When correlation collides with causation

This image was originally posted to Flickr by Senado Federal at https://flickr.com/photos/49143546@N06/30554309793. It was reviewed on  by FlickreviewR and was confirmed to be licensed under the terms of the cc-by-2.0.

Recently, an article caught my eye (link) entitled “Have Smartphones Destroyed a Generation?” It discusses the possible deleterious effects that smartphones may have on young people, or those born in the mid-1990s. The unique aspect of this cohort is that they were born after the start of the Internet age. For all their lives, they have been surrounded by screens. The gist of the article is that a wide variety of indicators for mental health disorders (such as anxiety and depression) have risen after wide adoption of the iPhone in 2007. Therefore, there is a link between smartphone use and mental health disorders among young people. I want to focus on one aspect: depression and smartphone use.

When we ask a research question, we want to be as specific as possible: is this new, or incident depression? What type of smartphone use are we talking about? Social media? Google Maps? Yelp reviews? What about amount of smartphone use? An hour, two hours a day? Who would we consider? Children? Teenagers? What other characteristics might we consider? And perhaps most important of all: Who would be in the control group?

The point of getting into all this nitty gritty stuff is so that we want to control for as many extraneous factors as possible and only examine the association of interest. But what should we control for? A conceptual framework of depression and smartphone use might help us. We want something that might explain the links. For example, perhaps excessive smartphone use leads to less sleep which leads to more depressive symptoms. We proceed cautiously and are transparent about any possible errors in our thinking.

A control group is extremely important. We need to find a similar group of people which would help control for possible factors. The controls could be a similar group of young people who don’t use smartphones and look at their incident rates of depression. You can see where things become difficult. Children and adolescents who don’t use smartphones are as rare as polar bears in the Sahara. Perhaps instead, we could define low smartphone use. Maybe we could use a historical control, like, for example, people born in the 80s.

A useful concept for causation is the Bradford Hill criteria. These are 9 criteria used in epidemiology to determine whether connections are causal. They are:

Strength: Is excessive smartphone use associated with a large increase in depression rates?

Consistency: Do studies with different groups of people in different settings show the same effect?

Specificity: Is there no other explanation for the causes of depression for young people?

Temporality: Did depression occur before smartphone use? Or did smartphone use occur before depression?

Biological gradient: Does greater use of smartphones lead to greater severity of depression?

Plausibility: Is there a plausible mechanism? For example, maybe social media affects the neurotransmitter systems in the brain that lead to depression.

Coherence: Can this be observed in a controlled setting?

Experiment: What if researchers control the amount of smartphone use? Would that change severity of depression?

Analogy: If smartphone use causes anxiety, then it’s more likely to cause depression.

Causation is definitely a complicated concept and it would take multiple studies to even start to explain it. But if we want to get beyond the complaints about “Correlation is not causation!” then we need to apply concepts such as the Bradford Hill criteria.

One digression: I dislike the “Correlation is not causation!” complaint because it assumes that researchers of a particular study did not think about possible confounders for their effect of interest. Of course they would try to control for confounding! If there is a strong enough association between depression and smartphone use while controlling for other factors, then it strongly hints at causation and should be investigated further.

Further reading: Elhai JD, Dvorak RD, Levine JC, Hall BJ. Problematic smartphone use: A conceptual overview and systematic review of relations with anxiety and depression psychopathology. J Affect Disord. 2017;207:251-259. DOI: http://dx.doi.org/10.1016/j.jad.2016.08.030

Summary: There is a link between smartphone use and mental health disorders. I provide some thinking on how it may be investigated further.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s