Pure Pedantry

I saw this news story in Nature a couple days ago about finding a gene for “ruthlessness.” I realized that I always say the same thing about these behavioral genetics stories — stories where they claim to find a gene for ____ (where blank is a behavioral abstraction like empathy). These studies are notoriously misinterpreted by the media, so I figure I will reiterate some caveats to remember about them.

In this study, Knafo et al. compared performance in the Dictator game to genotype for a particular allele of the vasopressin receptor AVPR1a.

In the Dictator game — which isn’t much of a game — the dictator decides the allocation of a reward to a receiver who they don’t have to interact with after the game. In this experiment, the participants start with 50 shekels (it was done in Israel). They can keep them all, divide them among himself/herself and the receiver, or give them all away. On a purely maximizing reward basis, you would expect that the dictator would keep all the money. There are no negative consequences to keeping all the money. However, you find when you test them that lots of people will divide the money between himself/herself and the passive receiver. Thus, the results in the Dictator game suggest that people account for the value to others in their perception of maximizing reward.

There is a great deal of variation in performance in the Dictator game — both between individuals and between groups. Some groups give more money away, suggesting that they indulge in more altruistic and pro-social behavior.

AVPR1a is one of the receptors for the peptide vasopressin. Although vasopressin is typically associated with a role in regulating water retention, it has many other functions. It has been shown in animals — in this case voles — that the presence or absence of vasopressin receptors in certain parts of the brain determines whether a species of vole mates for life or is polygamous. These results and others in animals suggest that AVPR1a may have a role in regulating pro-social behavior. (Reviewed here.)

The authors of this study compared AVPR1a allelic variants in 200 college students to performance in the Dictator game. AVPR1a has a long and a short allele that affect their expression. The significant result of this study is that participants in the Dictator game were also significantly more likely to keep all the money if they have a short allele of AVPR1a. (Giving away money was not related to gender.)

So far, so good. I believe these results. I have no reason to believe the authors weren’t methodical, and these results are in that sense correct. Further, this research is interesting because the authors can relate it to work in animals. That isn’t typically the case.

However, I want to emphasis that all behavioral genetics studies have the same set of interpretational caveats.

1) It’s all about replication.

It was my experience working in a psychiatric genetics lab that studies like this tend to be done in relatively inbred populations to increase the chance of getting statistical significance. By using an inbred population, you can help ensure that the other genes in the subject population are relatively similar. This lets you isolate the gene you are studying. However, this can also be a problem because other characteristics common in your experimental group could affect the outcome: common environment, epistatic interactions with genes related to your gene of interest, etc. If these characteristics determine the presence or absence of a trait and they correlate with your gene of interest, you may think that your gene is important when that is not the case.

The take home message is that you can never be sure your gene-behavior association is real until you repeat the study in multiple different study populations. If you find that the same gene is associated with your behavior in many different populations, that would give you much more confidence that your result is real.

2) It’s all about the mechanism.

Correlation is not causation. One of my pet peeves about behavioral genetics papers is that they point out associations between genes and behaviors, but the mechanism for how the gene influences the behavior is vague. At least in this paper, there is some validated animal studies that look at receptor distributions in animals and correlate those with behavior. That’s good.

On the other hand, I view papers like this with a great deal of reticence until they can show me how their gene directly results in a change in behavior.

3) Genetics are not causal for most behaviors.

Behavior results from an interaction between environment, genetics, cognition, and chance. There are very few genes that imply a behavioral destiny. For example, if you have the gene for Huntington’s disease, you are going to get Huntington’s disease. That is destiny. But for most psychiatric disorders, there is a strong element of environment in it. For these, genes confer susceptibility.

(I talk more about this here, although I would note that Razib doesn’t like my example of Huntington’s disease. I will always accede to Razib’s considerably greater experience in the finer details of population genetics. However, I think that my statement that “most psychiatric disorders are the result from a gene-environment interaction” is a fair one.)

The same point can be made for behaviors that are not pathological in the sense of psychiatric disorders. Personality traits are significantly less heritability than psychiatric disorders. Further, heritability as a measure is not the percentage of a trait that is genetic. Even highly heritable traits can be modified by environment. (For an explanation why read this post by Razib or this longer post at Three Toed Sloth.)

Whenever you interpret a paper like this, remember that having or not having this gene variant does not mean you will necessarily be “ruthless.” It might mean you have that capacity or you that you have a greater risk, but it is not a certainty.

4) The results of behavior experiments do not imply abstractions like “altruism” or “ruthlessness.”

Here is my biggest caveat about stories like this.

In behavior experiments — both in people and animals — you have to design some tractable task the participants can do that is related to the behavior you want to study. In the case of this paper, they use the Dictator game to assess altruistic behavior in humans. The jargon is that they have operationalized a behavior.

However, the process of operationalizing requires a certain abstraction from reality. Altruistic or dictatorial behaviors are complex things involving many parts, some of which are not captured in this experiment. Ruthlessness could involve the refusal to help, but it also involves the desire to do harm. No part of this experiment measures that desire, so this experiment is not sufficient to capture “ruthlessness” in the abstract.

Thus, when you read the results of an experiment like this, you need to screen out attempts to associate abstract concepts with the results. The subjects in this study were more likely to give money away if they had a long version of the gene. That is it. This does not mean that the subjects were more “altruistic.” It means that according to the author’s measure of altruism, they performed more highly. The author’s measure of altruism is imperfect, so you don’t know whether your results will generalize to other aspects of altruism.

Specifically in this case, I take issue with comparing the refusal to give away money and ruthlessness. Not giving away money has no serious consequences in this experiment. It is hardly ruthless in the sense of maliciously hurting someone. (Further the news article describing this result has pictures of Hitler and Saddam Hussein. That is just ham-fisted.) I can only conclude that the author of the news article over-interpreted these findings from the name of the behavioral test.

The full effect of how single experiments do not capture abstractions can be observed when you do other experiments on subjects who give away more money. The Dictator game suggests that these individuals are more altruistic — at least that is one interpretation. However, the author of the news article above mentions this other result:

Researchers should nevertheless be careful about using the relatively blunt tool of the Dictator Game to draw conclusions about human generosity, says Nicholas Bardsley at the University of Southampton, UK, who studies such games.

His research suggests that players who routinely give money away as Dictators are also perfectly happy to steal money off other players in games that involve taking rather than giving. This suggests that the apparently more altruistic players in Ebstein’s game may in fact be motivated by a desire simply to engage fully with the game, perhaps just because they feel that that is what’s expected of them.

In a case where they can take away money from people, the supposedly altruistic people from the Dictator game take more — they are more likely to coerce others. This brings the interpretation of them as being altruistic into question. Maybe they are just more willing to interact in general. Maybe they view the money more as collective property.

The point of my comment here is that you could make an argument that a gene determines altruism, but you can’t do it from just one experiment.
You need a variety of behavioral tests to capture all the aspects of something that complicated. Once you have clearly defined what you mean by altruism and shown that a gene affects it in a variety of contexts, then (maybe) you can say you have a gene for altruism.


I am not suggesting with these caveats that Knafo et al. is a bad paper. They have done and interesting experiment and shown an interesting result. Nor am I suggesting that the Nature coverage was particularly eggregious. Though their emphasis was a tad more exuberant than mine would have been, they commit no serious errors.

I just wanted to list these caveats because the media notoriously over-interprets the significance of behavioral genetics studies. The next time you hear a story for a gene for _____ in the NYTimes or something, remember these caveats before you buy into the hype.

Corpus Callosum has more on this story.

Comments

  1. #1 gfb
    April 9, 2008

    “most psychiatric disorders are the result from a gene-environment interaction”

    not to nitpick… but, interactions imply something very specific in both statistical and quantitative-genetic terms. i think that you meant to say that “… disorders are the result of genotype-environment interplay

  2. #2 Matt
    April 10, 2008

    You make a good point in that some individuals may be just trying to balance payoffs amongst all players. This might also make a difference in the conclusion drawn in this study.

  3. #3 Matt W.
    April 10, 2008

    I’m glad you posted these thoughts. I’ve actually worked with a behavioral researcher with similar sentiments about behavior-genetic twin studies for ADHD. He’s a meta-analyst, and yes, that has it’s own issues, but whenever he runs the numbers for behavior-genetic studies of ADHD, genes almost never account for more than 15% of the variance, which suggests there are other causes at work than merely genes. Good thoughts!

  4. #4 windy
    April 13, 2008

    3) Genetics are not causal for most behaviors.

    You are making a false dichotomy between ‘causal’ and ‘affected by the environment’. We say that tobacco smoking is causal for lung cancer, although most smokers don’t get lung cancer. The same for drinking alcohol and traffic accidents, etc. Otherwise, good points.

New comments have been disabled.