"THE NURTURE ASSUMPTION" (book review) by Arnold Kling

"The Nurture Assumption," by Judith Rich Harris, raises issues at two levels. At a substantive level, it questions the role that parenting style plays in childhood development. At a methodological level, it raises the question about what counts as valid empirical research in developmental psychology.

The base camp from which Harris begins is the findings of behavioral genetics. Studies of identical twins reared apart tend to show that about half of the variation in personality characteristics can be explained by heredity. This leaves half to be explained by other factors. Many psychologists take it as given that these other factors are parental behavior It is this presumption that Harris calls "The nurture assumption."

The first 50 pages of the book evaluate the evidence that parental behavior affects the personalities of children. Harris argues that the evidence is anecdotal and that the studies that are most careful about controlling for genetic effects produce quantitatively unimpressive results for the effect of parents on children. She sets the bar fairly high:

"A correlation of .19, even if it is significant in the statistical sense, is all but useless." (p. 19)

Harris also mentions the problem of reverse causation, or what economists would call "simultaneous equations bias." Reverse causation can occur when a parent's behavior toward a child depends on the characteristics of the child. As Harris puts it, "well-treated children tend to be nicer than those who are treated harshly, but that could be due to child-to-parent effects."

Next, Harris introduces an alternative to the nurture assumption. She argues that in the evolution of the human species, the ability to form groups and engage in group rivalry have been favorably selected. We needed a strong sense of tribal identity in order to fight off other tribes. Harris argues that children learn group dynamics from other children. Because this learning is what is important for survival, what children learn from their peers has stronger effects than what they learn from their parents.

In a way, this theoretical discussion is the most interesting part of the whole book. I was tempted to try to apply it to the context of business organizations. For example, Harris asserts that as groups become large, they tend to split up and feud. Does this tendency help to explain why large corporations lose cohesiveness? She also says that the tendency for a group to fracture may be reduced if the group is motivated by rivalry with another group. Does this imply that an organization can strengthen its culture by fostering an "us vs. them" attitude toward other firms?

To my untrained eye, the empirical evidence that Harris presents to support her theory of peer influence, or "group socialization," is as anecdotal and weak as the support for parental effects that she castigates in the first part of the book. Nowhere does she cite a study showing that variation in peer group characteristics explains a large share of the variation in personality. After having watched Harris dismiss a correlation of .19 as puny, one might be expecting to see 20 percent or more of the variation explained here. No such luck.

From an economist's perspective, I suspect that the attempt to explain personality differences suffers from the "errors in variables" problem. When the dependent variable is measured with error, this reduces the proportion of the variation that can be accounted for by any hypothesis. Some of the variation in the dependent variable is purely random, introduced my measurement error, and this random component will always be left unexplained by the model.

When the causal variables in the model are measured with error, the problem is even worse. Random noise in the independent variables will bias downward the statistical estimate of the size and significance of the causal relationship.

Measurement error is likely to be a major issue in this sort of research. The concepts to be measured, such as "intelligence" or "aggressiveness," are fuzzy to begin with. Moreover, the measurement tools--personality tests, self-reporting of moods and behavior, etc., are not terribly reliable. Trying to study relationships among these indicators is like trying to study the causes of variations in "length" using data obtained from rulers that have a 20 percent margin of error.

Imagine that we are trying to explain the aggressiveness of children based on the behavior or parents (or the behavior of peers for that matter.) Suppose that genetic factors explain half of the variation in measured aggressiveness of children. Much of the remaining variation may reflect measurement error. Combine this with the fact that the measure of parental behavior or peer behavior is likely to contain large random errors. There is good reason to be pessimistic that one will see quantitatively significant results, even if the underlying hypothesis is correct.

An implication of the errors-in-variables problem is that the effect of heredity on personality may be greater than is generally reported. If IQ measures intelligence with error, then the correlation between heredity and a "true" measure of intelligence would be higher than the correlation between heredity and IQ. The more ambiguous and poorly-measured the concept, the more the observed correlations understate the proportion of variation explained by heredity.

In macroeconomic time series, there is not enough valid "signal" in the data to enable one to confirm a specific, complex model of the determinants of unemployment, inflation, and so forth. Nonetheless, many economists believe in a specific, complex model. I know I do. However, my level of confidence in the model is low, because I understand the weakness of the evidence.

Similarly, there may not be enough "signal" in personality measurements to confirm models in which personality is affected by parental behavior. However, one can believe such a model, even if it difficult to prove statistically. One also can believe that peer groups affect personality, even though I suspect that this, too, will prove difficult to verify. Whatever one chooses to believe, it would be a mistake to place too much confidence in those beliefs. That is the main take-away I draw from this fascinating book.