Resolving the causal riddle
But there are always questions and more experiments...
Thanks to everyone for sending their resolutions to yesterday’s causal riddle. I think everyone got to more or less the answer I was looking for.
In the case of learning Japanese, we strongly believe that the counterfactual outcome is impossible. The scenario in which someone who has never been exposed to spoken Japanese and has never taken any Japanese classes spontaneously begins to speak Japanese is not plausible.
We have no such certainty about depression. There are a variety of explanations that feel plausible to explain why a person stops feeling depressed. Perhaps they bought a new bed and are sleeping better. Perhaps they received a promotion at work. Perhaps they made a new friend. Maybe those were the cause, not the drug. We also have a cultural belief, though one that is not necessarily “scientific” nor helpful to the depressed, that severe depression can be simply transitory.
Once you see it, the riddle seems simple. On Twitter (LOL Substack no longer lets you embed tweets. Thanks again, Elon!). Jared Huling thought “I'm not sure I'd even consider this a riddle.” I don’t really disagree with Jared! But I want to understand why we think it’s so simple. The reason we attribute causality is our beliefs. We have strong beliefs about how language can be acquired. We have strong suspicions about the effectiveness of SSRIs.
Why do we believe in certain processes and not others?
The simplest answer is one of rates. We no of no one who has spontaneously acquired a foreign language (the stories about spontaneous language acquisition are all apocryphal). We immediately think we know people who have cured their depression without SSRIs. If we set some “likelihoods” based on rates, any decision framework would lead us to certainty for language acquisition and uncertainty for depression.
But there are other reasons why we don’t believe in SSRIs. Though widely prescribed for forty years, SSRIs remain a treatment of considerable debate and controversy. Disquieting research has percolated its way into the popular press, alarming us that we lack strong evidence that drugs like Prozac cure depression and lack a plausible mechanism of how they operate. Perhaps we’d be more willing to attribute a causal impact to the drugs if we could see some concrete chemical change in the brain. If we could see the mechanism that makes SSRIs work, would that change our beliefs?
On the other hand, we think we understand the mechanism of language acquisition. A student memorizes some words. Maybe they learn some new alphabet and grammar rules. They practice conversation over and over with their instructor. Over time, skill develops, and eventually they have fluency. Is the mechanism I just described really more detailed than what we have for medications? Psychiatrists prescribe specific medications for specific indications. A person takes the medication, and over time they feel differently. Would we believe in SSRIs more if the treatment of depression was more interactive?
Does some of the difference in our beliefs arise from differences in evaluation and measurement? With languages, we have a concrete way to evaluate fluency. There are online tests that will examine basic competency. And we trust native speakers of the language to assess others’ ability to speak their language. With mental health, we do not have a cut-and-dry measurement process for distinguishing the afflicted and cured. But this disparages people’s ability to know if they are depressed. Just because we can’t measure it with tests does not deny the reality of a person’s depression.
I’m asking these leading questions because I think the answer is maybe. With language competency, checking in on gradual progress and adjusting lessons accordingly improves the rate at which you learn. What would a more interactive treatment plan look like for depression? How would we evaluate it? In previous blogs, I discussed how common endpoints in clinical trials take the form “did a bad event occur during the trial?” How can we conceptualize evaluations beyond such simple binaries? I will spend the next few blogs describing cases where researchers have answered these questions. The answers to these questions start to lead us down the path out of randomized trialomania in causal inference.