• We’ve found that many teenagers tend to be curious and open to having discussions about mental health and social media. However, that openness quickly closes if they feel that they’re being “given a lecturing” about social media. Starting the lesson with the examples (in the lesson slides and video) of past concerns about radio, television, and video games provides a helpful “we know some concerns have been overblown in the past” framing to the lesson. Then, presenting the more recent statistics about mental health declines feels less like a lecture, and more like the presentation of a compelling question: are these trends just associated with the spread of social media, or are they causally connected?
• Many students are frequent social media users. Bringing that expertise into the room can feel empowering, and we find that students are often excited to dive into this lesson and share their own experiences. At the same time, other students may not feel as comfortable discussing their own experience with social media, especially as it relates to mental health. We recommend keeping this lesson’s discussions at the statistical level, while also allowing students to share their own experiences according to their own level of comfort.
• To motivate the more complex experimental designs (randomized block and matched pairs), it’s helpful to explicitly motivate the problem of variation. In an ideal world, we could give someone a treatment, then rewind time, give them a different treatment, and then measure the difference in their outcomes. Because this isn’t possible, we have to randomly assign different individuals to different treatments. The problem with this approach is that different treatment groups could be composed of very different individuals, making the effect of treatment difficult to detect. Randomized block and matched pairs designs attempt to mitigate this problem by making the treatment groups as comparable as possible, making any treatment effect easier to detect.

First, download this lesson's Handout Key and read through its Discussion Question section. Then, check out our model discussion norms and the additional background notes below.

• This question represents another exploration of the concept of generalizability. Because the experiment is performed with college students, the results can’t be fully generalized to other age groups. So, students have to make their own inferences about how the treatment effect might differ for high school students. Surfacing this limitation during discussion can be helpful for reinforcing the concept of generalizability.
• Because of the generalizability limitation mentioned above, answers to the Discussion Question can (and should) vary. The important point to emphasize is that students should provide robust reasoning for their opinion, grounded in the context of the lesson and by statistical reasoning.
• Although they’re less frequent, experimental studies of social media and mental health have been performed with high school aged individuals (or with mixed samples of high school and college aged individuals). For example, Kleemans, Daalmans, Carbaat, & Anschütz (2018) and Davis & Goldfield (2024). That said, because these studies are so few in number and test different treatment conditions, it’s difficult to use them to make strong inferences about this lesson’s Discussion Question.

• The opening of the lesson mentions increased rates of major depressive episodes, anxiety disorders, and emergency department visits for self-harm among young people. It’s possible that the first two indicators could be attributed to more frequent detection and reporting of mental health disorders, rather than true increases in the underlying prevalence of these disorders. However, the same can’t be said about the third metric, since standards for reporting among emergency departments have not changed over the time periods mentioned in the lesson.
• Students may wonder: In the Iowa State social media experiment, why did researchers only ask participants to limit their social media use, rather than fully controlling it? In the full paper, the researchers state, “One consistent finding has been that complete abstinence from social media may not be sustainable for the average user. A less strict approach is to limit social media use by monitoring. Monitoring limited usage, as opposed to abstinence, may be more sustainable and practical.” So, researchers wanted to know if a more sustainable approach that required less coercion – merely telling folks to reduce their social media time – would still produce an effect. And it did.
• Students interested in this topic further can explore the increasing body of experimental research on social media use and mental health, including the following articles: Brailovskaia, Swarlik, Grethe, Schillack, & Margraf (2023), Davis & Goldfield (2025), Graham, Mason, Riordan, Winter, & Scarf (2021), Hunt, All, Burns, & Li (2021), Kleemans, Daalmans, Carbaat, & Anschütz (2018), Lambert, Barnstable, Minter, Cooper, & McEwan (2022), Thai, Davis, Mahboob, Perry, Adams, & Gold (2023), and Yuen, Koterba, Stasio, et al. (2019).

• The lesson discusses the matched pairs design of having each participant engage in both treatment conditions, with the order of those conditions randomized. A matched pairs design can also be applied to situations in which different participants are paired together, before each participant is randomly assigned to a different treatment. For example, many experiments seek to recruit identical twins – who have almost identical DNA – to serve as matched pairs in experiments. In fact, some Universities even have twin registries that help connect researchers with twins for studies.
• Randomized block and matched pairs designs can be connected to a statistical idea that arises later in the course: power. By reducing variation between treatment groups, these experimental designs increase the power to detect significant differences between treatment groups. This can help reduce study costs, as researchers can recruit fewer participants, while still retaining the power to detect treatment effects. That said, because students have not yet been introduced to the concept of power, it’s best to wait to surface this connection until later in the course.
• Although the Iowa State researchers did not implement a randomized block design or a matched pairs design, they used advanced data analysis methods (beyond the scope of AP Stats) that mimic these designs. For example, researchers used multiple regression models to control for gender when measuring treatment effects. This is similar to comparing treatment effects only within gender blocks. In addition, researchers had every student take a pre and post psychological evaluation. The pre scores were used as a regression control, allowing researchers to control for individuals’ baseline evaluations when measuring treatment effects. This is similar to a matched pairs comparison at an individual level.