Tag Archives: studies

Before You Get Too Excited About That Trigger Warning Study…

STUDY: Trigger Warnings Are Harmful To College Students says the Daily Wire, describing a study whose participants’ average age was 37 and which did not measure harm.

You can find the study involved here. A group of Harvard scientists asked 370 people on Mechanical Turk to read some disturbing passages – for example, a graphic murder scene from Crime and Punishment. Half the participants received the following trigger warning before the passage:

TRIGGER WARNING: The passage you are about to read contains disturbing content and may trigger an anxiety response, especially in those who have a history of trauma

Participants were asked to rate their anxiety before and after reading the passages. After they had finished, they were asked to fill out a bunch of questionnaires that measured their opinions about how trauma worked.

The researchers found that people who received the trigger warning were 5% more likely to endorse the idea that they were vulnerable to trauma, and also 5% more likely to endorse the belief that people with trauma could suffer persistent negative effects from that trauma. There were some subgroup and moderation analyses which I ignore for the usual reasons.

What might be some causes for concern with this study?

First, Stuart Ritchie points out that the results are statistically weak. Most of the results have p-values around 0.05, and are not corrected for multiple testing. That means it hasn’t been formally proven whether or not the results are random chance. I don’t like haggling over whether something is just above or just below a significance threshold. But if you do like that kind of haggling, this study doesn’t survive it very well.

Second, the participants’ average age is 37. There’s nothing wrong with 37 year olds, but everyone who says that they’ve proven trigger warnings don’t work on college students either hasn’t read the study, or is adding their editorial opinion.

Third, trigger warnings are aimed at people with trauma. The authors excluded people with trauma from the study. The study was not an attempt to test whether trigger warnings could ever be beneficial, nor to quantify the net (benefit minus harm). It just tested whether there might be nonzero harm. Even if trigger warnings harm healthy people, they might still help the people with trauma who they are intended to help.

But a much more fundamental concern is that this isn’t measuring anything we care about.

Some people read a trigger warning saying disturbing passages could cause emotional harm. Then they read a disturbing passage. Then, on a test, they were slightly more likely to agree with the statement that disturbing passages could cause emotional harm. Of note, they did not claim that they themselves had been harmed or triggered by the passage. In fact they specifically denied this; there was no difference in anxiety between the two groups after reading the passage. They just agreed, in a theoretical sense, that trauma was harmful.

The most boring possible explanation is that if you gave someone a passage saying “TRIGGER WARNING: SOME SNAKES BEAR LIVE YOUNG”, then made them read a book about torture-murder, then asked them whether some snakes could bear live young, they would be more likely to answer yes. Reading the statement about snakes doesn’t cause the snakes to bear young. It just convinces the reader of it.

In the same way, the trigger warning tells respondents that the passage can cause anxiety, and links this to a history of trauma. It doesn’t say outright that passage –> anxiety –> permanent trauma, but it kind of implies it. It certainly primes any pre-existing beliefs along those lines that participants might have heard. And it potentially induces demand effects – which Mechanical Turk is infamous for – by letting you know what the experimenters are thinking.

For this to result in actual harm, two extra things would have to happen.

First, an effect that happens a split second after you read a trigger warning stating something would have to stick around to become a permanent part of the psyche. This is definitely not how these things work. For example, in 2016 I made a group of people read either an essay warning of the dangers of artificial intelligence, or a control essay that didn’t mention the topic. Then I surveyed people on how dangerous they thought AI was. Unsurprisingly, the people who read the control essay were less worried. The effect stuck around after a month on another survey that explicitly reminded them of the essay they’d read, but when I covertly surveyed them again two years later, there was no difference in their beliefs. They’d forgotten the whole thing. If you think about it, both the experimental group and the control group in this new experiment must have seen lots of trigger warnings throughout their lives. The effects of all of them paled in comparison to having seen a trigger warning a minute earlier in the same experiment. Hours or days later, the effect of this one will fade away too.

(This is the kind of measurement I’ve condemned as streetlight psychology, after the story of the drunk who searched for his keys under a streetlight – not because he’d lost them there but because that was where it was easy to look. It’s been used before to study eg video game violence. If you make someone play a violent video game and then test them for violentness a few minutes later while the adrenaline is still in their system, they’ll be more violent. But this is little different from showing that people are sad minutes after they watch a sad movie. If you’re asking whether years of playing a video game will cause violent criminality (or years of watching sad movies will cause clinical depression), that’s a very different question. The trigger warning study has the same problem.)

But even if reading a trigger warning has lasting effects on your beliefs about trauma, we still have to prove that those beliefs cause you to be more easily traumatized. One of the effects in the study was people saying they believed they were more easily traumatized – but how much do we trust that belief? If I made people read a passage saying that cancer was very common and even healthy people often got it, I could certainly make them express a belief that they were more worried about and susceptible to cancer. But would that belief cause cancer? I realize some psychological constructs do work like this, where believing that you’re vulnerable makes you more vulnerable. But a lot of psychological constructs also don’t work like this; it can’t be voodoo all the way down.

So, in conclusion: this study tests an accommodation meant to prevent trauma in college students on a population of non-traumatized 30 and 40-somethings. It finds it does not increase anxiety, but may change beliefs about trauma in the very short term, although those changes do not meet strict standards of statistical significance. It is currently unclear whether those changed beliefs last longer than a few minutes, or whether they have any effect on anything.

I think about this Steven Kaas tweet a lot:

A lot of people care a lot about whether trigger warnings are helpful or harmful in college courses. This study answers (or “answers”) the superficially similar-sounding question of whether they affect certain beliefs about trauma in the very short term. I won’t say it has literally zero bearing on the original question, but if you treat it as having any specific positive amount of bearing, you’re probably wrong.

I don’t want to blame the researchers, who are experts in the psychology of trauma and doing important work. This isn’t an awful study from the perspective of psychologists trying to start a research program that might be suggestive of something after a decade or two. It is an awful study in the context of anything that anybody might report upon or use to form an opinion. Blaming the reporting would be better, but the right-wing sources that exaggerated this are still light-years behind the left in their ability to falsify and weaponize study results and I would feel bad saying anything that seemed to single them out as particularly culpable. Let’s just agree everything is terrible all the time.

As for trigger warnings themselves, I’ll repeat my own proposal: have them, but put them in the Boring Legalese Page of the book, the one where they list the ISBN number and the city where the publishing company has its headquarters and something something Library Of Congress. Make them matter-of-fact, like “Content: rape, murder, ethnic slurs”. You don’t need to embellish with “AND THEREFORE YOU SHOULD BE ANXIOUS” or “SO PEOPLE WITH A TRAUMA HISTORY SHOULD BEWARE” the way this experiment’s warning did. Just put a list of things people might want to know about on the Boring Legalese Page page, and let the couple percent of people with a trauma history check it before reading if they want. If it’s a worksheet or syllabus, put it in small print somewhere consistent. If your opposition to (or support for) trigger warnings is any more impassioned than that, I am suspicious of your motives and think maybe you should sit this one out.