Every breath you take: the effect of postpartum maternal smoking on childhood asthma.

Sabia, Joseph J.

1. Introduction

In June 2006, the U.S. Surgeon General's Office released a report titled, "The Health Consequences of Involuntary Exposure to Tobacco Smoke," which summarizes evidence from the public health and medical literatures on the adverse effects of exposure to secondhand tobacco smoke. Upon presenting the report to the media, Surgeon General Richard Carmona offered the following comments:

 The scientific evidence is now indisputable: secondhand tobacco
 smoke is not a mere annoyance. It is a serious health hazard that
 can lead to disease and premature death in children and non-smoking
 adults ... Children exposed to secondhand smoke are at an increased
 risk for Sudden Infant Death Syndrome (SIDS), acute respiratory
 infections, ear problems, and more severe asthma (Carmona 2006).

Carmona further emphasized that "there is no safe amount of secondhand smoke" to which a child can be exposed. Many states have adopted a variety of antismoking policies based on this evidence. (1) The Surgeon General's report concludes that while there is some medical evidence of a causal link between environmental tobacco smoke and asthma among school-age children, "the evidence is suggestive, but not sufficient to infer a causal relationship between secondhand smoke exposure from parental smoking and the onset of childhood asthma" (Report of the Surgeon General 2006, p. 9). Examining the appropriateness of inferring a causal relationship between parental smoking and childhood asthma requires more careful empirical attention than has been seen in the public health literature. This is the key contribution of the current study.

Today, nearly 5 million children have asthma, with the most rapid rate of increase occurring for those under the age of five (American Academy of Pediatrics 2000). Asthma remains the leading cause of pediatric emergency room use and hospital admission (National Institute of Environmental Health Sciences 2007). Over the last decade, federal agencies have turned their attention toward better surveillance of and more research on asthma. The Department of Health and Human Services (DHHS) issued a 1997 report titled "Action against Asthma," which set goals for reductions in asthma prevalence by 2010. In May 2000, the President's Task Force on Environmental Health Risks and Safety Risks to Children--a joint venture of the DHHS and the Environmental Protection Agency--issued a report titled "Asthma and the Environment: A Strategy to Protect Children," which set forth similar goals. The National Heart, Lung and Blood Institute's Asthma Education and Prevention Program has also produced several reports on this topic. Each of these agencies, along with the recent report from the Surgeon General, has cited exposure to secondhand smoke as a strong correlate of childhood asthma.

This study carefully examines the relationship between postpartum maternal smoking and childhood asthma among populations at highest risk of asthma--racial minorities and children of less-educated parents. In contrast to previous studies in the public health literature, this study more carefully addresses unmeasured heterogeneity in attempting to isolate the causal effect of secondhand smoke on childhood asthma. The findings presented here suggest that postpartum maternal smoking may have important adverse effects on children's health. Confirming several previous studies in the literature, cross-section estimates reflect consistent evidence of a positive association between maternal smoking and childhood asthma. Postpartum maternal smoking is associated with a 2 to 3 percentage point higher probability of asthma by age one and a 4 to 5 percentage point higher probability of asthma by age three. Importantly, these results are robust to the inclusion of individual fixed effects. Difference-in-difference models that compare the change in asthma rates for children whose mothers began smoking between their child's first and third birthday with mothers who remained nonsmokers find that maternal smoking is associated with a 4 to 7 percentage-point increase in asthma rates. This result is robust to several falsification tests, suggesting that other time-varying unmeasured maternal health input choices cannot explain this positive relationship.

2. Asthma-Smoking Literature

Asthma is a chronic inflammatory lung disease characterized by repeated instances of breathlessness, wheezing, and coughing (Department of Health and Human Services 1997). This disease has been discussed quite extensively in the medical and public health literatures but has largely been ignored in the economics literature. But as policymakers consider policy alternatives to reduce asthma rates--such as raising cigarette taxes--economists will be increasingly called upon for cost-benefit analyses. As a primer for such future research, a 1998 DHHS study estimated the annual cost of asthma, including medical costs and productivity losses, to be over $11 billion per year. Death rates due to asthma are on the rise, with an over 300% increase from 1977 to 1995. Race differences also persist. In 1995, the death rate among African Americans (11.5 per million) was over four times higher than that among whites (Mannino, Homa, and Pertowski 1998).

The medical community has reached a consensus that asthma is a disease of airway inflammation "resulting from a complex interplay between environmental exposures and genetic and other factors" (Asthma Priority Area Workgroup 2000, p. 9). Secondhand tobacco smoke from cigarettes, cigars, and pipes is composed of over 3800 unique chemical compounds (National Research Council 1986). Concentrations of suspended particulate matter are two to three times higher in homes with smokers than homes without smoking, and this particulate matter is believed to be negatively associated with respiratory health (Lefcoe and Inculet 1971; Dockery et al. 1982; American Academy of Pediatrics 2000).

While there is minimal understanding of the environmental causes of asthma, the environmental triggers of child asthma have become more well known. Among the principal allergens that trigger asthma attacks among young children are house dust mites, cockroaches, mold, and animal hair (for examples, see Bierman 1996; Warner et al. 1996; National Asthma Education and Prevention Program 1997; Platts-Mills and Carter 1997; and Institute of Medicine 2000). Air pollutants, including ozone and upper respiratory viruses, are also positively correlated with asthma attacks (Koren 1995; Busse, Gern, and Dick 1997).

Several studies have examined the relationship between exposure to secondhand tobacco smoke and the development of asthma in young children. However, almost all are cross-sectional studies that do not account for unobservable characteristics that may be correlated with both maternal smoking and the child's development of asthma, thus raising doubts as to the appropriateness of a causal interpretation of estimates.

Using the Third National Health and Nutritional Examination Survey, Gergen et al. (1998) examine the impact of environmental tobacco smoke on the respiratory health of a nationally representative sample of children aged two months to five years. They found that among children two months to two years of age, maternal smoking of greater than 20 cigarettes per day is associated with a significantly higher rate of diagnosed asthma. Weizman et al. (1990) find similar results using the National Health Interview Survey. (For a review of much of the medical literature, see Strachan and Cook 1998.)

A series of smaller, localized studies have also found a positive association between parental smoking and child asthma (for example, Burchfield et al. 1986; Evans, Levison, and Feldman 1987; O'Connor et al. 1987; Murray and Morrison 1989; Krzyzanowski, Quackenboss, and Lebowitz 1990). It is difficult to interpret these studies causally--and to generalize the results of local studies--given the failure to carefully address potential biases caused by unmeasured heterogeneity.

No large-scale studies in the economics literature have examined the impact of secondhand tobacco smoke on child asthma rates. Most studies have examined the impact of prepartum maternal smoking on infant health outcomes, usually focusing on birth weight (see, for example, Evans and Ringel 1999), and have explored how increases in cigarette taxes could lead to improved health outcomes for children (Ringel and Evans 2001; Colman, Grossman, and Joyce 2003). Other work has examined the impact of air pollutants on child asthma (Neidel12004). Less attention has been paid to the negative health effects associated with exposure to household-level secondhand smoke.

This paper contributes to the existing empirical literature in three key ways. First, while much of the health economics literature has focused on the effect of smoking during pregnancy on child health outcomes, this study examines whether there are important adverse child health effects of postpartum maternal smoking. Second, in examining the relationship between postpartum maternal smoking and childhood asthma, this study improves upon the public health literature by more carefully addressing potential biases that may result from unobserved heterogeneity. Finally, this study examines the effect of parental smoking on childhood asthma among populations at high risk of developing asthma--racial minorities and children of less-educated mothers.

3. Methodology

Cross-Section Estimates

The existing public health literature on the relationship between parental smoking and childhood asthma has generally presented cross-section estimates of the child health production function. To replicate the existing literature, probit and linear probability models of the following form are estimated (2):

[A.sup.*.sub.ijt] = [[beta].sub.0] + [[beta].sub.1] [S.sub.jt] + [X'.sub.it][[beta].sub.3] + [[epsilon].sub.ijt], (1)

where [A.sup.*] is a latent propensity asthma for infant i with mother j at time t. Then, [A.sub.ijt] = 1 if [A.sup.*] > 0 and [A.sub.ijt] = 0 if [A.sup.*] [less than or equal to] 0. S is a measure of parental smoking, [X.sub.i] is a vector of child characteristics, [X.sub.j] is a vector of mother characteristics, and [epsilon] is the error term.

The cross-section estimate of the effect of parental smoking on child asthma ([[beta]sub.1]) will only be unbiased in the absence of unmeasured characteristics associated with both parental smoking and childhood asthma. If, for example, mothers who smoke make other unobservable health decisions that increase the probability of child asthma--such as expending low effort to keep the home environment free of other indoor allergens--then the estimate of [[beta].sub.1] may be biased toward adverse health effects. Alternatively, if mothers have information on their children's health endowments--for example, that their children are genetically prone to developing asthma--and this information is unobserved to researchers, then the mothers who choose to smoke may be those whose children have the healthiest endowments. This type of selection will lead to downwardly biased estimates of the effect of secondhand smoke on childhood asthma.

Difference-in-Difference Strategy

This study's key methodological contribution to the existing secondhand smoke-asthma literature is to better control for unobserved heterogeneity in the child health production function by including individual fixed effects. Difference-in-difference estimates of the child health production function will control for time-invariant unmeasured heterogeneity. Given longitudinal data on adolescents in two periods (at age one and age three), a difference-in- difference model of the following form may be estimated:

[A.sub.it] + 2 - [A.sub.it] = ([[beta].sub.0t] + 2 - [[beta].sub.0t] + [[beta].sub.1] ([S.sub.j] + 2 - [S.sub.jt]) + ([X.sub.it] + 2 - [X.sub.it])'[[beta].sub.2] + ([X.sub.jt] + 2 - [X.sub.jt])'[[beta].sub.3] + ([[upsilon].sub.it] + 2 - [[upsilon].sub.it]). (2)

In this model, child asthma is measured at age one (period t) and age three (period t + 2), and the sample conditioned on non-smokers at period t. The difference-in-difference estimate ([gamma]) will be an unbiased estimate of the effect of postpartum maternal smoking on child asthma if there are no time-varying unobservables correlated with changes in smoking behavior and with asthma diagnoses, E[([[upsilon].sub.it+2] - [[upsilon].sub.it]) | ([S.sub.it+2] - [S.sub.it])=0]. The unobserved heterogeneity described to this point--mother's knowledge of child's unobserved health endowment and mother's unobserved propensity for engaging in other asthma-inducing activities--have been fixed. But if, for example, a family experiences an unobserved financial stressor between period t and t + 2, then this might be correlated with both an increase in smoking and an increase in asthma, with the latter not caused by smoking but by reduced investments in household cleanliness.

An alternate identification strategy to the fixed effects approach is to model the unmeasured heterogeneity via instrumental variables. Such an approach would require instruments that provide exogenous variation in parental smoking that are uncorrelated with unmeasured determinants of child asthma. While this approach is explored, concern about the lack of valid exclusion restrictions in the data results in my relying more heavily on the fixed effects identification strategy. The credibility of the identification assumption underlying the difference-in-difference (DD) model is examined through a set of falsification tests.

Antitests

One way to examine the appropriateness of interpreting DD estimates causally is through a series of falsification tests. This can be done by estimating Equation 2 using child health outcomes that are not expected to be causally influenced by exposure to secondhand smoke. The absence of a statistically significant relationship between the onset of smoking and, for example, child injuries, accidents, physical disabilities, and overall health would help to validate a causal interpretation of DD estimates of the asthma equation.

Similarly, estimating the relationship between the onset of other "unhealthy" parental behaviors correlated with smoking, but not expected to causally influence asthma, would provide an additional validity test for a causal interpretation of DD estimates of the effect of smoking on asthma. The consumption of alcohol and drugs, while correlated with smoking, should not directly influence childhood asthma. Thus, the absence of a significant relationship between alcohol or drug consumption and asthma would suggest that time-varying unobserved heterogeneity may not be an important source of bias. These falsification tests, while not conclusive, will provide some suggestive evidence of the credibility of the DD identification strategy.

4. Data

This study uses data from the Fragile Families and Child Wellbeing Study, a national survey that follows a birth cohort of parents and their children over a three-year period. The Fragile Families and Child Wellbeing Study was developed by the Center for Research on Child Wellbeing at Princeton University and the Social Indicators Survey Center at Columbia University. At the child's birth, 4898 new mothers were interviewed: 3712 were unwed and 1186 were married; 3830 interviews with fathers were completed. Initial interviews with brand new mothers and fathers took place between February 1998 and September 2000. Parents were interviewed immediately after the birth of the child (initial interview), 12 months later (one-year follow-up), and 36 months later (three-year follow-up), with mothers and fathers interviewed separately. (3) The Fragile Families study includes a stratified random sample of all U.S. cities with populations of 200,000 or more. (4) First, cities were sampled; then hospitals within cities were sampled; and finally, births within hospitals were sampled. (5) The data are representative of nonmarital births in U.S. cities with populations over 200,000.

This data set is useful in examining the relationship between parental smoking and childhood asthma for three reasons. First, the survey is targeted at less-educated, low-income families composed of racial minorities. Given the relatively high rates of asthma among children in these subpopulations, these data will allow me to explore the effect of secondhand smoke on asthma rates of the most at-risk populations. Second, the Fragile Families data have measures of child and infant health as well as parental smoking behavior at multiple points in time, permitting the estimation of DD models, which represents an important improvement to estimation strategies in the current public health literature. Third, multiple asthma measures are available, which allows me to test the sensitivity of results to varying definitions of the dependent variable.

The analysis for this paper focuses on the second and third waves of data, which correspond to outcomes when the child is approximately 12 months old and 36 months old, respectively. For the remainder of this paper, outcomes at the child's first birthday are referred to as baseline outcomes. The means and standard deviations of baseline asthma measures, maternal smoking, and other sociodemographic characteristics used in this analysis are found in Table 1A for high-risk asthma populations: newborn boys, blacks, Hispanics, and households with mothers who have less than a high school education. (6) Table 1B shows the correlation among a set of alternate asthma measures. Child asthma rates at age three, along with means of key control variables, are found in Table 1C.

Dependent Variables

There are three measures of child asthma available in the second and third waves of the Fragile Families data. The first measure asks whether a doctor or other health professional had ever told the mother that her child had asthma. Nearly 14% of mothers in the sample reported being told by a health professional that her one-year-old child had asthma. Asthma diagnosis rates were higher for blacks (17.5%) and Hispanics (13.9%) than for whites (5.7%). Boys were more likely to be diagnosed with asthma (16.5%) than girls (10.3%). Children whose mothers had less than a high school education had among the highest asthma rates (18.4%). By age three, the percentage of all mothers in the sample who had ever reported receiving an asthma diagnosis was 26%, with nearly a third of single mothers with less than a high school education reporting an asthma diagnosis.

It is important to note that mothers who received an asthma diagnosis are, by definition, also those who had taken their child to a doctor. This form of selection may be an important concern. Children who do have asthma may be undiagnosed because they have not been to a doctor. While there is no statistically significant difference in reported asthma rates by insurance status (12% of uninsured mothers report an asthma diagnosis by age one, compared to 14% of insured mothers (7)), selection may bias estimates of the effect of parental smoking on the probability of an asthma diagnosis if unobserved parental characteristics associated with smoking are also associated with frequency of doctor visits. For example, the estimated effect of parental smoking on child asthma may be biased toward zero if parents who smoke invest less in child health and do not take their children to health professionals for regular physical examinations. Conversely, estimates could be upwardly biased if smoking parents believe that smoking may have adverse respiratory effects on their children and thus choose to have them examined more frequently by medical professionals.

Estimates of asthma rates in the Fragile Families survey are, as expected, much higher that those reported in the nationally representative National Maternal and Infant Health Survey (NMIHS). The NMIHS shows that 11% of mothers of two- to four-year-olds report "they have been told by a doctor, nurse or health care provider that the [child] has asthma." This compares to the 13% of children with asthma by age one (and nearly a quarter of children by age three) observed in the full Fragile Families study. This difference can be explained by the Fragile Families study's focus on low-income, unmarried, less-educated, welfare-prone racial minorities--all populations with the highest asthma rates.

A second measure of asthma available in the Fragile Families data is whether the child has ever had an asthma attack. According to the data, 8.1% of mothers reported that their child had had an asthma attack by age one, and 15.9% reported an asthma attack by age three, each lower than the mean rate of asthma diagnoses. At age one, the rate was higher for blacks (10.8%) and Hispanics (7.7%) than for whites (3.1%). Moreover, boys had almost twice the rate of asthma attacks (10.4%) of girls (5.6%). These race- and sex-specific differences persisted at age three. Note that this measure of asthma does not specify whether a doctor or the parent has determined that the child had an asthma attack. Thus, this may introduce measurement error to the extent that mothers misdiagnose asthma attacks.

A final measure of asthma is whether the mother has ever had to take the child to an emergency room because of an asthma attack. In the sample, 7.6% of mothers reported taking their child to get urgent emergency care due to an asthma attack by age one. This percentage increased to just over 13% by age three. Similar differences exist by race and sex as described in the previous asthma measures. An affirmative response to getting emergency care for one's child's asthma attack requires the interaction of the identification of a severe attack and the mother taking immediate action to obtain emergency medical care for the child. Thus, this asthma measure may suffer from similar selection issues to those described in the professional asthma diagnosis measure described previously.

In summary, one important advantage of the Fragile Families data is the presence of multiple measures of child asthma, which allows me to test the sensitivity of findings to alternate definitions. (8) However, in addition to the selection issue just discussed, there is another important limitation to these data. The recent medical literature (see, for example, Martinez et al. 1995) suggests there is difficulty in accurately diagnosing asthma in young children. While wheezing during the first six years of life often precipitates an asthma diagnosis, Martinez et al. (1995) find that the majority of infants with wheezing problems do not have increased risks of asthma later in life, but rather simply have "transient conditions associated with diminished airway function at birth" (p. 133). This finding suggests that asthma rates of young children may be overreported.

While this paper focuses on asthma as a child health outcome that may be affected by exposure to secondhand smoke, this respiratory illness is not the only one that may be affected. An increase in coughing, wheezing, and ear infections may also be observed because of exposure to secondhand smoke. Importantly, if wheezing is associated with greater exposure to secondhand smoke, then the impact of maternal smoking on asthma may be upwardly biased. In the presence of this form of measurement error, it may be that estimates of the effects of smoking on asthma may be more appropriately interpreted as the effects of smoking on wheezing and other respiratory illnesses. This measurement problem is not unique to the Fragile Families data, (9) but it is important to note this limitation prior to proceeding with the analysis.

Smoking Behavior

Across waves, mothers and fathers are asked questions about their smoking behavior. (10) Responses showed that 19.6% of mothers smoked during pregnancy and 27.4% smoked in the year after their child's birth. This measure is generally consistent with that of other large data sets. Colman, Grossman, and Joyce (2003) report that in the Pregnancy Risk Assessment Monitoring System, approximately 22% of new mothers smoke. By the child's third birthday, approximately 34% of mothers had smoked since their child's birth.

Smoking rates differed significantly by race. During their child's first year of life, 37% of white mothers smoked, while only 18.7% of Hispanics and 27.2% of blacks did so; 38.7% of mothers with less than a high school education were postpartum smokers. Similar race and education-specific patterns existed for smoking patterns by the child's third birthday.

Most mothers were "light" smokers. Of mothers who smoked after their child's birth but before their first birthday, more than 75% smoked less than one-half a pack per day, 21% smoked one pack a day, and 4% smoked more than one pack a day. There was not a significant difference in maternal smoking by higher-order birth--28.6% of mothers with two or more children smoked, as did 25.6% of mothers with only one child. Similar patterns existed for smoking by the child's third birthday. Of mothers who smoked between the child's first and third birthdays, 78% reported smoking approximately less than one-half a pack per day.

By the child's first birthday, approximately 41% of fathers in the sample smoked, while 19.8% of households had a father residing there who smoked. The proportion was higher for whites (27%) than blacks (16.4%), and 38.1% of households had at least one parent who smoked. The correlation between mothers' smoking and fathers' smoking was 0.34.

One limitation of these data is that we do not have a report of the quantity of secondhand smoke to which the child is exposed in multiple waves. However, in wave 3, there are a few more precise measures of smoking behavior that may be correlated with childhood asthma. Mothers are asked how many smokers are in the household, whether the child is exposed to time in the same room as a smoker, and whether the mother smokes in the home. Means of these variables are available in Table 1C.

Other Demographic and Economic Variables

Other independent variables used to measure inputs in the child health production function at baseline are listed in Table 1A. These variables include economic and demographic characteristics: race, sex of the child, birth weight of the child, whether the mother has other biological children, marital status, household income, labor market participation, mother's age, whether the mother has health care coverage, whether the mother has a physical limitation, welfare participation, child's age, work stressors, and financial stressors. Table 1C includes an additional independent variable measured only at period t + 2: whether the mother takes medication for an asthma condition. This is the sole measure available that captures the child's genetic predisposition to asthma.

5. Empirical Findings

Tables 2-8 present the major findings of this study. All estimates in the tables are marginal effects, with t-statistics presented in parentheses. The results in these tables are limited to estimates of the relationship between smoking and asthma. Parameter estimates on control variables for cross-section and DD models are available for a subset of models in Appendixes A and B.

Cross-section Estimates

Table 2 presents probit estimates of the association between parental smoking and asthma by age one using wave 2 data. Column 1 presents the simple correlation, and column 2 adds controls for mother's age, age squared, race, sex, mother's marital status, employment status, education, household income, whether there are other children in the household, and whether the mother reports a physical disability. Column 3 includes these socioeconomic characteristics plus the child's birth weight, which serves as a control for unobserved child health endowment.

The evidence from these models suggests that there is a positive relationship between parental smoking and infant asthma. Moreover, the association is driven by postpartum maternal smoking. This is the case for two key reasons. First, the Fragile Families sample is focused on unmarried mothers, many of whom live apart from the infant's father. Second, mothers are more likely to serve as primary caregivers; thus, their smoking may be more highly correlated with the infant's exposure to secondhand smoke. Controlling for socioeconomic covariates (column 2), mothers' postpartum smoking is associated with a 3.4 percentage point higher probability of an asthma diagnosis at age one. When the infant's birth weight is added to the model to control for the infant's unobserved health endowment, the magnitude and significance of the parameter falls, though it is still marginally significant (column 3). Because mothers' smoking behavior seems to drive the association between parental smoking and infant asthma in this sample, this smoking measure is used in the remainder of the analysis. (11)

Table 3A presents probit estimates of the association between an asthma diagnosis and quantities of cigarettes consumed by the mother. Models for populations at high-risk of asthma--boys, African Americans, Hispanics, and less educated mothers--are estimated separately. The evidence suggests that relative to nonsmokers, mothers who smoked less than half a pack per day were more likely to have infants diagnosed with asthma. However, somewhat surprisingly, the probability of asthma did not rise with increasing quantities of cigarette consumption, except for Hispanics and less-educated mothers. This may be true because the proportion of mothers who reported smoking more than one pack per day is very small (1.0%). Hence, any maternal smoking is used as the key explanatory variable of interest in most of the subsequent tables.

In Table 3B, the estimates of the association between postpartum maternal smoking and three measures of infant asthma during the first year of the child's life are presented. Across all definitions of the dependent variable, there is consistent evidence of a positive relationship between postpartum maternal smoking and childhood asthma, with estimates around 3 percentage points.

Of particular note is the strong relationship between maternal smoking and infant asthma for Hispanics. Maternal smoking after pregnancy is associated with an 11.8 percentage point increase in the probability of a professional asthma diagnosis, an 11.1 percentage point increase in the probability of a parental asthma diagnosis, and an 8.7 percentage point increase in emergency asthma care. As noted in Table 1, Hispanics have the lowest educational attainment levels, with 42.4% of mothers having less than a high school education. Hence, in column 6, models are estimated for those mothers with low education levels, and there is evidence of a large significant effect.

These findings suggest strong evidence of a positive relationship between postpartum maternal smoking and infant asthma, especially for those populations at high risk for asthma. However, one criticism of the estimates presented in Tables 2 and 3 is that asthma diagnoses tend to materialize at later ages than age one. The American Lung Association suggests that asthma diagnoses by age two are more reliable and appropriate. In Table 4A, linear probability model estimates of the relationship between postpartum maternal smoking and child asthma by age three are presented. The smoking measure is whether the mother reported any smoking following the birth of the child. A key additional covariate included in the set of socioeconomic characteristics in these models is a measure of whether the mother is currently taking medication for an asthma condition. (12) This is the only available measure that allows us to examine a specific genetic connection between maternal asthma and the probability of child asthma. As expected, maternal consumption of asthma medication is positively associated with the probability of a child asthma diagnosis. (13)

As in Table 3B, Table 4A shows consistent evidence of a positive relationship between postpartum maternal smoking and child asthma by age three. Postpartum maternal smoking is associated with a 3 to 4 percentage-point higher probability of an asthma diagnosis and a 4 percentage-point higher probability of using emergency care services following a severe asthma attack. Again, the relationships are especially strong among traditionally high-risk populations across all asthma definitions. (14)

In contrast to earlier waves, wave 3 of the Fragile Families data does contain detailed information on the child's exposure to smokers in the household. In column 6 of Table 4A, estimates of the relationship between a household smoker spending positive hours in the same room as the child and the probability of asthma are presented. There is consistent evidence of a positive relationship between a child's exposure to a smoker in the same room and childhood asthma. Column 7 conditions the sample on all smokers and estimates whether the presence of more than one smoker in the household is associated with a higher probability of asthma. There is some evidence that exposure to more smokers in the home is associated with a greater likelihood of asthma. These results suggest evidence that increased exposure to secondhand smoke may increase the likelihood of asthma. (15)

In summary, the findings in Tables 2-4A suggest consistent evidence of a positive relationship between postpartum maternal smoking and child asthma across various model specifications and definitions of the dependent variable. These results are consistent with much of the public health and medical literature (Burchfield et al. 1986; Evans, Levison, and Feldman 1987; O'Connor et al. 1987; Murray and Morrison 1989; Krzyzanowski, Quackenboss, and Lebowitz 1990; Weizman et al. 1990; Gergen et al. 1998; Strachan and Cook 1998; Sabia and Lin 2006).

While there is strong evidence of a positive relationship between postpartum maternal smoking and child asthma, there is little evidence that smoking during pregnancy alone has an impact on child asthma, as shown in Table 4B. The association between maternal smoking and childhood asthma appears to occur after birth, when the infant may be breathing in dangerous particles from secondhand smoke. (16) Given the strong positive association between pre- and postpartum maternal smoking, the lack of significant results for prepartum smoking suggests that secondhand smoke--and not other unobserved health behaviors associated with postpartum smoking--may be a cause of childhood asthma. However, these results are merely suggestive; cross-section estimates of the relationship between postpartum maternal smoking and child asthma may still be upwardly biased in the presence of unmeasured heterogeneity. Next, we turn to fixed effects estimates of the child health production function.

Difference-in-Difference Estimates

Table 5 presents DD estimates of the relationship between the onset of maternal smoking and whether the child has ever had asthma. (17) The sample is restricted to mothers who reported that they were nonsmokers through the second wave of data (age one of the child). (18) This restriction is employed to identify the effects of contemporaneous exposure to secondhand smoke. These estimates may actually understate the true full effects of smoking if lagged maternal smoking affects asthma diagnoses or if smoking effects are cumulative.

It is important to note that because this analysis is conducted on a select sample, the results may not be generalizable to a broader group of mothers. The selected sample consists of approximately 70% of the full sample. In the selected sample, 66% of mothers reported not smoking during the pregnancy or at the child's first or third birthday and about 5% who did not report smoking prior to the child's first birthday reported smoking between the child's first and third birthdays. Thus, the 14% of mothers who smoked continuously (from the pregnancy through the child's third birthday) are excluded from the sample, as are the 6% who did not smoke during pregnancy but did smoke prior to both the child's first and third birthdays, and the 8% who reported quitting smoking sometime during the three years. Later, we examine whether the results are robust to include quitters in the fixed effects sample, as well as more precise measures of between wave changes in quantities of cigarettes consumed.

The DD strategy in Table 5 compares the change in asthma rates of children whose mothers began smoking by the child's third birthday (wave 3) with the change in asthma rates of children whose mothers remained nonsmokers. Parameter estimates on the full set of time-varying controls are available in Appendix B. (19) The findings in Table 5 suggest consistent evidence that the onset of maternal smoking is associated with a significant increase in child asthma, with a change in asthma diagnoses of approximately 7 percentage points. Across two of three definitions of asthma, maternal smoking is consistently associated with a significant increase in ever reporting asthma. (20,21) As in the ordinary least squares (OLS) models, the relationship between maternal smoking and childhood asthma is especially strong for less-educated, high-risk populations. Comparing DD models to OLS models suggests little evidence of unobserved heterogeneity bias, with parameter estimates statistically equivalent.

These findings suggest that the significant positive relationship between postpartum maternal smoking and childhood asthma cannot be explained by fixed unobserved heterogeneity. However, a few important caveats are necessary before concluding the presence of a causal link between postpartum maternal smoking and child asthma. First, the asthma measures used in the analysis are imperfect inasmuch as they may also measure the frequency with which mothers obtain medical care services for their children. If mothers who begin smoking also avail themselves of more medical services for their children than mothers who remain nonsmokers, they are more likely to report an asthma diagnosis. However, in results not presented but available upon request, there is little evidence of a significant relationship between maternal smoking and the number of times that the child had a routine checkup in the last 12 months.

Second, because medical professionals may have some difficulty in diagnosing childhood asthma, it may be the case that maternal smoking is associated with other child respiratory illnesses that may not necessarily lead to asthma at later ages. As Martinez et al. (1995) suggest, wheezing is one such childhood illness that may, in fact, be transient and not result in future asthma. In fact, evidence from the National Maternal Infant Health Survey suggests that maternal smoking is positively associated with a number of childhood respiratory illnesses, including wheezing. Thus, to the extent that maternal smoking induces greater wheezing in children that is misdiagnosed as asthma, the findings here may reflect negative health effects that are more transient than asthma.

Finally, while fixed effects models do control for fixed unobserved heterogeneity, DD estimators could lead to upwardly biased estimates of the effect of maternal smoking on child asthma in the presence of time-varying unobservables that are correlated with both the onset of smoking and childhood asthma, such as unobserved financial stressors, which could directly affect child health production. (22)

One method of addressing this concern with time-varying unobservables is to model unmeasured heterogeneity via instrumental variables (IV). This approach was attempted, with state cigarette taxes, state-level antismoking sentiment, individual-specific exogenous stressors, and first-stage heteroskedasticity each considered as potential exclusion restrictions. The results of IV models are generally consistent with OLS and DD models, which may add to our confidence in causally interpreting DD estimates. (23) However, there remain important concerns about the validity of the exclusion restrictions available in the data. Hence, the focus of the remaining analysis will focus on the credibility of the fixed effects identification strategy. In Table 6, a series of "antitests" are presented to examine whether time-varying unmeasured heterogeneity is an important source of bias in DD estimates.

Falsification Tests

Table 6 presents DD estimates of the relationship between the onset of maternal smoking and other child health outcomes that are not expected to be causally affected by exposure to secondhand tobacco smoke but could be correlated with other unmeasured health input choices that are also associated with smoking. These health outcomes include the following: overnight stays in the hospital, doctor visits for accidents or injuries, emergency room visits for accidents or injuries, physical disabilities, and an overall assessment of health. If the onset of smoking is significantly associated with these other health outcomes, time-varying unmeasured heterogeneity may be an important source of bias in the DD estimates presented in Table 5.

Row 1 presents estimates of the effect of the onset of maternal smoking on asthma diagnoses. As before, there is a significant positive relationship. However, rows 2-7 reflect that the estimated effect of smoking on other health outcomes not believed to be causally affected by smoking are not significant. This evidence lends credibility to a causal interpretation of the results in Table 5 and suggests that time-varying factors related to household health production are not important sources of unmeasured heterogeneity bias. If these factors were important, one would expect to find significant effects for other health-related outcomes.

Table 7 presents another set of falsification tests. In Table 7, the relationship between the onset of alcohol or drug consumption and asthma is estimated. While alcohol consumption and drug use are each expected to be correlated with smoking, one would not expect them to causally affect asthma. The first five columns of Table 7 present estimates of the effect of the onset of any alcohol consumption on the probability of asthma. Across definitions of asthma, there is no evidence of a significant effect of drinking on asthma. Columns 6-10 present estimates of the effect of the onset of binge drinking--defined as consuming four or more drinks in a single sitting in the last month--and asthma. Again, none of the estimated effects is significant.

The final five rows of Table 7 present estimates of the effect of the onset of illegal drug use on the asthma rates. As shown previously, while drug use is expected to be correlated with smoking, one would not expect its use to causally affect asthma. Across specifications, there is no evidence that the onset of alcohol or drug use is associated with an increase in asthma rates. Taken together, the results in Tables 6 and 7 add credibility to the identification assumption underlying the DD approach.

Intensity of Smoking

One concern with the smoking onset estimates in Table 5 is that these models capture the relatively rare event of nonsmoking morns beginning to smoke between their newborn child's first and third birthdays. Thus, perhaps the results are not generalizable to a broader population of single mothers. Moreover, the previous estimates do not account for changes in the quantity of cigarettes consumed but rather only whether the mother smokes. (24) In Table 8, both starters and quitters are included in the sample, and the relationship between the change in quantity of cigarettes smoked and an asthma diagnosis is estimated. The results are consistent with those in Table 5; an increase in cigarette consumption is associated with significantly higher asthma rates. Moreover, in contrast to the OLS estimates in Table 3A, an increase of cigarette consumption of greater than one-half pack per day is associated with a significant increase in asthma diagnoses. (25)

An important limitation to the fixed effects results is they do not explicitly control for reverse causality. One might imagine that an asthma diagnosis could induce mothers to cut back on cigarettes consumed or to quit smoking entirely. This potential source of bias is exacerbated by the lack of precise information on the timing of the asthma diagnoses and on quitting behavior. In results not presented here, there is some evidence to suggest that smoking cessation may be one response to a child asthma diagnoses. (26) This form of endogeneity bias would be expected to bias fixed effects estimates downward since nonsmoking mothers might remain nonsmokers if their child develops asthma. In this sense, the fixed effects estimates presented in Tables 5 and 8 may be considered lower bound estimates.

Taken together, the empirical findings of this study suggest that the significant positive relationship between postpartum maternal smoking and childhood asthma is robust to controls for fixed individual unmeasured heterogeneity, especially for populations that are at high risk for developing child respiratory disease.

6. Conclusions and Policy Implications

The public health literature as well as a recent Surgeon General's report claims that secondhand tobacco smoke may be one of the triggers of childhood asthma. The evidence presented here suggests that postpartum maternal smoking is significantly positively associated with childhood asthma among populations with a high risk of asthma. This finding is robust to the inclusion of individual fixed effects, reflecting that fixed unobserved heterogeneity cannot fully explain the relationship. Difference-in-difference estimates suggest that the onset of maternal smoking is associated with a 7 percentage-point increase in asthma diagnoses by age three and a 4 to 5 percentage-point increase in having to obtain emergency care for a severe asthma attack. The magnitude of the relationship is largest for those households where the mother has less than a high school education, suggesting that the least educated smoking mothers may expose their children to the most secondhand smoke.

While this study has endeavored to establish more credible evidence of a causal link between parental smoking and childhood asthma, an important caveat to these findings is that the results do not suggest that secondhand smoke is the predominant cause of asthma. It is fairly well established in the medical literature that there are other strong predictors of childhood asthma: genetic predisposition, viral infections, and exposure to other indoor and outdoor pollutants. Clearly, exposure to secondhand smoke cannot explain racial differences in asthma rates because postpartum smoking rates among racial minorities is lower than among whites even though childhood asthma rates are higher. Race-specific differences in baseline asthma rates may be due to differences in genetic predisposition to asthma, greater viral infections, or greater exposure to indoor or outdoor allergens.

Moreover, one should exercise care in interpreting these findings as a rationale to support policies to reduce smoking, such as higher cigarette taxes. While secondhand tobacco smoke is a classic negative externality that could, in principle, justify an increase in cigarette taxes on efficiency grounds, it is not clear that the external costs are not already being internalized within the family. Altruistic mothers may internalize the health costs to their children and choose to smoke anyway because they receive a high level of utility from cigarette consumption (Viscusi 1995). Hence, a policy recommendation to raise cigarette taxes to combat child asthma requires evidence that (i) the externality is not already being internalized, (ii) higher cigarette taxes will reduce postpartum maternal smoking, (iii) a reduction in maternal smoking will reduce child asthma rates, and (iv) the health benefits to children from raising cigarette taxes outweigh the costs imposed on other smokers who do not impose negative externalities on others. This paper has presented evidence in support of the third condition. Even under the assumption that conditions (i) and (iv) are satisfied, it is not clear that higher cigarette taxes would have a significant impact on postpartum maternal smoking. The empirical evidence is quite mixed.

A few recent papers have examined the effect of cigarette taxes on maternal smoking and on subsequent child outcomes. Colman, Grossman, and Joyce (2003) estimated the impact of cigarette taxes on smoking before, during, and after pregnancy. While the authors find that a 10% increase in cigarette taxes increases the probability of pregnant women quitting by 10%, they also find that half of all quitters resume smoking within six months of delivery and 75% resume smoking within a year. This finding may suggest that higher cigarette taxes may not have a large impact on new mothers' decision to smoke. Furthermore, while Evans and Ringel (1999) and Ringel and Evans (2001) find that cigarette taxes are a useful policy tool for deterring smoking by pregnant women and improving birth outcomes, (27) there is little evidence on whether cigarette taxes reduce exposure of children to secondhand smoke and improve their postbirth health outcomes. Thus, the evidence is insufficient to conclude that an increase in cigarette taxes would enhance efficiency. (28)

While this study has presented evidence of adverse consequences of postpartum maternal smoking on childhood asthma among at-risk populations, future work utilizing more precise measures of exposure to secondhand smoke and additional child respiratory illnesses would provide a more complete picture of the potential adverse health effects of household smoking. Moreover, longitudinal data that follow children through older ages would allow examination of whether there are long-run adverse health effects.

Appendix A
LPM Estimates of Relationship between Asthma Diagnosis at Age Three and
Postpartum Maternal Smoking (a)

 All Boys
 (1) (2)

Maternal smoking 0.031 * (1.63) 0.052 ** (1.93)
Birth weight (kg) -0.053 *** (3.69) -0.059 *** (3.00)
Mother on asthma meds 0.210 *** (4.12) 0.300 *** (4.17)
Child boy 0.075 *** (4.41) --
Mother's age -0.006 (0.94) -0.016 (1.50)
Mother's age squared 0.000 (1.13) 0.000 (1.59)
Child's age (months) 0.015 *** (3.64) 0.014 ** (2.25)
Black mother 0.067 *** (2.76) 0.079 ** (2.27)
Hispanic mother 0.103 *** (3.87) 0.119 *** (3.10)
Other race mother 0.064 (1.23) 0.109 (1.53)
Married -0.011 (0.39) -0.013 (0.33)
Father present in house -0.057 ** (2.42) -0.086 *** (2.57)
Mother high school -0.045 ** (2.04) -0.056 * (1.77)
Mother some college -0.045 * (1.81) -0.047 (1.32)
Mother college grad -0.045 (1.14) -0.056 (0.97)
Other children 0.065 *** (3.06) 0.089 *** (2.89)
Household income (000s) -0.0001 (0.43) 0.000 (0.29)
Mother health limitation 0.021 (0.69) -0.053 (1.20)
Mother no insurance -0.040 (1.14) -0.060 (1.16)
Mother employed -0.036 ** (1.94) -0.052 ** (1.93)
AFDC receipt 0.029 (1.27) 0.023 (0.70)
Employment-related stress 0.022 (1.27) 0.017 (0.69)
Borrow money 0.023 (1.15) 0.053 * (1.85)
N 2504 1318

 Blacks Hispanics
 (3) (4)

Maternal smoking 0.026 (0.92) 0.120 *** (2.68)
Birth weight (kg) -0.070 *** (3.35) -0.033 (1.00)
Mother on asthma meds 0.208 *** (3.03) 0.364 *** (3.08)
Child boy 0.086 *** (3.37) 0.070 * (1.86)
Mother's age -0.000 (0.01) -0.026 ** (2.20)
Mother's age squared 0.000 (0.12) 0.0005 ** (2.43)
Child's age (months) 0.011 * (1.82) 0.015 * (1.67)
Black mother -- --
Hispanic mother -- --
Other race mother -- --
Married -0.027 (0.57) -0.020 (0.40)
Father present in house -0.062 * (1.85) -0.049 (0.97)
Mother high school -0.053 * (1.64) -0.025 (0.53)
Mother some college -0.085 ** (2.24) -0.007 (0.13)
Mother college grad -0.069 (0.92) -0.017 (0.15)
Other children 0.058 * (1.72) 0.090 ** (1.98)
Household income (000s) -0.0002 (0.01) -0.001 (0.59)
Mother health limitation -0.006 (0.14) 0.128 * (1.76)
Mother no insurance -0.081 (1.36) 0.040 (0.60)
Mother employed -0.031 (1.02) -0.028 (0.67)
AFDC receipt -0.003 (0.09) 0.015 (0.27)
Employment-related stress 0.049 * (1.87) -0.020 (0.50)
Borrow money 0.030 (1.04) 0.084 * (1.86)
N 1258 570

 Less Educated
 (5)

Maternal smoking 0.079 ** (2.31)
Birth weight (kg) -0.044 (1.55)
Mother on asthma meds 0.212 ** (2.27)
Child boy 0.108 *** (3.25)
Mother's age -0.020 (1.27)
Mother's age squared 0.0004 (1.55)
Child's age (months) 0.015 ** (1.93)
Black mother 0.047 (0.92)
Hispanic mother 0.088 * (1.65)
Other race mother 0.067 (0.54)
Married 0.065 (1.20)
Father present in house -0.078 * (1.85)
Mother high school --
Mother some college --
Mother college grad --
Other children 0.075 * (1.63)
Household income (000s) -0.002 (1.54)
Mother health limitation 0.050 (0.97)
Mother no insurance -0.023 (0.39)
Mother employed -0.042 (1.16)
AFDC receipt 0.051 (1.32)
Employment-related stress 0.014 (0.35)
Borrow money -0.002 (0.06)
N 804

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses. All models also include dummy variables
for type of housing unit.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Appendix B
DD Estimates of Effect of Postpartum Maternal Smoking on Asthma
Diagnosis (a)

 All Boys
 (1) (2)

Maternal smoking 0.074 ** (2.35) 0.054 (1.25)
Child's age (months) 0.003 (1.44) 0.003 (0.94)
Married -0.027 (1.31) -0.050 * (1.70)
Father present in house -0.022 (1.25) -0.023 (0.99)
Other children -0.012 (0.58) -0.021 (0.75)
Household income (000s) 0.000 (0.12) 0.000 (0.27)
Mother health limitation -0.003 (0.12) 0.014 (0.42)
Mother no insurance 0.007 (0.32) 0.027 (0.95)
Mother employed -0.015 (1.14) -0.030 * (1.66)
AFDC receipt -0.027 * (1.62) -0.028 (1.26)
Mother's age -0.012 ** (2.01) -0.013 (1.49)
Mother's age squared 0.0002 * (1.66) 0.000 (1.26)
Employment-related stress -0.004 (0.06) 0.011 (0.13)
Borrow money -0.011 (0.77) 0.000 (0.02)
N 1550 816

 Blacks Hispanics
 (3) (4)

Maternal smoking 0.096 ** (2.17) 0.118 * (1.81)
Child's age (months) -0.001 (0.17) 0.010 ** (2.03)
Married -0.025 (0.76) -0.046 (1.33)
Father present in house -0.026 (1.11) -0.036 (0.84)
Other children 0.031 (0.97) 0.001 (0.01)
Household income (000s) 0.000 (0.65) -0.001 (1.19)
Mother health limitation 0.007 (0.22) -0.008 (0.15)
Mother no insurance 0.007 (0.21) 0.021 (0.59)
Mother employed -0.005 (0.26) -0.038 (1.41)
AFDC receipt -0.033 (1.55) -0.039 (0.95)
Mother's age -0.009 (0.78) -0.013 (1.45)
Mother's age squared 0.000 (0.74) 0.000 (1.30)
Employment-related stress -0.015 (0.08) 0.013 (0.13)
Borrow money -0.018 (0.92) 0.013 (0.42)
N 773 394

 Less Educated
 (5)

Maternal smoking 0.142 *** (2.75)
Child's age (months) 0.001 (0.32)
Married 0.070 (1.43)
Father present in house -0.077 ** (2.11)
Other children 0.022 (0.50)
Household income (000s) 0.000 (0.03)
Mother health limitation 0.002 (0.04)
Mother no insurance 0.025 (0.61)
Mother employed -0.015 (0.54)
AFDC receipt 0.004 (0.14)
Mother's age -0.008 (0.61)
Mother's age squared 0.000 (0.36)
Employment-related stress -0.068 (0.61)
Borrow money -0.037 (1.21)
N 402

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses.

* Significant at 10%

** Significant at 5%

*** Significant at 1%

Appendix C1
OLS and Fixed Effects Estimates of Association between Asthma at Age
Three and Onset of Postpartum Maternal Smoking (a)

 Girls

 OLS DD
 (1) (2)

Asthma diagnosis 0.090 (1.38) 0.090 * (1.91)
Asthma attack 0.077 (1.46) 0.044 (1.23)
Emergency care 0.071 (1.48) 0.058 * (1.86)
N 734 734

 White

 OLS DD
 (3) (4)

Asthma diagnosis 0.064 (0.66) -0.006 (0.08)
Asthma attack 0.145 * (1.89) 0.053 (0.97)
Emergency care 0.179 *** (3.21) 0.073 ** (2.23)
N 328 328

 [greater than or equal
 to] HD educ

 OLS DD
 (5) (6)

Asthma diagnosis 0.019 0.012
Asthma attack -0.044 -0.057
Emergency care -0.044 -0.020
N 1148 1148

(a) Coefficient estimates presented; absolute values of t-statistics
in parentheses.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Appendix C2
Changes in Onset of Smoking and Asthma Variables across Waves (a)

 All

 Wave 1 Wave 2

Postpartum smoking 0.053 (0.224)
Ever asthma diagnosis 0.127 (0.333) 0.209 (0.407)
Ever asthma attack 0.080 (0.272) 0.133 (0.339)
Ever emergency care 0.071 (0.257) 0.110 (0.313)
N 1500 1500

 Less Educated

 Wave 1 Wave 2

Postpartum smoking 0.000 0.096 (0.295)
Ever asthma diagnosis 0.167 (0.374) 0.265 (0.442)
Ever asthma attack 0.103 (0.305) 0.167 (0.374)
Ever emergency care 0.096 (0.295) 0.144 (0.352)
N 402 402

(a) Standard deviations appear in parentheses.

Appendix D
IV Estimates of Association between Asthma at Age Three and Postpartum
Maternal Smoking (a,b,c)

 All

 Lewbel
 OLS IV IV
 (1) (2) (3)

Asthma diagnosis 0.037 ** 0.275 * 0.065
 (1.95) (1.77) (1.12)
 Joint F-test F = 13.7 F = 14.3
 SarganOveridtest p = 0.45 p = 0.68
 BSHeterotest -- p = 0.00
Asthma attack 0.024 0.236 * -0.020
 (1.45) (1.74) (0.42)
 Joint F-test F = 12.7 F = 14.4
 SarganOveridtest p = 0.38 p = 0.82
 BSHeterotest -- p = 0.00
Emergency care 0.046 *** 0.259 * 0.048
 (3.02) (1.89) (1.04)
 Joint F-test F = 10.8 F = 13.7
 SarganOveridtest p = 0.53 p = 0.75
 BSHeteroTest -- p = 0.00
N (b) 2504 2504 2504

 Boys

 Lewbel
 OLS IV IV
 (4) (5) (6)

Asthma diagnosis 0.062 ** 0.567 ** 0.095
 (2.30) (2.37) (1.09)
 Joint F-test F = 7.0 F = 6.8
 SarganOveridtest p = 0.23 p = 0.24
 BSHeterotest -- p = 0.00
Asthma attack 0.035 0.341 * 0.031
 (1.46) (1.70) (0.41)
 Joint F-test F = 6.9 F = 6.8
 SarganOveridtest p = 0.55 p = 0.39
 BSHeterotest -- p = 0.00
Emergency care 0.073 *** 0.404 ** 0.177 **
 (3.22) (1.95) (2.35)
 Joint F-test F = 5.8 F = 6.3
 SarganOveridtest p = 0.57 p = 0.66
 BSHeteroTest -- p = 0.00
N (b) 1318 1318 1318

 Blacks

 Lewbel
 OLS IV IV
 (7) (8) (9)

Asthma diagnosis 0.033 0.459 0.096
 (1.16) (1.58) (1.08)
 Joint F-test F = 4.6 F = 7.4
 SarganOveridtest p = 0.44 p = 0.74
 BSHeterotest -- p = 0.00
Asthma attack 0.001 0.297 -0.059
 (0.05) (1.19) (0.77)
 Joint F-test F = 4.3 F = 7.3
 SarganOveridtest p = 0.69 p = 0.91
 BSHeterotest -- p = 0.00
Emergency care 0.030 0.322 0.052
 (1.27) (1.19) (0.70)
 Joint F-test F = 3.5 F = 7.3
 SarganOveridtest p = 0.50 p = 0.91
 BSHeteroTest -- p = 0.00
N (b) 1258 1258 1258

 Hispanics

 Lewbel
 OLS IV IV
 (10) (11) (12)

Asthma diagnosis 0.132 *** 0.280 0.188 ***
 (3.02) (1.22) (2.89)
 Joint F-test F = 6.8 F = 24.5
 SarganOveridtest p = 0.49 p = 0.211
 BSHeterotest -- p = 0.00
Asthma attack 0.116 *** 0.179 0.142 ***
 (3.l2) (0.90) (2.56)
 Joint F-test F = 6.5 F = 24.1
 SarganOveridtest p = 0.12 p = 0.02
 BSHeterotest -- p = 0.00
Emergency care 0.116 *** 0.215 0.157 ***
 (3.40) (1.20) (3.04)
 Joint F-test F = 6.6 F = 22.6
 SarganOveridtest p = 0.10 p = 0.04
 BSHeteroTest -- p = 0.00
N (b) 570 570 570

 Less Educated

 Lewbel
 OLS IV IV
 (13) (14) (15)

Asthma diagnosis 0.085 ** 0.331 0.050
 (2.53) (1.36) (0.24)
 Joint F-test F = 5.4 F = 1.2
 SarganOveridtest p = 0.95 p = 0.67
 BSHeterotest -- p = 0.91
Asthma attack 0.051 * 0.266 0.056
 (1.77) (1.22) (0.32)
 Joint F-test F = 4.9 F = 1.2
 SarganOveridtest p = 0.35 p = 0.59
 BSHeterotest -- p = 0.91
Emergency care 0.085 *** 0.323 -0.004
 (3.05) (1.48) (0.02)
 Joint F-test F = 4.5 F = 1.3
 SarganOveridtest p = 0.56 P = 0.19
 BSHeteroTest -- p = 0.91
N (b) 804 804 804

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses.

(b) Sample sizes reported are for asthma diagnosis outcome. Sample
sizes for other asthma outcomes are approximately 2% smaller because
of missing observations. All models include full set of observables
described in appendix.

(c) While state-specific variations in cigarette taxes and antismoking
sentiment provide one set of plausible instruments, IV models using
these measures as exclusion restrictions did not successfully meet
standard relevance and exogeneity requirements. The psychology and
public health literatures suggest another set of instruments that may
provide exogenous variation in F smoking-individual measures of stress
(see Conway et al. 1981, Rose et al. 1983, House et al. 1986,
Pomerleau and Pomerleau 1987, Green and Johnson 1990, Perkins and
Grobe 1992, Frone et al. 1994, Todd et al. 1996, Schachter 1997, and
Todd, 2004 for a discussion of the psychological and physiological
link between stressors and smoking). F-tests of the joint
F-significance of the instruments suggest that the instruments are
jointly significant at the I% level in each of the models; however,
F-statistics are generally small. Given the possibility of weak
instruments, this suggests some caution in interpreting findings
(Stock and Staiger 1997). Sargan tests of overidentifying restrictions
fail to reject the hypothesis that the exclusion restrictions are
uncorrelated with the error of the asthma equation. The results of the
standard IV models suggest a significant positive relationship between
postpartum maternal smoking and childhood asthma at age three,
findings consistent with the sign on the OLS estimates. However, the
magnitudes of the IV estimates are much larger than OLS estimates.
One x explanation for this finding is that cross-section estimates
of the relationship between maternal smoking and childhood asthma are
biased downward. This may occur if the types of mothers that choose to
smoke have children with the lowest unobserved propensity for
developing asthma. Given that the standard IV estimates appear to be
implausibly large, it may be that family stress is, in fact,
correlated with unmeasured determinants of child asthma, and Sargan
tests of overidentifying restrictions are insufficiently strong to
detect this correlation. An alternative IV approach includes stress
measures in both the smoking and asthma equations. Identification is
achieved via heteroskedasticity in the smoking equation (Lewbel, 2006).
Across these models, IV estimates show a consistent, though
imprecisely estimated, relationship between maternal smoking and
childhood asthma. The magnitudes of these IV estimates appear
plausible, and Hausman tests reveal that they are statistically
equivalent to OLS estimates. Conservatively, maternal smoking is
associated with a 4 percentage point higher probability of asthma,
with the greatest estimated magnitude among racial minorities. Taken
together, the IV results are generally consistent with a causal
interpretation of the positive relationship between maternal smoking
and childhood asthma, but the possible lack of valid exclusion
restrictions and the reliance on technical identification suggests
that caution be taken in interpreting IV results.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Thanks to Bisakha Sen, Jenny Williams, Don Kenkel, Julie Hotchkiss, three anonymous referees, and participants at the Western Economic Association International meetings in June 2006 for useful comments and suggestions on earlier drafts of this paper. Thanks also to Nikki Williams for excellent editorial assistance.

Received April 2006; accepted January 2007.

References

American Academy of Pediatrics. 2000. Guide to your child's allergies and asthma. New York: Villard Books.

Asthma Priority Area Working Group, President's Task Force on Environmental Health Risks and Safety Risks to Children. 2000. Asthma and the Environment: A Strategy to Protect Children. Accessed 24 August 2006. Available from: http://aspe.hhs.gov/sp/asthma/appxd.pdf.

Bierman, C. W. 1996. Environmental control of asthma. Immunology and Allergy Clinics of North America 16:753-64.

Burchfield, C. M., M. W. Higgins, J. B. Keller, W. F. Howatt, W. J. Butler, and I. T. T. Higgins. 1986. Passive smoking in childhood: Respiratory conditions and pulmonary function in Tecumseh, Michigan. American Review of Respiratory Disease 133:966-73.

Busse, W. W., J. E. Gern, and E. C. Dick. 1997. The role of respiratory viruses in asthma. Ciba Foundation Symposium 206:208-13.

Carmona, Richard. 2006. New Surgeon General's Report Focuses on the Effects of Secondhand Smoke. Accessed 31 July 2006. Available from: http://www.hhs.gov/news/press/2006pres/20060627.html.

Colman, Greg, Michael Grossman, and Ted Joyce. 2003. The effect of cigarette taxes on smoking before, during and after pregnancy. Journal of Health Economics 22:1053-72.

Conway, T. L., R. R. Vickers, H. W. Ward, and R. H. Rahe. 1981. Occupational stress and variation in cigarette, coffee, and alcohol consumption. Journal of Health and Social Behavior 22:155-65.

Department of Health and Human Services. 1997. Action against Asthma: A Strategic Plan for the Department of Health and Human Services. Accessed 21 July 2006. Available from: http://aspe.hhs.gov/sp/asthma/overview.htm.

Dockery, D. W., J. H. Ware, B. G. Ferris, Jr., F. E. Speizer, N. R. Cook, and S. M. Herman. 1982. Change in pulmonary function in children associated with air pollution episodes. Journal of the Air Pollution Control Association 32:937-42.

Evans, D., J. Levison, and C. H. Feldman. 1987. The impact of passive smoking on emergency room visits of urban children with asthma. American Review of Respiratory Disease 135:567-72.

Evans, William N., and Jeanne S. Ringel. 1999. Can higher cigarette taxes improve birth outcomes? Journal of Public Economics 72:135-54.

Frone, M. R., M. L. Cooper, and M. Russell. 1994. Stressful life events, gender, and substance use: An application of Tobit regression. Psychology of Addictive Behaviors 8:59-70.

Gergen, Peter J., Jean A. Fowler, Kurt R. Maurer, William W. Davis, and Mary D. Overpeck. 1998. The burden of environmental tobacco smoke exposure on the respiratory health of children 2 months through 5 years of age in the United States: Third national health and nutrition examination survey, 1988 to 1994. Pediatrics 101:1-6.

Green, K. L., and J. V. Johnson. 1990. The effects of psychosocial work organization on patterns of cigarette smoking among male chemical plant employees. American Journal of Public Health 80:1368-71.

House, J. S., V. J. Strecher, H. L. Metzner, and C. A. Robbins. 1986. Occupational stress and health among men and women in the Tecumseh community health study. Journal of Health and Social Behavior 27:62-77.

Institute of Medicine, National Academy of Science. 2000. Clearing the air: Asthma and indoor air exposures 2000. Washington, DC: National Academy Press.

Koren, S. H. 1995. Associations between criteria air pollution and asthma. Environmental Health Perspectives 103:235-42.

Krzyzanowski, M., J. J. Quackenboss, and M. D. Lebowitz. 1990. Chronic respiratory effects of indoor formaldehyde exposure. Environmental Research 52:117-25.

Lefcoe, N. M., and I. I. Inculet. 1971. Particulates in domestic premises. Archives of Environment Health 22:230-8.

Levine, D. B., and E. C. Bryant. 1997. An examination of alternate approaches to selecting a sample of new births. Paper presented for The National Center for Education Statistics, Office of Educational Research and Improvement, U.S. Department of Education.

Lewbel, Arthur. 2006. Identification of heteroskedastic endogenous or mismeasured regressor models. Working Paper, Boston College.

Mannino, D. M., D. M. Homa, and C. A. Pertowski. 1998. Surveillance for asthma--United States, 1960-1995. CDC Surveillance Summaries 47(SS-1):1-28.

Martinez, F. D., A. L. Wright, L. M. Taussig, C. J. Holberg, M. Halonen, and W. J. Morgan. 1995. Asthma and wheezing in the first six years of life. New England Journal of Medicine 332:133-8.

Murray, A. B., and B. J. Morrison. 1989. Passive smoking by asthmatics: Its greater effect on boys than on girls and on older than on younger children. Pediatrics 84:451-9.

National Asthma Education and Prevention Program. 1997. Guidelines for the diagnosis and management of asthma, National Institutes of Health, National Heart, Lung, and Blood Institute. NIH Publication No. 97-4051: 1-146.

National Institute of Environmental Health Sciences. 2007. Background: Research Studies on Asthma, Accessed April 20, 2008. Available from: http://www.niehs.nih.gov/health/topics/ conditions/asthma/studies/background.cfm.

National Research Council. 1986. Environmental tobacco smoke." Measuring exposures and assessing health effects. Washington, DC: National Academy Press.

Neidell, Matthew J. 2004. Air pollution, health, and socio-economic status: The effects of outdoor air quality on childhood asthma. Journal of Health Economics 23:1209-36.

O'Connor, G. T., S. T. Weiss, I. B. Tager, and F. E. Speizer. 1987. The effect of passive smoking on pulmonary function and nonspecific bronchial responsiveness in a population-based sample of children and young adults. American Review of Respiratory Disease 135:800-4.

Perkins, K. A., and J. E. Grobe. 1992. Increased desire to smoke during acute stress. British Journal of Addiction 87:1037-40.

Platts-Mills, T. A., and M. C. Carter. 1997. Asthma and indoor exposure to allergens. New England Journal of Medicine 336:1382-4.

Pomerleau, C. S., and O. F. Pomerleau. 1987. The effects of a psychological stressor on cigarette smoking and subsequent behavioral and physiological responses. Psychophysiology 24:278-85.

Report of the Surgeon General. 2006. The health consequences of involuntary exposure to tobacco smoke. Washington, DC: U.S. Department of Health and Human Services.

Ringel, Jeanne S., and William N. Evans. 2001. Cigarette taxes and smoking during pregnancy. American Journal of Public Health 91:1851-6.

Rose, J. E., S. Ananda, and M. E. Jarvik. 1983. Cigarette smoking during anxiety-provoking and monotonous tasks. Addictive Behaviors 8:353-9.

Sabia, Joseph J., and Tsui-Fang Lin. 2006. Child respiratory health and household smoking. Mimeograph, Department of Public Administration & Policy, American University, Working paper.

Schachter, S. 1977. Studies of the interaction of psychological and pharmacological determinants of smoking: Nicotine regulation in heavy and light smokers. Journal of Experimental Psychology 106:5-12.

Stock, James, and Douglas Staiger. 1997. Instrumental variables regression with weak instruments. Econometrica 65:557-86.

Strachan, David P., and Derek G. Cook. 1998. Parental smoking and childhood asthma: Longitudinal and case-control studies. Thorax 53:204-12.

Todd, Michael. 2004. Daily processes in stress and smoking: Evidence of negative events, nicotine dependence, and gender. Psychology of Addictive Behaviors 18:31-9.

Todd, M., L. Chassin, C. C. Presson, and S. J. Sherman. 1996. Role stress, role socialization, and cigarette smoking: Examining multiple roles and moderating variables. Psychology of Addictive Behaviors 10:211-21.

Viscusi, Kip. 1995. Cigarette taxation and the social consequences of smoking. In Tax policy and the economy 9, edited by James M. Poterba. Cambridge, MA: MIT Press.

Warner, A. M., B. Bjorksten, A. K. Munir, C. Moiler, C. Schou, and N. I. Kjellman. 1996. Childhood asthma and exposure to indoor allergens: Low mite levels are associated with sensitivity. Pediatric Allergy & Immunology 7:61-7.

Weizman, M., S. Gortmaker, D. K. Walker, and A. Sobol. 1990. Maternal smoking and childhood asthma. Pediatrics 85:505-11.

(1) The most common policies include workplace smoking bans, restaurant smoking bans, and school smoking regulations. More recently, several state court actions and legislative decisions have restricted smoking rights of foster parents. Maine, Oklahoma, and Vermont have each passed regulations forbidding foster parents from smoking around foster children, and recent court decisions in Virginia and Maryland have conditioned visitation rights among divorced parents on refraining from smoking when in their children's presence.

(2) Assume that a parent maximizes utility, which is a function of child health and private consumption, subject to income and time constraints. Equation 1 may be considered the dimension of child health production.

(3) Sampling of mothers takes place at hospitals, where there is a higher response rate than from using birth records. Work by Levine and Bryant (1997) found that the 1988 National and Maternal Infant Health Survey, which used birth record data, completed interviews with only 80% of mothers. Because the Fragile Families study focused on nonmarital births, birth record data would have likely led to an even smaller completion rate. While most existing data on unwed fathers are flawed because of high rates of missing fathers and lack of information on the selection process by which fathers participate in these surveys, the Fragile Families Study is a population-based survey with a low rate of missing fathers.

(4) The study's authors note that stratification was not geographic, but based on policy environments and labor market conditions.

(5) The 20 U.S. cities included in the survey were Oakland, San Jose, San Antonio, Austin, Corpus Christi, Milwaukee, Chicago, Indianapolis, Nashville, Detroit, Toledo, Pittsburgh, Richmond, Jacksonville, Newark, Boston, New York City, Philadelphia, Baltimore, and Norfolk.

(6) Means and standard deviations of variables for newborn girls, white non-Hispanics, and households with mothers who have at least a high school education are available upon request. Race is measured by reported race of the biological mother.

(7) For diagnoses by age three, 22% of uninsured mothers reported a professional diagnosis compared with 26% of insured mothers.

(8) The exact questions are (i) "Has a doctor or health care professional ever told you that your child has asthma?" (ii) "Since your child was born, has he had an episode of asthma or an asthma attack?" and (iii) "'Since the child was born, did your child have to visit an emergency room or urgent care center because of asthma?" The correlation coefficients between asthma measures are presented in Table 1B. Of all mothers who reported an asthma diagnosis, 62% reported that their child had an asthma attack. Of those who said their child had had an asthma attack, 100% had received a professional diagnosis. Similarly, of all mothers reporting that they had taken their infant for urgent emergency care because of an asthma attack, 100% had received a diagnosis; 79% of mothers who made a diagnosis of an asthma attack had taken their child for urgent emergency care. A similarly strong correlation exists for asthma rates at age three.

(9) For example, a similar measurement problem exists in the National Maternal and Infant Health Survey.

(10) The cigarette consumption questions are slightly different in each wave of data. In the first wave--which asks questions about smoking behavior during the pregnancy--there are four categories of answers: "'none," "'less than one pack a day,'" "one or more but less than two packs," and "two or more packs a day." In the second wave, the categories include "none," "less than half a pack a day," "about a pack," "a pack and a half," "about two packs," and "more than two packs." Postpartum maternal smoking questions are asked about cigarette consumption in the last month.

(11) Alternate subsequent models using "any parent in the household smoking" produce similar findings to those presented.

(12) This piece of data is only available in wave 3 of the data (when the children are approximately three years old). This is why these data are not included in the analysis of asthma by age one, using wave 2 data.

(13) Parameter estimates on this and other coefficients are available in Appendix A.

(14) The findings in Table 4A do not appear to be driven by smoking mothers being more able to avail themselves of more medical care services than nonsmokers, thus resulting in a higher probability of an asthma diagnosis. In wave 3, mothers were asked whether they have ever been prevented from taking their children to a doctor or to the hospital because of financial constraints. When I restrict the sample to those mothers who reported that they had not felt such financial constraint, the results are similar to those presented here. However, this finding does not rule out the possibility that there are other mother-specific unobservable preferences that are positively correlated with smoking and frequent doctor visits. It simply suggests that financial constraints are not driving the results.

(15) A second way to examine exposure to secondhand smoke would be to interact mother's employment outside the home and smoking. One might expect that employed mothers who smoke are less likely to expose their children to smoke than unemployed mothers who smoke. While the sign on the interaction of these variables was negative, it was not significant in any specification.

(16) This is not to suggest that maternal smoking during pregnancy may not have adverse effects on newborns' health. As noted previously, these data (and other research) show a strong negative association between prepartum smoking and low birth weight.

(17) The coding of this variable assumes that if a mother reported her child had ever had asthma at age one, the child was coded as having asthma at age three.

(18) Mothers who smoked during pregnancy are excluded from the fixed effects sample. However, results are robust to their inclusion.

(19) Included among these time-varying observables are insurance coverage, employment, presence of other children, household income, marital status, mother's physical limitation, AFDC participation, work-related stress, borrowing money from friends, and feeling trapped by one's family situation. Controls for age are also included.

(20) Alternative models that restrict the sample to only those who

were non-smokers during pregnancy, but include mothers who smoked during the first year of the child's life, produce similar results to those presented here, though the magnitude of the estimated relationship between smoking and asthma are smaller, as expected. This is because infants who develop asthma between age one and three are assumed not to be affected by mothers who are smoking in both waves of data.

(21) Appendix CI presents OLS and fixed effects estimates for non high-risk children: newborn girls, white non Hispanics, and children in households with mothers who have at least a high school education. Appendix C2 shows between wave changes in smoking and asthma measures in the DD sample.

(22) Similarly, fixed effects estimates could be biased downward if mothers acquire new information about their children's health endowments between age one and age three and change their smoking behavior accordingly. For example, mothers who discover that their children are not particularly susceptible to respiratory illnesses may start smoking, while those that find out that their children do have more fragile health conditions may choose to remain nonsmokers.

(23) See Appendix D for IV estimates using individual-specific stressers as instruments. Footnote c in this table explains these results in more detail.

(24) As Appendix C2 notes, only 5.4% of the sample of all mothers and 10% of less-educated mothers began smoking between waves; 16.9% of mothers in the sample changed their quantity of cigarettes smoked between waves. Approximately 7% reported a change in smoking one pack or more of cigarettes per day.

(25) An explanation for the difference in findings is that there is greater between-wave variation in quantities of cigarettes consumed in the DD sample than in the OLS sample.

(26) These regression results are available upon request of the author.

(27) These authors find that a 10% increase in cigarette taxes reduces the smoking participation of pregnant women by 5%. Further, they find that a 1-cent increase in state cigarette taxes is associated with a 0.21 gram increase in birth weight.

(28) Additionally, in a general equilibrium framework, higher cigarette taxes may cause mothers to substitute away from cigarette consumption and toward drug or alcohol consumption. While this may improve asthma outcomes, it could lead to worse effects on child health and functioning.

Joseph J. Sabia, American University, Department of Public Administration & Policy, School of Public Affairs, Ward Circle Building, Washington, DC 20016, USA; E-mail sabia@american.edu.

Table 1A. Means and Standard Deviations of Variables at Baseline

 All Boys
 (1) (2)

Dependent variables
 Asthma diagnosis 0.136 (0.343) 0.165 (0.372)
 Asthma attack 0.081 (0.273) 0.104 (0.306)
 Emergency care 0.076 (0.266) 0.096 (0.295)

Smoking variables (unconditional)

 Any parent smokes 0.495 (0.500) 0.503 (0.500)
 Any parent in household smokes 0.381 (0.486) 0.382 (0.486)
 Father smokes 0.411 (0.492) 0.420 (0.494)
 Father in household smokes 0.198 (0.398) 0.200 (0.400)
 Prepartum maternal smoking 0.196 (0.397) 0.184 (0.387)
 Postpartum maternal smoking 0.274 (0.446) 0.271 (0.445)
 Mother smokes less than half 0.205 (0.404) 0.203 (0.403)
 pack/day
 Mother smokes one pack/day 0.059 (0.235) 0.058 (0.233)
 Mother smokes more than one 0.010 (0.100) 0.010 (0.102)
 pack/day

Smoking variables (conditional)

 Mother smokes less than half 0.749 (0.434) 0.748 (0.435)
 pack/day
 Mother smokes one pack/day 0.215 (0.411) 0.214 (0.411)
 Mother smokes more than one 0.037 (0.189) 0.038 (0.192)
 pack/day

Socioeconomic variables

 Mother's marital status 0.251 (0.434) 0.244 (0.430)
 Mother working 0.558 (0.497) 0.561 (0.496)
 Mother with less than high 0.305 (0.461) 0.302 (0.459)
 school education
 Mother high school education 0.313 (0.464) 0.319 (0.466)
 Mother some college 0.267 (0.442) 0.263 (0.441)
 Mother's age 26.44 (6.050) 26.35 (6.030)
 Mother health limitation 0.069 (0.253) 0.069 (0.253)
 Other biological kids 0.616 (0.486) 0.613 (0.487)
 No health insurance 0.091 (0.288) 0.088 (0.283)
 Household income (000s dollars) 32.24 (34.27) 31.81 (34.57)
 Sex of child (boy= 1) 0.529 (0.499) --
 Black, non-Hispanic 0.490 (0.500) 0.492 (0.500)
 Hispanic 0.236 (0.424) 0.231 (0.421)
 Other race 0.034 (0.180) 0.034 (0.182)
 Birth weight (grams) 3233.8 (609.1) 3278.9 (619.3)
 Child's age (months) 15.1 (3.580) 15.0 (3.490)
 Father present in household 0.577 (0.494) 0.574 (0.495)
 Employment related stress 0.385 (0.487) 0.381 (0.486)
 Borrow money from relatives 0.237 (0.425) 0.230 (0.421)
N 3245 1717

 Blacks Hispanics
 (3) (4)

Dependent variables
 Asthma diagnosis 0.175 (0.380) 0.139 (0.346)
 Asthma attack 0.108 (0.311) 0.077 (0.267)
 Emergency care 0.108 (0.311) 0.068 (0.252)

Smoking variables (unconditional)

 Any parent smokes 0.537 (0.499) 0.403 (0.491)
 Any parent in household smokes 0.374 (0.484) 0.313 (0.463)
 Father smokes 0.450 (0.498) 0.343 (0.475)
 Father in household smokes 0.164 (0.371) 0.186 (0.389)
 Prepartum maternal smoking 0.204 (0.403) 0.096 (0.294)
 Postpartum maternal smoking 0.272 (0.445) 0.187 (0.390)
 Mother smokes less than half 0.210 (0.407) 0.161 (0.368)
 pack/day
 Mother smokes one pack/day 0.052 (0.221) 0.022 (0.148)
 Mother smokes more than one 0.010 (0.100) 0.003 (0.063)
 pack/day

Smoking variables (conditional)

 Mother smokes less than half 0.770 (0.422) 0.859 (0.349)
 pack/day
 Mother smokes one pack/day 0.194 (0.396) 0.121 (0.327)
 Mother smokes more than one 0.037 (0.189) 0.020 (0.141)
 pack/day

Socioeconomic variables

 Mother's marital status 0.133 (0.340) 0.224 (0.417)
 Mother working 0.576 (0.494) 0.493 (0.500)
 Mother with less than high 0.313 (0.464) 0.424 (0.495)
 school education
 Mother high school education 0.367 (0.482) 0.285 (0.452)
 Mother some college 0.263 (0.440) 0.246 (0.431)
 Mother's age 25.76 (5.790) 25.80 (5.620)
 Mother health limitation 0.078 (0.268) 0.060 (0.238)
 Other biological kids 0.661 (0.474) 0.601 (0.490)
 No health insurance 0.075 (0.264) 0.135 (0.342)
 Household income (000s dollars) 24.70 (25.34) 26.37 (26.07)
 Sex of child (boy= 1) 0.531 (0.499) 0.518 (0.500)
 Black, non-Hispanic -- --
 Hispanic -- --
 Other race -- --
 Birth weight (grams) 3215.6 (626.1) 3321.8 (563.3)
 Child's age (months) 16.1 (3.640) 14.1 (3.070)
 Father present in household 0.440 (0.497) 0.657 (0.475)
 Employment related stress 0.380 (0.485) 0.362 (0.481)
 Borrow money from relatives 0.263 (0.441) 0.200 (0.401)
N 1590 764

 <HS Ed
 (5)

Dependent variables
 Asthma diagnosis 0.184 (0.388)
 Asthma attack 0.101 (0.302)
 Emergency care 0.107 (0.309)

Smoking variables (unconditional)

 Any parent smokes 0.645 (0.479)
 Any parent in household smokes 0.492 (0.500)
 Father smokes 0.534 (0.499)
 Father in household smokes 0.226 (0.419)
 Prepartum maternal smoking 0.294 (0.456)
 Postpartum maternal smoking 0.387 (0.487)
 Mother smokes less than half 0.276 (0.447)
 pack/day
 Mother smokes one pack/day 0.090 (0.286)
 Mother smokes more than one 0.021 (0.144)
 pack/day

Smoking variables (conditional)

 Mother smokes less than half 0.710 (0.454)
 pack/day
 Mother smokes one pack/day 0.235 (0.425)
 Mother smokes more than one 0.055 (0.228)
 pack/day

Socioeconomic variables

 Mother's marital status 0.095 (0.293)
 Mother working 0.361 (0.480)
 Mother with less than high --
 school education
 Mother high school education --
 Mother some college --
 Mother's age 23.84 (5.270)
 Mother health limitation 0.107 (0.309)
 Other biological kids 0.660 (0.474)
 No health insurance 0.117 (0.322)
 Household income (000s dollars) 16.75 (18.06)
 Sex of child (boy= 1) 0.524 (0.500)
 Black, non-Hispanic 0.502 (0.500)
 Hispanic 0.327 (0.469)
 Other race 0.020 (0.141)
 Birth weight (grams) 3199.2 (581.8)
 Child's age (months) 15.1 (3.550)
 Father present in household 0.465 (0.499)
 Employment related stress 0.369 (0.483)
 Borrow money from relatives 0.266 (0.442)
N 990

Standard deviations in parentheses.

Table 1B. Correlation Coefficients between Asthma Measures at Baseline

 Asthma Diagnosis Asthma Attack Emergency Care

Asthma diagnosis 1.000 -- --
Asthma attack 0.750 *** 1.000 --
Emergency care 0.725 *** 0.793 *** 1.000
N 3245 3245 3245

*** Significant at 1%.

Table 1C. Means and Standard Deviations of Key Variables at Age Three

 All Boys
 (1) (2)

Dependent variables
 Asthma diagnosis (a) 0.256 (0.437) 0.291 (0.455)
 Asthma attack (a) 0.159 (0.366) 0.190 (0.392)
 Emergency care (a) 0.133 (0.339) 0.164 (0.370)
Smoking variables
 Prepartum maternal smoking 0.208 (0.406) 0.192 (0.394)
 Any postpartum maternal 0.337 (0.473) 0.336 (0.473)
 smoking
 Number of smokers in 1.49 (0.746) 1.48 (0.737)
 household (b)
 Mother smokes in home (b) 0.417 (0.493) 0.400 (0.419)
 Smoker in room with child 0.225 (0.418) 0.232 (0.422)
Maternal asthma measure
 Mother taking medication 0.030 (0.172) 0.032 (0.176)
 for asthma
N 2504 1318

 Blacks Hispanics
 (3) (4)

Dependent variables
 Asthma diagnosis (a) 0.297 (0.457) 0.284 (0.451)
 Asthma attack (a) 0.189 (0.392) 0.176 (0.381)
 Emergency care (a) 0.172 (0.378) 0.138 (0.345)
Smoking variables
 Prepartum maternal smoking 0.219 (0.414) 0.100 (0.300)
 Any postpartum maternal 0.346 (0.476) 0.261 (0.440)
 smoking
 Number of smokers in 1.41 (0.677) 1.47 (0.785)
 household (b)
 Mother smokes in home (b) 0.515 (0.500) 0.307 (0.463)
 Smoker in room with child 0.282 (0.450) 0.111 (0.314)
Maternal asthma measure
 Mother taking medication 0.038 (0.192) 0.026 (0.160)
 for asthma 1258 570
N
 <HS Ed
 (5)

Dependent variables
 Asthma diagnosis (a) 0.331 (0.471)
 Asthma attack (a) 0.201 (0.401)
 Emergency care (a) 0.178 (0.383)
Smoking variables
 Prepartum maternal smoking 0.308 (0.462)
 Any postpartum maternal 0.461 (0.499)
 smoking
 Number of smokers in 1.58 (0.825)
 household (b)
 Mother smokes in home (b) 0.456 (0.499)
 Smoker in room with child 0.298 (0.458)
Maternal asthma measure
 Mother taking medication 0.035 (0.183)
 for asthma
N 804

Standard deviations in parentheses.

(a) Asthma variables capture whether there was any asthma detected at
age one or at age three. b Measures are conditional on it being
a smoking household.

Table 2. Probit Estimates of Association between Postpartum Parental
Smoking and Child Asthma Diagnosis at Age One (Dependent Variable = 1
if Asthma Diagnosis; 0 if not) (a,b)

 Socioeconomic
 No Covariates Covariates
 (1) (2)
Smoking covariate
 Either parent smokes 0.046 *** (3.36) 0.022 * (1.89)
 Either parent in 0.037 ** (2.26) 0.032 ** (1.89)
 household smokes
 Father smokes 0.037 *** (3.20) 0.007 (0.60)
 Father in household 0.001 (0.14) -0.008 (0.15)
 smokes
 Mother smokes 0.46 ** (2.42) 0.034 ** (1.93)
N 2344 2344

 Covariates and
 Birth Weight
 (3)
Smoking covariate
 Either parent smokes 0.011 (0.84)
 Either parent in 0.025 * (1.70)
 household smokes
 Father smokes 0.003 (0.27)
 Father in household 0.010 (0.57)
 smokes
 Mother smokes 0.028 * (1.63)
N 2344

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses. Each estimate presented in this
table comes from a separate regression model.

(b) Model (1) is a simple correlation; (2) includes controls for age,
age squared, race, sex, marital status, employment status, education,
household income, whether higher-order birth, and whether mother
reports a disability; (3) adds birth weight to the preceding
covariates.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Table 3A. Probit Estimates of Association between Child Asthma
Diagnosis at Age One and Quantity of Postpartum Cigarettes Consumed
(Dependent Variable = 1 if Asthma Diagnosis; 0 if not) (a)

 Socioeconomic Covariates and
 Covariates Birth Weight
 (1) (2)

Independent variable
 <Half pack/day (b) 0.037 ** (2.40) 0.029 * (1.88)
 One pack/day (b) 0.240 (0.98) 0.011 (0.49)
 >One pack/day (b) -0.001 (0.03) -0.010 (0.24)
N 3245 3245

 Boys Blacks
 (3) (4)

Independent variable
 <Half pack/day (b) 0.043 * (1.87) 0.014 (0.62)
 One pack/day (b) 0.150 (0.32) -0.028 (0.99)
 >One pack/day (b) -0.009 (0.89) -0.072 (0.88)
N 1717 1590

 Hispanics Less Educated
 (5) (6)

Independent variable
 <Half pack/day (b) 0.098 *** (3.92) 0.046 * (1.68)
 One pack/day (b) 0.217 *** (2.93) 0.092 * (1.64)
 >One pack/day (b) 0.410 ** (2.48) 0.075 (1.14)
N 764 990

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses.

(b) Omitted category is nonsmokers.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Table 3B. Robustness of Estimated Effect of Any Postpartum Smoking at
Age One to Definitions of Child Asthma a(Dependent Variable = 1 if
Asthma; 0 if Not)

 Socioeconomic Include Birth
 Covariates Weight
 (1) (2)

Dependent variable
 Asthma diagnosis 0.034 *** (2.52) 0.025 * (1.87)
 Asthma attack 0.017 * (1.61) 0.013
 Emergency care 0.023 ** (2.38) 0.017 * (1.80)
N 3245 3245

 Boys Blacks
 (3) (4)

Dependent variable
 Asthma diagnosis 0.037 * (1.85) 0.003 (0.16)
 Asthma attack 0.026 * (1.60) -0.022 (1.27)
 Emergency care 0.029 ** (1.95) -0.004 (0.17)
N 1717 1590

 Less
 Hispanics Educated
 (5) (6)

Dependent variable
 Asthma diagnosis 0.118 *** (3.59) 0.054 ** (2.03)
 Asthma attack 0.111 *** (4.43) 0.014 (0.70)
 Emergency care 0.087 *** (3.87) 0.019 (0.96)
N 764 990

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Table 4A. LPM Estimates of Association between Postpartum Maternal
Smoking and Asthma at Age Three (a,b) (Dependent Variable = 1 if
Asthma; 0 if Not)

 Any Postpartum Smoking

 All Boys
 (1) (2)

Dependent variable
 Asthma diagnosis 0.031 * (1.63) 0.052 * (1.91)
 Asthma attack 0.019 (1.15) 0.028 (1.17)
 Emergency care 0.043 *** (2.83) 0.069 *** (2.98)
[N.sup.b] 2504 1318

 Any Postpartum Smoking

 Blacks Hispanics
 (3) (4)

Dependent variable
 Asthma diagnosis 0.026 (0.92) 0.120 *** (2.68)
 Asthma attack -0.006 (0.24) 0.098 ** (2.31)
 Emergency care 0.026 (1.09) 0.107 *** (3.01)
[N.sup.b] 1258 570

 Smoker in Room
 Any Postpartum w/ Child for
 Smoking Positive Hours

 Less Educated All
 (5) (6)

Dependent variable
 Asthma diagnosis 0.079 ** (2.31) 0.083 *** (3.91)
 Asthma attack 0.043 (1.46) 0.056 *** (3.03)
 Emergency care 0.079 *** (2.75) 0.090 *** (5.15)
[N.sup.b] 804 2551

 2-4 Smokers
 in House vs.
 1 Smoker

 All
 (7)

Dependent variable
 Asthma diagnosis 0.048 * (1.63)
 Asthma attack 0.020 (0.79)
 Emergency care 0.009 (0.38)
[N.sup.b] 948

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses.

(b) Sample sizes reported are for asthma diagnosis outcome. Sample
sizes for other asthma outcomes are approximately 2% smaller because
of missing observations.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Table 4B. Probit Estimates of Association between Prepartum Maternal
Smoking and Asthma at Age Three. a (Dependent Variable = 1 if
Asthma; = 0 if Not)

 All Boys
 (1) (2)

Dependent variable
 Asthma diagnosis -0.008 (0.33) 0.013 (0.39)
 Asthma attack -0.021 (0.20) -0.001 (0.02)
 Emergency care 0.01 (0.52) 0.048 * (1.66)
[N.sup.b] 2539 1302

 Blacks Hispanics
 (3) (4)

Dependent variable
 Asthma diagnosis 0.005 (0.14) 0.006 (0.08)
 Asthma attack -0.038 (1.29) 0.013 (0.58)
 Emergency care -0.005 (0.19) 0.055 (1.02)
[N.sup.b] 1274 585

 Less Educated
 (5)

Dependent variable
 Asthma diagnosis 0.019 (0.49)
 Asthma attack -0.005 (0.14)
 Emergency care 0.014 (0.42)
[N.sup.b] 818

(a) All estimates presented are marginal effects with absolute values
of t-statistics in parentheses.

(b) Sample sizes reported are for asthma diagnosis outcome. Sample
sizes for other asthma outcomes are approximately 2% smaller because
of missing observations.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Table 5. DD Estimates of Association between Onset of Postpartum
Maternal Smoking and Childhood Asthma' (Dependent Variable = 1 if
Asthma; 0 if Not)

 All

 OLS DD
 (1) (2)

Dependent variable
 Asthma diagnosis 0.063 (1.38) 0.074 ** (2.35)
 Asthma attack 0.024 (0.61) 0.011 (0.54)
 Emergency care 0.054 (1.52) 0.042 * (1.88)
[N.sup.b] 1550 1550

 Boys

 OLS DD
 (3) (4)

Dependent variable
 Asthma diagnosis 0.040 (0.61) 0.054 (1.25)
 Asthma attack -0.012 (0.21) -0.021 (0.56)
 Emergency care 0.042 (0.78) 0.018 (0.57)
[N.sup.b] 816 816

 Blacks

 OLS DD
 (5) (6)

Dependent variable
 Asthma diagnosis 0.066 (0.97) 0.096 ** (2.17)
 Asthma attack -0.004 (0.08) 0.000 (0.00)
 Emergency care 0.009 (0.17) 0.012 (0.34)
[N.sup.b] 773 773

 Hispanics
 OLS DD
 (7) (8)

Dependent variable
 Asthma diagnosis 0.058 (0.62) 0.118 * (1.81)
 Asthma attack -0.023 (0.31) 0.048 (0.89)
 Emergency care 0.070 (1.01) 0.117 *** (2.60)
[N.sup.b] 394 394

 Less Educated

 OLS DD
 (9) (10)

Dependent variable
 Asthma diagnosis 0.173 ** (2.24) 0.142 *** (2.75)
 Asthma attack 0.144 ** (2.19) 0.091 ** (2.11)
 Emergency care 0.209 *** (3.40) 0.115 *** (3.06)
[N.sup.b] 402 402

Table 6. Antitest DD Estimates of Association between Onset of
Postpartum Maternal Smoking and Several Health Outcomesa

 All Boys
 (1) (2)

Dependent variable
 Asthma diagnosis 0.053 ** (1.95) 0.042 (1.08)
 Overnight stay in hospital 0.023 (0.57) -0.098 * (1.67)
 General health assessment -0.101 (1.10) -0.189 (1.41)
 (= 1 excellent; = 5 poor)
 Whether visited doctor 0.032 (0.78) 0.050 (0.82)
 for accident or injury
 Number of doctor visits 0.091 (1.01) 0.144 (1.13)
 for accident/injury
 Whether visited emergency 0.046 (1.20) 0.060 (1.05)
 room for accident/injury
 Whether child has -0.009 (0.46) -0.030 (1.08)
 physical disability
N 1640 842

 Blacks Hispanics
 (3) (4)

Dependent variable
 Asthma diagnosis 0.074 ** (1.96) 0.083 (1.36)
 Overnight stay in hospital 0.043 (0.84) 0.021 (0.23)
 General health assessment -0.223 * (1.82) -0.051 (0.24)
 (= 1 excellent; = 5 poor)
 Whether visited doctor 0.038 (0.69) -0.041 (0.52)
 for accident or injury
 Number of doctor visits 0.088 (0.71) 0.006 (0.04)
 for accident/injury
 Whether visited emergency 0.003 (0.05) 0.158 ** (1.96)
 room for accident/injury
 Whether child has 0.008 (0.29) -0.039 (1.29)
 physical disability
N 819 403

 Less Educated
 (5)
Dependent variable
 Asthma diagnosis 0.137 *** (2.87)
 Overnight stay in hospital -0.129 * (1.90)
 General health assessment -0.192 (1.10)
 (= 1 excellent; = 5 poor)
 Whether visited doctor 0.023 (0.39)
 for accident or injury
 Number of doctor visits 0.175 (0.90)
 for accident/injury
 Whether visited emergency 0.018 (0.30)
 room for accident/injury
 Whether child has -0.046 (1.41)
 physical disability
N 425

Table 7. Antitest DD Estimates of Association between Onset of Alcohol
Consumption and Childhood Asthma (Dependent Variable = 1 if Asthma; 0
if not) (a)

 Onset of Any Alcohol

 All Boys Blacks Hisp. <HS
 (1) (2) (3) (4) (5)

Dependent
 variable
 Asthma 0.003 0.013 0.006 0.008 0.006
 diagnosis (0.17) (0.65) (0.27) (0.24) (0.22)
 Asthma -0.009 0.002 -0.001 -0.012 -0.015
 attack (0.90) (0.12) (0.59) (0.48) (0.64)
 Emergency -0.003 0.018 -0.004 -0.008 0.000
 care (0.31) (1.09) (0.27) (0.34) (0.01)
[N.sup.6] 1504 794 781 380 514

 Onset of Binge Drinking

 All Boys Blacks Hisp. <HS
 (6) (7) (8) (9) (10)

Dependent
 variable
 Asthma -0.030 -0.036 0.005 -0.087 -0.049
 diagnosis (0.98) (0.86) (0.01) (1.58) (0.90)
 Asthma -0.023 -0.009 -0.005 -0.027 -0.016
 attack (0.90) (0.26) (0.32) (0.62) (0.35)
 Emergency -0.007 0.002 0.011 -0.025 -0.036
 care (0.32) (0.07) (0.27) (0.65) (0.88)
[N.sup.6] 1504 794 781 380 514

 Onset of Drug Use

 All Boys Blacks Hisp. <HS
 (11) (12) (13) (14) (15)

Dependent
 variable
 Asthma -0.029 -0.045 -0.024 -0.030 -0.026
 diagnosis (1.21) (1.32) (0.65) (0.49) (0.57)
 Asthma -0.021 -0.091 ** -0.032 -0.021 -0.018
 attack (1.07) (2.24) (1.00) (0.42) (0.47)
 Emergency -0.033 * 0.018 -0.023 -0.050 -0.029
 care (1.84) (1.09) (0.79) (1.13) (0.83)
[N.sup.6] 1504 794 781 380 514

(a) Coefficient estimates presented; absolute values of t-statistics
in parentheses.

(b) Sample sizes reported are for asthma diagnosis outcome. Sample
sizes for other asthma outcomes are approximately 2% smaller because
of missing observations.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.

Table 8. Individual Fixed Effects Estimates of Effect of between Wave
Changes in Quantity of Cigarette Consumption on Diagnosis of Asthma
(Dependent Variable = 1 if Asthma Diagnosis; 0 if Not) (a)

 All Boys
 (1) (2)

<Half pack/day (b) 0.012 (0.69) -0.019 (0.78)
One pack/day (b) 0.062 ** (2.30) 0.055 (1.43)
>One pack/day (b) 0.130 ** (2.20) 0.140 ** (1.98)
N 2293 1183

 Blacks Hispanics
 (3) (4)

<Half pack/day (b) 0.032 (1.24) -0.014 (0.36)
One pack/day (b) 0.071 * (1.85) 0.153 ** (2.03)
>One pack/day (b) 0.120 (1.54) 0.059 (0.20)
N 1137 510

 Less Educated
 (5)

<Half pack/day (b) 0.050 * (1.74)
One pack/day (b) 0.111 *** (2.63)
>One pack/day (b) 0.112 (1.29)
N 691

(a) Coefficient estimates presented; absolute values of t-statistics
in parentheses.

(b) Omitted category is nonsmokers.

* Significant at 10%.

** Significant at 5%.

*** Significant at 1%.