We need better measures to understand the influence of social media on substance use

Abstract

Social media use is extremely common, with an estimated 63.9% of the global population using social media in 2024 [1]. Adolescents and young adults (aged 16–24 years) stand out compared to other age groups, and spend approximately 180 minutes per day on social media [2]. Together with the reported impact on wellbeing, policymakers globally are considering drastic changes to limit youth access to social media platforms [3]. For instance, Australia recently passed a law to ban people under 16 years of age from using social media and France, the United Kingdom and Norway are also considering implementing age restrictions [3].

To justify these bans, policymakers have cited evidence of the link between social media use and negative health outcomes among young people, including the impact on substance use [4]. Indeed, a recent systematic review and meta-analysis found a positive link between social media use and alcohol, drugs, tobacco and vaping [5]. However, there is an ongoing debate about the quality of the evidence assessing the impact of social media use, with concerns about: small effects, a lack of theoretical models, the design (often cross-sectional), failing to control for confounders or consideration of the positive effects of social media use [6-8].

One of the most critical methodological limitations is with how social media use or exposure to certain content on social media has been measured [7, 8]. Recent reviews have found that most studies used self-reported social media use (e.g. how much time/how often do you use social media) and exposure to substance use (e.g. how often do you see alcohol on social media) [4, 5]. For example, in a recent meta-analysis, only six of 94 studies used an objective assessment of social media use [7]. Unfortunately, self-reports for digital media use in general and exposure to substance-related content have seldom shown comparable accuracy to objective measures [8]. In this editorial, we briefly overview the discrepancy between self-report and objective measures and identify alternative options for measuring social media use or exposure to substance-related content.

Parry et al. [8] meta-analysed 45 studies that measured both self-report and logged digital media use (including 4 that measured social media use) and found only a modest correlation between the two measures. They concluded that ‘when asked to estimate their usage, participants are rarely accurate.’ More recent studies have found similar discrepancies when comparing objective and self-reported social media use [9-11]. There are a number of reasons why self-report may be inaccurate, like recall bias (social media is used so frequently that it is unmemorable) or limitations with the measures used (e.g. asking participants to report their social media use averaged over longer time periods, assessing use of all social media platforms with one question, limited response options, lack of using existing validated questions) [7].

For researchers aiming to measure social media use, there are several solutions to obtain more accurate or objective estimates. Most smartphones monitor screen and app use, and researchers can ask participants to use these logs to inform their self-report estimates or to upload screenshots of app use in a survey [12]. Additionally, researchers have developed apps (e.g. eMoodie, Minuku) that can be installed to not only monitor time and app use (when used, how long accessed), but can also be integrated with other data streams such as surveys [13, 14]. For example, the eMoodie app not only passively collected time spent and frequency of social media app usage, but also sent twice daily surveys to assess psychological constructs such as affect and mental wellbeing. Although these approaches (and those outlined below) are important, one key limitation is that they require more effort from participants, which may impact who is willing to take part in research and potentially introduce participant selection bias.

There is limited evidence regarding the accuracy of self-reported exposure to substance-related content on social media, and further research is needed in this area. In our work with other digital media, we found that the actual occurrences of exposure to alcohol was 5× higher when participants estimated alcohol content in their favourite movies and nearly 2× higher when they estimated alcohol content immediately after having seen a 10-minute video [15, 16]. Although there is limited evidence from self-reports of exposure on social media, one study found that participants were even inaccurate when recalling how often they themselves had posted about alcohol, finding that there was only a medium correlation between self-reports and objective measures [17].

Collecting objective exposure to substance-related content is difficult, but recently researchers have established some exciting and innovative solutions. Angus et al. [18] developed a browser plugin and a mobile application that can automatically identify and collect the ads people are exposed to on-line. Similarly, the Stanford Screenomics project developed an app with open-source code that can take intermittent screenshots while people are using their smartphone [19]. Collecting images is particularly exciting because it allows researchers to not only consider how often people are exposed to substance-related content, but to also understand the nature of that content (e.g. positive/social depictions) or the source of the content (e.g. ads, influencer posts, posts from friends). Considering the nature and source can help determine what type of exposure is related to more substance use or what type of exposure may be preventative.

Although collecting and analysing images is exciting, there are ethical implications to collecting this content and researchers should be guided by ethics boards and current best practices (see Angus et al. and Carrière et al., for best practices) [18, 20]. However, future applications could ensure that data never leaves a participant's device. For example, thanks to improvements in artificial intelligence (AI), future computational applications could use approaches like browser plugins to passively count exposure to certain content [21].

There are also relatively simple solutions to improve reporting that may be more accessible for researchers and do not involve additional computational resources. For example, Rutherford et al. [22] asked participants to ‘scroll through’ social media for 30 minutes and take screenshots of alcohol-related ads. To avoid collecting images, this approach could be adapted so participants are prompted to scroll through for shorter periods and manually count and describe substance-related content they see. This places more burden on participants and still involves self-report, but the focus on specific periods of time could produce more accurate estimates than current self-report methods and does not require participants to upload potentially sensitive data.

It is important for future research to consider both self-report and objective measures to help to understand the interplay between the two [23]. In the context of exposure to substance-related content on social media, very few studies have included both self-report and objective measures and doing so will allow us to determine the accuracy of self-reported exposure or which self-report questions yield more accurate estimates. Furthermore, it may be important to understand whether the discrepancy between self-report and objective measures is systematic and what implications this may have for interpreting the current evidence [23]. For example, self-reported exposure may be very similar to social normative perceptions. That is, those who believe substance use is more common may also believe that they see more oftensubstance-related content on social media (even though they do not).

The literature on the impact of social media use and exposure on substance use is currently dominated by self-report, which tends to be inaccurate. With today's technology (e.g. screen recordings, apps, AI) objective measurement can now be more easily obtained. This is particularly important given the desire of policymakers to regulate social media. Therefore, objective data can help shine light on the true amount of social media use and the exposure to unhealthy content including substance use to inform policy.

Benjamin Riordan: Conceptualization (lead); writing—original draft (lead). Damian Scarf: Conceptualization (supporting); writing—original draft (supporting). Megan Strowger: Writing—original draft (supporting); writing—review and editing (supporting). Gedefaw Alen: Conceptualization (supporting); writing—review and editing (supporting). Taylor Winter: Conceptualization (supporting); writing—review and editing (supporting). Emmanuel Kuntsche: Conceptualization (supporting); writing—review and editing (lead).

None.