Vandell's article is a thoughtful and welcome response to the challenge I first issued to academic psychology 5 years ago (Harris, 1995). Unfortunately, in her review of my work, she has erroneously attributed to me some notions I do not hold. Responding to her critique involves not only dealing with our genuine disagreements but also clearing up some misunderstandings. The first part of my response, therefore, is a critique of Vandell's critique of my critique of contemporary research on parenting. Then I discuss the reasons why Vandell and I came up with different answers to the following questions:

1. Do parenting behaviors have any lasting effects on child outcomes?

2. Are learned social behaviors, and the cognitions and emotions associated with them, specific to the social context in which they were learned? Vandell correctly attributed to me the proposition that dyadic relationships are context-specific and do not generalize, but this proposition follows from a broader one: that all learned behaviors are context-specific.

3. What are the experiences that do have lasting effects on child outcomes, and what role do peers play in these experiences?

Although my answers to Questions 1 and 3 have received the most attention, my answer to Question 2 is the most novel part of my theory and may prove to be the most important. Even if future research does not uphold my answer to Question 3 (and it might not), there is already ample evidence to support my answer to Question 2. The belief that children find it easier to reuse something they learned in the past than to learn something new -- something tailored specifically to their current situation -- has long been an impediment to scientific progress.

One previous impediment, the denial of genetic influences on behavior, has been overcome. Contemporary developmentalists admit that genes influence child outcomes; this is not an issue on which Vandell and I disagree. Our disagreement is about environmental influences on development. "Environmental influences on what?" Vandell (2000, p. 699) asked, and she had a good point. It was a mistake on my part to use one word -- personality -- to cover all the outcome variables that group socialization (GS) theory was designed to explain. In fact, the theory applies to the outcomes of two distinguishable processes: personality development and socialization. Both occur in childhood and result in habitual patterns of behavior (primarily social behavior) not readily modifiable in adulthood, but the term personality refers to ways in which people chronically differ in behavior, whereas socialization refers to a process by which children adopt culturally approved patterns of behavior and end up behaving more alike. As I will explain, GS theory postulates different processes to account for these two aspects of development.

Vandell claimed that I have constructed a "straw man" -- that I have accused developmentalists of holding a view of development that has been dead for years. But "this view of the child as a blank slate on which parents are free to create" (Vandell, 2000, p. 700) is not what I accused developmentalists of holding. The "blank slate" view of development -- the idea that babies are born with no innate knowledge, no built-in predispositions -- is indeed long dead. The mental set I call the "nurture assumption," on the other hand, is still very much alive.¹ It is the pattern of thought that underlies the use of the loaded term nurture -- a word that means "to take care of" or "to rear" -- as a substitute for the neutral term environment. Environment poses a question; nurture thinks it already knows the answer.

Though developmentalists recognize that parents are not the only influences on children and that peers, teachers, neighborhoods, and culture also have an impact, the nurture assumption makes its presence known even when these other influences are being considered. Yes, peers are important, but parental influence is seen as primary because early experiences with parents supposedly influence later relationships with peers (Sroufe, Egeland, & Carlson, 1999; Vandell, 2000, pp. 703, 705), and the right sort of parenting can supposedly keep an adolescent from joining the wrong sort of peer group (Lykken, 1997; Steinberg, 1997). Yes, the culture is important, but culture is thought of as something that is passed on from parents to children. The world outside the home is seen as influencing the child through its effects on the parents; according to Bronfenbrenner (1986), the question is, "How are intrafamilial processes affected by extrafamilial conditions?" (p. 723).

"To put it plainly," Vandell (2000) wrote, "I am hard pressed to find a single developmental psychologist who subscribes to the nurture assumption as articulated by Harris, in which parenting has the same effects on all children, regardless of temperament, developmental status, or competence" (p. 700). I have never accused developmentalists of subscribing to anything of the sort.² In my 1995 article I stated that there was good evidence that parents do not treat their children all alike and then discussed the hypothesis (citing Hoffman, 1991, and Kagan, 1984) that even if the parents do treat their offspring alike, "different children might experience or interpret environmental events in different ways" (Harris, 1995, p. 460). I revisited that hypothesis in my book (Harris, 1998) and again concluded that it could not resolve the discrepancies between a growing volume of data and the beliefs of most psychologists.

The view of development that Vandell defended is well known to me: Infants are not all alike at birth but come with built-in predispositions; a parenting style that works well for one child might not work well for another; parent-child relationships are a two-way street; and there are many influences on a child's development but family influences are generally the most important. This is precisely the view I held for many years and faithfully relayed to the readers of the child development textbooks I coauthored (e.g., Harris & Liebert, 1991). In 1994 I abandoned the last of these propositions (but retained the other three) after a long, hard look at the evidence. As Wade observed (quoted in Tavris, 1998), the attempts to fit the discrepant evidence into the prevailing viewpoint were "like trying to fit a double-sized sheet onto a queen-sized bed. One corner fits, but another pops out" (p. 14). The strenuous efforts still being made to keep that sheet in place are testimony to the power of the nurture assumption.

For example, developmentalists tend to be very critical of evidence that goes against the assumption, far less critical of evidence that supports it. Vandell made it clear in her discussion of Loehlin's (1997) work that she understands the hazards of using the same informants to provide the data on both sides of a correlation; response biases (shared method variance) can produce spurious correlations. Yet she cited the work of Steinberg, Lamborn, Darling, Mounts, and Dornbusch (1994) as support for the statement that "adolescents' problem behaviors were related to parenting styles" (Vandell, 2000, p. 701). Steinberg et al. (1994; Lamborn, Mounts, Steinberg, & Dornbusch, 1991) found significant correlations between adolescents' descriptions of their parents' child-rearing methods and replies by the same adolescents to questions about their own behaviors and attitudes.³ Evidence that such within-informant correlations can be misleading was provided by Pike, Reiss, Hetherington, and Plomin (1996), who asked both parents and adolescents to report on the parent's negative behavior toward the adolescent and on the adolescent's antisocial behavior. The researchers found significant correlations between the parents' two reports and the adolescents' two reports but negligible correlations between informants.

Response biases are only one of the sources of spurious or misleading correlations between parental behaviors and child outcomes discussed in my book (Harris, 1998); six others are listed in my article (Harris, 1995, pp. 477-479) and are also discussed in the book. Though direct genetic effects and gene-environment correlations may account for many of the reported associations between parenting and child behavior, I have never claimed (and it is highly unlikely) that they account for all of them.

Although genetic factors are not the only source of misleading correlations, they are the most pervasive. I will therefore consider them first.

Contemporary developmentalists acknowledge that babies are born with distinctive characteristics that make certain developmental outcomes more likely. It is also accepted that parental behavior is, in part, a reaction to the child's past and current behavior (Bell, 1979; Scarr & McCartney, 1983) and physical appearance (Langlois, Ritter, Casey, & Sawin, 1995). Every published report of a correlation between parental behavior and child outcome now contains a disclaimer admitting that the direction of effects is uncertain and that the correlation could be due in part to what I call a child-to-parent effect (Harris, 1998, p. 27), a type of evocative gene-environment correlation (Scarr & McCartney, 1983). It is never admitted, however, that the correlation could be due entirely to the child-to-parent effect and that the parent-to-child effect could be zero, even though that possibility cannot be rejected because neither effect has been measured.

Another kind of gene-environment correlation -- the passive kind -- is unlikely to be mentioned at all. There is a perplexing gap between the understanding that babies are born with innate characteristics and the understanding that they inherit these characteristics from their parents. In fact, the word heredity is seldom used nowadays; it has been replaced by words like nature and genetic, which acknowledge children's genes without acknowledging their source. Children share 50% of their genes with each of their biological parents, which means that for genetic reasons alone, children born with a predisposition to be timid (Kagan, 1994) are more likely to be raised by timid parents, and children born with a predisposition to be aggressive (Deater-Deckard & Plomin, 1999) are more likely to be reared by aggressive parents.

Longitudinal designs. As Vandell (2000, pp. 700-701) noted, one of the ways researchers attempt to deal with ambiguities in their data is by studying children's development over time. The reasoning behind the longitudinal method is that by "controlling" for the child's characteristics at Time 1, changes between Time 1 and Time 2 can presumably be attributed to environmental influences. For example, in an unpublished study by Arcus (1991) described by Kagan (1994), infants were tested at 4 months and judged to have "high reactive" or "low reactive" temperaments. High-reactive babies are more likely to become timid children, but not all the high-reactive babies in this sample were judged to be timid at 21 months. The failure of some of the babies to fulfill the prediction of the early test was attributed by Kagan (1994, pp. 204-206) to the way their mothers treated them in the intervening 17 months: High-reactive infants whose mothers used a relatively tough child-rearing style were less likely to become timid toddlers than those whose mothers used a wimpy style.

The problem is that a measure made at one stage of development is not an adequate control for genetic influences at a later stage of development, for two reasons. First, the early measure may lack validity: It may not do a good job of measuring whatever it is that influences behavior at the later stage. Second, genetic influences can produce change as well as stability. An individual may have blond hair at age 5, brown hair at age 25, and no hair at age 75, and these changes are ordained by the genes. As we get older we become more like our parents because our parents, too, went through these changes. Longitudinal studies using genetically informative designs (Plomin et al., 1993; Reiss, 2000) have shown that genetic influences that affect behavior or adjustment at one age are not necessarily the same genetic influences that affect these variables at another age. There are genes that kick in early and genes that bide their time.

Intervention studies. A rigorous way to test hypotheses about development is to do an experiment: an intervention. As Vandell (2000, p. 701) mentioned, it is the random assignment of families to the intervention or control conditions that gives this design its power. Because the families in the two groups can be regarded as genetically equivalent, the design provides a control for genetic factors. Thus, intervention studies are an excellent way to test predictions made by different theories of development. From propositions I will discuss later, GS theory generates the following prediction: Home-based interventions aimed at improving parents' child-rearing style can improve children's behavior at home, and school-based interventions can improve children's behavior in school, but home-based interventions will not improve children's behavior in school.

Unfortunately, due to methodological shortcomings, the majority of published intervention studies cannot be used to test this prediction. Some lacked a control group (e.g., Patterson, 1974). In others, the child's behavior was assessed only at home (e.g., Johnson & Christensen, 1975), usually on the basis of parents' reports. In Webster-Stratton's (1984) study, hailed by Patterson, Dishion, and Chamberlain (1993, p. 57) as a "major contribution to the field," the children's behavior was rated only at home in the postintervention assessment. It was rated at school in the 1-year follow-up, but there was no control group for the 1-year follow-up. Perhaps it is too harsh to call a failure to obtain an objective measure of the child's behavior outside the home a "methodological shortcoming," but the philosophy behind the intervention work of Patterson and his colleagues (e.g., Patterson et al., 1993) is that antisocial behavior outside the home has its origins in the home.

Still other studies cannot be used to test the predictions of GS theory because the home-based intervention was combined with a school-based intervention -- a method that can improve the child's behavior in both places but that makes it impossible to assess the effects of the home-based intervention on the child's behavior in school. In Webster-Stratton's (1998) recent study, for example, the intervention took place in a Head Start classroom as well as in the home; teachers as well as parents were taught strategies for managing children's troublesome behavior. The postintervention assessment of the children's classroom behavior was made by the same teachers. ⁴

A study that meets all the criteria for an experimental test of my prediction was carried out recently by Forgatch and DeGarmo (1999). There was a control group, the intervention occurred only at home, and the children's behavior was assessed both at home and at school. Ratings at school were made by teachers who were blind to the children's experimental group assignment. But, "contrary to expectations" (p. 718), the researchers found no differences between the teachers' postintervention ratings of children in the intervention and control groups.

The results that Forgatch and DeGarmo (1999) reported in their abstract came from a post hoc data analysis -- a path analysis, which showed that "improved parenting practices correlated significantly with improvements in teacher-reported school adjustment" (p. 711). The path analysis turned an experimental study into a correlational study and showed, as correlational studies almost always do, that the "better" parents in the group had "better" children. The point of having a control group is that it provides a control for genetic influences, because the two groups being compared are presumed to be genetically equivalent. But this advantage is lost if researchers start making comparisons within a group. The select subgroup of parents whose parenting practices improved cannot be presumed to be genetically equivalent to those who did not improve. The fact that there was no mean difference in school behavior between the children in the intervention and control groups means that the intervention did not produce a net improvement in the school behavior of the children in the intervention group. (The results reported by Cowan & Cowan, in press, are also based on path analyses. For my critique of Dishion & Andrews, 1995, see Harris, in press.)

The outcome of many years of research on parent-training interventions by Forehand and his colleagues (e.g., Forehand & Long, 1988) was summed up by Wierson and Forehand (1994). Parents, they said, found the training to be effective in improving their children's behavior at home. "However," Wierson and Forehand admitted, "research has been unable to show that child behavior is modified at school" (p. 148). Similar conclusions have been reached in regard to programs designed to improve the school adjustment and academic performance of children living in poverty: "There is little evidence that parenting programs produce the hoped-for linkage between changed parent behaviors and improved child outcomes" (St.Pierre & Layzer, 1998, p. 7; see also White, Taylor, & Moss, 1992; Zaslow, Tout, Smith, & Moore, 1998).

Gene-environment interactions. Behavioral genetic studies estimate genetic and environmental influences on developmental outcomes by measuring pairs of siblings. The results of these studies are monotonously consistent, regardless of the outcome measure used and the kinds of sibling pairs who participate -- twins or ordinary siblings, reared together or apart, biologically related or not. Genetic effects generally account for 35% to 65% of the variance among the participants, the effects of being reared in the same home account for 0 to 10%, and the balance remains unaccounted for (see Harris, 1995, 1998; Plomin, DeFries, McClearn, & Rutter, 1997; Rowe, 1994). With few exceptions, being reared in the same home does not make siblings of any kind more alike.

These results show not only that siblings are different but that the home environment has had little or no net effect on the measured outcomes. If being reared by conscientious parents, for example, tended to make children more conscientious, then two children reared by conscientious parents should, on average, both be more conscientious than two reared by careless parents. Therefore, two children reared in the same home should be significantly more alike in conscientiousness than two reared in different homes, which is exactly what the studies do not find (Bouchard, 1994). The same results also rule out the possibility that being reared by conscientious parents makes children less conscientious on average. The bottom-line effect of the shared home environment on conscientiousness is not noticeably different from zero. That is why knowledgeable developmentalists (e.g., Collins, Maccoby, Steinberg, Hetherington, & Bornstein, 2000; Vandell, 2000) are now showing a diminished interest in looking for main effects and a keen interest in gene-environment interactions. If being reared by conscientious parents makes children with one kind of temperament more conscientious and those with another kind of temperament less conscientious, then parents might have an influence after all, even in the absence of main effects.

There are four problems with this approach. First, any explanation that is used to account for the unexplained variance -- the nongenetic differences between children reared in the same home -- should apply to all kinds of sibling pairs, including identical twins, because the conclusion about the ineffectiveness of the shared environment holds for all of them. The environmental influences that fail to make ordinary siblings more alike also fail to make identical twins more alike (Bouchard, Lykken, McGue, Segal, & Tellegen, 1990). About 50% of the variance in adult personality is left unexplained by identical genes plus the experiences shared by reared-together twins.

Consider a pair of reared-together identical twins both born with high-reactive temperaments. In all probability their mother will treat them similarly (Reiss, 2000); it is unlikely that she would be tough with one, wimpy with the other, or leave one at home and take the other to the playground. Can we, then, attribute the differences in the twins' adult personalities to the fact that they responded differently to their mother's treatment? But why should they respond differently if they were born with the same (or very similar) temperaments? Gene-environment interactions cannot account for the personality differences between identical twins because identical twins have the same genes. Any phenotypic differences between them have to be due to differences in environment.

The second problem is that the environmental factors examined in the studies usually cited as evidence for gene-environment interactions -- for example, high family conflict (Bergeman, Plomin, McClearn, Pedersen, & Friberg, 1988), low socioeconomic status (Cloninger, Sigvardsson, Bohman, & von Knorring, 1982), or criminal parents (Mednick, Gabrielli, & Hutchings, 1987) -- cannot account for differences between siblings reared together. These factors differ for children reared in different families but not for those reared in the same family. The study by Bergeman et al., cited by Vandell, involved identical twins reared apart. In an adoption study using within- rather than between-family comparisons, Plomin, DeFries, and Fulker (1988) performed hundreds of analyses in a search for significant gene-environment interactions and found only about as many as would be expected by chance.

When found, gene-environment interactions often fail to hold up (Wachs & Plomin, 1991), and that is the third problem. The interactions reported by Bergeman et al. (1988) have not been replicated. However, one type of gene-environment interaction has proven to be replicable, which brings us to the fourth problem: Demonstrating a gene-environment interaction is not the same as demonstrating parental influence -- environment does not necessarily mean nurture. Studies have shown that children who inherit predispositions toward criminal behavior (Cloninger et al., 1982; Mednick et al., 1987), schizophrenia (Tienari et al., 1994), or alcoholism (Cloninger et al., 1982; McGue, 1999) are more likely to fall prey to these risks if they are reared in adverse circumstances. But these studies always involve between-home rather than within-home comparisons and cannot pinpoint which of the many differences between a favorable home and an unfavorable one is having the effect. It is likely, for instance, that these homes differ in socioeconomic status (whether or not this variable was recorded), which means that the unfavorable home is apt to be more crowded and located in a different sort of neighborhood. Early exposure to infections or toxins has been proposed as a possible etiological trigger for schizophrenia (Andreasen, 1999).

For alcoholism (McGue, 1999) and criminal behavior (see Harris, 1998), the evidence strongly suggests that the operative environmental influences are in the neighborhood, not in the home. Shared environmental influence has been found to contribute substantially to adolescent delinquency (Rowe, 1997) -- this is one of the exceptions -- but siblings who share a home also share a neighborhood and perhaps a peer group. Rowe has shown that the likelihood that two adolescent siblings will both be delinquent (or both be law-abiding) is greater if they are close in age and if they spend a lot of time together -- factors that make it more probable that they belong to the same peer group.

Within-family environmental differences. Vandell and I and psychologists of every persuasion agree that each child in the family grows up in a unique home environment, as a result of differential treatment by parents and asymmetric relationships between siblings. Can these within-family environmental differences account for the unexplained differences between siblings?

It is a difficult question to answer. If parents treat their children differently and the children are different, how can we tell if the parents are causing the differences or reacting to preexisting differences between the children? Vandell (2000) cited a study by Rodgers, Rowe, and Li (1994) to support her carefully phrased statement that "differences in home environments for children in the same family are related to differences in child and adult mental health outcomes" (p. 702), but as Rodgers et al. (1994) admitted, "Of course, the causal direction is ambiguous" (p. 381).

A major study designed to resolve the ambiguity was launched in 1987; its results are described in an important new book by Reiss (2000). Reiss and his colleagues studied 720 pairs of same-sex adolescent siblings (identical and fraternal twins and full, half and stepsiblings) over a 3-year period, from early to mid-adolescence. The adolescents' behavior and adjustment, and the behavior of their parents, were judged by observers, the parents, and the adolescents themselves; these multiple sources of information were combined to produce measures of unusual reliability.

Reiss has confessed that he was "shocked" by the results of his study (Paul, 1998, p. 46). The results are indeed shocking to believers in the nurture assumption (and Reiss, in spite of everything, remains a believer). Virtually all correlations between parental behaviors and child outcomes were accounted for by genetic factors. The parents did indeed treat their children differently but they were reacting to genetic differences among the children, rather than causing the differences. The study confirmed the finding that the environment not shared by siblings was by far the largest (in many cases, the sole) nongenetic contributor to the adolescents' behavior and adjustment, but it eliminated all of the following as possible sources of nonshared environmental influence: "differential marital conflict about the adolescent versus the sib, differential parenting toward siblings, and asymmetrical relationships the sibs construct with each other" (Reiss, 2000, p. 407). Ruling out the last factor indicates that differences in age -- that is, birth order -- cannot account for the differences between siblings, a finding consistent with the results of a recent meta-analysis by Turkheimer and Waldron (2000).

Turkheimer and Waldron (2000) and Reiss (2000) came to the same conclusion: that developmentalists and behavioral geneticists alike have so far been unable to identify the important environmental influences on development, the sources of the nongenetic variation in personality and behavior. These researchers also agreed that it is fruitless to try to find them in studies that provide no way of controlling for genetic influences.

As I stated earlier, Vandell's claim that I attribute all correlations between parenting and child outcomes to direct genetic effects or gene-environment correlations is incorrect. In this section and the next, I discuss two other entries from my list of six "confounding factors that can be mistaken for evidence of effects of the home environment" (Harris, 1995, p. 477).

The results of Reiss's (2000) study seem to imply that genetic factors can indeed account for almost all correlations between parental behaviors and child outcomes, but the participants in that study were overwhelmingly white and middle class. In studies that mix together families from different ethnic groups or socioeconomic classes, there is another source of misleading correlations between parents and offspring: the effects of shared culture or subculture.

The fact that parents and children are usually members of the same culture is generally attributed to parental influence; according to the prevailing view, children learn culturally appropriate behaviors and attitudes from their parents. Group socialization theory offers an alternative explanation: Culture is not passed directly to individual children from individual parents but to a group of children from a group of parents -- their own parents and the parents of their peers. Because most children are reared by parents who belong to the same culture as the parents of their peers, most children end up sharing a culture with their parents. Only in atypical circumstances is it possible to test the predictions made by these alternative views of cultural transmission (see Harris, 1998, chap. 9).

Most contemporary studies include participants from various socioeconomic classes and ethnic groups. The effects of shared culture -- what I call parents'-group-to-children's-group effects (Harris, 1995) -- can produce misleading results because different cultural groups are likely to vary both in child-rearing styles and in the behavioral norms of the children's peer groups. For example, African American parents tend to use more physical punishment than European American parents (Deater-Deckard, Dodge, Bates, & Pettit, 1996), and African American children who live in low-income neighborhoods tend to be more aggressive than European American children (Kupersmidt, Griesler, DeRosier, Patterson, & Davis, 1995). But African American children are not significantly more aggressive than European American children if they live in neighborhoods composed primarily of middle-class European Americans (Kupersmidt et al., 1995). Because most African American children do not live in middle-class European American neighborhoods, researchers can easily get the impression, if data from different cultural groups are combined without taking neighborhood into account, that punitive parenting produces aggressive children. Or, if many Asian American families are included in the study, researchers may conclude that punitive parenting leads to academic achievement (though to be honest I have not seen anyone make this claim). Asian American parents, too, use more physical punishment than European American parents (Chao, 1994, Table 1; Kelley & Tseng, 1992).

As shown by the last example, if data from different subcultural groups are considered separately, the results may be difficult to explain in terms of prevailing theories. The conclusions drawn by Steinberg and his colleagues (e.g., Steinberg et al., 1994) about the beneficial effects of authoritative parenting on adolescents appear to hold only for White families, not for African Americans or Asian Americans (Steinberg, Dornbusch, & Brown, 1992). I have previously offered a possible explanation for these findings (Harris, 1995, pp. 479-480; 1998, pp. 47-49).

One reason why people believe so strongly in the nurture assumption is that they can see parents influencing their children. They observe the child of permissive parents being obnoxious, and the child of abusive parents looking cowed and fearful, in the presence of their parents. The fundamental attribution error (see Myers, 1999) causes observers to assume that these children will be obnoxious or fearful in other social contexts too. This section describes how that assumption has misled researchers; in a later section I will summarize the evidence against it.

In an article in a medical journal, Straus, Sugarman, and Giles-Sims (1997) reported that children whose mothers administered frequent spankings tended to become more aggressive over a 2-year period. The researchers concluded that a reduction in parental punitiveness could "reduce the level of violence in American society" (p. 761). Largely unnoticed, another article on the effects of punishment, in the same issue of the same journal, came to a different conclusion. Gunnoe and Mariner (1997) used a longitudinal design similar to that of Straus et al. but found no overall tendency for children who were spanked to become more aggressive. The apparent inconsistency can be explained by context effects. The method used by Straus et al. to assess aggressiveness (mothers' reports) was a measure of how the children behaved at home; the method used by Gunnoe and Mariner (children's reports of how many fights they got into at school) was a measure of how they behaved outside the home. The proposition that the effects of spankings are limited to the context in which they were administered makes both results understandable.

Context effects also solve the puzzle of birth order. Why do most psychologists continue to believe that birth order has noticeable and lasting effects on personality, even though most studies of adult personality offer no support for this belief (Harris, 1998, Appendix 1)? The answer is that birth order effects are real, but they are tied to the context of the family. We ordinarily know the birth order only of people we are close to, and these are the people we are most likely to see with their parents and siblings. Birth order studies in which parents are asked to rate their children's personalities, or adults are asked to compare themselves with their siblings, generally do yield significant birth order effects; studies that use other methods generally do not (Harris, 2000a). In a study in which pairs of siblings were rated both by parents and by teachers, parents judged the older sibling to be more aggressive than the younger one, but teachers judged them to be about the same (Deater-Deckard & Plomin, 1999).

The assumption that experiences at home affect behavior or adjustment in other contexts also causes researchers to overlook the fact that family misfortunes such as divorce have repercussions on children's lives outside the home and to assume that adverse outcomes are the results of experiences at home. A parental divorce often involves moving to a different residence, and moving disrupts the child's life outside the family. Even when other demographic differences are controlled, children who have experienced frequent residential moves have higher rates of social, behavioral, and academic problems (Eckenrode, Rowe, Laird, & Brathwaite, 1995; Wood, Halfon, Scarlata, Newacheck, & Nessim, 1993). McLanahan and Sandefur (1994) found that controlling for two factors -- number of residential moves and differences in family income -- can erase most of the differences in outcome between children reared in single-parent and two-parent families. (For my explanation of the remaining differences, see Harris, 1998, pp. 305-309; 2000b.)

I have focused on the evidence that appears to favor the prevailing view of parental influence; for the evidence against it, see The Nurture Assumption (Harris, 1998). The evidence that appears to favor the prevailing view is inadequate to support it, because all reported correlations between parenting behaviors and child outcomes reported by developmentalists could be due to some combination of direct genetic effects, gene-environment correlations, parents'-group-to-children's-group effects, context effects, and response biases. When Vandell (2000) pointed out that the contribution of confounding effects to these correlations "does not negate the possibility that parents' behaviors also influence or affect children" (p. 702), she was suggesting that I cannot use the ambiguity of the evidence to disprove the theory of parental influence, and she was right. But in most realms of scientific inquiry, theories are tested against the null hypothesis and accepted only when there is sufficient unambiguous evidence to reject it. In developmental psychology, it works the other way: The theory that parental behaviors have effects on child outcomes is accepted a priori, and someone who dares to question it is required to back up her skepticism with evidence that the null hypothesis is true (which, of course, is impossible to produce). My position -- admittedly an unpopular one (N = 1) -- is that those who wish to find out more about child development should stick to the null hypothesis of zero parental influence until it can be rejected on the basis of solid evidence.

I do not deny that parents can influence their children's behavior in any context in which they habitually interact, or that the behaviors (and their associated cognitions and emotions) acquired in these contexts may last a lifetime. What I deny is that these behaviors are automatically carried along to other contexts and retained there, whether or not they are appropriate in the new context. Sometimes they are appropriate and the child can retain a previously learned behavior; a child who learned to speak English at home does not have to learn it again in the day-care center. But often these behaviors are inappropriate, and in this case the home behaviors are discarded and new ones acquired. Because developed societies require very different behaviors in the home and in public (Dencik, 1989), and people everywhere make distinctions between kin and nonkin (Pinker, 1997), children behave differently in different contexts (Goldsmith, 1996).

Group socialization theory allows for the possibility that there may be transitory effects of the home environment. When children are placed in a new situation, they have no recourse but to fall back on what they learned previously at home or elsewhere; they will use whatever seems most relevant to their current circumstances. Vandell (2000, p. 702) cited a study by Kochanska (1997) that illustrates this point. Kochanska tested 4- and 5-year-olds in novel situations; in one test, for example, the children were left alone in an unfamiliar laboratory room after a researcher showed them how to play a game and told them not to cheat. Kochanska found that children were more likely to display a "conscience" if their mothers had used a method of child-rearing that was well-suited to their child's temperament. I have no doubt that mothers who adjust their methods of child-rearing to their child's characteristics will have children who are better behaved at home, and that patterns of behavior acquired at home will be used by default in novel situations that resemble the home (in Kochanska's laboratory, the mother was nearby and no peers were present). It takes repeated experiences in a new situation to acquire new behaviors.

The question, then, is: Does what the child learned at home generalize to familiar settings outside the home? Many years ago, Hartshorne and May (1930/1971) tested children's ability to resist temptations to lie or cheat at home, in school, and on the playground; they found only low correlations between honesty in the various settings. (The correlations were due to the fact that children with higher IQs and those from higher socioeconomic classes were somewhat less likely to cheat.) Hartshorne and May concluded that "the normal unit for character education is the group or small community" (p. 197), rather than the family.

Patterns of socialized behavior may be entirely learned (e.g., whether a child speaks English or Japanese) or influenced by heritable characteristics (conscientiousness, as well as IQ, might play a role in honesty). I will begin with the example of language and accent because it is easier to explain how my model works without the additional complication of genetic influences, and because there are ample data -- no need to rely on anecdote. Then I will turn to more complex cases in which genetic influences are involved and behavior is an outcome of personality as well as socialization.

In developed societies, most children learn their first language at home and bring this language with them to adulthood; they discover that the language of their home is also used outside the home, so they keep using it. Is this a parent-to-child effect or a parents'-group-to-children's-group effect? The answer becomes clear only when one considers children whose parents speak a language (or have an accent) that is different from the one used by the child's peers.

Under such circumstances, children bring with them to adulthood the language and accent of their peers, not those of their parents.⁵ Evidence comes from studies of hearing children reared by deaf parents and of deaf children reared by hearing parents (see Harris, 1998) and from studies of the children of immigrants (e.g., Bickerton, 1983). Bickerton interviewed adults whose parents had immigrated to Hawaii around the turn of the last century. Though the parents came from all over the world and spoke a variety of languages, their adult offspring all spoke the same language -- a creole, created in their childhood peer groups. No trace of their parents' language or accent was detectable in their speech; the language of their peers had become their "native language" (p. 119). Bickerton's findings also rule out the possibility that children get their language from the community as a whole, adults as well as peers. The adults in these communities used a pidgin, not the creole, to communicate with each other.

Insights into typical development can be gained by studying individuals who develop in atypical ways. Baron-Cohen and Staunton (1994) studied the accents of children and young adults with autism. These were the offspring of immigrants; their mothers spoke English with a foreign accent. The nonautistic siblings of these children spoke English without a foreign accent, but 83% of the children and young adults with autism had retained the foreign accent of their mothers. Baron-Cohen and Staunton attributed this outcome to "a lack of the normal drive to identify with peers" (p. 241).

The example of language demonstrates that, although many different factors may exert their influence on developmental outcomes, each may exert its influence only within its own domain (Bugental, 2000). The result is not a blend but what psycholinguists call code-switching -- switching back and forth between separate programs, the way the children of immigrants switch back and forth between the language of their home and the language they speak (without a foreign accent) outside the home. What makes it look as though developmental outcomes result from many different environmental influences blended together is the fact that most social behaviors are influenced by genes as well as by experiences. Genetic influences transcend context.

Supportive evidence comes from studies using a technique called multivariate genetic analysis, which makes it possible to estimate how much genetic and environmental influences each contribute to the correlations found between behavior in different contexts (Harris, 2000a; Saudino, 1997). For example, children who are shy in one social context are not necessarily shy in others, but some children are shy in every context (Rubin, Hastings, Stewart, Henderson, & Chen, 1997). Multivariate genetic analysis shows that the genetic component of shyness is responsible for the correlation across contexts; differences in shyness from one context to another are due primarily to environmental influences (Cherny, Fulker, Corley, Plomin, & DeFries, 1994).

Shyness is a personality trait rather than a socialized behavior. Personality theorists (e.g., Mischel, 1973) debated for years about the degree to which personality is situation-specific. The lack of agreement was the result, I believe, of a failure to distinguish between genetic and environmental components of personality. According to GS theory, consistencies in behavior across contexts are primarily due to the genetic component of personality.⁶ Modern personality tests yield stable measures because they are designed to find cross-contextual consistencies in behavior. McCrae and Costa (1999) recently concluded that the components of their five-factor model of personality reflect "endogenous basic tendencies" (p. 144) with a biological basis.

Most of the behaviors thought of as socialized (or unsocialized) are influenced by genetic factors as well as by learning, so correlational studies that provide no control for genetic factors can produce results -- correlations between a child's behavior in different contexts -- that look like generalization. The correlations, however, are often remarkably weak. Dishion, Duncan, Eddy, Fagot, and Fetrow (1994) studied a type of social behavior they called "coercive" (i.e., hostile, uncooperative, or bossy). They observed school-age children with their parents and with their peers and recorded instances of coercive behavior. Behavior in the two contexts was "only weakly correlated" (p. 260); the correlation was .19. As I explained in my book (Harris, 1998, p. 62), this correlation might be due to genetic influences. A child who has inherited a predisposition to be disagreeable or aggressive will have a tendency to behave that way both at home and on the playground. Though Vandell (2000, p. 701) has quoted out of context my statement about a correlation of .19 being "all but useless,"⁷ the correlation is weak enough that genetic influences -- which, when measured, tend to bite off sizable chunks of variance -- could account for all of it. It provides scant support for the proposition (Patterson et al., 1993) that coercive behavior is learned at home and then carried to the playground.

The best evidence against that proposition is that interventions designed to teach parents better ways of dealing with their noncompliant children do succeed in reducing the children's coercive behavior at home but do not produce significant improvements in their behavior in school (Wierson & Forehand, 1994). Wierson and Forehand's conclusion supports my proposition that behaviors acquired at home do not generalize to familiar outside-the-home settings.

Vandell (2000) wrongly attributed to me the belief that infants' early relationships "have no import for psychological characteristics after infancy" (p. 700). Actually, I have said (Harris, 1995, p. 477n; 1998, p. 153) that the formation of an early attachment -- any attachment -- appears to be necessary for optimal social development, just as early exposure to a language -- any language -- is necessary for optimal language development. What has no import for later psychological characteristics, according to GS theory, is how well or poorly these early relationships go. Human babies, I maintain, are smart enough to keep their various relationships straight. The expectation that Mommy will come running when they cry does not lead them to expect a similar response from babysitters, siblings, or peers.

The various relationships of a given child will all be affected by genetically influenced characteristics such as temperament and appearance. Nevertheless, the evidence indicates that these relationships are, to a surprising degree, independent of each other. Infants who behave in a somber, subdued fashion in the presence of their depressed mothers behave normally in the presence of nondepressed caregivers; their subdued behavior is "specific to their interactions with their depressed mothers" (Pelaez-Nogueras, Field, Cigales, Gonzalez, & Clasky, 1994, p. 358). Secure attachments to one caregiver are not good predictors of secure attachments to another (Cox, Owen, Henderson, & Margand, 1992; Goossens & van IJzendoorn, 1990); nor are secure attachments to parents good predictors of successful relationships with peers (Howes, Matheson, & Hamilton, 1994; Lamb & Nash, 1989). The quality of sibling relationships does not predict the quality of relationships with peers (Stocker & Dunn, 1990); sometimes the correlations are negative (East & Rook, 1992; Stocker, 1994).

Though at first glance it seems counterintuitive that children should fail to apply what they learned in their sibling relationships to their relationships with peers, the tendency not to generalize is adaptive. Children who are dominated by older siblings at home would be handicapped, not helped, by the expectation that they will also be dominated by their peers. In fact, even in early childhood, later-born children are no more likely than firstborns to adopt a submissive role in peer interactions (Abramovitch, Corter, Pepler, & Stanhope, 1986).

The finding that sibling and peer relationships are independent provides strong support for my proposition that patterns of social behavior acquired in dyadic relationships are context- and relationship-specific. The extension of this proposition to long-term personality development is supported by the evidence that birth order effects do not show up on most tests of adult personality but do show up on tests that evoke the family context (Harris, 2000a).

Evolutionary psychology (Cosmides & Tooby, 1992; Geary & Bjorklund, 2000; Pinker, 1997) has given us a new view of the human mind -- a mind composed of modules that each serve a specific adaptive purpose. According to GS theory, dyadic relationships and group processes are governed by separate mental modules (Harris, 1995; see also Bugental, 2000). The two modules can issue conflicting commands; they produce different behaviors and emotions (compare speaking to a group with speaking to an individual). The capacity to form dyadic relationships is present from birth; identification with a group develops more slowly.

"Harris has proposed a highly compartmentalized self," Vandell (2000, p. 703) observed correctly. The visual system is also highly compartmentalized -- there are modules for detecting movement, for location, for recognition of faces, and so on (Farah, 1992; Rao, Rainer, & Miller, 1997) -- yet we see the world as a seamless whole.

Evidence that dyadic relationships and group processes are handled by different mental modules was cited by Vandell: Children's friendships and their status in the peer group have different short- and long-term correlates (e.g., Bagwell, Newcomb, & Bukowski, 1998; Vandell & Hembree, l994). Because qualities such as intelligence, cheerfulness, and a pleasant appearance are assets in every area of a child's life, success in these two domains of social endeavor is correlated (and success in either is correlated with academic performance). However, there are many children who have close friendships but low status in the group and many with the opposite pattern, and the two domains make independent contributions to children's satisfaction with their lives (Parker & Asher, 1993; Vandell & Hembree, 1994).

Having a friend undoubtedly makes a child happier and less lonely, but is there any good evidence that dyadic relationships with peers have long-term effects on personality or socialized behavior? As Vandell (2000, p. 705) acknowledged, most studies of children's friendships suffer from cause-or-effect ambiguities. Most also fail to distinguish between friendships and group status. Of the studies Vandell cited, only Bagwell et al. (1998) distinguished the two domains and looked at their long-term correlates. The results were consistent with GS theory: Only peer acceptance or rejection was "uniquely associated with variations in overall life status adjustment" in adulthood; friendship had "unique predictive implications only for positive relations with family members" (Bagwell et al., 1998, p. 150). Childhood friendships also made a small unique contribution to "general self-worth" in adulthood, but this measure was an assessment of self-perceived competence in eight different areas, five of which would be affected by success in the relationship domain.

If relationships with parents and siblings -- children's earliest and most enduring relationships -- have only context- and relationship-specific effects, the implication is that the same is true of dyadic relationships with friends. The evidence Vandell presented is inadequate to support her belief that children's friendships (as opposed to group acceptance or status) have long-term effects on developmental outcomes.

Socialization, as I noted in the introduction, makes children more alike; personality refers to ways in which they differ. According to GS theory, socialization results from assimilation to a group and contrast effects between groups. Nongenetic differences in personality, on the other hand, are attributed primarily to differentiation within groups.

Cultures differ in the social behavior they demand; thus, children have to learn through experience how people like them are expected to behave in their society. The problem is that acceptable behavior in every society differs for children and adults. Children who imitated the behavior of their parents would not make a success of childhood (Harris, 1999).

According to GS theory, children solve this dilemma by means of a cognitive process called self-categorization (Turner, 1987). Children sort people into cognitive categories such as kid and adult or boy and girl (the relevant categories depend upon the social context), and then decide which one they belong in. The "group" in GS theory is actually a social category. Its members need not congregate in one place or even interact; a child can identify with a group even if its members reject her.

When people identify with a group, they take on its behaviors and attitudes (Turner, 1987). Thus, according to the theory, the way children learn to behave outside the home is by identifying with a group of others they perceive to be similar to themselves and taking on the behaviors and attitudes of that group. To a large extent, at least in the early years, this means taking on the behaviors and attitudes the majority of the children learned at home -- a parents'-group-to-children's-group effect that can be identified as such only by observing the small minority of children who learned something different at home (and controlling for heredity).

This view of socialization agrees with descriptions of childhood in hunter-gatherer and tribal societies. According to Eibl-Eibesfeldt (1989), once children in these societies have been weaned, they spend most of their time in the local play group. "Thus," he observed, "the child's socialization occurs mainly within the play group" (p. 600).

Children's groups. Vandell (2000, p. 703) regarded it as "surprising" that I have not relied more heavily on evidence documenting the influence of peer groups. Although I have cited many of the studies she mentioned, it is true that I do not rely on them heavily. Like the studies of parenting and friendship discussed earlier, most of these studies are correlational and provide no controls for genetic effects, so the results are uninterpretable. That is why the Robbers Cave study (Sherif, Harvey, White, Hood, & Sherif, 1961) is so important -- it was an experiment. The researchers selected 22 boys who were as much alike as possible and arbitrarily divided them into two groups. The two groups quickly developed contrasting behavioral norms. The Rattlers saw themselves as tough and manly; the Eagles saw themselves as pure and religious. The war that broke out between them is evidence of children's deep-rooted predisposition to ally with a group and to feel strongly about their group alliances, even when the groups are created arbitrarily.

Most children's groups form spontaneously and consist of individuals who were similar to begin with (Rowe, 1994), which makes it difficult to assess their effects. A study that circumvents this problem was carried out by Kindermann (1993). Vandell (2000) reported that Kindermann "observed that the motivational profiles of students' peer networks at the beginning of the school year predicted changes in the students' academic engagement from fall to spring" (p. 704), but Kindermann did more than that: He observed the effects of changes in group membership over the course of the school year. When a child switched from a clique of academic achievers to a clique of nonachievers (or vice versa), the child's attitude toward schoolwork shifted to match that of the new group. The design of this study provides a control for the effects of the child's IQ and the parents' attitudes, because neither was likely to change over the course of a school year.

When children divide or are divided into two groups, GS theory predicts that any preexisting differences between the groups will be widened by contrast effects between groups and assimilation within them. Thus, putting antisocial youth together for a group intervention is likely to make them more antisocial, a prediction that has recently been confirmed by Dishion, McCord, and Poulin (1999).

The teacher. Vandell seems to feel that I underestimate the importance of teachers, but there is a lengthy discussion in my book (Harris, 1998, pp. 243-247) about the dramatic impact of one of the teachers she mentioned, "Miss A" (Pedersen, Faucher, & Eaton, 1978). I attribute the influence of charismatic teachers like Miss A not to their ability to "set the stage for peer interactions" (Vandell, 2000, p. 706) but to their ability to forge a diverse bunch of students into a united group and give them a common goal. When children divide up into proschool and antischool factions, GS theory predicts that average academic performance will go down.⁸ Smaller classes reduce the likelihood that a division will occur, and children generally do better in smaller classes (Mosteller, 1995). Some teachers (I suspect that Miss A was one of them) have a magical ability to keep a classroom united.

What sent me on a quest for a new theory was the realization that existing theories could not account for the nongenetic variation in adult personality. Ironically, my solution to this puzzle has turned out to be the most speculative aspect of GS theory -- the part for which there is the least evidence. Though most of the existing data are consistent with the theory, they cannot be used to test it.

Within any group, individuals vary in status -- social power, not necessarily correlated with niceness -- and in the way they are typecast or labeled by their groupmates. These differences will occur even between identical twins who belong to the same peer group. If the differences in status and typecasting are persistent, GS theory predicts they will leave permanent marks on the personality. ⁹

One reason it is difficult to test this prediction with existing data is that standard sociometric ratings measure children's likability rather than their social power within their own clique. The second problem is that the direction of effects is ambiguous. Although many studies have found correlations between rejection (i.e., dislike) by peers and later personality problems (Bagwell et al., 1998; Parker & Asher, 1987), it is seldom possible to tell whether the rejection caused the later personality problems or early signs of these problems led to the rejection. Rowe (1989) has shown that sociometric status is influenced by genetic factors. Vandell (2000, p. 704) noticed that I made little use of the existing data and correctly guessed one of my reasons.

Because status and typecasting are determined partly by the group's reaction to characteristics an individual was born with (Rowe, 1989), the best way to test this aspect of GS theory is to study within-group processes in children's groups using genetically informative longitudinal designs. Such studies have not yet been done; kudos to Vandell for calling for them.

In the meantime, I have searched for ways of estimating group status that are independent of preexisting personality characteristics and have come up with one; unfortunately, it works only for boys. A boy's status in his peer group is determined in part by his size. Small, slow-maturing boys generally have lower status (Savin-Williams, 1979; Weisfeld & Billings, 1988). Consistent with my predictions, studies have shown that children who are small for their age tend to have more than their share of psychological problems (Richman, Gordon, Tegtmeyer, Crouthamel, & Post, 1986). The best study, though, is an old one, greatly in need of replication. Jones (1957) did a longitudinal study comparing males who were slow developers -- smaller than most of their peers throughout childhood -- with early developers. The two groups ended up almost equal in mean adult height, but there were significant differences in their adult personalities. The early developers were more dominant, relaxed, and poised; the slow developers were more impulsive, attention-seeking, and touchy. Perhaps personality, and not size per se, is the reason why tall men command higher salaries than short ones (Pinker, 1997).

I have no data other than anecdotes to support my prediction that how one is typecast or labeled by one's peers also has lasting effects on personality. This is a virgin area for patient researchers who are willing to wait 10 years or more to collect their data.

Do parenting behaviors have any lasting effects on child outcomes? Vandell (2000) views parenting effects as "conditional" (p. 700), but the evidence for conditional effects is no more convincing than the evidence for main effects. At best, one can answer this question only with "not proved." Despite herculean efforts by researchers, the efficacy of parenting has still not been proved.

Also not proved is the proposition that children learn things from one relationship or in one context that they automatically carry with them to new ones. A good deal of evidence supports my proposition that learned behavior is tailored to fit specific relationships and contexts. The implication is that if parenting behaviors do have lasting effects, the effects are specific to the context in which the behaviors were experienced. Because children are destined to play out their adult lives in other contexts, what they learn in these other contexts will be more important in the long run. Thus, the answer to the question "What are the experiences that do have lasting effects?" should be sought outside the child's home and family of origin.

The alternative view of development I have presented is not in accord with people's firmly held beliefs, but firmly held beliefs can get in the way of scientific inquiry. There is a large volume of evidence (I have only scratched the surface of it here) that does not fit the prevailing view. Group socialization theory can account for that evidence because that is what it was designed to do. Whether it can account for evidence produced by future research remains to be seen.

I thank Charles Harris, David Rowe, and Carol Tavris for their helpful comments on earlier versions of this article and Joan Friebely for her assistance in library research. Correspondence concerning this article should be addressed to Judith Rich Harris, 54 Crawford Road, Middletown, New Jersey 07748. Electronic mail may be sent to 72073.1211@compuserve.com.

1. "Reflecting her graduate training in traditional learning theory," Vandell stated, "Harris presented John Watson and B. F. Skinner as primary sources for the nurture assumption" (Vandell, 2000, p. 700). In fact, I laid the nurture assumption squarely at the door of Sigmund Freud (Harris, 1998, p. 4). As for my "training in traditional learning theory," perhaps if I had been more receptive to it, my graduate school career might have been more successful. My research, before I became interested in development, was in perception, information processing, and psychophysical scaling (e.g., Harris & Harris, 1984; Harris, Shaw, & Altom, 1985; Stevens & Harris, 1962 ). back

2. Vandell (2000) claimed that I accuse developmentalists of holding a deterministic view of parental influence; she also claimed that my own theory is deterministic (p. 699). Both charges are unfounded. All contemporary theories of development, mine included, are probabilistic: Certain genetic and environmental inputs make certain outcomes more or less likely. back

3. Steinberg and his colleagues found that teenagers who gave unfavorable reports of their parents' child-rearing methods and who reported more problem behavior (Lamborn et al., 1991) reported even more problem behavior 1 year later (Steinberg et al., 1994). Cowan, Powell, and Cowan (1998) pointed out that the use of a longitudinal design does not solve the problem of interpreting correlations: "The fact that a parent's behavior at Time 1 is highly correlated with a child's behavior at Time 2 does not establish parental influence, because the parent's Time 1 behavior may have been influenced by the child's behavior prior to the Time 1 assessment or by other factors in the parent's experience" (pp. 7-8). Interpreting such correlations as evidence of parental influence is even less justifiable when all the researchers know about the parent's behavior at Time 1 is what the child told them. back

4. When classroom behavior was judged by the teachers who took part in the intervention, there was a significant postintervention difference between the children in the intervention and the control groups. But a year later, when the children's behavior was judged by different teachers in a different classroom, no difference was found between the groups, even though the children in the intervention group were still behaving better at home (Webster-Stratton, 1998). back

5. I am assuming that the children were exposed to the peers' language before they reached puberty and that their parents' language was rare in their community. In communities in which there are many immigrants with the same native language, children may grow up bilingual, because they share both their languages with their peers. Likewise, they may retain a "foreign" accent, also shared with their peers. back

6. Not all similarities in behavior across contexts are due to genetic factors; they may be due to similar experiences. A child who has been rejected by potential social partners in several different contexts may be shy in all of them, just as a child who has learned that English is understood both at home and in school will speak English in both places. back

7. That statement was taken from a paragraph (Harris, 1998, p. 19) in which I was explaining, for the benefit of a general audience, that a correlation might be statistically significant and yet have little or no practical value in predicting behavior. I later (Harris, 1998, p. 62) returned to the correlation of .19 and showed how GS theory accounts for it. back

8. The antischool group will decline in academic performance; the proschool group may get better. Average academic performance will go down because ceiling effects limit the improvement of the proschool group. back

9. Differences in status and typecasting also occur in the home and also have persistent effects, but these effects are specific to the family context. Differences in status between nontwin siblings can be measured in adulthood in birth order studies that evoke that context (Harris, 2000a). back