Teen smoking and birth outcomes.

• Author: Walker, Mary Beth

1. Introduction

   The Centers for Disease Control and Prevention (CDC) reports that the incidence of low birth weight births (infants weighing less than 2500 grams at birth) is on the rise, and that very young mothers (those 15 years old and under) are two to three times more likely to have a low birth weight baby than their counterparts aged 24-34 years. The incidence of low birth weight for all teens is 23% higher than for the population as a whole (CDC 2006a). A recent study by Chen et al. (2007) concludes that low birth weight and other adverse birth outcomes observed in teen pregnancies cannot be fully attributed to known risk factors, such as low socioeconomic status and inadequate prenatal care.

   Low birth weight is correlated with a number of adverse outcomes for children, including future health problems and poorer educational outcomes. Low birth weight infants account for large public health expenditures: studies show that more than one third of the total dollar amount spent in the United States on health care during the first year of life can be attributed to low birth weight, even though low birth weight infants account for fewer than 10% of all births in the United States (Lewitt et al. 1995). (1)

   The presence of a link between birth weight and smoking has long been accepted. In a 2001 report, the Surgeon General stated that "Infants born to women who smoke during pregnancy have a lower average birth weight and are more likely to be small for gestational age than infants born to women who do not smoke. Low birth weight is associated with increased risk for neonatal, perinatal, and infant morbidity and mortality. The longer the mother smokes during pregnancy, the greater the effect on the infant's birth weight" (CDC 2001). Multiple studies have shown that tobacco use during pregnancy is correlated with lower birth weights; see, for example, Evans and Ringel (1999), Abrevaya (2006), and Abrevaya and Dahl (2008). Shiono and Behrman (1995) report that smoking during pregnancy accounts for 20% of low birth weight births, making it the single most important modifiable risk factor for low birth weight in developed countries (Kramer 1987).

   We also observe that the incidence of teen smoking is relatively high: In 2004, 21.7% of all high school students reported smoking cigarettes, while the incidence of cigarette smoking among nonteens was 20.9%. (2) Data from the state of Georgia (1994-2002) indicate that approximately 22.1% of nonblack teen mothers report that they smoked during their pregnancies; whereas, only 11.7% of nonblack older mothers report smoking (Table 1).

   Could the observed differences in birth weights for babies born to teen mothers and babies born to nonteens be attributable, at least in part, to differences in the effects of smoking on infant health for these two groups? The issue is complicated by several factors. First, there are the physiological effects of nicotine on the fetus; medical research indicates that nicotine itself is a neuroteratogen, affecting nervous system development (see Roy, Seidler, and Slotkin 1998; Slotkin 1998; Law et al. 2003). Smoking also interferes with the function of the placenta, which may lead to malnutrition (Law et al. 2003). Then, too, teen mothers will not have sustained the same physical damage from smoking as adult women, simply because the teens have not had the same length of exposure to tobacco. These causal effects do not suggest any reason to suspect substantial differences in the impacts on babies born to smoking teens or smoking adults.

   However, smoking during pregnancy not only has a direct physical effect on the health of the fetus, but it also serves as a possible signal for other unhealthy behaviors that are not usually measured in our data sets. Although not all studies use methods to account for the possible correlation of maternal tobacco use with these other unobservable influences, in recent work researchers do recognize the endogeneity of tobacco use in birth outcome models; see, for example, Almond, Chay, and Lee (2005) or Abrevaya (2006). Simply stated, the hypothesis is that women who choose to smoke during pregnancy, despite the considerable evidence that relates smoking to poor birth outcomes, could be likely to engage in other risky behaviors. Use of tobacco could provide a signal of the mother's attitude or concern for a healthy birth, and these unobservable factors could also affect the pregnancy outcome.

   Perhaps some fraction of the difference in birth outcomes for teens and nonteens results from systematic differences in either the extent of these unobserved behaviors or the correlation of these behaviors with tobacco use. Thus, obtaining empirical evidence of the causal effect of maternal tobacco use on birth weight for both teen mothers and older mothers could provide some useful information on the signal provided by tobacco use, such as the teen mother's attitude or concern for a healthy baby relative to a nonteen mom. In this paper, we provide estimates of the impact of maternal tobacco use on birth outcomes for teen mothers and older mothers, using a unique data set of the entire population of births in the state of Georgia over the period 1994-2002. We use three different estimation methods that rely on different assumptions regarding the unobserved components of maternal behavior, in the hope of obtaining estimates of the causal effect of smoking on birth weights. The results of the alternative estimators suggest that both ordinary least squares (OLS) and matching estimators, which rely on observable characteristics to estimate the causal link between birth weight and smoking, may overstate the impact of smoking on birth weight. The fixed effects estimates, which control for unobservables, suggest that there are some differences of the impact of smoking on birth weight for teen and nonteen mothers, but that the effect is substantially smaller than what is found in the other estimations.

   Evidence that the impact of smoking on birth weight for teens and nonteens differs can inform future research into both teen smoking and teen pregnancy, as well as the policies and programs aimed at the teen population. Currently many of the antismoking campaigns and programs are focused on teenagers. For example, the national campaign "Healthy People 2010" lists tobacco use as one of its 10 high-priority public health issues, targeting a 50% reduction in tobacco use for teens. Evidence to justify and reinforce these efforts could be useful in the general policy debate regarding tobacco use.

   Section 2 of this paper reviews the literature. Section 3 discusses the empirical strategy. Section 4 introduces the data used in the analyses. Section 5 presents the results, and Section 6 concludes the paper.

2. Previous Literature

   Across the United States, teen births are on the decline. Southern states continue to have the highest teen birth rates in the nation. In 1990, the national teen birth rate (births per 1000 females ages 15-17) was 37, and in Georgia it was 50. In 2004, these figures were 22.1 and 29.3, respectively (CDC 2006a). More detailed data on births in Georgia reflect some startling statistics regarding teen pregnancies. If we consider all births to mothers below the age of 19, 4% of those births are to mothers younger than age 15 (at time of delivery) and 26% to mothers ages 15-16. In 2002, 9% of live infant births were of low birth weight, an increase from 8.5% in 1998. Of teen births in 2000, 82% were covered by Medicaid. (3)

   The previous literature most relevant to our work are the recent studies that recognize the endogeneity of tobacco use in birth outcome models and use various techniques to account for this estimation problem. In a randomized experiment, Permutt and Hebel (1989) considered the impact of "stop smoking" counseling on birth weights for a group of smoking mothers. The control group for comparison was a group of smoking mothers who did not receive counseling. The authors found a negative effect of smoking on birth weight of about 400 grams, using a sample of 935 mothers. This is quite a large effect given that the normal birth weight is 3500 grams. This study is unique in its natural experiment approach, but the causal effect of smoking is estimated imprecisely because of a small sample size.

   Abrevaya (2006) estimates the causal effect of smoking during pregnancy on birth weight and gestation length in weeks using panel data techniques. This study is an interesting departure from the rest of the literature because it employs a panel data analysis using a sample of mothers with multiple births during the sample period. Since there are no individual identifiers in the data set that would allow the author to uniquely identify a mother (for example, her Social Security number), he employs a matching strategy to determine which individual mothers experienced multiple births during the time period considered. The results from the fixed effects models indicate that the effect of smoking on birth outcomes is smaller than those obtained from the OLS models, suggesting a strong negative correlation between the omitted variables and the smoking indicators. Our study is similar to this one in that one of our identification strategies relies on the variation in the smoking behavior of mothers who give multiple births during the period analyzed. Because our data are drawn from administrative records, we identify each mother perfectly. We are also able to control for a much larger set of variables.

   Almond, Chay, and Lee (2005) is another recent study on the effects of maternal smoking during pregnancy on health outcomes of singleton births controlling for a wide set of background characteristics. The authors compare the hospital costs, health outcomes, and infant mortality rates between heavier and lighter infants from all twin pairs born in the United States. In order to identify the causal effect of smoking...