Abstract
We examined the possibility that the association between maternal smoking during pregnancy and childhood behavioral problems is the result of confounding. Data from the first three waves of the Fragile Families and Child Wellbeing Study were analyzed. We estimated the association between maternal smoking during pregnancy and externalizing problems in three-year olds using a propensity score matching approach. After successfully matching children based on their mother's propensities to smoke during pregnancy, the results indicate that maternal cigarette smoking is related to childhood externalizing behavioral problems, but only among mothers who smoked more than a pack per day while pregnant. At lower levels of exposure, the association between exposure to cigarette smoke in utero and externalizing behavioral problems in childhood can be explained by confounding. The results of this study support prevention efforts intended to reduce prenatal exposure to cigarette smoke, especially by mothers who smoke heavily.
Keywords:
Notes
1. Standardized bias statistics are appropriate when evaluating the success of matching with dichotomous covariates. In general, standardized bias statistics less than 20% suggest successful matching on a particular measure (Loughran et al. 2009). Stated differently, a standardized difference value greater than 20% for any covariate would suggest that the covariate is out of balance. After matching, all of the standardized bias estimates were less than 20%.
2. In order to further examine the effects of maternal smoking on externalizing behavioral problems we constructed an alternative version of the heavy maternal smoking treatment indicator. This alternative measure was coded such that 1 = heavy smoking mothers and 0 = non-smoking mothers. In other words, light smokers were excluded from the analysis. We then examined PSM models using the altered version of the heavy smoking variable. The results of the PSM analysis revealed that prior to matching, heavy maternal smoking significantly influenced externalizing behavioral problems in children (Mean difference = 5.82, t-value = 3.77, p ≤ 0.05). After matching, the effects of heavy maternal smoking were reduced to non-significance (Mean difference = 2.64, t-value = 1.23, p > 0.05). This result departs from our previous finding that heavy maternal smoking influenced adverse behavior before and after matching. It is important to point out, however, that we constructed the original heavy smoking measure in accordance with prior research (Boutwell and Beaver 2010). Furthermore, given that the mean differences remained large (for the alternative smoking measure) even after matching, we opted to retain that measure in the current study.