Abstract

This study evaluates the relationship between social media usage and state sad and anxious mood in a sample of young adults (N = 670) in the years 2019 through 2021 using cross-sectional, experimental, and ecological momentary assessment approaches. In our cross-sectional data, objectively measured usage of Instagram and Twitter was not associated with state sad or anxious mood after adjusting for multiple comparisons. We used an experimental design to test the effect of low versus high acute social media usage on state mood and did not find an effect of time spent using social media on state sad or anxious mood. In our ecological momentary assessment sample, participants reported their time spent on social media and state sad and anxious mood six times per day over 7 days, and we found a weak, positive, bidirectional relationship between social media usage and state sad and anxious mood. In this study, evidence for a relationship between social media usage and sad and anxious mood was inconsistent, and effects were small in magnitude. Time spent on social media sites may be less important for mood than other aspects of social media usage, such as the type of content accessed.

Keywords: social media, sad mood, anxious mood, ecological momentary assessment, multimethod

Disclosures: The author(s) declare that there were no conflicts of interest with respect to the authorship or the publication of this article. The authors have no financial support nor funding information to disclose.

Data Availability: All data have been made publicly available through the Open Science Framework (https://doi.org/10.17605/OSF.IO/MGJDS) and can be accessed at https://osf.io/mgjds/?view_only=dcf1fd1686f24697a4d 05ace4f4e4a12. Data from the present study are part of a larger data set from which the methods and results have been published that can be located at E. L. Unruh-Dawes et al. (2022). This article is discussed in greater detail in the current article.

Open Science Disclosures: The data are available at https://osf.io/mgjds/?view_only=dcf1fd 1686f24697a4d05ace4f4e4a12

Open Access License: This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND). This license permits copying and redistributing the work in any medium or format for noncommercial use provided the original authors and source are credited and a link to the license is included in attribution. No derivative works are permitted under this license.

Correspondence concerning this article should be addressed to Emma Unruh-Dawes, Department of Psychology, Oklahoma State University, 116 Psychology Building, Stillwater, OK 74078, United States. Email: [email protected]

The popularity of social media (SM) has increased substantially in the past 10 years and has become a popular platform for sharing content and information with friends and family, connecting with others through personal updates and posting photos and videos, and sharing and receiving news information (Alhabash & Ma, 2017; Pittman & Reich, 2016). Young adults are some of the most frequent users of SM, with 88% of young adults in the United States using at least one SM platform. During this period of rapid increase in SM usage among young adults, the rate of depression and anxiety symptoms has also increased among this group. Rates of depressive episodes are the highest among emerging adults (aged 18–25; National Institute of Mental Health, 2022), and the incidence of major depressive episodes has increased substantially from 2009 to 2017 (Twenge et al., 2019). Further, anxiety is common among young adults, and the incidence of anxiety symptoms nearly doubled among young adults from 2008 to 2018 (Goodwin et al., 2020).

Although there are many possible explanations for the increases in depression and anxiety disorders among young adults, because the increase has occurred over the same period as increases in the use of SM among young adults, SM has been implicated as a potential contributor (e.g., Twenge et al., 2019). Supporting this trend, the use of multiple SM sites is associated with increased depression and anxiety symptoms in adults aged 19–32 (Primack et al., 2017). Further, a recent systematic review of adolescent users aged 13–18 revealed that checking SM messages or likes, SM investment, and addictive or problematic usage were all associated with depression and anxiety disorders, though there was substantial variability between studies in the strength of these effects (Keles et al., 2020). Taken together, multiple studies have linked SM usage to the increases in depression and anxiety disorders and symptoms among young adults (Bettmann et al., 2021; Keles et al., 2020; Primack et al., 2017; Twenge et al., 2019).

However, many studies remain inconclusive or demonstrate positive effects of SM on mental health (e.g., Akram & Kumar, 2017; Deters & Mehl, 2013). For example, a meta-analysis of 70 studies supported the benefits of SM on reduced depression and anxiety disorders through social connection (Seabrook et al., 2016). Other meta-analyses have found no support for SM use increasing internalizing symptoms in youth (Ferguson et al., 2024). In addition to social connectivity, communication about mental health via SM can increase feelings of support and may increase help-seeking (Halsall et al., 2019; O’Reilly, 2020). Certain longitudinal and experimental studies regarding these relationships have not found that SM use predicts greater depression symptoms (Ferguson, 2024; Heffer et al., 2019; Jensen et al., 2019; Vally & El Hichami, 2019).

The contradictory findings regarding the relationship between SM and mental health are likely at least partially due to methodological limitations, specifically: (a) an overfocus on broad constructs such as “depression” or “anxiety” as outcomes rather than more specific factors that may be more directly linked to SM usage, (b) the use of cross-sectional data at the expense of experimental and longitudinal designs, (c) the use of self-report in evaluating SM usage, and (d) either grouping all SM sites together or focusing only on Facebook, which may overlook potential differences between sites in their relationship with mental health outcomes.

Social Media and State Mood

Overall, there is a high degree of inconsistency in the literature, with some reviews and meta-analyses showing an association between SM usage and depression and anxiety symptoms and disorders and others not supporting this relationship (Heffer et al., 2019; Huang, 2017; Jensen et al., 2019; Keles et al., 2020; McCrae et al., 2017; Vahedi & Zannella 2021; Vally & El Hichami, 2019; Yoon et al., 2019). However, within the evidence that demonstrates a relationship between SM use and negative mental health, it is unclear how, exactly, SM usage may contribute to negative mental health states. Both depression and anxiety disorders are broad, heterogeneous constructs (Fried, 2017; Nelemans et al., 2014). As such, there is unlikely to be a direct relationship between SM usage and depression and anxiety disorders broadly. Instead, SM usage may lead to specific psychopathogenic processes and outcomes that then contribute to depression and anxiety. For example, disturbed sleep (G. C. Adams & McWilliams, 2015), low self-esteem (Woods & Scott 2016), social comparison (Vogel et al., 2014), negative body image (Fardouly et al., 2018), loneliness (Pittman & Reich, 2016), and fear of missing out (Roberts & David, 2020) have all been investigated as links between SM usage and negative mental health outcomes.

Negative mood states such as feeling sad or anxious are core symptoms of depressive and anxiety disorders, respectively (American Psychiatric Association, 2013), and these mood states play a key role in the development and maintenance of their respective psychopathologies (Barlow, 2000; Georgiades et al., 2006). Although not always indicative of psychopathology, negative mood states may be a particularly important factor to investigate in the relationship between SM usage and negative mental health. Negative state mood has been associated with both the use of Instagram (Brown & Tiggemann, 2016; Frison & Eggermont, 2017) and Twitter (Jeri-Yabar et al., 2019) in certain circumstances, and the present study seeks to expand on these findings.

In addition to the possibility that SM usage may lead to negative mood states, it may be the case that negative mood may result in increased SM usage. According to mood management theory, individuals hedonistically pursue external stimuli, oftentimes using different forms of media such as movies, video games, music, and SM, to improve negative mood (Nabi, 2009). For example, a meta-analysis found that the use of Facebook is often motivated by the desire to escape an unpleasant mood (Ryan et al., 2014). At the same time, Facebook usage decreases feelings of happiness in some users and has been associated with greater negative affect (Abbasi & Drouin, 2019; Papacharissi & Mendelson, 2011). These results may be reconciled in that unhappy people may pursue online interactions and experiences with the intention of improving mood, but, ultimately, those experiences result in worsening mood (Bruni & Stanca, 2008).

Instrumental models of emotion regulation (e.g., Tamir, 2009) posit that, rather than always attempting to downregulate negative emotion, individuals may be motivated to maintain negative mood states under some circumstances. This is consistent with evidence that anxious and depressed individuals select media that aligns with their current mood with the goal of maintaining that mood state (Strizhakova & Krcmar, 2007). Further, there is also evidence that adolescents reliably choose media that aligns with their mood, regardless of whether they are experiencing low mood or high mood (Dillman Carpentier et al., 2008). Thus, it may be the case that individuals in a negative mood state will be more likely to use SM either to change or maintain their mood state.

Cross-Sectional Data in Social Media and Mental Health

A common methodological limitation among studies in this area is the broad use of cross-sectional data and, consequently, a lack of experimental and longitudinal designs. A recent systematic review indicated that, of 13 studies examining the influence of SM usage on negative mental health outcomes in adolescents, 12 used cross-sectional data, and only one used a longitudinal design (Keles et al., 2020). Other reviews similarly demonstrate a preponderance of cross-sectional data (Huang, 2017; Keles et al., 2020; Seabrook et al., 2016; Twomey & O’Reilly, 2017; Vahedi & Zannella, 2021; Yin et al., 2019; Yoon et al., 2019). This reliance on cross-sectional data limits our understanding of causality between the variables and may at least be partially responsible for the contradictory results in studies examining SM usage, depression, and anxiety disorders and symptoms. The present study addresses this limitation by using an experimental manipulation of SM usage and using ecological momentary assessment (EMA) to collect longitudinal data.

Self-Report of Social Media Usage

The research examining SM and mental health almost exclusively uses self-report to collect information about SM usage rates (Keles et al., 2020; Krause et al., 2021; Prizant-Passal et al., 2016; Vahedi & Zannella, 2021; Yin et al., 2019; Yoon et al., 2019). In fact, we found only two studies that used objective measures of SM usage in the context of mental health (Hunt et al., 2018; Sewall et al., 2022). Self-reported behaviors are subject to social-desirability bias (Fisher & Katz, 2000) and a variety of memory biases (Shiffman et al., 2008). For example, retrospective reports of smoking behavior consistently failed to match data collected in real time (Shiffman et al., 1997). Similarly, even daily diary self-reported caloric intake often underestimates objectively measured energy expenditure (Stea et al., 2014). To address these limitations in the context of SM usage and mental health, the present study collected baseline SM usage data directly from participants’ mobile phones.

Combination of Social Media Sites and Primary Focus on Facebook

Much SM research has focused on the effects of SM usage broadly or focusing on the effects of Facebook usage specifically. For example, in a recent systematic review of SM usage on depression and anxiety disorders in adolescents, 10 of 13 studies combined all SM sites together rather than asking participants about individual SM sites (Keles et al., 2020). In this review, the three studies that did not combine sites examined only Facebook. In a meta-analysis examining time spent on SM sites and psychological well-being, 12 of 67 studies combined SM sites together, and 51 examined only Facebook (Huang, 2017). Other reviews also demonstrate that a large majority of studies either combined SM sites or examined only Facebook (Krause et al., 2021; Seabrook et al., 2016; Twomey & O’Reilly, 2017; Vahedi & Zannella, 2021; Yin et al., 2019; Yoon et al., 2019). Combining all SM sites assumes that they have similar effects on users and obscures potential differences between sites. Although there has been progress by examining the unique effects of Facebook, this effort should be extended to other SM sites. Considering the major differences in the types of content and interactions that users experience on different SM sites, it is very likely that there are also differences in the effects each platform may have on mental health. For example, Instagram usage is primarily motivated by social interaction, self-expression, and escapism, whereas motivations for Twitter usage typically consist of information gathering and sharing, and pursuing news information (Holton et al., 2014; Kircaburun et al., 2020; Lee et al., 2015). Although these previous studies refer to differences in motivations for the use of certain SM sites, it is likely that because of these differences in motivations, there are differences in the effects of the site and the behaviors that the user is engaging in.

The present study focuses on Instagram and Twitter (rebranded as X in July 2023) due to the high usage rates of these sites among young adults and their differences in content and interaction style. Instagram is a photo-sharing platform where users can add photos to a timeline where other users can view, like, share, and comment on the photo. Instagram receives approximately 2 billion active daily users (Statista, 2023) and is most popular among emerging adults with 71% of this age group reporting use of the application (Statista, 2021a).

Twitter allows users to compose 280-character “tweets” that contain words and have options to contain other features such as links, photos, and videos to interact with others. Tweets are posted to a chronological timeline where other users can interact using mentions, replies, and hashtags, as well as follow accounts to read others’ tweets. Twitter reported approximately 368.4 million active daily users in 2022, and although less popular than Instagram among emerging adults, Twitter is still a popular SM site for this age group, with 42% of emerging adults reporting using Twitter (Statista, 2021b, 2022).

The Present Study

To better understand the potential associations between SM usage and negative mental health, we conducted a multimethod study examining the relationships between Instagram usage, Twitter usage, and state sad and anxious mood. We designed our study to address some of the notable methodological limitations of previous work in this area, namely the nearly exclusive use of cross-sectional data, use of self-report for measuring SM usage, focus on broad and heterogeneous disorders rather than emotional processes that lead to the development of such disorders, and grouping all SM sites together. As such, we evaluated both Instagram and Twitter usage via data taken from participants’ mobile phones and used three methods of data collection: cross-sectional baseline data, experimental manipulation, and longitudinal EMA.

Based on prior work demonstrating a relationship between SM usage and negative mood states (Anixiadis et al., 2019; Brown & Tiggemann, 2016; Park, 2015), (Hypothesis 1) we expected higher baseline Instagram usage (measured via data from participants’ phones) to be associated with higher state sad and anxious mood. (Hypothesis 2) We expected the same relationship for Twitter usage, such that baseline Twitter usage (measured via data from participants’ phones) would be associated with higher state sad and anxious mood. Similarly, (Hypothesis 3) we expected participants randomized to a 30-min in-lab Instagram or Twitter usage session to demonstrate higher levels in state sad and anxious mood compared to those randomized to a 5-min session. In our EMA data, (Hypothesis 4) we expected SM usage since the prior assessment to be positively associated with state sad and anxious mood at the current assessment. (Hypothesis 5) We also expected sad and anxious mood at a given assessment to be positively associated with greater SM usage between that assessment and the subsequent assessment.

Method

Participants who were 18 years of age or older were recruited from a large Midwestern university using a psychology student participant pool. The present study and all data collection methods were approved by the institutional review board of the university. Participants who indicated on a prescreening survey that they used Instagram, Twitter, or both “once or twice per week” or more frequently were invited to participate. Participants provided informed consent to the study procedures before providing baseline cross-sectional data, participating in an experimental manipulation, and providing EMA data. Detailed methods about each of these study aspects are provided below. Participants received partial course credit for participation and had no knowledge of their SM usage data being collected as part of the study before participation in the study. Data collection took place in the years 2018 through 2021, and all study staff who interacted directly with participants were blinded to the hypotheses.

Data from the present study are part of a larger data set from which one other article has been published (see E. L. Unruh-Dawes et al., 2022). In this previously published article, relationships between Instagram and Twitter use, suicidal ideation, and interpersonal variables including perceived burdensomeness and thwarted belongingness were examined. No analyses in the current article are duplicated from the prior publication.

We powered our study to detect the baseline cross-sectional relationship between SM usage and state mood. We recruited 674 participants, of whom four did not participate, resulting in a total sample of 670. The four excluded participants terminated participation before complete data could be collected. An a priori power analysis (G*Power; Faul et al., 2009), using an effect size of d = 0.69 for the relationship between Facebook usage and negative mood (Fardouly et al., 2018), indicated that 94 participants would be needed to detect an effect of this size with a power of .8 and an α of .05. Thus, we were overpowered to detect our main effects.

Cross-Sectional Data

Materials

Social Media Usage

The overwhelming majority of emerging adults with access to the internet use their SM accounts through mobile smartphones (Villanti et al., 2017). To obtain an objective measure of SM usage, a research assistant collected “usage times and percentages” from the “Settings” application in the participants’ phones. We recorded the time spent on Instagram and Twitter for the past 7 or 10 days. Prior to iOS 12.0, iPhones recorded time spent on sites over the prior 7 days. After the release of iOS 12.0 on September 17, 2018, iPhones recorded time over the prior 10 days. All units of time collected were converted into minutes to standardize the measurements. Time spent from the past 7 or 10 days was averaged by the number of days to create a measure of “average daily usage” (in minutes per day) for each application. Participants without iPhones were excluded from the study because other smartphone operating systems did not offer the ability to extract the number of minutes spent on individual sites. This resulted in our sample consisting of only iPhone users. Our small percentage of non-iPhone users (4.1%; see the Participants section) is consistent with the broader young adult population (Statista, 2019).

Participants also completed self-report measures of Instagram usage and Twitter usage, wherein they estimated the number of times they used each application per day and how many minutes they used the application each time they used it. By multiplying these numbers, we obtained a self-report estimate of the number of minutes the participants used Twitter and Instagram per day.

State Sad and Anxious Mood

State sad and anxious mood was measured using three items per mood state from the Profile of Mood States (POMS; McNair et al., 1971) and the Positive and Negative Affect Schedule (PANAS; Watson et al., 1988). “Blue” (POMS), “Sad” (PANAS), and “Down” (PANAS) were used to assess state sad mood. “Worried” (POMS), “Anxious” (PANAS), and “Nervous” (POMS) were used to assess state anxious mood. Participants were asked to evaluate their current mood (“right now”). The internal consistency for the sad and anxious mood items was Cronbach α = .86 and .82, respectively.

Demographics

Participants reported their age, gender, and race/ethnicity.

Procedure

After providing informed consent, participants were seated at a computer in the laboratory and completed questionnaires through Qualtrics, an online survey administration platform. After completing questionnaires, a research assistant collected Instagram and Twitter usage data from the “Settings” feature of the participant’s mobile phone.

Participants

Of the 670 participants, 28 (4.1%) used mobile phones, from which we were unable to collect relevant SM usage data. We also excluded participants with fewer than 7 days of SM usage data available (n = 30, 4.4%). Finally, some participants used only Instagram or only Twitter, leaving a total of 606 and 591 participants with data available for cross-sectional analyses for Instagram and Twitter, respectively. Importantly, participants included in analyses did not differ in state sad mood, t(662) = 1.13, p = .26, or state anxious mood, t(663) = 1.10, p = .27, from those excluded from the analyses.

Participants had a mean age of 19.1 years (SD = 1.3), and the sample was 74.5% female, 25.2% male, and .2% transgender, and .2% self-reported their gender as not falling within the provided categories. Participants were primarily White (77.3%), and 6.4% of the sample were Native American or Alaskan Native, 6.2% Black or African American, 4.9% identified with multiple races, 2.3% identified as a race not listed, 2.1% Asian, 0.7% Native Hawaiian or other Pacific Islander, and 0.2% chose not to answer. Additionally, 6.7% identified as White Hispanic and 3.8% as non-White Hispanic.

In-Lab Social Media Usage

Materials

State Sad and Anxious Mood

State sad and anxious mood was measured using the same instrument as the cross-sectional portion of the study. The internal consistency for the sad and anxious mood items was Cronbach α = .84 and .88, respectively.

Manipulation Check

Participants indicated on a 0–100 slider scale the percentage of time that they were using the application consistent with the instructions (see the Procedure section).

Procedure

A randomization schedule was created prior to study initiation. After completing the questionnaires outlined above for the cross-sectional data, participants engaged in SM usage on their phones in the lab. Participants could be assigned to one of four groups: using Instagram for 30 min, Twitter for 30 min, Instagram for 5 min, or Twitter for 5 min. However, participants could only be assigned to conditions for which they currently had an application on their mobile phones. That is, participants who only had Instagram on their mobile phones could be assigned to one of the two Instagram conditions but not the Twitter conditions. The converse was true for participants with only Twitter. Participants with both applications could be assigned to any one of the four conditions. This created an imbalance among the conditions, with 103 participants assigned to use Instagram for 5 min, 126 assigned to use Instagram for 30 min, 111 assigned to use Twitter for 5 min, and 114 assigned to use Twitter for 30 min. We selected 30 min of SM usage time as our active condition as 30 min is one of the most common usage times for young adults (de Zúñiga et al., 2012; Jelenchick et al., 2013).

While using their SM in the lab, participants were told to “use the application in [their] normal manner” and not to exit the application until their allotted time was up. For the 5-min condition, participants completed a word search as a filler task during the remaining 25 min to control for the time spent in the lab, cognitive effort, and so on between conditions. Immediately following their in-lab SM usage or their SM use plus filler task, participants completed the measures of state sad and anxious mood and the manipulation check. Social media usage times were based off of previous literature that has shown that 0–30 min of SM usage per day is one of the more common usage time frames, and therefore, we chose 30 min to represent the higher end of this spectrum and 5 min as the lower end of this spectrum (Primack et al., 2017). Five minutes rather than 0 min was selected to examine short-term use rather than complete abstinence from SM.

Participants

Of the 670 participants, 216 (32.2%) indicated on the manipulation check that they spent less than 90% of their time on task during the SM usage manipulation and were excluded, leaving 454 participants for analyses.

Notably, participants included in analyses reported significantly lower baseline state anxious mood (M = 2.83, SD = 2.6), t(664) = 2.49, p = .013, Cohen’s d = .21, and state sad mood (M = 0.94, SD = 1.8), t(663) = 2.76, p = .006, d = .24, compared to those excluded from the analyses (anxious mood: M = 3.45, SD = 3.2; sad mood: M = 1.44, SD = 2.3). This potential selection bias is explored in more detail in the Results and Discussion sections.

Participants had a mean age of 19.1 years (SD = 1.4), and the sample was 72.9% female, 26.4% male, and .2% transgender, and .4% self-reported their gender as not falling within the provided categories. Participants were primarily White (81.1%), and 6.2% were Native American or Alaskan Native, 4.8% Black or African American, 4.2% identified with multiple races, 1.3% identified as a race not listed, 1.5% Asian, and 0.9% Native Hawaiian or other Pacific Islander. Additionally, 5.7% identified as White Hispanic and 3.3% as non-White Hispanic.

EMA

Materials

Social Media Usage

SM usage was evaluated with a single question that asked, “How much have you used social media since your last survey?” Participants could respond in 30-min increments from 0 to 300 min. Participants were instructed to use their best estimate and to round to the nearest 30-min increment.

State Sad and Anxious Mood

State sad and anxious mood were each measured with a single question of “How ____ are you feeling right now?” with the blank completed with “sad” and “anxious,” respectively. Participants responded on a 0–10 scale with anchors of not at all and very, respectively. Single-item questions were chosen to assess state mood in order to reduce participant burden and increase response rates.

Procedure

At the conclusion of the in-lab study procedures, participants were provided instructions regarding the EMA portion of the study. Participants chose a 12-hr window of time (e.g., 10 a.m.–10 p.m.) during which they would complete their EMA surveys each day. Participants completed six surveys per day for seven consecutive days. Each survey was sent at a random time in a 2-hr window (e.g., one survey was sent between 10 a.m. and noon, another between noon and 2 p.m.), resulting in a pseudorandom survey pattern. Participants provided researchers with their mobile phone numbers and mobile carriers (AT&T, Sprint, etc.). A web platform custom-built for one author (T.T.W.) sent each participant a text (in the pseudorandom pattern mentioned above) with an embedded link for the survey. Once the 7 days had been completed, the text messages were no longer sent, and participants received compensation via class credit.

Participants

Of the 670 participants, 63 were excluded because their mobile carriers did not accept text messages from our EMA web platform. Of the 528 participants who began the EMA portion of the study, 321 participants were excluded because they did not complete at least four EMA surveys per day. This cutoff was chosen to ensure sufficient data for analyses. This resulted in a final sample of 207 participants with valid EMA data. Participants included in analyses did not differ in state sad mood, t(667) = .31, p = .76; state anxious mood, t(668) = .70, p = .48; or mean daily Instagram usage, t(632) = .49, p = .62, from those excluded from the analyses. There was a nonsignificant difference between groups in mean daily Twitter usage, t(617) = 1.93, p = .054, d = .17, with those excluded from analyses using Twitter for fewer minutes per day (M = 17.3, SD = 21.5) compared to those included in analyses (M = 20.9, SD = 21.7).

Participants had a mean age of 19.2 years (SD = 1.3), and the sample was 72.1% female, 26.9% male, and .5% transgender, and .5% self-reported their gender as not falling within the provided categories. Participants were primarily White (82.2%), and 6.7% were Native American or Alaskan Native, 3.4% Black or African American, 4.3% identified with multiple races, 0.5% identified as a race not listed, 1.9% Asian, and 1% Native Hawaiian or other Pacific Islander. Additionally, 7.2% identified as White Hispanic and 1% as non-White Hispanic.

Analytic Plan

We used zero-order correlations to examine the cross-sectional relationships between Instagram usage, Twitter usage, state sad mood, and state anxious mood (Hypotheses 1 and 2).

To test Hypothesis 3, we used a 2 (social media platform: Instagram, Twitter) × 2 (social media session length: 30 min, 5 min) × 2 (time: pre, post) × 2 (mood: sad, anxious) mixed-model analysis of variance (ANOVA). The critical interaction for this hypothesis is the Session Length by × Time interaction, though we examined all main effects and interactions.

Given the nested nature of the EMA data (multiple assessments of SM usage and state mood within each participant), we used multilevel modeling to evaluate the relationship between SM usage and state mood. For Hypothesis 4, we examined the fixed effect of SM usage since the previous survey on current state sad mood and current state anxious mood. For Hypothesis 5, we examined fixed effects of state sad and anxious mood at time t on SM usage between time t and t + 1.

Transparency and Openness

All data have been made publicly available through the Open Science Framework and can be accessed at https://osf.io/mgjds/?view_only=dcf1fd1686f24697a4d05ace4f4e4a12. This study’s design and analyses were not preregistered.

Results

Cross-Sectional Results

Objective Instagram usage was not significantly associated with state sad mood (p = .576) or state anxious mood (p = .277). Objective Twitter usage was not significantly associated with state sad mood (p = .156) but was significantly negatively associated with state anxious mood (p = .014). If adjusted for multiple comparisons and an α level of .0125 was used, the correlation would no longer be significant.

Exploratory Cross-Sectional Results

Objective measures of Instagram and Twitter usage explained approximately 5%–10% of the variance (r² = .049 and .099, respectively) in their respective self-report measures. See Table 1 for correlations between these variables.

In-Lab Social Media Usage Results

The 2 (social media platform: Instagram, Twitter) × 2 (social media session length: 30 min, 5 min) × 2 (time: pre, post) × 2 (mood: sad, anxious) mixed-model ANOVA did not reveal a statistically significant interaction in the critical Session Length × Time interaction, F(1, 450) = 1.04, p = .309, η_p² = .002. There was a main effect of time, F(1, 450) = 117.2, p < .001, η_p² = .207, with participants demonstrating higher state sad and anxious mood premanipulation (M = 1.89, SD = 1.96) compared to postmanipulation (M = 1.14, SD = 1.73). There was also a main effect of mood type, F(1, 450) = 271.3, p < .001, η_p² = .376, with participants reporting higher anxious mood (M = 2.20, SD = 2.19) compared to sad mood (M = 0.82, SD = 1.58). There was a statistically significant interaction between time and mood type, F(1, 450) = 117.1, p < .001, η_p² = .207, that was driven by a larger decrease in anxious mood from pre- to postmanipulation (M = 1.24, SD = 2.11) compared to the decrease in sad mood (M = 0.25, SD = 1.29). There were no other significant main effects or interactions. Specifically, the interaction between time and condition on sad mood, F(1, 450) = .508, p = .677, η_p² = .003, and on anxious mood, F(1, 450) = .204, p = .894, η_p² = .001, was not statistically significant. The interaction between time and condition was not significant, F(1, 450) = .275, p = .844, η_p² = .002.

Exploratory In-Lab Social Media Use Analyses

As noted in the Method section, participants excluded due to low time spent on task during the social media manipulation reported higher premanipulation state sad and anxious mood compared to participants included in the above analyses. This indicates the potential for selection bias (Winship & Mare, 1992), and as such, we conducted additional analyses to explore how this potential selection bias may have affected our results. First, in the entire sample of 670 participants, there are small but statistically significant negative associations between time spent on task and premanipulation anxious mood (r = −.083, p = .033) and sad mood (r = −.116, p = .003). When time spent on task is added as a covariate in the 2 (social media platform) × 2 (social media session length) × 2 (time) × 2 (mood) mixed-model ANOVA using all 670 participants, the pattern of results is identical to the original results, though the effect sizes for the significant effects are diminished. If inclusion versus exclusion based on manipulation check data is included as a between-subjects variable in the 2 (social media platform) × 2 (social media session length) × 2 (time) × 2 (mood) mixed-model ANOVA using all 670 participants, again the pattern of results is the same with minor changes in the sizes of the statistically significant effects. As expected, there is a main effect of manipulation check group, F(1, 656) = 90.9, p = .007, η² = .011, but there are no statistically significant interactions with manipulation check group (all ps > .09).

EMA Results

Our EMA data contained 7,629 observations across 207 participants. For the effects of SM usage on sad mood, the intraclass correlation indicated that 55.5% of the variance in sad mood was due to between-person variation, leaving 44.5% of the variance explained by within-person variation. The fixed effect of SM usage on sad mood was statistically significant, β = .11 (95% CI [.07, .15]), p < .001, R_marginal² = .011, indicating a positive relationship between the two such that higher SM usage at time t was associated with higher sad mood at time t + 1.

The intraclass correlation for anxious mood indicated that 59.3% of the variance was due to between-person variation, leaving 40.7% explained by within-person variation. The fixed effect of SM usage on anxious mood was statistically significant, β = .11 (95% CI [.07, .16]), p < .001, R_marginal² = .008, indicating a positive relationship between the two such that higher SM usage at time t was associated with higher anxious mood at time t + 1.

The intraclass correlation for SM indicated that 49.8% of the variance was due to between-person variation, leaving 50.2% explained by within-person variation. The fixed effect of sad mood on SM usage was statistically significant, β = .012 (95% CI [.007, .017]), p < .001, indicating a positive relationship between the two such that higher sad mood at time t was associated with higher SM usage at time t + 1, but the effect of anxious mood was not statistically significant, β = .002 (95% CI [−.0002, .005]), p = .07. The marginal R² for the fixed effects of sad and anxious mood on SM usage was .01. When correcting for multiple comparisons, our results would remain significant. Importantly, effect sizes below .10 should be interpreted with caution as these sizes may not be able to be differentiated from statistical noise (Ferguson & Heene, 2021).

For descriptive purposes, intraindividual correlations for each of these analyses are shown in Figure 1. This shows the distribution of the correlation between SM usage and state mood within each participant (Table 2).

Discussion

We took a multimethod approach to examine the relationship between social media (SM) and mood in young adults. Of our five hypotheses, only two (those regarding our EMA data) were supported. Beginning with the results from our cross-sectional data, we did not find significant positive cross-sectional associations between objectively measured Instagram or Twitter usage and state sad mood or state anxious mood (Hypotheses 1 and 2). These cross-sectional results are consistent with the one study we are aware of that used an objective measure of SM usage (Sewall et al., 2022) but are inconsistent with the broader literature that has revealed small, positive associations between SM use and negative mental health outcomes (Keles et al., 2020; Meier & Reinecke, 2021; Primack et al., 2017; Twenge et al., 2019). Our cross-sectional results are in line with previous research that has also found no associations between SM and anxiety and depression symptoms and disorders (Heffer et al., 2019; Jensen et al., 2019; Vally & El Hichami, 2019). Although the present study did not examine anxiety and depression disorders or symptoms, sad and anxious moods are often relevant variables in the development and maintenance of anxiety and depression disorders, and therefore, comparing studies that do examine anxiety and depression highlights important differences that might be characteristic of such mental health factors.

One potential explanation for the contrast between our cross-sectional results and those of most prior studies is our use of more objective measures of Instagram and Twitter usage. As noted in our introduction, most studies collect rates of SM usage using self-report (Keles et al., 2020; Krause et al., 2021; Prizant-Passal et al., 2016; Vahedi & Zannella, 2021; Yin et al., 2019; Yoon et al., 2019). We found only modest correlations between objective measures of Instagram and Twitter usage and their self-reported counterparts, which indicates that self-reported usage is inaccurate and potentially susceptible to a range of biases. This inaccuracy has the potential to substantially alter research results. Conducting additional analyses regarding self-reported social media use and comparing to objectively measured social media use is important considering the emerging body of literature showing that self-reported and objectively measured social media use differ due to bias and error with self-reporting social media use (Parry et al., 2021).

Though we found nonsignificant results in our cross-sectional data examining differences between sites, it is worthwhile to note that we did find different patterns of results for Twitter and Instagram. As noted in our introduction, most prior work has examined SM broadly without differentiating between sites or has focused solely on Facebook. Our cross-sectional results indicate that different sites may have differing relationships with mental health-related variables and, thus, should be examined separately when possible.

To our knowledge, the present study is the first to investigate the effects of an SM usage manipulation on state sad mood and state anxious mood. In contrast to our Hypothesis 3, we did not find an increase in state sad or anxious moods following acute 30-min usage of Instagram or Twitter compared to a 5-min usage control group. Instead, there was a main effect of time on mood where, regardless of condition, participants demonstrated a decrease in sad and anxious mood from pre- to postmanipulation. Potential interpretations of these experimental results include differences between the immediate effects of SM usage on mood versus the long-term effects and methodological factors that may have contributed to our results. For example, it is possible that Instagram and Twitter usage have positive effects on state mood immediately following their use due to potential distraction and therefore have positive effects for mood but have potentially deleterious effects in the long term. It may also be the case that 30 min is not long enough to provoke a mood response compared to 5 min of usage. Future work should consider extending SM usage manipulations beyond the common usage times to determine if the decrease in negative mood changes after a certain point in time.

Our experimental results also demonstrated potential selection bias in that a higher premanipulation state sad and anxious mood was associated with less time on task during the SM manipulation. Follow-up analyses indicated that this likely did not strongly affect the experimental results. However, the negative (though weak) relationship between premanipulation state mood and time spent on SM during the task raises intriguing questions. Though all participants, on average, demonstrated decreases in state sad and anxious mood over the course of the task, it could be that participants with higher state sad and anxious mood expected time on SM to worsen their mood and thus avoided it. Though speculative, this would be consistent with mood management theory (Nabi, 2009), wherein the individuals with a higher negative mood engaged in other tasks on their cell phone that they expected would improve their negative mood.

Our EMA results supported Hypotheses 4 and 5. Specifically, we found that greater social media (SM) usage positively predicts subsequent sad and anxious mood and that sad mood predicts greater SM usage. It should be noted that these were small effects, predicting approximately 1% of the within-person variability in sad mood, anxious mood, and SM usage. Only our EMA results support prior work suggesting a positive relationship between SM usage and negative mood (Brown & Tiggemann, 2016; Frison & Eggermont, 2017; Jeri-Yabar et al., 2019).

The EMA results are also consistent with prior work that has found a bidirectional relationship between SM usage and mood (e.g., McCrae et al., 2017). Considering the effect of state sad mood on SM usage, participants with higher state sad mood may have been pursuing experiences or interactions with the goal of improving mood as outlined by mood management theory (Nabi, 2009) or, consistent with instrumental models of emotional regulation (Tamir, 2009), they may have been motivated to maintain their negative mood states rather than attempting to downregulate them. However, given that state mood predicted a very low proportion of variability in subsequent SM usage, it may be that habits and external factors drive usage patterns more than mood (cf. Mazar & Wood, 2022).

The small effect in our EMA results may also be due to issues with self-report. Although EMA data collection has been shown to reduce the error incurred by longer term retrospective self-report by using repeated measurements, there remains a risk of self-report error (Shiffman et al., 1997). We note that our results that indicated a relationship between mood and SM usage (cross-sectional and EMA), all relied on self-report of SM usage, which may indicate that this effect may be explained by mood-related memory or response biases.

It is also important to note that effect sizes below .10 may be indistinguishable from statistical noise, and it may not be possible to determine whether such effects represent true relationships or statistical “crud” (Ferguson & Heene, 2021). While we have put our EMA results in the context of the broader literature above, we acknowledge that these results should be interpreted with extreme caution unless or until they are replicated.

Limitations and Constraints on Generality

The results of the present study should be interpreted considering the methodological limitations. Age cohorts differ in SM usage rates and the age at which they were first exposed to SM. The present study used a young adult, college student sample, and as such, results may not generalize to other age groups. We did not control for family environment or academic stress, and therefore, some of our effects could be inflated due to this. It should also be noted that we included only iPhone users in our study. However, this excluded only around 4% of eligible participants. Furthermore, although there are stereotypes related to Android users versus iPhone users, such as iPhone users being more sociable or extroverted than Android users, those stereotypes are largely untrue (Shaw et al., 2016). There are documented differences between Android and iPhone users as it relates to security awareness and only minute differences in personality (Götz et al., 2017; Reinfelder et al., 2014). Despite this, our exclusion of non-iPhone-using participants was unlikely to systematically bias our results. Regarding our SM usage manipulation, as stated previously, the 30-min use group and the 5-min use group were chosen based on previous literature; however, due to these usage differences, complete abstinence of SM use was not measured. To reduce participant burden, we did not inquire separately about Instagram and Twitter usage in the EMA portion of the study. This limits our ability to draw conclusions about what specific sites were related to state mood in the EMA data. It may also be the case that purely collecting information about “last week’s SM use” and not collecting information about the activities that occurred during the in-lab SM use limits our knowledge about broader SM use patterns of participants. Additionally, with our EMA data, sad and anxious moods were measured using a single item, which might limit the full spectrum of such mood states. There were many participants who did not answer enough of the EMA events to be considered for analysis. Unfortunately, this issue is common within EMA research (Colombo et al., 2018). Self-reported SM usage was not strongly associated with actual SM usage, therefore weakening the conclusions that can be drawn from the EMA analyses. Effect sizes below .10 should be interpreted with caution as these size effects cannot be differentiated from statistical noise. Importantly, the SARS-CoV-19 (COVID-19) global pandemic represented a change in SM usage habits among SM users, and therefore, rates of SM usage prior to and following the global pandemic should be interpreted with caution (Brailovskaia & Margraf, 2021).

Strengths

Despite the limitations, the present study has notable strengths. We explored the relationship between mood and SM usage using a rigorous multimethod approach utilizing cross-sectional, experimental, and longitudinal EMA data. In our cross-sectional data, we collected objective measures of Instagram and Twitter usage directly from participants’ cell phones rather than relying on self-report. In the cross-sectional and experimental data, we investigated hypothesized relationships within specific SM sites (Instagram and Twitter) rather than asking about “social media” generally, which allowed us to gain a clearer understanding of how these individual sites affect the user based on their individual affordances.

Conclusions

In contrast with prior work (Bettmann et al., 2021; Keles et al., 2020; Primack et al., 2017; Twenge et al., 2019), we found only a weak positive relationship between SM usage and negative mood. Furthermore, this weak effect was confined to self-report measures of SM usage; objectively measured SM usage and in-lab acute usage were not associated with negative mood states. As such, we found little support for a major role of SM usage broadly on negative mood states. However, it may still be the case that negative interactions on SM (e.g., bullying) and interaction with specific types of content result in negative mood. Future work would do well to move away from measuring raw time spent on SM and focus more on the relationship between the interactions that users are experiencing while using SM, the content that they are exposed to while using SM, and specific mood states while using SM.

Supplementary Materials

https://doi.org/10.1037/tmb0000142.supp

Copyright © The Author(s) 2024

Received November 3, 2023
Revision received August 16, 2024
Accepted August 16, 2024