People often argue whether depression is, or is not, caused by a “chemical imbalance”.
Much of what happens in our brains is chemical, why would depression not be? If by “imbalance” we happen to mean “a state of brain chemicals that the patient doesn’t want”, as opposed to, say, some specific theory that is now discredited like “not enough serotonin” (i.e., the low serotonin myth), then depression can reasonably be thought of as a “chemical imbalance”.
Disagreement about whether depression is a chemical imbalance may stem from confusion about what it means for something to be a “model” for something else. For instance, “chemical imbalance” is a model for depression. But it’s just one model, and it is often not the most useful one. Here are all the models I’ve seen people use for depression:
(1) Neurology – caused by a chemical imbalance of neurotransmitters in your brain; that’s why anti-depressants are effective for many people.
(2) Beliefs – caused by unhelpful perceptions of yourself, your future, or the world; that’s why cognitive therapy is helpful for many people.
(3) Circadian rhythm/sleep – that’s why some people may find chronotherapy (involving sleep and light) or sleep apnea treatment effective.
(4) Relationships – that’s why socially isolated people often feel depressed and why Interpersonal Therapy may be effective.
(5) Nutrition – that’s why people with depression are sometimes found to have vitamin deficiencies or harmful diets and may feel better if these are corrected.
(6) Society – that’s why groups of people who are oppressed, shunned, or in poverty are more likely to be depressed, and improving these social problems may resolve the depression.
(7) Behavior – that’s why depressed people are sometimes found stuck in harmful behavioral feedback loops and why Behavioral Activation for depression is helpful.
(8) Trauma – that’s why people who were unloved or abused as children may be more likely to be depressed, and why people often show depressive symptoms for a while after a loved one dies.
(9) Meaning – that’s why depressed people sometimes feel that nothing matters, and why techniques from the ACT approach that get you to take action towards what you value may be useful.
(10) Genetics, that’s partly why depression tends to run in families and helps explain why one person gets depressed in a particular circumstance that another person doesn’t get depressed in.
These can’t all be true, can they? I believe they can, in the sense that each of these is partially correct, or to be more precise, each of these is a model for depression that is better than useless. What’s nice about this set of models is that they are complementary: they each capture some distinct aspect of depression, and each will be especially useful in certain circumstances. In particular, a number of these refer to different things that can cause or trigger depression. Not all of these causes will be in play in any particular case, so sometimes one model will apply more than other.
There are many ways to model any system. A good model typically is not one that fully explains every detail of a system but rather one that explains important aspects of the system in important cases of interest. It is not a contradiction to find that there are multiple good models for the same system that are very different from each other. They may each capture different important attributes of the system, or may each be accurate in different situations, or they may just represent similar information in very different seeming ways. Even if one model is much better than the others on average, there may be specific cases where the alternative models make better predictions. And realizing that may allow us to combine models to create an even better one.
Consider the game of Pac-Man as an example. One way to model the game is to think of the ghosts as sentient beings that are trying to touch you, which will cause you to die. This is the intuitive model that players often have of the game, and despite being completely untrue on one level (the ghosts are not sentient), it is quite a useful model that enables you to play the game effectively.
However, if you train a reinforcement learning algorithm to play Pac-Man, it will produce a very different model of the system (one that may well contain no direct notion of ghosts that are out to get you, and likely the model it produces will be one that you’ll have a hard time understanding). Yet, using this model, the software may play Pac-Man as well as you do or perhaps even better. So whose model is right, the intuitive human one or that of the reinforcement learning algorithm? Neither is “right”, and neither is “wrong”; they both capture important elements of the game in different ways.
A more complete model of Pac-Man than either the intuitive human one from a reinforcement learning algorithm’s is the computer code of Pac-Man, which determines the gameplay. From the point of view of you playing the game effectively, it MIGHT help to some extent to study the code in advance to learn potential quirks of the ghosts’ behavior, but mostly the code would not be useful while you play, and you’ll find it a lot easier to think of ghosts being out to get than to mentally reference the underlying algorithms. So your “ghosts are out to get me” model may be more helpful to you than the “source code of the game” model, despite the latter being a much more accurate description.
But even the source code is not the ultimate model of your Pac-Man experience, as the ULTIMATE model would include the hardware that code runs on (maybe the behavior of that code varies slightly on different hardware), and even the people around who might distract you during the game, and ultimately the laws of physics on which the behavior of the hardware and everything else depends.
As statistician George Box once wrote, “The most that can be expected from any model is that it can supply a useful approximation to reality: All models are wrong; some models are useful.”
So let’s stop asking if depression “is” a chemical imbalance. Instead, let’s ask: is it useful to sometimes model depression as a chemical imbalance? I think the answer there is yes; this is one model that should be used among other models in certain specific instances. It may sometimes be a useful model, but other times other models of depression will be much more useful. And, of course, much of what happens in our brains can be modeled as chemical changes. So the question is: when is this a useful way to think about what’s happening chemically in the brain? When does such a model give us insight?
If a depressed person responds well to anti-depressants, we can use our Neurology model of depression to try to understand this (though that still may be very tricky to do). On the other hand, if a person’s depression lifts after they stop believing that they are inherently worthless, we can use our Beliefs model to try to understand this. And if a person’s depression improves when they adjust their sleep patterns, we can use our Circadian rhythm model to try to explain this. And so on.
Ideally, we’d try to figure out the most effective model for a given person’s depression so that we can accurately model their specific depression, not merely depression in general.
Comments