Ethan KrosspersName, Philippe VerduynpersName, Emre DemiralppersName, Jiyoung ParkpersName, David Seungjae LeepersName, Natalie LinpersName, Holly ShablackpersName, John JonidespersName, Oscar YbarrapersName (2013date). “Facebook Use Predicts Declines in Subjective Well-Being in Young Adults.” PLoS ONEtitle·j.
Object of study — the temporal association between Facebook useterm and affective and cognitive well-beingterm in young adults
Research question — whether naturally occurring Facebook useterm predicts later changes in momentary affectterm·var and life satisfactionterm·var
Contribution — a smartphone experience-sampling study using lagged models to distinguish temporal predictionterm from reverse ordering and direct social interactionterm
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | declining subjective well-being following more frequent Facebook use — “Over 500 million people interact daily with Facebook. Yet, whether Facebook use influences subjective well-being over time is unknown. We addressed this issue using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience. We text-messaged people five times per day for two-weeks to examine how Facebook use influences the two components of subjective well-being: how people feel moment-to-moment and how satisfied they are with their lives. Our results indicate that Facebook use predicts negative shifts on both of these variables over time. The more people used Facebook at one time point, the worse they felt the next time we text-messaged them; the more they used Facebook over two-weeks, the more their life satisfaction levels declined over time. Interacting with other people “directly” did not predict these negative outcomes. They were also not moderated by the size of people's Facebook networks, their perceived supportiveness, motivation for using Facebook, gender, loneliness, self-esteem, or depression. On the surface, Facebook provides an invaluable resource for fulfilling the basic human need for social connection. Rather than enhancing well-being, however, these findings suggest that Facebook may undermine it.” | |
| claim-type | prediction — “Within a relatively short timespan, Facebook has revolutionized the way people interact. Yet, whether using Facebook predicts changes in subjective well-being over time is unknown. We addressed this issue by performing lagged analyses on experience sampled data, an approach that allowed us to take advantage of the relative timing of participants' naturally occurring behaviors and psychological states to draw inferences about their likely causal sequence [17] , [18] . These analyses indicated that Facebook use predicts declines in the two components of subjective well-being: how people feel moment to moment and how satisfied they are with their lives.” | |
| method — evidence | — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | |
| object of study — evidence | — “Over 500 million people interact daily with Facebook. Yet, whether Facebook use influences subjective well-being over time is unknown. We addressed this issue using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience. We text-messaged people five times per day for two-weeks to examine how Facebook use influences the two components of subjective well-being: how people feel moment-to-moment and how satisfied they are with their lives. Our results indicate that Facebook use predicts negative shifts on both of these variables over time. The more people used Facebook at one time point, the worse they felt the next time we text-messaged them; the more they used Facebook over two-weeks, the more their life satisfaction levels declined over time. Interacting with other people “directly” did not predict these negative outcomes. They were also not moderated by the size of people's Facebook networks, their perceived supportiveness, motivation for using Facebook, gender, loneliness, self-esteem, or depression. On the surface, Facebook provides an invaluable resource for fulfilling the basic human need for social connection. Rather than enhancing well-being, however, these findings suggest that Facebook may undermine it.” | |
| research question — evidence | — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | |
| contribution — evidence | — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | |
| sense | prediction as temporal statistical forecasting rather than a cognitive expectation or neural prediction-error signal — “We examined the relationship between Facebook use and affect using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affect (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook use (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affect at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook use at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook use during the time period separating T 1 and T 2 predicted changes in affect over this time span.” | senses[0] |
| sense | affect as a self-reported good-to-bad momentary state used to operationalize affective well-being — “Participants always answered the affect question first. Next the worry and loneliness questions were presented in random order. The Facebook use and direct social interaction questions were always administered last, again in random order. Our analyses focused primarily on affect (rather than worry and loneliness) because this affect question is the way “affective well-being” is typically operationalized.” | senses[1] |
| sense | control as adjustment for prior affect and competing social or emotional explanations — “An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook use. In contrast to Facebook use, “direct” social interaction did not predict changes in cognitive well-being, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-being, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interaction did not substantively alter the significant relationship between Facebook use and affective well-being, B = .05, χ 2 = 10.78, p <.01.” | senses[2] |
| control | prior affect in the lagged model — rules out stable or preceding affect as the source of later affect differences — “We examined the relationship between Facebook use and affect using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affect (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook use (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affect at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook use at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook use during the time period separating T 1 and T 2 predicted changes in affect over this time span.” | controls[0] |
| control | reverse temporal pathway — held — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” — rules out worse affect causing increased Facebook use — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” | controls[1] |
| control | direct social interaction comparison — held — “An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook use. In contrast to Facebook use, “direct” social interaction did not predict changes in cognitive well-being, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-being, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interaction did not substantively alter the significant relationship between Facebook use and affective well-being, B = .05, χ 2 = 10.78, p <.01.” — rules out all social interaction as the source of declining well-being — “An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook use. In contrast to Facebook use, “direct” social interaction did not predict changes in cognitive well-being, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-being, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interaction did not substantively alter the significant relationship between Facebook use and affective well-being, B = .05, χ 2 = 10.78, p <.01.” | controls[2] |
| control | loneliness adjustment — held — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” — rules out Facebook use acting only as a proxy for loneliness — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | controls[3] |
| audit | Facebook use —causes→ declines in affective well-being — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span reports that Facebook use PREDICTED declines; a prediction typed as cause. — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concept_edges[1] |
| relation | Facebook use —causes→ declines in affective well-being — Earlier Facebook use statistically predicts later worse affect; causality remains provisional. — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concept_edges[1] |
| weight | Facebook use —causes→ declines in affective well-being — 0.65 — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concept_edges[1] |
| audit | Facebook use —causes→ cognitive well-being — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span reports that Facebook use PREDICTED declines in cognitive well-being; a prediction typed as cause, and the direction is unsigned. — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concept_edges[2] |
| relation | Facebook use —causes→ cognitive well-being — Average Facebook use predicts later life-satisfaction decline; causality remains provisional. — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concept_edges[2] |
| weight | Facebook use —causes→ cognitive well-being — 0.65 — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concept_edges[2] |
| outcome | control — held — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” | controls[1] |
| outcome | control — held — “An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook use. In contrast to Facebook use, “direct” social interaction did not predict changes in cognitive well-being, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-being, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interaction did not substantively alter the significant relationship between Facebook use and affective well-being, B = .05, χ 2 = 10.78, p <.01.” | controls[2] |
| outcome | control — held — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | controls[3] |
| limitation | Results from young adults may not generalize to other ages or social networks. — “Although these findings raise numerous future research questions, four stand out as most pressing. First, do these findings generalize? We concentrated on young adults in this study because they represent a core Facebook user demographic. However, examining whether these findings generalize to additional age groups is important. Future research should also examine whether these findings generalize to other online social networks. As a recent review of the Facebook literature indicated [2] “[different online social networks] have varied histories and are associated with different patterns of use, user characteristics, and social functions (p. 205).” Therefore, it is possible that the current findings may not neatly generalize to other online social networks.” | limitations[0] |
| limitation | Naturalistic lagged ordering does not establish definitive causality without an intervention. — “Finally, although the analytic approach we used in this study is useful for drawing inferences about the likely causal ordering of associations between naturally occurring variables, experiments that manipulate Facebook use in daily life are needed to corroborate these findings and establish definitive causal relations. Though potentially challenging to perform—Facebook use prevalence, its centrality to young adult daily social interactions, and addictive properties may make it a difficult intervention target—such studies are important for extending this work and informing future interventions.” | limitations[1] |
| limitation | Observed effect sizes were statistically significant but relatively small. — “Two caveats are in order before concluding. First, although we observed statistically significant associations between Facebook usage and well-being, the sizes of these effects were relatively “small.” This should not, however, undermine their practical significance [29] . Subjective well-being is a multiply determined outcome—it is unrealistic to expect any single factor to powerfully influence it. Moreover, in addition to being consequential in its own right, subjective well-being predicts an array of mental and physical health consequences. Therefore, identifying any factor that systematically influences it is important, especially when that factor is likely to accumulate over time among large numbers of people. Facebook usage would seem to fit both of these criteria.” | limitations[2] |
| limitation | A single bipolar affect scale cannot distinguish less positive from more negative affect. — “Second, some research suggests that asking people to indicate how good or bad they feel using a single bipolar scale, as we did in this study, can obscure interesting differences regarding whether a variable leads people to feel less positive, more negative or both less positive and more negative. Future research should administer two unipolar affect questions to assess positive and negative affect separately to address this issue.” | limitations[3] |
| limitation | The study did not identify the mechanism linking Facebook use to well-being decline. — “Second, what mechanisms underlie the deleterious effects of Facebook usage on well-being? Some researchers have speculated that online social networking may interfere with physical activity, which has cognitive and emotional replenishing effects [27] or trigger damaging social comparisons [8] , [28] . The latter idea is particularly interesting in light of the significant interaction we observed between direct social contact and Facebook use in this study—i.e., the more people interacted with other people directly, the more strongly Facebook use predicted declines in their affective well-being. If harmful social comparisons explain how Facebook use predicts declines in affective well-being, it is possible that interacting with other people directly either enhances the frequency of such comparisons or magnifies their emotional impact. Examining whether these or other mechanisms explain the relationship between Facebook usage and well-being is important both from a basic science and practical perspective.” | limitations[4] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | Greater Facebook use predicted worse affect later the same day after controlling earlier affect. — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” | content_tags.findings[0] |
| finding | Earlier affect did not predict subsequent Facebook use in the reverse pathway. — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” | content_tags.findings[1] |
| finding | Greater average Facebook use predicted declining life satisfaction across fourteen days. — “To examine how Facebook use influenced “cognitive well-being,” we analyzed whether people's average Facebook use over the 14-day period predicted their life satisfaction at the end of the study, controlling for baseline life satisfaction and average emotion levels over the 14-day period. The more participants used Facebook, the more their life satisfaction levels declined over time, B = −.012, β = −.124, t (73) = −2.39, p = .02, (see Figure 1 , bottom).” | content_tags.findings[2] |
| finding | Direct social interaction predicted improved affect rather than the Facebook-associated decline. — “An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook use. In contrast to Facebook use, “direct” social interaction did not predict changes in cognitive well-being, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-being, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interaction did not substantively alter the significant relationship between Facebook use and affective well-being, B = .05, χ 2 = 10.78, p <.01.” | content_tags.findings[3] |
| finding | Controlling loneliness did not remove Facebook use as a predictor of affective or cognitive decline. — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | content_tags.findings[4] |
| finding | Network size, support, depression, loneliness, gender, self-esteem, timing, and use motivation did not significantly moderate the association. — “Next, we examined whether a number of theoretically relevant individual-difference variables including participants' number of Facebook Friends, their perceptions of their Facebook network support, depressive symptoms, loneliness, gender, self-esteem, time of study participation, and motivation for using Facebook (e.g., to find new friends, to share good or bad things, to obtain new information) interacted with Facebook use to predict changes in affective or cognitive well-being ( Text S6 ). In no case did we observe any significant interactions ( p s>.16).” | content_tags.findings[5] |
| quantity | 82 — participants recruited — “Eighty-two people ( M age = 19.52, SD age = 2.17; 53 females; 60.5% European American, 28.4% Asian, 6.2% African American, and 4.9% other) were recruited for a study on Facebook through flyers posted around Ann Arbor, Michigan. Participants needed a Facebook account and a touch-screen smartphone to qualify for the study. They received $20 and were entered into a raffle to receive an iPad2 for participating.” | content_tags.quantities[0] |
| quantity | 5 times per day — smartphone experience-sampling prompts — “Participants were text-messaged 5 times per day between 10am and midnight over 14-days. Text-messages occurred at random times within 168-minute windows per day. Each text-message contained a link to an online survey, which asked participants to answer five questions using a slider scale: (1) How do you feel right now? ( very positive [0] to very negative [100]; M = 37.47, SD = 25.88); (2) How worried are you right now? ( not at all [0] to a lot [100]; M = 44.04, SD = 30.42); (3) How lonely do you feel right now? ( not at all [0] to a lot [100]; M = 27.61, SD = 26.13); (4) How much have you used Facebook since the last time we asked? ( not at all [0] to a lot [100]; M = 33.90, SD = 30.48); (5) How much have you interacted with other people “directly” since the last time we asked? ( not at all [0]to a lot [100]; M = 64.26, SD = 31.11). When the protocol for answering these questions was explained, interacting with other people “directly” was defined as face-to-face or phone interactions. An experimenter carefully walked participants through this protocol to ensure that they understood how to answer each question and fulfill the study requirements.” | content_tags.quantities[1] |
| quantity | 14 days — experience-sampling duration — “Participants were text-messaged 5 times per day between 10am and midnight over 14-days. Text-messages occurred at random times within 168-minute windows per day. Each text-message contained a link to an online survey, which asked participants to answer five questions using a slider scale: (1) How do you feel right now? ( very positive [0] to very negative [100]; M = 37.47, SD = 25.88); (2) How worried are you right now? ( not at all [0] to a lot [100]; M = 44.04, SD = 30.42); (3) How lonely do you feel right now? ( not at all [0] to a lot [100]; M = 27.61, SD = 26.13); (4) How much have you used Facebook since the last time we asked? ( not at all [0] to a lot [100]; M = 33.90, SD = 30.48); (5) How much have you interacted with other people “directly” since the last time we asked? ( not at all [0]to a lot [100]; M = 64.26, SD = 31.11). When the protocol for answering these questions was explained, interacting with other people “directly” was defined as face-to-face or phone interactions. An experimenter carefully walked participants through this protocol to ensure that they understood how to answer each question and fulfill the study requirements.” | content_tags.quantities[2] |
| quantity | B = .08 — Facebook-use coefficient predicting worse subsequent affect — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” | content_tags.quantities[3] |
| quantity | p < .0001 — significance of the lagged affect result — “We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affect predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).” | content_tags.quantities[4] |
| quantity | B = -.012 — Facebook-use coefficient predicting life-satisfaction decline — “To examine how Facebook use influenced “cognitive well-being,” we analyzed whether people's average Facebook use over the 14-day period predicted their life satisfaction at the end of the study, controlling for baseline life satisfaction and average emotion levels over the 14-day period. The more participants used Facebook, the more their life satisfaction levels declined over time, B = −.012, β = −.124, t (73) = −2.39, p = .02, (see Figure 1 , bottom).” | content_tags.quantities[5] |
| variable | Facebook use (time-varying predictor) — “We examined the relationship between Facebook use and affect using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affect (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook use (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affect at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook use at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook use during the time period separating T 1 and T 2 predicted changes in affect over this time span.” | content_tags.variables[0] |
| variable | affect (momentary affective-well-being outcome) — “Participants always answered the affect question first. Next the worry and loneliness questions were presented in random order. The Facebook use and direct social interaction questions were always administered last, again in random order. Our analyses focused primarily on affect (rather than worry and loneliness) because this affect question is the way “affective well-being” is typically operationalized.” | content_tags.variables[1] |
| variable | life satisfaction (cognitive-well-being outcome) — “To examine how Facebook use influenced “cognitive well-being,” we analyzed whether people's average Facebook use over the 14-day period predicted their life satisfaction at the end of the study, controlling for baseline life satisfaction and average emotion levels over the 14-day period. The more participants used Facebook, the more their life satisfaction levels declined over time, B = −.012, β = −.124, t (73) = −2.39, p = .02, (see Figure 1 , bottom).” | content_tags.variables[2] |
| variable | direct social interaction (alternative social-exposure control) — “An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook use. In contrast to Facebook use, “direct” social interaction did not predict changes in cognitive well-being, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-being, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interaction did not substantively alter the significant relationship between Facebook use and affective well-being, B = .05, χ 2 = 10.78, p <.01.” | content_tags.variables[3] |
| variable | loneliness (time-varying alternative explanation and covariate) — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | content_tags.variables[4] |
| definition | affective well-being — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | content_tags.definitions[0] |
| definition | cognitive well-being — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | content_tags.definitions[1] |
| definition | direct social interaction — “Participants were text-messaged 5 times per day between 10am and midnight over 14-days. Text-messages occurred at random times within 168-minute windows per day. Each text-message contained a link to an online survey, which asked participants to answer five questions using a slider scale: (1) How do you feel right now? ( very positive [0] to very negative [100]; M = 37.47, SD = 25.88); (2) How worried are you right now? ( not at all [0] to a lot [100]; M = 44.04, SD = 30.42); (3) How lonely do you feel right now? ( not at all [0] to a lot [100]; M = 27.61, SD = 26.13); (4) How much have you used Facebook since the last time we asked? ( not at all [0] to a lot [100]; M = 33.90, SD = 30.48); (5) How much have you interacted with other people “directly” since the last time we asked? ( not at all [0]to a lot [100]; M = 64.26, SD = 31.11). When the protocol for answering these questions was explained, interacting with other people “directly” was defined as face-to-face or phone interactions. An experimenter carefully walked participants through this protocol to ensure that they understood how to answer each question and fulfill the study requirements.” | content_tags.definitions[2] |
| definition | prediction — “We examined the relationship between Facebook use and affect using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affect (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook use (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affect at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook use at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook use during the time period separating T 1 and T 2 predicted changes in affect over this time span.” | content_tags.definitions[3] |
| concept | prediction — not defined in this paper's own words; lagged association controlling prior outcome — “We examined the relationship between Facebook use and affect using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affect (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook use (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affect at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook use at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook use during the time period separating T 1 and T 2 predicted changes in affect over this time span.” | concepts[0] |
| concept | affective well-being — not defined in this paper's own words; momentary felt state — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | concepts[1] |
| concept | cognitive well-being — life satisfaction — “So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .” | concepts[2] |
| concept | Facebook use — not defined in this paper's own words; self-reported use between prompts — “We examined the relationship between Facebook use and affect using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affect (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook use (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affect at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook use at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook use during the time period separating T 1 and T 2 predicted changes in affect over this time span.” | concepts[3] |
| concept | direct social interaction — not defined in this paper's own words; face-to-face or telephone contact — “Participants were text-messaged 5 times per day between 10am and midnight over 14-days. Text-messages occurred at random times within 168-minute windows per day. Each text-message contained a link to an online survey, which asked participants to answer five questions using a slider scale: (1) How do you feel right now? ( very positive [0] to very negative [100]; M = 37.47, SD = 25.88); (2) How worried are you right now? ( not at all [0] to a lot [100]; M = 44.04, SD = 30.42); (3) How lonely do you feel right now? ( not at all [0] to a lot [100]; M = 27.61, SD = 26.13); (4) How much have you used Facebook since the last time we asked? ( not at all [0] to a lot [100]; M = 33.90, SD = 30.48); (5) How much have you interacted with other people “directly” since the last time we asked? ( not at all [0]to a lot [100]; M = 64.26, SD = 31.11). When the protocol for answering these questions was explained, interacting with other people “directly” was defined as face-to-face or phone interactions. An experimenter carefully walked participants through this protocol to ensure that they understood how to answer each question and fulfill the study requirements.” | concepts[4] |
| concept | loneliness — not defined in this paper's own words; felt social isolation — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concepts[5] |
| concept | social comparison — not defined in this paper's own words; possible emotional mechanism of Facebook effects — “Second, what mechanisms underlie the deleterious effects of Facebook usage on well-being? Some researchers have speculated that online social networking may interfere with physical activity, which has cognitive and emotional replenishing effects [27] or trigger damaging social comparisons [8] , [28] . The latter idea is particularly interesting in light of the significant interaction we observed between direct social contact and Facebook use in this study—i.e., the more people interacted with other people directly, the more strongly Facebook use predicted declines in their affective well-being. If harmful social comparisons explain how Facebook use predicts declines in affective well-being, it is possible that interacting with other people directly either enhances the frequency of such comparisons or magnifies their emotional impact. Examining whether these or other mechanisms explain the relationship between Facebook usage and well-being is important both from a basic science and practical perspective.” | concepts[6] |
| concept | affect question — not defined in this paper's own words; single bipolar affect item — “Participants always answered the affect question first. Next the worry and loneliness questions were presented in random order. The Facebook use and direct social interaction questions were always administered last, again in random order. Our analyses focused primarily on affect (rather than worry and loneliness) because this affect question is the way “affective well-being” is typically operationalized.” | concepts[7] |
| concept | declines in affective well-being — not defined in this paper's own words; worsening momentary affect — “We first examined whether worry or loneliness predicted changes in Facebook use over time (i.e., T 1 worry [or T 1 loneliness] predicting T 1–2 Facebook use, controlling for T 0–1 Facebook use). Worry did not predict changes in Facebook use, B = .04, χ 2 = 2.37, p = .12, but loneliness did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for loneliness renders the relationship between Facebook use and changes in affective and cognitive well-being non-significant—what one would predict if Facebook use is a proxy for loneliness. This was not the case. Facebook use continued to predict declines in affective well-being, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-being, B = −.012, β = −.126, t (72) = 2.34, p = .02, when loneliness was controlled for in each analysis. Neither worry nor loneliness interacted significantly with Facebook use to predict changes in affective or cognitive well-being ( p s>.44).” | concepts[8] |
Full text — tagged
Facebook Useterm Predicts Declines in Subjective Well-Being in Young Adults Abstract Over 500 million people interact daily with Facebook. Yet, whether Facebook useterm influences subjective well-being over time is unknown. We addressed this issue using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience. We text-messaged people five times per day for two-weeks to examine how Facebook useterm influences the two components of subjective well-being: how people feel moment-to-moment and how satisfied they are with their lives. Our results indicate that Facebook useterm predicts negative shifts on both of these variables over time. The more people used Facebook at one time point, the worse they felt the next time we text-messaged them; the more they used Facebook over two-weeks, the more their life satisfactionterm·var levels declined over time. Interacting with other people “directly” did not predict these negative outcomes. They were also not moderated by the size of people's Facebook networks, their perceived supportiveness, motivation for using Facebook, gender, lonelinessterm, self-esteem, or depression. On the surface, Facebook provides an invaluable resource for fulfilling the basic human need for social connection. Rather than enhancing well-being, however, these findings suggest that Facebook may undermine it.
Introduction Online social networks are rapidly changing the way human beings interact. Over a billion people belong to Facebook, the world's largest online social network, and over half of them log in daily [1] . Yet, no research has examined how interacting with Facebook influences subjective well-being over time. Indeed, a recent article that examined every peer-reviewed publication and conference proceeding on Facebook between 1/2005 and 1/2012 (412 in total) did not reveal a single study that examined how using this technology influences subjective well-being over time ( Text S1 ) [2] .
Subjective well-being is one of the most highly studied variables in the behavioral sciences. Although significant in its own right, it also predicts a range of consequential benefits including enhanced health and longevity [3] – [5] . Given the frequency of Facebook usage, identifying how interacting with this technology influences subjective well-being represents a basic research challenge that has important practical implications.
This issue is particularly vexing because prior research provides mixed clues about how Facebook useterm should influence subjective well-being. Whereas some cross-sectional research reveals positive associations between online social network use (in particular Facebook) and well-being [6] , other work reveals the opposite [7] , [8] . Still other work suggests that the relationship between Facebook useterm and well-being may be more nuanced and potentially influenced by multiple factors including number of Facebook friends, perceived supportiveness of one's online network, depressive symptomatology, lonelinessterm, and self-esteem [9] , [10] , [11] .
So, how does Facebook usage influence subjective well-being over time? The cross-sectional approach used in previous studies makes it impossible to know. We addressed this issue by using experience-sampling, the most reliable method for measuring in-vivo behavior and psychological experience over time [12] . We text-messaged participants five times per day for 14-days. Each text-message contained a link to an online survey, which participants completed using their smartphones. We performed lagged analyses on participants' responses, as well as their answers to the Satisfaction With Life Questionnaire (SWLS) [13] , which they completed before and immediately following the 14-day experience-sampling period, to examine how interacting with Facebook influences the two components of subjective well-being: how people feel (“affective” well-being) and how satisfied they are with their lives (“cognitive” well-being) [14] , [15] . This approach allowed us to take advantage of the relative timing of participants' natural Facebook behavior and psychological states to draw inferences about their likely causal sequence [16] – [19] .
Methods Participants Eighty-two people ( M age = 19.52, SD age = 2.17; 53 females; 60.5% European American, 28.4% Asian, 6.2% African American, and 4.9% other) were recruited for a study on Facebook through flyers posted around Ann Arbor, Michigan. Participants needed a Facebook account and a touch-screen smartphone to qualify for the study. They received $20 and were entered into a raffle to receive an iPad2 for participating.
Ethics Statement The University of Michigan Institutional Review Board approved this study. Informed written consent was obtained from all participants prior to participation.
Materials and Procedure Phase 1 Participants completed a set of questionnaires, which included the SWLS ( M = 4.96, SD = 1.17), Beck Depression Inventory [20] ( M = 9.02, SD = 7.20), the Rosenberg Self-Esteem Scale [21] ( M = 30.40, SD = 4.96), and the Social Provision Scale [22] ( M = 3.55, SD = .34), which we modified to assess perceptions of Facebook support. We also assessed participants' motivation for using Facebook by asking them to indicate whether they use Facebook “to keep in touch with friends (98% answered yes),” “to find new friends (23% answered yes),” “to share good things with friends (78% answered yes),” “to share bad things with friends (36% answered yes),” “to obtain new information (62% answered yes),” or “other: please explain (17% answered yes).” Examples of other reasons included chatting with others, keeping in touch with family, and facilitating schoolwork and business.
Phase 2 Participants were text-messaged 5 times per day between 10am and midnight over 14-days. Text-messages occurred at random times within 168-minute windows per day. Each text-message contained a link to an online survey, which asked participants to answer five questions using a slider scale: (1) How do you feel right now? ( very positive [0] to very negative [100]; M = 37.47, SD = 25.88); (2) How worried are you right now? ( not at all [0] to a lot [100]; M = 44.04, SD = 30.42); (3) How lonely do you feel right now? ( not at all [0] to a lot [100]; M = 27.61, SD = 26.13); (4) How much have you used Facebook since the last time we asked? ( not at all [0] to a lot [100]; M = 33.90, SD = 30.48); (5) How much have you interacted with other people “directly” since the last time we asked? ( not at all [0]to a lot [100]; M = 64.26, SD = 31.11). When the protocol for answering these questions was explained, interacting with other people “directly” was defined as face-to-face or phone interactions. An experimenter carefully walked participants through this protocol to ensure that they understood how to answer each question and fulfill the study requirements.
Participants always answered the affect questionterm first. Next the worry and lonelinessterm questions were presented in random order. The Facebook useterm and direct social interactionterm questions were always administered last, again in random order. Our analyses focused primarily on affectterm·var (rather than worry and lonelinessterm) because this affect questionterm is the way “affective well-beingterm” is typically operationalized.
Phase 3 Participants returned to the laboratory following Phase 2 to complete another set of questionnaires, which included the SWLS ( M = 5.13, SD = 1.26) and the Revised UCLA Lonelinessterm Scale [23] ( M = 1.69, SD = .46). Participants' number of Facebook friends ( M = 664.25, SD = 383.64) was also recorded during this session from participants' Facebook accounts ( Text S2 ).
Results Attrition and compliance Three participants did not complete the study. As the methods section notes, participants received a text message directing them to complete a block of five questions once every 168 minutes on average (the text message was delivered randomly within this 168-minute window). A response to any question within a block was considered “compliant” if it was answered before participants received a subsequent text-message directing them to complete the next block of questions. Participants responded to an average of 83.6% of text-messages (range: 18.6%–100%). Following prior research [24] , we pruned the data by excluding all of the data from two participants who responded to <33% of the texts, resulting in 4,589 total observations. The results did not change substantively when additional cutoff rates were used.
Analyses overview We examined the relationship between Facebook useterm and affectterm·var using multilevel analyses to account for the nested data structure. Specifically, we examined whether T 2 affectterm·var (i.e., How do you feel right now? ) was predicted by T 1–2 Facebook useterm (i.e., How much have you used Facebook since the last time we asked? ), controlling for T 1 affectterm·var at level-1 of the model (between-day lags were excluded). Note that although this analysis assesses Facebook useterm at T 2 , the question refers to usage between T 1 and T 2 (hence the notationT 1–2 ). This analysis allowed us to explore whether Facebook useterm during the time period separating T 1 and T 2 predicted changes in affectterm·var over this time span.
When non-compliant cases were observed, we used participants' responses to the last text message they answered to examine the lagged effect of Facebook useterm on well-being to maximize power. So, if we were interested in examining whether T 2–3 Facebook useterm predicted T 3 Affectterm·var controlling for T 2 Affectterm·var, but did not have data on T 2 Affectterm·var, then we used T 1 Affectterm·var instead. Excluding trials in which participants did not respond to the previous texts (rather than following the aforementioned analytical scheme) did not substantively alter any of the results we report.
Significance testing of fixed effects was performed using chi-squared distributed (df = 1) Wald-tests. All level-1 predictors were group-mean centered, and intercepts and slopes were allowed to vary randomly across participants (see Table 1 for zero-order correlations). We tested for moderation by examining whether each moderator variable was related to the slope of T 1–2 Facebook useterm when predicting T 2 affectterm·var, controlling for T 1 affectterm·var.
Within-person and between-person zero-order correlations.
Data from one person who scored 4SDs above the sample mean on the BDI were excluded from the BDI moderation analyses; data from one person who scored 4SDs above the sample mean on number of Facebook friends were excluded from the moderation analyses based on Facebook friends.
The relationship between mean Facebook useterm and life satisfactionterm·var was assessed using OLS regressions because these data were not nested. Both unstandardized ( B ) and standardized ( β ) OLS regression coefficients are reported (see Text S3 ).
Facebook useterm and well-being Affective well-beingterm We examined whether people's tendency to interact with Facebook during the time period separating two text messages influenced how they felt at T 2 , controlling for how they felt at T 1 . Nested time-lag analyses indicated that the more people used Facebook the worse they subsequently felt, B = .08, χ 2 = 28.90, p <.0001, (see Figure 1 , top). The reverse pathway (T 1 Affectterm·var predicting T 1–2 Facebook useterm, controlling for T 0–1 Facebook useterm) was not significant, B = −.005, χ 2 = .05, p = .82, indicating that people do not use Facebook more or less depending on how they feel (see Text S4 , S5 ).
Facebook useterm predicts declines in affectterm·var and life satisfactionterm·var over time. Interacting with Facebook during one time period (Time 1–2 ) leads people to feel worse later on during the same day (T 2 ) controlling for how they felt initially (T 1 ); values are regression weights from multilevel analyses (Panel A). Average Facebook useterm over the course of the 14-day experience-sampling period predicts decreases in life satisfactionterm·var over time; values are standardized regression weights from OLS regression analysis (Panel B). * p <.05, ** p <.01, *** p <.001.
Cognitive well-beingterm To examine how Facebook useterm influenced “cognitive well-beingterm,” we analyzed whether people's average Facebook useterm over the 14-day period predicted their life satisfactionterm·var at the end of the study, controlling for baseline life satisfactionterm·var and average emotion levels over the 14-day period. The more participants used Facebook, the more their life satisfactionterm·var levels declined over time, B = −.012, β = −.124, t (73) = −2.39, p = .02, (see Figure 1 , bottom).
Alternative explanations An alternative explanation for these results is that any form of social interaction undermines well-being. Because we also asked people to indicate how frequently they interacted with other people “directly” since the last time we text messaged them, we were able to test this idea. Specifically, we repeated each of the aforementioned analyses substituting “direct” social interaction for Facebook useterm. In contrast to Facebook useterm, “direct” social interaction did not predict changes in cognitive well-beingterm, B = −.006, β = −.059, t (73) = 1.04, p = .30, and predicted increases (not decreases) in affective well-beingterm, B = −.15, χ 2 = 65.30, p <.0001. Controlling for direct social interactionterm did not substantively alter the significant relationship between Facebook useterm and affective well-beingterm, B = .05, χ 2 = 10.78, p <.01.
Another alternative explanation for these results is that people use Facebook when they feel bad (i.e., when they are bored lonely, worried or otherwise distressed), and feeling bad leads to declines in well-being rather than Facebook useterm per se. The analyses we reported earlier partially address this issue by demonstrating that affectterm·var does not predict changes in Facebook useterm over time and Facebook useterm continues to significantly predict declines in life satisfactionterm·var over time when controlling for affectterm·var. However, because participants also rated how lonely and worried they felt each time we text messaged them, we were able to test this proposal further.
We first examined whether worry or lonelinessterm predicted changes in Facebook useterm over time (i.e., T 1 worry [or T 1 lonelinessterm] predicting T 1–2 Facebook useterm, controlling for T 0–1 Facebook useterm). Worry did not predict changes in Facebook useterm, B = .04, χ 2 = 2.37, p = .12, but lonelinessterm did, B = .07, χ 2 = 8.54, p <.01. The more lonely people felt at one time point, the more people used Facebook over time. Given this significant relationship, we next examined whether controlling for lonelinessterm renders the relationship between Facebook useterm and changes in affective and cognitive well-beingterm non-significant—what one would predict if Facebook useterm is a proxy for lonelinessterm. This was not the case. Facebook useterm continued to predict declines in affective well-beingterm, B = .08, χ 2 = 27.87, p <.0001, and cognitive well-beingterm, B = −.012, β = −.126, t (72) = 2.34, p = .02, when lonelinessterm was controlled for in each analysis. Neither worry nor lonelinessterm interacted significantly with Facebook useterm to predict changes in affective or cognitive well-beingterm ( p s>.44).
Moderation Next, we examined whether a number of theoretically relevant individual-difference variables including participants' number of Facebook Friends, their perceptions of their Facebook network support, depressive symptoms, lonelinessterm, gender, self-esteem, time of study participation, and motivation for using Facebook (e.g., to find new friends, to share good or bad things, to obtain new information) interacted with Facebook useterm to predict changes in affective or cognitive well-beingterm ( Text S6 ). In no case did we observe any significant interactions ( p s>.16).
Exploratory analyses Although we did not have a priori predictionsterm about whether Facebook useterm and direct social contact would interact to predict changes in affective and cognitive well-beingterm, we nevertheless explored this issue in our final set of analyses. The results of these analyses indicated that Facebook useterm and direct social contact interacted significantly to predict changes in affective well-beingterm, B = .002, χ 2 = 19.55, p <.0001, but not changes in cognitive well-beingterm, B = .000, β = .129, t (71) = .39, p = .70. To understand the meaning of the former interaction, we performed simple slope analyses. These analyses indicated that the relationship between Facebook useterm and declines in affective well-beingterm increased linearly with direct social contact. Specifically, whereas Facebook useterm did not predict significant declines in affective well-beingterm when participants experienced low levels of direct social contact (i.e., 1 standard deviation below the sample mean for direct social contact; B = .00, χ 2 = .04, p = .84), it did predict significant declines in well-being when participants experienced moderate levels of direct social contact (i.e., at the sample mean for direct social contact; B = .05, χ 2 = 11.21, p<.001) and high levels of direct social contact (i.e., 1 standard deviation above the sample mean for direct social contact; B = .10, χ 2 = 28.82, p<.0001).
Discussion Within a relatively short timespan, Facebook has revolutionized the way people interact. Yet, whether using Facebook predicts changes in subjective well-being over time is unknown. We addressed this issue by performing lagged analyses on experience sampled data, an approach that allowed us to take advantage of the relative timing of participants' naturally occurring behaviors and psychological states to draw inferences about their likely causal sequence [17] , [18] . These analyses indicated that Facebook useterm predicts declines in the two components of subjective well-being: how people feel moment to moment and how satisfied they are with their lives.
Critically, we found no evidence to support two plausible alternative interpretations of these results. First, interacting with other people “directly” did not predict declines in well-being. In fact, direct social network interactions led people to feel better over time. This suggests that Facebook useterm may constitute a unique form of social network interaction that predicts impoverished well-being. Second, multiple types of evidence indicated that it was not the case that Facebook useterm led to declines in well-being because people are more likely to use Facebook when they feel bad—neither affectterm·var nor worry predicted Facebook useterm and Facebook useterm continued to predict significant declines in well-being when controlling for lonelinessterm (which did predict increases in Facebook useterm and reductions in emotional well-being).
Would engaging in any solitary activity similarly predict declines in well-being? We suspect that they would not because people often derive pleasure from engaging in some solitary activities (e.g., exercising, reading). Supporting this view, a number of recent studies indicate that people's perceptions of social isolation (i.e., how lonely they feel)—a variable that we assessed in this study, which did not influence our results—are a more powerful determinant of well-being than objective social isolation [25] . A related question concerns whether engaging in any Internet activity (e.g., email, web surfing) would likewise predict well-being declines. Here too prior research suggests that it would not. A number of studies indicate that whether interacting with the Internet predicts changes in well-being depends on how you use it (i.e., what sites you visit) and who you interact with [26] .
Future research Although these findings raise numerous future research questions, four stand out as most pressing. First, do these findings generalize? We concentrated on young adults in this study because they represent a core Facebook user demographic. However, examining whether these findings generalize to additional age groups is important. Future research should also examine whether these findings generalize to other online social networks. As a recent review of the Facebook literature indicated [2] “[different online social networks] have varied histories and are associated with different patterns of use, user characteristics, and social functions (p. 205).” Therefore, it is possible that the current findings may not neatly generalize to other online social networks.
Second, what mechanisms underlie the deleterious effects of Facebook usage on well-being? Some researchers have speculated that online social networking may interfere with physical activity, which has cognitive and emotional replenishing effects [27] or trigger damaging social comparisonsterm [8] , [28] . The latter idea is particularly interesting in light of the significant interaction we observed between direct social contact and Facebook useterm in this study—i.e., the more people interacted with other people directly, the more strongly Facebook useterm predicted declines in their affective well-beingterm. If harmful social comparisonsterm explain how Facebook useterm predicts declines in affective well-beingterm, it is possible that interacting with other people directly either enhances the frequency of such comparisons or magnifies their emotional impact. Examining whether these or other mechanisms explain the relationship between Facebook usage and well-being is important both from a basic science and practical perspective.
Finally, although the analytic approach we used in this study is useful for drawing inferences about the likely causal ordering of associations between naturally occurring variables, experiments that manipulate Facebook useterm in daily life are needed to corroborate these findings and establish definitive causal relations. Though potentially challenging to perform—Facebook useterm prevalence, its centrality to young adult daily social interactions, and addictive properties may make it a difficult intervention target—such studies are important for extending this work and informing future interventions.
Caveats Two caveats are in order before concluding. First, although we observed statistically significant associations between Facebook usage and well-being, the sizes of these effects were relatively “small.” This should not, however, undermine their practical significance [29] . Subjective well-being is a multiply determined outcome—it is unrealistic to expect any single factor to powerfully influence it. Moreover, in addition to being consequential in its own right, subjective well-being predicts an array of mental and physical health consequences. Therefore, identifying any factor that systematically influences it is important, especially when that factor is likely to accumulate over time among large numbers of people. Facebook usage would seem to fit both of these criteria.
Second, some research suggests that asking people to indicate how good or bad they feel using a single bipolar scale, as we did in this study, can obscure interesting differences regarding whether a variable leads people to feel less positive, more negative or both less positive and more negative. Future research should administer two unipolar affect questionsterm to assess positive and negative affectterm·var separately to address this issue.
Concluding Comment The human need for social connection is well established, as are the benefits that people derive from such connections [30] – [34] . On the surface, Facebook provides an invaluable resource for fulfilling such needs by allowing people to instantly connect. Rather than enhancing well-being, as frequent interactions with supportive “offline” social networks powerfully do, the current findings demonstrate that interacting with Facebook may predict the opposite result for young adults—it may undermine it.
Supporting Information (DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
Acknowledgments We thank Emily Kean for her assistance running the study and Ozlem Ayduk and Phoebe Ellsworth for their feedback.
Funding Statement The authors have no support or funding to report.
References Associated Data Supplementary Materials (DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.
(DOCX) Click here for additional data file.