Can Frequent Social Media Use Cause Anxiety & Depression?

Written by Alec McClean and Jakey Lebwohl on October 8, 2025. Posted in Current News

Many studies have found a correlation between heavy social media use and higher levels of internalizing disorders (e.g., anxiety and depression) in adolescents, especially in girls

However, most of these studies are cross-sectional: they measure social media use and depression symptoms at the same point in time.

Whether the association observed in cross-sectional studies extends to longitudinal studies — which follow the same participants over months or years and measure both social media use and mental health — is a critical question.

Luckily, more and more data is available from such longitudinal studies, and researchers are leveraging these datasets to estimate whether earlier levels of one variable predict later changes in the other.

In this post, we provide a review of six relevant longitudinal studies through the lens of two contrasting hypotheses from researchers: social psychologist Jonathan Haidt, author of The Anxious Generation, and developmental psychologist Candice Odgers.

Our jumping off point is the 2024 debate between Haidt and Odgers.

The first hypothesis, put forth by Haidt in “The Anxious Generation” and his subsequent work, we’ll call forward predictability: earlier social media use predicts subsequent poor mental health.

Haidt proposes that the social pressures and addictive design of social media pulls some teens (especially girls) to heavy use, which puts them at higher risk for depression and anxiety by multiple pathways, including social comparison, perfectionism, and emotional contagion. Haidt, Zach Rausch, and Jean Twenge compiled empirical evidence regarding this claim in a collaborative review document, starting in 2019.

The second hypothesis, forwarded by Odgers, we will refer to as reverse predictability: earlier poor mental health predicts subsequent social media use, but not the other way around.

Odgers made this case in a high profile critique of “The Anxious Generation” in Nature:

When associations over time are found, they suggest not that social-media use predicts or causes depression, but that young people who already have mental-health problems use such platforms more often or in different ways from their healthy peers. [Emphasis added]

This argument, which directly challenges Haidt’s argument that heavy social media use predicts future mental health problems, has since been echoed by several researchers and writers.

Below we review the evidence Odgers puts forth in favor of the Reverse viewpoint, and we introduce three high-quality studies that provide evidence for the Forward hypothesis. Then we discuss the limitations of current longitudinal research and what these limitations mean for policymakers.

Before we get to the studies themselves, it’s important to understand the basics of a longitudinal study and the scope of our analysis. We recommend all readers start with the first two subsections below: “Longitudinal Studies: The Basics” and “Our Focus: Social Media Use, Internalizing Disorders, and Adolescents”.

After that, readers interested in more technical details can continue with the remaining subsections on quantifying predictability and the distinction between prediction and causation — but those are optional and may be skipped.

Longitudinal Studies: The Basics

Longitudinal studies follow the same participants over time, measuring social media use (SMU) and depression (or other mental health variables) at multiple “waves,” often spaced months or a year apart.

Researchers try to leverage this design to estimate whether earlier levels (or changes) of one variable predict later levels (or changes) in the other. In most studies there are four possible pathways of interest:

1. The association of past social media use (SMU) with future SMU
2. The association of past depression with future depression
3. The association of past SMU with future depression
4. The association of past depression with future SMU

The picture below is a simple illustration of these pathways, adapted from Figure 1 in Nagata et al. 2025. This post focuses on pathway 3 (forward) and pathway 4 (reverse) pathways.

Our Focus: Social Media Use, Internalizing Disorders, and Adolescents

Because Odgers directly challenges the core arguments of The Anxious Generation (TAG), we focus here on evaluating her claims about reverse predictability in that context. In TAG, Haidt claims that heavy social media use causes increases of internalizing disorders (e.g., anxiety, depression, self-harm) among adolescents.

He further argues that this causal relationship helps explain the relatively sudden population-wide increase in internalizing disorders that began in multiple Western nations in the early 2010s.

We therefore evaluate studies by how well they test these claims: Does the study provide evidence that social media use is linked to internalizing disorders among adolescents? Studies that include adults, or that look at other outcomes such as life satisfaction, are less useful.1

Figure 1. Between 2010 and 2019, internalizing disorders such as anxiety, depression, and anorexia became more prevalent compared to other mental illnesses.

We Focus on Forward vs. Reverse Predictability, but There are Two Other Options

As well as arguing for reverse predictability, Odgers and others sometimes argue that there’s no relationship between social media and depressive symptoms. In other words, they sometimes argue that there’s no relationship along pathways 3 and 4 in the diagram above.

A final possibility is that there are associations in both directions; we will refer to this as reciprocal predictability, or reciprocity. A possible explanation for reciprocity could be the following: teens who are prone to depression are more strongly motivated to turn to social media, and that increased time on social media makes them even more depressed.

The possibility of reciprocity is worth noting, because a reciprocal relationship would still mean social media use is associated with harm.

For simplicity, this post focuses on forward and reverse predictability.

Quantifying Predictability

There are myriad ways to define and estimate a measure of predictability or association. In these studies, the researchers typically codify the relationship through an estimate from a statistical model. The measure of predictability therefore depends on the statistical model, rather than being defined independently from it.

Rather than dive into the validity of this approach or whether the proposed statistical models are correct, we focus merely on interpreting the results as they are presented in each paper; in essence, we give the researchers the benefit of the doubt that their proposed statistical model is relevant and useful.

This may not be the case, but interrogating the models in detail is outside of the scope of this review.

We Focus on “Predictability,” Not Causality

We focus on “predictability” and associations rather than “causality” and causal effects. We follow the conventions in this literature, where researchers typically estimate whether changes in social media use at one time point are associated with, or predict, changes in depression at a later time point, and vice versa.

While the underlying debate centers on causal claims (TAG argues that heavy social media use causes depression and anxiety), the longitudinal studies we examine are primarily designed to test predictive relationships over time.

Translating these predictive relationships into causal claims requires strong assumptions. None of the studies we review assess whether these causal assumptions were violated or were satisfied, making it difficult to determine whether their results directly support forward or reverse causality hypotheses.

Instead, to simplify our discussion, and because researchers in this area focus primarily on prediction rather than causation, we also focus on predictive associations and do not evaluate the plausibility of assumptions required to make causal inferences from these longitudinal data.

Reverse predictability

In May 2024, Odgers provided a useful summary of her reverse predictability argument in The Atlantic in an essay titled “The Panic Over Smartphones Doesn’t Help Teens”:

When associations are found, things seem to work in the opposite direction from what we’ve been told: Recent research among adolescents—including among young-adolescent girls, along with a large review of 24 studies that followed people over time—suggests that early mental-health symptoms may predict later social-media use, but not the other way around. [Emphasis added]

To support this claim, Odgers points to three studies often cited as evidence for reverse predictability: Puukko et al. (2020), a six-year longitudinal study of adolescents; Heffer et al. (2019), a two-year longitudinal study; and Hancock et al. (2022), a review and meta-analysis of 226 studies, including 24 that were longitudinal.

Let’s take a close look at each of these studies to assess what they can — and cannot — tell us about the relationship between adolescent depression and social media use.2

Study 1. Puukko et al. 2020

Puukko and colleagues (2020) conducted a six-year longitudinal study of 2,891 Finnish adolescents, tracking “active social media use” and depressive symptoms.

They concluded that “depressive symptoms predicted small increases in active social media use during both early and late adolescence, whereas no evidence of the reverse relationship was found.” On the surface, this supports the reverse predictability hypothesis.

However, crucially, this study did not measure the participants’ time spent on social media. Instead, they focused on the amount of “active social media use,” which they define as “socially-oriented…use, such as sending messages, sharing updates, and liking other people’s doings….”

Here are the items they used, which were answered on seven-point Likert scales ranging from 1 (“never”) to 7 (“all the time”):

“I follow my friends’ profiles, pictures, and updates,”
“I update my status and share content with others,”
“I chat (e.g., Whatsapp, Facebook, email),”
“I share pictures and picture updates of my doings taken with my phone (e.g., Instagram).”

This changes the implications of the research’s findings. The study can give us interesting insight into the interaction between depression and a specific type of “socio-digital participation” over time, but it cannot tell us whether or not social media use generally predicts depression.

To illustrate how the particular measure used in the study could lead to an erroneous conclusion about social media, consider the following analogy. Imagine we are investigating whether heavy sugar consumption is unhealthy.

Now suppose that one study asked people only about their fruit consumption. Fruit is indeed high in sugar, but consuming whole fruits (which include fiber and many micronutrients) is among the healthiest ways to consume sugar.

Furthermore, merely eating a lot of fruit will not necessarily make a person a heavy consumer of sugar overall. If it turned out that people who eat a lot of fruit are just as healthy as those who eat little, this would not disconfirm the hypothesis that high sugar intake is unhealthy.

We’d want to find studies that included all sugar intake, including those forms that are expected to be less healthy, such as in candy, soda, and ultra-processed desserts.

Study 2. Hancock et al. 2022

Hancock and colleagues conducted a meta-analysis of studies linking social media use with six forms of well-being: eudaimonic well-being, hedonic well-being3, social well-being, depression, anxiety, and loneliness. The headline was that overall, social media use showed little association with these outcomes.

Yet the study did find significant correlations with anxiety (r = .13) and depression (r = .12). Hancock et al. also provide an analysis restricted to longitudinal studies, which is ostensibly why Odgers cited the study. But here, complications surface.

For these longitudinal studies, Hancock did not report results for depression specifically. Instead, the authors combined outcomes into two broad categories: “psychological well-being” (eudaimonic, hedonic, social) and “psychological distress” (depression, anxiety, loneliness).

Between these two, “psychological distress” is more relevant to our hypotheses. When they analyzed the psychological distress category, they found no evidence of an effect in either direction. They wrote:

“For psychological distress, neither the effect size of Use → Well-being (r = -.04, 95 percent CI = [-.14, .05], p = .28) nor the effect size of Well-being → Use was significantly different from zero (r = .00, 95 percent CI = [-.10, .11], p = .99).”

In other words, the social media use → psychological distress estimate was -0.04 (CI: [-0.14, 0.05]), while the psychological distress → social media use estimate was 0.00 (CI: [-0.10, 0.11]).

Therefore, taken at face value, this study cannot be used as evidence for either forward or reverse predictability, and instead corresponds to evidence of no relationship between social media use and mental health.

Additionally, undermining the study’s relevance to the question of social media’s effect on depression in adolescents, these studies did not focus solely on our population of interest — adolescents.

It’s worth noting, however, that in their analysis of the larger set of cross-sectional studies, they report that correlations were larger for the minority of studies that examined adolescents, compared to young adults.

Taken together, these two aspects make it clear that these results are not relevant to the specific hypotheses about the effect of social media use on internalizing disorders in adolescents.

This is taken from a long document, read the rest here afterbabel.com

Header image: Blogspot.com

Editor’s note: your PSI editors suggest one of, if not the most important, reasons for such increases in anxiety and depression in younger people is the incessant fearmongering and lies about the climate.

Please Donate Below To Support Our Ongoing Work To Defend The Scientific Method

PRINCIPIA SCIENTIFIC INTERNATIONAL, legally registered in the UK as a company incorporated for charitable purposes. Head Office: 27 Old Gloucester Street, London WC1N 3AX.