A blog about problems in the field of psychology and attempts to fix them.

Sunday, November 27, 2011

Modeler's Hippocratic Oath

This will be familiar to some people already. In an essay put out in 2009 Emanual Derman and Paul Wilmott write a manifesto aimed at the financial modelers, but with points that can apply to any modelers. The highlight of the article is the Modeler's Hippocratic Oath. Of course, the original Hippocratic oath was the medical doctors pledge (to the Greek gods) to do no harm (sort of). What would it look like if modern modelers were forced to make such a pledge? Maybe something like this:

  • I will remember that I didn’t make the world, and it doesn’t satisfy my equations.
  • Though I will use models boldly to estimate value, I will not be overly impressed by mathematics.
  • I will never sacrifice reality for elegance without explaining why I have done so.
  • Nor will I give the people who use my model false comfort about its accuracy. Instead, I will make explicit its assumptions and oversights.
  • I understand that my work may have enormous effects on society and the economy, many of them beyond my comprehension.

What a great idea! Even the parts that are more clearly aimed at economic modelers would work pretty well for psychological modelers. In the second line, we might change 'value' to 'values' and in the final line we might or might not remove explicit mention of the economy. In any case, we would have an oath that in no way stops mathematical modeling, but serves to remind us to be explicit about our assumptions, and to dissuade others from having false confidence in our models.

Psychology has its share of formal mathematical models of the type of the type Derman and Wilmott are concerned with, but it also has other types of models. Psychologists increasingly rely on agent-based models to simulate behavior, and in the future (one hopes) will rely on agent-based models to update their hypothesis-testing methods. Psychologists also rely on many forms of statistical modeling, such as factor analysis, structural-equation modeling, hierarchical linear modeling, neural network models. Some of these contain painfully arbitrary elements. For example, there is no hard and fast rule for determining the number of factors indicated by a factor analysis, and structural equation models are often built to test a specific theory in a way that makes unpredicted results unlikely to create problems. Also, almost all of these methods have a tenuous connection to the psychology of an individual person. Even a very good structural equation model, quite worthy of publication, might be very poor at predicting any individual outcome - e.g. whether a particular person might become a drug user.

I am not against doing any of this work, but I would like a bit of that oath mixed in. Make the assumptions and oversights explicit, remind people about the inaccuracies of the model if they seem to be getting to confident. 

The neural-network models are particularly troublesome. They have the additional flaw of containing the work "neural" in the name, so people get confused and think that they work like brains. But neural network models are just fancy regression equations, very useful for some purposes, but not at all a biologically sophisticated simulation of neural functioning. For example, one critical element of brain function is the synchronous or asynchronous arrival of action potentials at given locations (e.g. what gets screwed up in Multiple Sclerosis, when you start loosing your mylen sheaths). Again, I am not advocating that we stop neural-network modeling; I am advocating that the modeler be taught to feel obliged to interrupt and correct anyone who starts to imply that the models work in exactly the same way that actual nervous systems work.

By the way, another line from the essay is:
The most important question about any financial model is how wrong it is and how useful it is despite its assumptions.
Yes, yes! Not true models are useful, but only if we remember that they are not true. One of Skinner's criticisms of the use of mental terms as place-holders (e.g. using them as we await further investigation) was that the next generation of scientists always seemed to forget they were place-holders. This has happened with several major psychological phenomenon in recent times. For example, the "central executive" was supposed to be a place-holder term for all the things the brain did that we couldn't yet put in a more specific part of a model. It was kind of like cognitive psychology's "black box", only with the hope that the box would shrink over time as we got better models for specific mental functions. However, very quickly people forget that the central executive wasn't a thing and there are now countless studies about the role and function of the central executive in various processes.

That last part might not be on topic completely, I am a bit feverish... but I'd love to know what people think about the idea of the modeler's oath.


  1. Hi Eric,
    Interesting post. I like the idea of a modeler's oath. It is so easy to reify models we become invested in that we need reminders to maintain perspective.
    In regards to your comment about central executive, I would agree with the problem of the placeholder or little homonculus...that said, I do think the idea of executive processes carries functional value. When my kids are acting impulsively, I will often say "Activate frontal lobe!" :o)...


  2. Gregg,
    Agreed, there are things happening that are worth study, but there is not an executive at the center of it all (not even a Senior Assistant to the Executive Vice-President).

    This creates a problem though. What do we call the processes until they are individuated sufficiently, if the place holders don't work (because they get reified). The only solution I can think of would be to make an unmistakable place-holder label. You would have a diagram with "memory", "attention", etc., then "UNIDENTIFIED PROCESSES". Of course, then you might as well have a section of your diagram labeled "Here be dragons"!

    (Of course, this is ignoring the problems with this type of cognitive modeling - for example, the assumption that attention and memory are easily differentiated.)

  3. Very interesting topic. I like how two people respond in your article. Good job.

  4. Eric,

    I'm totally with you on the idea of Modeler's Hippocratic Oath, but you've mischaracterized the role of the central executive in working memory.

    First, it was never intended to represent "all the things in the brain" we hadn't fully understood. It was based on the observation that storage demands and manipulation demands of tasks requiring short-term memory exerted experimentally separable influences on performance. The role of the central executive was originally to place-hold yet-unspecified cognitive control processes that might account specifically for manipulation components of performance, as opposed to storage. So yes, it served a place-holder role (didn't we used talk about this back in the Davis days?), but the place being held was much more circumscribed than what you've described.

    Second, the central executive "box" most certainly HAS shrunk over time, in the sense that potential functions have been taken out of it and put elsewhere, and also in the sense that the nature of the major "control" function has been boiled down -- most of the current models of working memory that reference a central executive (Although the Cowan and Engle models are probably more clear about this than, say, Baddeley), explicitly describe it's function as that of switching active representations in and out of attention. That is, it's not just "manipulation" any more, but a particular form of manipulation.

    Moreover, recent studies of the central executive do not simply say "central executive" and wave their hands...they attempt to manipulate and measure specific executive functions and characterize their relationship to things like memory retention, task-switching and inhibitory control. I fail to see how these efforts should count as perpetuating a "place-holder" view the central executive -- their purpose, and their effect on theory, has been exactly the opposite.

  5. Joel,

    Quite right! Rereading that part of my post, I certainly overstated the case... but I still think the situation is worse than you are making it out to be. Granted: The "central executive" was a place-holder for a not fully specified, but certainly limited, number of cognitive abilities; that the goal was to specify the abilities more fully, and thereby eventually shrink the placeholder-function out of existence; and that some cognitive psychologists have actually advanced that function.

    Even granted all that, I still think there are many people who explicitly in their writing, or implicitly in their research strategies, have forgotten the placeholder function and reified the central executive as a coherent, studiable thing in itself. This is what Skinner warned against: Not that research strategies using hypothetical constructs are inherently flawed in their proposal, but that too much of the profession inevitably loses track of the game plan.

    I'm having trouble finding some of the better examples I have seen recently, but a quick search reveals that it is possible to have an "isolated central executive deficit" (apparently this is common in people with severe traumatic brain injury), several studies divided deficits into two categories "central executive and non-executive functions", and studies interested in the role of the central executive in manipulation of representations, storage, decision making, personality, and several other processes. Overall, the picture I get (admittedly as a person outside the sub-field) is that the term central executive is used in many different ways, at least some of the people seem to believe there is a thing in the head called a central executive, and many seem to think it is virtuous to multiply the things attributed to the central executive, rather than reduce the things attributed to it. Again, that isn't to say that no one is using the term as it was intended, but that the field as a whole is not.

    Do you really think I am over exaggerating? Or is it just that I am being unfair to the smaller group of scientists who are doing good work with the concept?

  6. Say, you got a nice blog article. Keep writing.