A long line of empirical research has revealed that antisocial conduct, including early childhood behavioral problems, is heavily influenced by genetic factors. Results culled from behavioral genetic research, for example, have consistently indicated that at least 50 percent of the variance in antisocial phenotypes is attributable to genetic influences, with environmental factors explaining the remaining variance [1,2]. More specifically, the environments that have been identified as being the most salient to human development are nonshared environments, which are environments that are unique to each sibling [3,4]. Shared environments, or environmental factors that are equally experienced by all siblings from the same household, tend to have very little, if any, effect on behavioral phenotypes [5–8]. The differential effects of shared environments versus nonshared environments have been highlighted in the behavioral genetic literature. To illustrate, the results of four meta-analyses have converged to show that shared environments explain between 0–10 percent of the variance in antisocial phenotypes, while nonshared environments account for about 40 percent of the variance [9–12].

Given that the nonshared environment explains such a large percentage of the variance in behavioral phenotypes, there has been great interest in trying to identify the specific nonshared environments that are related to antisocial behaviors. Much of this research has focused on differential parental treatment, differential exposure to antisocial peer groups, and differential stochastic experiences [1,13–15]. But, perhaps one of the most salient nonshared environments—and one that has also been the source of a considerable amount of research attention—is the prenatal environment [16]. Compromised prenatal environments have been linked with a wide array of maladaptive outcomes, including reduced cognitive abilities, mental retardation, mental illnesses, and even antisocial behaviors [17,18].

One of the main ways in which prenatal environments are thought to affect later-life behavioral problems is through exposure to toxins, such as teratogens and neuroteratogens [16,19]. Exposure to cigarette smoke in utero represents what is perhaps the most widely researched toxin as it relates to childhood conduct problems [20]. Prenatal exposure to some of the agents found in cigarette smoke, especially nicotine, has been found to disrupt normal fetal brain growth, to be linked to a smaller head circumference, to produce learning and memory impairments, and to generally reduce executive control processes ([19], for a general overview see [21]). These findings are particularly important because there is a solid knowledge base linking neuropsychological deficits and executive dysfunctions to antisocial behaviors, including externalizing problem behaviors in childhood [17,22,23]. As a result, the effect that prenatal exposure to cigarette smoke has on behavioral problems may be mediated by neuropsychological impairments.

A considerable amount of research has examined the connection between prenatal exposure to cigarette smoke and behavioral problems in childhood, delinquency in adolescence, and crime in adulthood. In general, the results of at least 40 studies have revealed that exposure to cigarette smoke in utero is a risk factor for neurological dysfunction, as well as antisocial behaviors across all stages of the life course [22,24]. For example, Wakschlag and her colleagues [25] recently examined a nationally representative sample of respondents in order to explore the effects that prenatal exposure to cigarette smoke had on a range of antisocial outcomes. The study revealed that individuals exposed to cigarette smoke in utero were more likely to exhibit symptoms of oppositional defiant disorder and to have an early onset of delinquency.

The available evidence indicates a significant and positive association between prenatal exposure to cigarette smoke and later-life behavioral problems, leading some scholars to believe that this association is causal [25,26]. However, the methodologies that have been employed to estimate the effect of prenatal exposure to cigarette smoke on antisocial phenotypes are limited in their ability to rule out other rival explanations. Most notably, the association between prenatal exposure to cigarette smoke and behavioral problems later in life may be spurious owing to confounding. Ultimately, there are at least two reasons why this may be the case [27].

First, there are certain maternal characteristics that are correlated with both childhood behavioral problems and prenatal cigarette smoking. Specifically, maternal smoking is most common among women with antisocial traits [27]. Evidence is also accumulating that suggests the effects of prenatal smoking may cease to exert a significant effect on childhood behavioral problems after controlling for maternal antisocial behavior. Silberg and her colleagues [28] tested this possibility by analyzing a sample of over 500 male twins drawn from the Virginia Twin Study of Adolescent Behavioral Development. The findings of the study revealed that after controlling for familial factors, including maternal antisocial characteristics, prenatal smoking was no longer significantly associated with antisocial behavior in children.

The second reason that the relationship between maternal smoking and behavioral problems may be confounded is because of genetic factors. Specifically, researchers have suggested that maternal cigarette smoking during pregnancy might signify an underlying genetic risk for antisocial behavior [26,27,29]. A wealth of studies, examining thousands of twin pairs, has revealed that there is a sizeable genetic influence on the development of antisocial traits [10,12]. As a result, mothers who smoke while pregnant may be passing along the genetic propensity for antisocial phenotypes to their offspring. If this is the case, then the association between prenatal exposure to cigarette smoke and offspring behavioral problems may be the result of genetic transmission, not exposure to cigarette smoke in utero. Maughan and her colleagues [27] examined this possibility by employing a sample of twins drawn from the Environmental Risk (E-Risk) Longitudinal Twin Study. The results revealed the presence a significant dose-response effect of prenatal smoking on behavioral problems. Most notably, however, the researchers found that over half of the association between maternal smoking and behavioral problems was due to the influence of shared genetic factors. In other words, when a number of other risk factors were taken into account—including, maternal psychopathology, family disadvantage, and genetic effects—the strength of the relationship between maternal smoking and behavioral problems was either markedly reduced or completely eliminated. This study provides some initial evidence that the association between maternal smoking during pregnancy and offspring behavioral problems might be largely or partially driven by genetic confounds [but see 20].

Taken together, the available evidence suggests that the association between prenatal exposure to cigarette smoke and subsequent behavioral problems is complex and may be the result of confounding factors. However, much more research is needed that specifically addresses the issue of confounding before any definitive conclusions can be drawn. The current research seeks to provide a cautious step in this direction by examining the influence of prenatal exposure to cigarette smoke on externalizing behavioral problems in children by using propensity score matching (PSM) to help correct for confounding.

To examine the effects of prenatal smoking on childhood behavioral problems we employ propensity score matching analysis (PSM). PSM is a quasi-experimental design that has been used across disciplines in order to isolate treatment effects on a number of outcomes using observational data [52–54]. Matching based on propensity scores can be used in situations where a true experimental design is not feasible. For example, it would be both unethical and impractical to randomly assign mothers to smoking and non-smoking groups during their pregnancy.

The PSM analysis for the current study proceeded in a series of interrelated steps. First, we examined whether prenatal exposure to cigarettes was associated with increased externalizing behavioral problems by using the dichotomous treatment measure of maternal smoking (0 = non-smoker, 1 = smoker). Second, we examined whether the relationship between smoking and behavioral problems would remain after accounting for confounding among the variables by using PSM. To address this issue, propensity scores were estimated for the mothers in the study based on the fourteen covariates. Multivariate logistic regression models were used to generate log odds which were converted to conditional probabilities ranging in value from 0 to 1 (i.e., the probability that a mother smoked during her pregnancy). Subjects were matched based on these probabilities using 3-to-1 nearest neighbor matching with a matching caliper of 0.05. Third, and finally, we repeated the steps outlined above using the treatment variable of heavy maternal smoking (0 = mother smoked less than 10 cigarettes, including non-smokers, 1 = mother smoked 10 or more cigarettes per day).

The results of the pre- and post-matching independent samples t-tests for the fourteen covariates are presented in Table 1. Prior to matching on propensity scores, the subjects in the sample differed on eleven out of the fourteen covariates. The differences between subjects offer some initial evidence that the relationship between prenatal nicotine exposure and behavioral problems may be confounded (i.e., influenced by unmeasured variables). After matching based on propensity scores, we were able to eliminate significant differences between the treatment and control groups in the sample (i.e., the children of mothers who smoked during their pregnancy, and those who did not).

Once statistically equivalent groups were established, we were able to directly examine the effect of any exposure to prenatal smoking on externalizing behavioral problems in children. Figure 1 displays the effects of maternal smoking on externalizing behaviors both before and after matching.

Prior to being placed into matched groups, the children of mothers who smoked during their pregnancy exhibited higher levels of externalizing problem behaviors when compared to the children of non-smoking mothers (mean difference = 1.80, t-value = 6.98, P ≤ 0.05). Once subjects were placed into matched groups, prenatal exposure to cigarette smoke was no longer a significant predictor externalizing behavioral problems (mean difference = 0.30, t-value = 0.74, P > 0.05)

The results of the balancing estimates for the treatment indicator of heavy maternal smoking are presented in Table 2. This analysis revealed that prior to matching, the participants differed significantly on ten out of the fourteen covariates in the study. After respondents were matched on propensity scores, none of the covariates differed significantly between the treatment and control groups.

The association between exposure to high levels of maternal smoking and externalizing behavioral problems is presented in Figure 2. Prior to matching, the children exposed to high levels of cigarette smoke in utero also had significantly higher scores on the externalizing problem behaviors scale when compared to those children not exposed (mean difference = 1.72, t-value = 4.83, P ≤ 0.05). After matching on the fourteen covariates, the effect of heavy maternal smoking on externalizing behavioral problems was no longer statistically significant (mean difference = 0.70, t-value = 1.36, P > 0.05).

Researchers in the medical and behavioral sciences have long been concerned with the deleterious effects of prenatal smoking on offspring development. Early research on the topic demonstrated links between prenatal smoking and low birth weight, respiratory dysfunction, delayed motor development, and a host of other adverse health outcomes in children [19]. The consistent and robust body of findings linking maternal smoking during pregnancy with offspring’s health in early childhood prompted numerous warnings to expectant mothers cautioning them against the harmful effects of prenatal nicotine use on a developing fetus.

Subsequent research has demonstrated a statistically significant and consistent association between exposure to cigarette smoke in utero and behavioral problems over the life course. The bulk of this research, however, has failed to adequately address the methodological issues that may render the association between the prenatal smoking and externalizing problems spurious. We addressed this shortcoming in the literature by estimating the effect of prenatal exposure to cigarette smoke on childhood externalizing behavioral problems by employing propensity score matching (PSM). Consistent with the extant literature, our analysis of the ECLS-B data revealed a significant and positive association between maternal smoking and externalizing behavioral problems prior to directly modeling confounding factors. Once confounding was taken into account by estimating PSM models, the effect of prenatal exposure to cigarette smoke on childhood behavioral problems was no longer statistically significant. This null effect was observed for any exposure to cigarette smoke in utero as well as high levels of cigarette smoke exposure in utero.

The results of the current study stand in stark contrast to research revealing a direct effect of prenatal exposure to cigarette smoke on behavioral problems, leading to the question of what accounts for these disparate results. The main aspect that distinguishes the current study from most prior research is that we directly addressed confounding by using PSM techniques. Had we not modeled confounding directly, the results of our study would have been in line with those generated in previous research. We should add, however, that our findings dovetail with the results of some of the more methodologically rigorous studies that control for both genetic and environmental confounds [20,25,26]. These studies have either failed to detect any association between prenatal exposure to cigarette smoke and behavioral problems later in life or found very attenuated associations. Interestingly, research using animal models, where randomization is possible, has consistently revealed a link between prenatal exposure to tobacco smoke and subsequent impairments [22,23,25]. Perhaps the reason for these differential findings is tied to the outcome of interest—in the current study, the outcome of interest was externalizing behavioral problems early in life. More research is needed on this topic before any definitive conclusions can be drawn about the factors that might be accounting for the different findings that are generated from animal models and from those generated from human models using PSM.

Taken together, the results of our study, along with those that have adequately accounted for confounding [20,25,26,55], tend to call into question whether prenatal exposure to cigarette smoke has a significant main effect on behavioral problems. The null effect of prenatal exposure to cigarette smoke necessarily raises ethical questions about whether our study is advocating for women to continue to smoke throughout their pregnancy. There are at least four reasons why such a position is unfounded. First, replication studies that analyze other samples are needed before the results of the current study should be accepted. And, even if the results are replicated and prenatal exposure to cigarette smoke does not have an effect on behavioral problems, exposure to cigarette smoke in utero has been linked to various other outcomes, such as cognitive abilities [16]. As a result, reducing smoking among pregnant women should increase a range of prosocial and adaptive outcomes that cut across various domains of healthy human development.

Second, the current study only examined the main effect that prenatal exposure to cigarette smoke had on childhood behavioral problems. Existing research has indicated that exposure to cigarette smoke may have its strongest effects on antisocial phenotypes when it is paired with certain environments [18]. In this way, the effect of prenatal exposure to cigarette smoke is likely to be dependent on the presence of certain environmental factors. Third, and relatedly, there is some evidence to indicate certain genotypes may exacerbate or blunt the negative effects of prenatal exposure to cigarette smoke. For example, Hong et al. [56] found that polymorphisms in the GSTM1 and GSTT1 genes moderated the effect of prenatal exposure to cigarette smoke on birth weight. Whether other genetic polymorphisms might moderate the effect of prenatal exposure to cigarette smoke on antisocial behaviors remains an open-empirical question awaiting future research.

The fourth reason to use caution in interpreting the results of the current study is because we only examined behavioral problems measured in childhood. Perhaps the effects of prenatal exposure to cigarette smoke vary across different sections of the life course, with the effects becoming more intense over time. We do not know of any research testing this possibility, but future researchers should explore the potential age-graded nature of the effects of prenatal exposure to cigarette smoke on behavioral problems in childhood, adolescence, and adulthood.

Public policies aimed at reducing prenatal smoking draw heavily on research that has linked exposure to cigarette smoke in utero with myriad physiological, developmental, and cognitive delays in children. Our findings do not obviate the importance of these smoking prevention efforts. However, our results do suggest that the association between maternal smoking during pregnancy and childhood externalizing behavioral problems is highly complex, perhaps more complex than originally thought. As a result, much more research is needed to identify the various ways in which prenatal exposure to cigarette smoke may interfere with normal human development.