The paradox that went away
In 1998 a careful study found that the more people used the internet, the lonelier they became. By the time the follow-up was published four years later, the effect had gone. A short history of one finding, and what it suggests about reading the research on phones.
On Sunday 30 August 1998, the New York Times ran a story under the headline "Sad, Lonely World Discovered in Cyberspace." It reported that researchers at Carnegie Mellon University had followed ordinary families through their first years online and found that the more time people spent on the internet, the more depressed and lonely they became (Harmon, 1998). The paper behind it appeared in American Psychologist the following month, under a title that put the puzzle in a single question: a social technology that reduces social involvement and psychological well-being? (Kraut et al., 1998).
The finding was surprising, including to the people who made it. The project had been funded by the National Science Foundation and by technology companies including Intel, Hewlett-Packard, AT&T and Apple, and its designers had expected the opposite result. "We were shocked by the findings," Robert Kraut, the lead author, told the Times.
What makes the study worth returning to is not that it was wrong. It was a careful piece of work, and in one important respect better designed than much of what has been published about phones since: it logged how much people used the technology instead of asking them, and self-reported use tracks actual use poorly (Parry et al., 2021). It is worth returning to because of what happened next, which is unusual for a famous finding. The same team kept following the same people with the same measures, ran a second study besides, and the effect went away.
The study
The project was called HomeNet. In 1995 and 1996, the Carnegie Mellon team gave ninety-three Pittsburgh families a home computer and internet access, recruiting them through high school journalism programmes and community development organisations (Kraut et al., 2002). The 1998 analysis covered 169 people in seventy-three of those households, over their first one to two years online (Kraut et al., 1998).
Two features of the design still look good. The researchers did not ask people how much they used the internet; server software logged it — hours online, email sent and received, websites visited. And they did not compare heavy users with light users at a single moment, which would have left them unable to say whether the internet changed people or whether different people used the internet differently. They surveyed everyone before they went online and again afterwards, and asked whether heavier use predicted changes in loneliness, depression, social circles and family communication, once each person's starting point was taken into account.
The team expected the internet to widen people's networks and give them more support to draw on. Instead, participants who used it more became less socially involved and lonelier than those who used it less, and reported more symptoms of depression. Heavier use was also associated with less communication within the family, smaller social circles and more day-to-day stress (Kraut et al., 2002). All of this happened even though the participants mainly used the internet to communicate. That was the paradox: a technology used mainly for contact with other people seemed to be leaving its heavier users more isolated.
The effects on depression and loneliness were small. As the Times reported them, one hour a week online was associated with an increase of about one per cent on the depression scale and of less than half of one per cent on the loneliness scale. The effect on social circles was larger: each weekly hour went with the loss of 2.7 people from a circle that averaged sixty-six, around four per cent (Harmon, 1998). These were real associations, and not trivial ones. But the headline described a sad, lonely world, and the numbers described something more modest.
The study had critics. Some pointed out that there was no control group: everyone in the sample got the internet, so a general decline in well-being across the period, or ordinary regression to the mean, could have produced the pattern. Some questioned whether the measures captured what they claimed to; Shapiro raised both points (Shapiro, 1999). The authors' reply to the first was reasonable — anything affecting everyone should have affected heavy and light users alike, and the finding was about the difference between them (Kraut et al., 2002). But the most useful response had already begun. By the time the paper appeared in September 1998, the team had surveyed the original participants a third time and started a second study.
Going back
The follow-up, published in 2002 as "Internet Paradox Revisited," did two things (Kraut et al., 2002).
First, it reported a third survey of the original HomeNet participants, taken in February 1998, by which point they had been online for nearly two or nearly three years. Attrition was substantial — of 335 people originally eligible, fewer than half returned the third survey, and the analysis used the 208 who had completed at least two of the three. Across the longer period, the negative associations had largely disappeared. The one that remained was stress, which rose with internet use. For depression, the direction had reversed: heavier use predicted more depressive symptoms in the first period and fewer in the second. For loneliness, heavier use predicted more of it in the first period and was unrelated to it in the second.
Second, it reported a new study, run in 1998 and 1999 and designed to fix the original's weaknesses. The team recruited households in the Pittsburgh area who had just bought a new computer, and households who had just bought a new television as a comparison group. And they tried to run a true experiment: half the computer buyers were randomly offered free internet service, and the other half were paid instead.
The experiment failed, and the way it failed is the most instructive detail in the paper. Within twelve months, eighty-three per cent of the households who had not been given internet access had got it on their own, against ninety-five per cent of those who had. The control group had dissolved. Even among the television buyers, nearly three in ten were online by the end of the year. The researchers combined the groups and analysed the effect of internet use across 406 people in 216 households. One strength of the original was lost along the way: server logs existed only for the households given free access, so internet use in the new study was measured mainly by self-report, which the authors acknowledged might be too insensitive to pick up real differences.
In this sample, heavier use was associated with mostly good things. Over the year, heavier users' local and distant social circles grew more, they spent more time face to face with friends and family, and they reported more trust in other people and more positive mood. A few costs remained: stress again rose with use, and heavier users came to know less about their own city and to feel less committed to staying in it.
The average concealed a split. Extraverts who used the internet more became less lonely, felt better about themselves and reported less negative mood. Introverts who used it more showed the reverse on the same measures. The authors called this the "rich get richer" pattern: people who were already good at using social resources were best placed to use a new one.
What changed
The authors considered three explanations for why the same kind of study, run a few years apart, had produced opposite results. The participants might have learned to use the internet better. The samples might have differed. Or the internet itself might have changed.
The first could explain the shift within the original group but not the difference between the two studies, since both looked at people early in their time online. Measured differences between the samples in age, race and social resources did not appear to account for it either, though the authors noted that unmeasured differences might. They said plainly that their data could not settle the question, and called a change in the internet itself the most parsimonious explanation: "the Internet may have become a more hospitable place over time."
The reasoning is easy to follow. Between 1995 and 1998, the number of Americans with internet access at home more than quadrupled. A HomeNet family going online in 1995 was entering a space where, the authors point out, few of their family and friends had access. By 1998, many more of the people they knew were likely to be there, and new services, such as instant messaging with lists of friends who were online, made it easier to reach them. The internet also simply had more on it — more news, health and hobby information, the beginnings of online shopping. If the explanation is right, the 1998 study measured what a communication technology does to people while most of the people they want to communicate with are not yet on it.
That is a hypothesis, not a finding — nobody tested it directly. But if it holds, it is not a flaw in the original study so much as a property of the thing being studied. A communication technology would not have a fixed effect on the person using it, because much of what it does would depend on who else is using it, and that changes as the technology spreads.
Two decades later
Kraut stayed with the question. In 2016, Moira Burke and Kraut linked survey measures of well-being from 1,910 Facebook users to counts of what those users had actually done on the site, taken from server logs. The answer to the original paradox turned out to depend on what kind of contact was involved. Receiving composed, targeted messages from close ties was associated with improvements in well-being. Viewing friends' broadcasts to their whole audience, and receiving one-click feedback such as likes, were not (Burke & Kraut, 2016).
That is a more useful finding than either "the internet makes you lonely" or "the internet brings people together," and it echoes the 2002 paper's own unresolved question. Those authors wrote that no research, including their own, could yet tell whether the effects of the internet depended on what people did online or on what they gave up to be there (Kraut et al., 2002). It is the same question this site keeps arriving at by other routes.
Meanwhile, the broad association between internet use and well-being settled into something small. A 2010 meta-analysis of forty studies and more than 21,000 participants found an average correlation of about −0.04 to −0.05 — a detectable relationship, in the direction of harm, of a size that explains almost none of the variation between people (Huang, 2010).
Why this matters now
The psychologist Amy Orben has described what she calls the Sisyphean cycle of technology panics. A new technology arrives, concern follows, researchers are asked to investigate its effects on the young, and because there is no theory carried over from the last technology, the research seems to start again each time (Orben, 2020). The HomeNet story is a single, well-documented turn of that wheel, and it carries four lessons worth taking into any study of phones.
First findings may be made in an unusual world. If the authors' explanation is right, the earliest research on a communication technology studies people using it before most of the people they care about have arrived, and whatever it finds is partly a finding about that moment. Smartphones are now at the opposite extreme, which brings its own distortion: almost everyone is already on them, and the people in a study cannot simply step outside a network everyone else is still inside.
The control group disappears. The 2002 team tried to randomise internet access and watched eighty-three per cent of their control group go online anyway. Anyone trying to run a clean experiment on phones meets the same problem in a stronger form. Most trials since have worked around it by asking people to give something up for a few weeks, which is a different question from what owning the technology does over years.
Averages can hide opposite effects. In the 2002 sample, the same amount of internet use went with less loneliness for extraverts and more for introverts. An average across both groups would have described neither of them. Any study of phones that reports only an overall effect may be doing the same.
The first finding travels further than the follow-up. The 1998 paper has been cited roughly twice as often as the 2002 paper that revisited it, according to the OpenAlex citation index — about 5,000 citations against about 2,400 — and the gap holds for citations made after 2002, so it is not just the original's head start. A citation is not an endorsement, and many papers cite both. But the paper researchers cite more often is the one that found the paradox, not the one that watched it fade.
None of this means the concern was foolish, or that today's concerns about phones will dissolve in the same way. Some of the 2002 results point the other way: stress rose with use in both studies, and introverts did worse. And the phone is not the internet of 1995. It is with people everywhere, and the social world it connects is already fully built.
But the story is a good reason to ask, of any confident finding about what a technology does to people, when it was measured, who else was there at the time, and whether anybody went back.
References
- Burke, M., & Kraut, R. E. (2016). The relationship between Facebook use and well-being depends on communication type and tie strength. Journal of Computer-Mediated Communication, 21(4), 265–281.
- Harmon, A. (1998, August 30). Sad, lonely world discovered in cyberspace. The New York Times.
- Huang, C. (2010). Internet use and psychological well-being: A meta-analysis. Cyberpsychology, Behavior, and Social Networking, 13(3), 241–249.
- Kraut, R., Kiesler, S., Boneva, B., Cummings, J., Helgeson, V., & Crawford, A. (2002). Internet paradox revisited. Journal of Social Issues, 58(1), 49–74.
- 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(9), 1017–1031.
- Orben, A. (2020). The Sisyphean cycle of technology panics. Perspectives on Psychological Science, 15(5), 1143–1157.
- Parry, D. A., Davidson, B. I., Sewall, C. J. R., Fisher, J. T., Mieczkowski, H., & Quintana, D. S. (2021). A systematic review and meta-analysis of discrepancies between logged and self-reported digital media use. Nature Human Behaviour, 5(11), 1535–1547.
- Shapiro, J. S. (1999). Loneliness: Paradox or artifact? American Psychologist, 54(9), 782–783.