Internet, which was scored as 1) and social support (ISEL) scores all fell between 4 (0 was the minimum) and 16 (the maximum possible)

Table 1: Pearson Product Moment Correlation Coefficients


1 2 3 4 5 6 7 Mean  SD   


1. Internet Use   14.21   4.31
2. Depression (CES-D)  -.02     15.74 9.82
3. Internet Self-Efficacy   .65** -.11     36.11 11.76
4. General Stress (Hassles)  -.03    .43** -.04     85.02 19.61
5. Social Support (ISEL)  -.04   -.59**  .09   -.36**   13.50 2.84
6. Internet Stress  -.18*   .23** -.25**  .29** -.20**     14.74 4.77
7. Internet Experience   .22** -.18*   .33** -.11    .15   -.12     4.66 .67
8. E-mail Use   .20**  -.07     .15*    .10     .22**  -.06    .07   8.53 2.19

** p < .001; * p < .05

Our initial model in which Internet stress was linked to depression through self-efficacy was not supported; a goodness of fit test indicated that the data did not fit the model, c 2 (14) = 23.24, p < .05. Alternative models were evaluated for their statistical goodness of fit and also their correspondence to theoretically justifiable relationships within the social cognitive paradigm. A revised model is shown in Figure 2, with the significant paths (p < .05) indicated. In it, Internet use was related to depression through two routes: first, Internet usage is related to depression through self-efficacy (▀ = .59) and then through Internet stress (▀ = -.25). Internet stress was in turn related to hassles (▀ = .23) and hassles to depression (▀ = .25). Prior Internet experience was related to self-efficacy as predicted (▀ = .20). The expected relationship between prior experience and social support fell slightly below the level of significance used in this study (▀ = .12, p = .110). The hypothesized link between social support and self-efficacy was not confirmed.

Overall, Internet use was positively related to e-mail use (▀ = .31), which in turn was positively related to social support (▀ = .20). Social support had a significant and negative direct relationship to depression (▀ = -.50) and also acted on depression through hassles (▀ = -.31).

Figure 2: Final Path Model

figure2 (10332 bytes)

Note: * indicates significance at the .05 level, ** at the .01 level. Only statistically significant links (p < .05) are shown. Path coefficients are standardized betas (▀).

The results can be interpreted to indicate that the amount of Internet use was related to depression through two different mechanisms. Usage as well as prior Internet experience increased self-efficacy, which in turn decreased stress encountered online, a contributor to general life hassles related to depression, the central path in Figure 2. In the second mechanism, represented in the top part of Figure 2, as Internet use increased so did email sent to known associates, which increased social support, and in turn decreased depression. In other words, Internet use decreased depression through the use of electronic mail to obtain social support. Social support also reduced depression by acting on general life stress (hassles), and general stress increases depression. That means that social support can partially reverse the effects of Internet stress on general life stress and so on depression. But Internet usage can also increase depression by creating a new source of Internet stress, although that stress may be controlled by the development of Internet self-efficacy.

Given the substantial path coefficients and a favorable goodness of fit results, c2 (18) = 14.44, p = .700, it was concluded that the data fit this model. The variance in depression explained by this model was 37 percent, which can be characterized as a large effect (Cohen, 1988). Kraut et al.'s (1998) results had an R2 of .19, a moderate effect size.

In both studies the direct relationship between Internet usage and depression was low (r = -.02 in the present study, r = .18 in Kraut et al.). The reconstructed correlations, obtained by multiplying the path coefficients found along each path together, were -.008 for the path from Internet usage to depression through Internet self efficacy and -.031 for the path through social support. This suggests that the social support mechanism shown in the upper part of Figure 2 is more powerful than the self-efficacy mechanism in the lower path. However, although the relationships were significant, as indicated by statistically significant path coefficients, the magnitude of the effects on depression were slight across both paths.

The importance of prior Internet experience was further explored by comparing the Internet self-efficacy and Internet Stress scores of those with high and low levels of online experience. There was a significant difference, t (39, 131) = 3.60, p < .001 ( two-tailed), in self-efficacy between those with more than two years' prior experience with the Internet (M = 37.80, SD = 11.32)  and those with less than two years (M = 30.30, SD = 11.63). The amount of prior Internet experience was related to perceptions of Internet Stress somewhat differently: Those with the very least experience--a year or less--had the greatest Stress (M =  16.82, SD = 3.59) compared to those with more than one year's experience (M = 14.56, SD = 3.67), t (17, 153) = -2.37, p < .05.


We found support for two propositions that counter the more negative conclusions of Kraut et al.'s (1998) Internet Paradox study. First, Internet communication with people we know can alleviate depression, at least among socially isolated and moderately depressed populations, such as college students, who may tend to rely on social technologies to obtain social support. Second, stressful interactions with the Internet itself, rather than inadequate interactions with other people through the Internet, may lead to depression, but self-efficacy reverses the effect of that stress.

The latter findings pose a rival explanation for the Internet paradox, one first suggested by Hamman (1999). The novice users that Kraut et al. studied may never have achieved the necessary degree of self-efficacy needed to cope with the new sources of stress that the Internet introduced into their lives. All of Kraut et al.’s subjects had less than two years of experience at the time of the post-test.

Self-efficacy could be a third variable that accounts for concomitant increases in depression and in Internet use in Kraut et al. (1998). Inefficacious users in the HomeNet study may have spent more time on line than efficacious users as a consequence of their poor performance. It may have taken them longer to find what they were looking for, or they may have wasted time trying to resolve online problems rather than engaging in productive tasks. And the stress they encountered in the process made them more depressed.

Contrary to Kraut et al. (1998) the present research established a relationship between Internet usage and social support. Kraut et al. speculated that social support provided an explanatory link in the Internet/depression connection, reasoning that when people substitute shallow online relationships for supportive offline relationships, social support declines, causing depression. No significant effects due to social support were found in their study, however. The current results do show a relationship between Internet usage and social support, presumably through the exchange of e-mail with known associates. Only marginal support was found for the proposed mechanism that users gradually learn how to obtain social support through the Internet. The relationship of Internet experience to social support may also be mediated by self-efficacy relating to social support (cf. Holahan & Holahan, 1987a), or confidence in one’s ability to obtain social support online.

The present results tend to rule out another competing hypothesis about the role of social support in the Internet paradox. Shapiro (1999) speculated that young college students in the HomeNet sample experienced shrinking social networks as they made the transition to college and turned, unsuccessfully, to the Internet to fill the void (although the college students who left home were in fact dropped from the sample). In the present research we found evidence of a relationship between Internet use and a reduction in depression among college students. They thus may have used the Internet to obtain social support rather than to replace it. This view is consistent with a national survey in which Internet users reportedly improved their social relations (Pew Research Center, 2000) and with Wellman and Gulia’s (1999) evidence that the Internet is used to maintain real-world relationships.

Self-efficacy did not mediate depression in exactly the manner hypothesized, preceding Internet stress and hassles rather than following them. And, Internet stress preceded general life stress/hassles rather than following it, as hypothesized. However, it is possible to interpret this result within social cognitive theory: Self-efficacy improved performance (Bandura, 1982; 1997), and as performance improved users were less likely to encounter stressful negative outcomes. In other words, self-efficacious Internet users were less likely to make mistakes that were sources of stress when using the Internet and were better able to work around problems that were not of their own making. They therefore correctly perceived a reduced likelihood of encountering stressful situations on the Internet. Successful Internet use is perhaps such a critical domain of behavior for college students that stressors in that domain may contribute to a general feeling of "being hassled" and so to depression.

More puzzling was the failure of social support to enter into the relationship between Internet use, self-efficacy and depression. One possibility is that self-efficacy may reduce stress without mediation by social support, such as when users obtain help from online FAQ files or by gradually learning to solve their own problems. Self-efficacy should act in concert with social support, though, when novice Internet users obtain technical help or moral support from others for their Internet problems. In the present study, students completing Internet assignments could get help from the instructor, her teaching assistants, or peer tutors. However, the ISEL (Cohen et al., 1985) does not address specific support of this nature, only general social support.

The ISEL used here and in the original Internet Paradox study may inadequately reflect online social support generally. It also does not correspond well to a sociocognitive conceptualization of social support as an overall rewarding social environment (Bandura, 1997; see Silverman, 1999, for anecdotal examples of social support in an Internet newsgroup that can be understood in sociocognitive  terms). In this view, significant social support might come from weak ties with familiar strangers, unknown neighbors and "urban agents" (e.g. service role occupants such as teachers, bartenders, and perhaps online help attendants) even in mediated channels with few social cues (Adelman et al., 1987), sources that avoid impositions on close relationships (Walther & Boyd, in press). Not all social support is supportive; there are negative instances that detract from psychological well-being (Rook, 1984). The ISEL asks about the availability of people who can provide supportive but not specifically whether positive support is actually obtained. One of its subscales stresses tangible forms of support that can perhaps only be provided by physically proximate real-world associates (e.g., moving furniture) and it is that very dimension that may best buffer the effects of stressful life events (Cohen & Wills, 1985). The ISEL lacks an attachment dimension (intimate relationships providing security and safety) found in the Social Provisions Scale (SPS; Cutrona & Russell, 1987) that measured social support in the study from which we hypothesized the social support to self–efficacy relationship (Cutrona & Troutman, 1986).NOTE 4 The SPS or another more suitable measure of social support might produce the hypothesized relationship.

However, other commonly used measures of social support also limit responses to small networks of significant others (Norbeck et al., 1981; Oritt et al., 1985; Sarason et al., 1983), stress tangible forms of support that require geographic proximity (Barrera et al., 1981; Cutrona & Russell, 1987) or emphasize the availability of support as opposed to its actual provision (Cutrona & Russell, 1987). Affective reactions to (Norbeck et al., 1981) or satisfaction with the social support received (Oritt et al., 1985; Sarason et al., 1983) come closer to our conceptual definition, but reflect only the support forthcoming from a short list of close associates that might tend to exclude exclusively online relationships.

Finally, the CES-D may not be a valid measure of depression but rather an indicator of general psychological stress (Rierdan, 1999). On that basis, the initially hypothesized relationship among social support, stress and depression might still be observed by introducing a different measure of depression into the model.


A limitation of our study is that we relied upon self-reports of Internet behavior. However, our research showed a high (r = .65, p < .001, two-tailed) correlation between Internet usage recorded in a contemporaneous diary and retrospective recall of the same behavior. This is consistent with research comparing self- reports of computer system activity with electronic log data (Deane, Podd, & Henderson, 1998) and also with comparisons between self-reported and objectively measured activity in multi-user games (Zielke, Schildmann, & Wirausky, 1995). Within social cognitive theory it is the perception of behavior rather than the "actual" behavior that matters: "If [people] want to exert influence over their own actions, they have to know what they are doing" (Bandura, 1986, p. 336). Earlier we noted how an "actual" measure of e-mail use may have been a confounding factor in the original Internet paradox study.

The use of a convenience sample from an introductory college course poses another limitation. Other populations may yield different results and different causal mechanisms. The sample did include a wider range of Internet experience than was reflected in the original Internet Paradox study, but still not the full range of experience in the general Internet user population, excluding those with less than two months’ experience.

Finally, our cross sectional design limits our ability to make statements about causal relationships. Third variables arising from history and maturation cannot be ruled out. The exact relationship between social support and depression also differs between cross-sectional and prospective time series studies, possibly because of confounding between prior depressive symptoms and subscales of social support measures, notably the self-esteem dimension of the ISEL (Schonfeld, 1990).

For Further Research

Thus, an important direction for future research is to verify the causal mechanisms proposed here through longitudinal studies. In addition to verifying the possible causal relationship between general Internet use and measures of psychological well-being, it would be instructive to examine the impact of specific types of Internet use (e.g. e-mail, chat rooms, online research, entertainment). While the social displacement hypothesis--that strong face-to-face social ties are replaced by weak online ties--does not seem a viable mechanism in light of the current analysis, it may yet prove to be valid when examining the impact of online entertainment on social involvement, for example.

There is a need for controlled studies that do not draw exclusively upon naive populations, or else to extend them beyond two years, since novice users may not achieve the necessary levels of self-efficacy required to relieve the stress of their struggles to master the Internet in that time. Cross-lagged correlation techniques could be applied among more widely representative populations

The multidimensional nature of social support should be recognized in future research, as the subscales have been found to have differing relationships to depression (Hawkins, 1999; Schonfeld, 1991). Depression itself is multidimensionsal and the CES-D in particular has been found to have four factors (depressed affect, somatic disturbance, positive affect, and interpersonal difficulties; Lewis, 1995) which may be differentially affected by social support. Two different types of stressful events are also recognized, one arising from daily hassles as examined here and in Kraut et al. (1998), but another stemming from major life crises (such as the death of a spouse). Thus, interactions among differing dimensions of social support, depression and stress could be productively examined in relation to Internet communication. For example, social support for major life crises might be more forthcoming from online discussion groups organized around major life crises (e.g., cancer support groups) than from e-mail with known associates (e.g. Walther & Boyd, in press).

Further development of a measure of Internet stress is called for since the present one achieved a barely acceptable minimum level of internal consistency. Stressful events over which there is considerable volitional control (e.g., the receipt of unwanted e-mail) can be distinguished from those where control is limited (e.g., Internet brownouts). According to the goodness-of-fit hypothesis (Roberts, 1995), social support may be more important when the stressors cannot be altered. Sources of stress that are attributable to technology and those that are attributable to the behavior of other people on the Internet, such as the receipt of unwanted e-mail, could also be distinguished.

Further research should also explore other constructs suggested by social cognitive theory. The possibility of social self-efficacy was discussed previously. A distinction might be made between general Internet self-efficacy and coping self-efficacy, that is, beliefs in one's ability to successfully perform actions that alleviate specific sources of stress (Holahan & Holahan, 1987b).

Policy Implications

Kraut et al. (1998) posed some far-reaching suggestions for public policy that we would like to critique in light of the current findings. We agree that attention should to be devoted to fostering the use of the Internet as a medium of social exchange as well as a medium for commerce and information retrieval. However, their recommendation to improve search capabilities for finding people (p. 1030) could be counterproductive, at least for novice users, since it is likely to lead to stressful unwanted contacts. Thus, more powerful tools for filtering out unwanted e-mail are also needed.

We also do not find the hypothesis that the Internet inherently diminishes strong social ties to be entirely compelling. Kraut et al. seek policies that would encourage communication in pre-existing social groups (p. 1030). Such policies could make that hypothesis self-fulfilling at the expense of Internet users who seek rewarding social interaction (e.g. in illness-related support groups) unavailable in their own social circle. Walther and Boyd (in press) found that Internet social support offers benefits that face-to-face social networks cannot by providing anonymity, constant access to better quality expertise, and enhanced modes of expression, with less chance of embarrassment and without incurring an obligation to the support provider. This perspective highlights the need for policies that promote contacts outside of existing social networks.

But perhaps there should be much more concern about computer support as well as social support. The Internet has always been a rather hostile place for the "newbie." Support for new users may prove to be a critical factor in efforts to close the Digital Divide (Hoffman & Novak, 1998; NTIA, 1999). The novice users in the present sample had a powerful motivation to master the Internet, in that it was instrumental to their success in college, while the more experienced users may have been intrinsically motivated to adopt the Internet as an expression of a personal interest in computing while still in high school. The late adopters who must be introduced to the Internet to close the Digital Divide may more closely resemble the HomeNet sample. They may lack sufficiently compelling expectations of the outcomes of Internet usage to adopt it on their own and may fail to develop the sense of self-efficacy required to master their anxieties and persist in Internet use once introduced to it.

Current social policies aimed at closing the divide, such as the e-Rate program (USAC, 2000), focus on providing the technological means of Internet access, but not the technical and social support that may prove vital to the success of these efforts. The new E-Corps initiative (Corporation for National Service, 2000) aimed at staffing schools and libraries with tutors and technical support is a promising step in this direction. Presumably, these efforts will create what is referred to in social cognitive theory as an enactive mastery experience, through which novice users are guided to achieve gradual improvements in performance. Vicarious experience (seeing similar others succeed), verbal persuasion and control of physiological states are also effective ways of increasing self-efficacy (Bandura, 1997).

It might also be argued that the problem will solve itself as new users progress to become experienced ones. However, to take this approach would be to condemn new Internet users to years of unproductive effort. It also risks widening the Digital Divide if frustrated users, unable to use the Internet effectively to obtain desirable outcomes, abandon its use or fail to strive for new levels of attainment.


Adelman, M. B., Parks, M. R. & Albrecht, T. L. (1987). Beyond close relationships: Support in weak ties. In T. L. Albrecht & M. B. Adelman (Eds.), Communicating social support (pp 126-147). Newbury Park, CA: Sage.

Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84, 191-215.

Bandura, A. (1982). Self-efficacy mechanisms in human agency. American Psychologist, 37, 122-147.

Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory. Englewood Cliffs, N.J.: Prentice-Hall.

Bandura, A. (1989). Human agency in social cognitive theory. American Psychologist, 44, 1175-1184.

Bandura, A. (1997). Self-efficacy: The exercise of control. New York : W.H. Freeman.

Barrera, M., Sandler, I. N., & Ramsay, T. B. (1981). Preliminary development of a scale of social support: Studies on college students. American Journal of Community Psychology, 9, 435-447.

Charney, T., & Greenberg, B. S. (in press). Uses and gratifications of the Internet. In C. Lin. & D. Atkin, (Eds). Communication, technology and society: New media adoption and uses. Cresskill, NJ: Hampton Press.

Cobb, S. (1976). Social support as a moderator of life stress. Psychosomatic Science, 38, 300-314.

Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, N.J.: L. Earlbaum Associates.

Cohen, S., Mermelstein, R., Kamarck, T., & Hoberman, H. (1985). Measuring the functional components of social support. In I. G. Sarason & B. R. Sarason (Eds.), Social support: Theory, research and applications (pp. 73-94). The Hague, Holland: Martines Niijhoff.

Cohen, S., Sherrod, D. R., & Clark, M. S. (1986). Social skills and the stress-protective role of social support. Journal of Personality and Social Psychology, 50, 963-973.

Cohen, S., & Wills, T. A. (1985). Stress, social support and the buffering hypothesis. Psychological Bulletin, 98, 310-357.

Cook, T., & Cambell, D. (1979). Quasi-experimentation: Design & analysis issues for field settings. Chicago: Rand McNally.

Cutrona, C. E., & Troutman, B. R. (1986). Social support, infant temperament and parenting self-efficacy: A mediational model of postpartum depression. Child Development, 57, 1507-1518.

Corporation for National Service (2000). Creating digital opportunities for national service. National Service News, 103. Retrieved November 29, 2000 from the World Wide Web:

Cutrona, C. E., & Russell , D. W. (1987). The provisions of social support and adaptations to stress. Advances in personal relationships (Vol. 1, pp. 37-67). London: J. Kingsley Publishers.

Deane, F. P., Podd, J., & Henderson, R. D. (1998).  Relationship between self-report and log data estimates of information system usage.  Computers in Human Behavior, 14, 621-636.

D÷ring, N. (1996). FŘhren computernetze in die Veriensamung? Grupppendynamick, 27, 289-307.

Eastin, M. A., & LaRose, R. L. (2000). Internet self-efficacy and the psychology of the digital divide. Journal of Computer Mediated Communication, 6 (1). Retrieved November 29, 2000 from the World Wide Web:

Granovetter, M. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360-1380.

GVU (Graphic, Visualization and Usability Center) (1999). GVU's tenth annual WWW user's survey. Georgia Institute of Technology. Retrieved November 23, 1999 from the World Wide Web: (Questions 11, 30, 101, 102).

Hamman, R. B. (1999). Computer networks linking communities: A study of the effects of computer network use upon pre-existing communities. In U. Thiedke (Ed.), Virtualle Gruppen- Characteristika und Prolemdimensionen (Virtual groups: Characteristics and problematic dimensions). Wiesbaden, Germany: Westdeutscher Verlag. Retrieved November 23, 1999, from the World Wide Web:

Hashimoto, K., Kurita, H., Haratani, T., & Ishibachi, T. (1999). Direct and buffering effects of social support on depressive symptoms of the elderly with home help. Psychiatry and Clinical Neurosciences, 53, 95-100.

Hawkins, W. E., Tan, P. P., Hawkins, M. J., Smith, E. & Ryan, E. (1999). Depressive symptamatology and specificity of social support. Psychological Reports, 84, 1180-1186.

Heim, M (1993) The metaphysics of virtual reality. Oxford: Oxford University Press.

Hoffman, D. L., &. Novak, T. P. (1998, April 17). Bridging the racial divide on the Internet. Science, 280, 90-391.

Holahan, C. K., & Holahan, C. J. (1987a). Self-efficacy, social support and depression in aging: A longitudinal analysis. Journal of Gerontology, 42, 65-68.

Holahan, C. K., & Holahan, C. J. (1987b). Life stress, hassles and self-efficacy in aging: A replication and extension. Journal of Applied Social Psychology, 17, 574-592.

Joseph, S., & Lewis, C.A. (1995). Factor-analysis of the center for epidemiologic studies-depression scale. Psychological Reports, 76, 40-42.

Kanner, A. D., Coyne, J. C., Schaefer, C. & Lazarus, R. S. (1981). Comparisons of two modes of stress measurement: Daily hassles and uplifts versus major life events. Journal of Behavioral Medicine, 4, 1–39.

Katz, J. E., & Aspden, P. (1997). A nation of strangers? Communications of the ACM, 40, 81-86.

Kessler, R. C. (1997). The effects of stressful life events on depression. Annual Review of Psychology, 48, 191-214.

Kiesler, S. & Kraut, R. (1999). Internet use and the ties that bind. The American Psychologist, 54, 783-784.

Kraut, R., Patterson, M., Lundmark, V., Kiesler, S., Mukophadhyay, T., & Scherlis, W. (1998). Internet paradox: A social technology that reduces social involvement and psychological well-being? American Psychologist, 53, 1017-1031

Kraut, R., Mukhopadhyay, T., Szczypula, J., Kiesler, S., & Sherlis, B. (1999) Information and communication: Alternative uses of the Internet in households. Information Systems Research, 10, 287-303.

Kroker, A., & Weinstein M. A. (1994). Data trash: The theory of the virtual class. New York: St. Martin's Press.

LaRose, R. (1999). Understanding personal telephone behavior. In H. Sawhney & G. A. Barnett (Eds.), Progress in communication sciences: Vol. XV (pp. 1-18). Stamford, CT: Ablex.

McClendon, M. (1994). Multiple regression and causal analysis. Itasca, IL: F. E. Peacock Publishers, Inc.

McNally, S.T. & Newman, S. (1999). Objective and subjective conceptualizations of social support. Journal of Psychosomatic Research, 46, 309-314.

Nie, N. H., & Erbring, L. (2000). Internet and society. Retrieved May 10, 2000 from the World Wide Web:

Norbeck, J. S., Lindsey, A. M. & Carrieri, V. L. (1981). The development of an instrument to measure social support. Nursing Research, 30, 264-269.

NTIA (National Telecommunications and Information Administration) (1999). Falling through the net: Defining the digital divide. Retrieved November 23, 1999 from the World Wide Web:

Oritt, E. J., Paul, S. C., & Behrman, J. A. (1985). The perceived support network inventory. American Journal of Community Psychology, 13, 565-582.

Parks, M. R. & Floyd, K. (1996). Making friends in cyberspace. Journal of Communication, 46, 80-97.

Parks, M. R., & Roberts, L. D. (1998). ‘Making MOOsic’: The development of personal relationships on line and a comparison to their off-line counterparts. Journal of Social and Personal Relationships, 15, 519-537.

Pew Research Center (2000). Tracking online life: How women use the Internet to cultivate relationships with family and friends. Retrieved May 10, 2000, from the World Wide Web:

Radloff, L. (1977). The CES–D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385–401.

Rich, A. R., & Scovel, M. (1987). Causes of depression in college students: A cross-lagged panel correlational analysis. Psychological Reports. 60, 27-30.

Rierdan, J. (1999). Internet-depression link? The American Psychologist, 54, 781-782.

Riphagen, J. & Kanfer, A. (1997). How does e-mail affect our lives? National Center for Supercomputing Applications Retrieved October 15, 1999 from the World Wide Web:

Roberts, S. M. (1995). Applicability of the goodness-of-fit hypothesis to coping with daily hassles. Psychological Reports, 77, 943-954.

Rook, K. (1984). The negative side of social interaction: impact on psychological well-being. Journal of Personality and Social Psychology, 46, 1097-1108.

Sarason, I. G., Levine, H. M., Basham, R. B., & Sarason, B. R. (1983). Assessing social support: The social support questionnaire. Journal of Personality and Social Psychology, 44, 127-139.

Scherer, K. (1997). College life on-line: Healthy and unhealthy Internet use. Journal of College Student Development, 38, 655-665.

Shapiro, J. S. (1999). Loneliness: Paradox or artifact? The American Psychologist, 54, 782-783.

Shonfeld, I. S. (1991). Dimensions of functional social support and psychological symptoms. Psychological Medicine, 21, 1051-1060.

Silverman, T. (1999). The Internet and relational theory. The American Psychologist, 54, 780-781.

Stoll, C. (1995). Silicon snake oil: Second thoughts on the information superhighway. NY: Doubleday.

Turkle, S. (1995). Life on the screen: Identity in the age of the Internet. New York: Touchstone Books.

USAC (Universal Service Administrative Company) (2000). Schools and Libraries Program. Retrieved November 29, 2000 from the World Wide Web:

Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal and hyperpersonal interaction. Communication Research, 23, 3-43.

Walther, J. B., & Reid, L. D. (2000, February 4). Understanding the allure of the Internet. Chronicle of Higher Education, B4-B5.

Walther, J. B., & Boyd, S. (in press). Attraction to computer-mediated social support. In C.A. Lin & D. Atkin (Eds.), Communication technology and society: Audience adoption and uses of the new media. New York: Hampton.

Wellman, B. (1996). Are personal communities local? A Dumptarian reconsideration. Social Networks 18, 347-354.

Wellman, B., & Gulia, M. (1999). Virtual communities as communities: Net surfers don't ride alone. In. M. A. Smith & P. Kollock (Eds.), Communities in cyberspace (pp. 167-194). NY: Routledge.

Wynn, E., & Katz, J. E. (1997). Hyperbole over cyberspace: Self-presentation and social boundaries in Internet home pages and discourse. The Information Society, 13, 297-327.

Zielke, A., Schildmann, S., & Wirausky, H. (1995). Spiel- und Sozialverhalten im Morgengrauen [Play and social behavior in the MUD Morgengrauen]. Retrieved April 5, 2000 from the World Wide Web:

Copyright © 2001 Behavior OnLine, Inc. All rights reserved.

| Behavior OnLine Home Page |