In this paper, we also examine the impact of higher taxes on maternal smoking and birth outcomes, but we look at this issue using a different methodology. We
examine the smoking rates among pregnant women in four states before and after large, one-time tax hikes. There are four reasons we are interested in using these
experiments in this context. First, since previous estimates of demand elasticities are based only on marginal price or tax changes, these estimates may not accu-
rately predict the likely consequences of recent large tax hikes. Second, by focus- ing on one state at a time, we can use a more scientific method to choose a control
group. Third, by focusing on these large tax hikes and pregnant women, we can potentially identify the health benefits of tax hikes. Fourth, as we document in the
conclusion, large tax hikes are becoming more frequent.
Data for this project is taken from the 1990–97 Natality Detail files that contain a census of births in the United States and include measures of self-reported smok-
ing for births in most states. We find that increasing cigarette excise taxes reduces smoking participation among pregnant women but the size of the response differs
greatly across experiments with the smoking participation price elasticity of demand ranging from
−1.83 to −0.10. The weighted average across the four states is
−0.49, a number close to previous estimates. We also find that the tax hikes increased birth weight and decreased the risk of low birth weight and these
changes occur almost immediately after the tax hike is imposed. We use two-stage least squares 2SLS to calculate the impact of smoking on birth outcomes.
Pooling the data from states with precise first-stage estimates, we find that smok- ing during pregnancy reduces birth weight by 182 grams and increases the
chance of a low birth-weight estimate by 7 percentage points. Our results are statistically similar to OLS estimates, suggesting that the omitted variable bias
is not severe in single equation models where maternal smoking is treated as exogenous.
II. Background
Our goal is to examine whether maternal smoking changed after states adopted large cigarette excise tax hikes. Although there are a number of
states that have adopted large tax hikes over the past 10 years, only a small num- ber are eligible to examine as experiments. Our research design requires that we
have data on maternal smoking before and after the tax hike as well as data on the size and timing of tax hikes. The primary data for our analysis is the Natality
Detail File National Center for Health Statistics, various years that includes information about maternal demographic characteristics, as well as pregnancy and
infant health indicators for a census of all births in the United States. Starting in 1989, the Natality Detail File includes self-reported data on a mother’s smoking
behavior during pregnancy. At the time we started this project, Natality Detail files were available for births through 1997. Allowing at least two-years worth of data
before and after the tax hike, this restricts our attention to singular tax hikes in the 1991 through mid 1994 period. Using data about state cigarette excise tax rates
from the Tax Burden on Tobacco, we identified tax hikes of at least 14 cents per Lien and Evans
375
pack in states that had no other tax change in a four-year window.
2
Subsequently, we have identified four states Arizona, Illinois, Michigan, and Massachusetts that
we can examine. The dates and size of the tax hikes are listed in Table 1. Table 1 also includes some important descriptive statistics of births in the year prior
to the tax hike for each state. Of the four states, the highest percentage of smokers in the year prior to the tax hike was in Massachusetts 23 percent and the lowest was in
Arizona 12 percent. In contrast, these states had the highest and lowest low birth- The Journal of Human Resources
376
Table 1 Descriptive Statistics of Births, Natality Detail File
Arizona Illinois
Massachusetts Michigan
Date of tax hike 112994
71493 1193
5194 Change in tax centspack
18 to 58 30 to 44
26 to 51 25 to 75
Average retail price for pack 162.9
182.1 180.6
163.4 of cigarettes in fiscal year
of tax hike Descriptive statistics of births during the 12-month period before the tax-hike mean
standard deviation Number of births
× 1000 70.3
187.7 88.5
134.1 Smoker, indicator variable
0.123 0.152
0.233 0.196
0.329 0.359
0.423 0.397
Birth weight in grams 3317
3324 3395
3336 572
616 592
621 Low birth weight 2,500
0.068 0.077
0.059 0.080
grams indicator variable 0.251
0.266 0.236
0.267 Age of the mother
26.1 26.8
28.3 26.6
6.0 6.0
5.7 5.8
Black, indicator variable 0.04
0.23 0.10
0.21 0.19
0.42 0.30
0.40 Hispanic, indicator variable
0.34 0.15
0.10 0.03
0.48 0.36
0.30 0.18
Married, indicator variable 0.62
0.66 0.74
0.71 0.49
0.47 0.44
0.46
2. During this period Maryland, Rhode Island, Washington, and Washington, D.C. all had a tax hike greater than 14 cents. However, these states had more than one tax hike between 1991 and 1994 and are not evalu-
ated in this paper. North Dakota did have a 15-cent tax hike in 1993 but there were too few births in the state to use this experiment in our analysis. New York had a 17-cent tax increase in 1993, but the state did not
report smoking status in the Natality Detail data during 1991-93. We should also note that there was an increase in the Federal excise tax on cigarettes of 4 cents per pack that went into effect in 1993. This tax hike
is common to all states in our sample over this period and is captured by the year effects in our analysis.
weight rates, respectively. This result is driven by the low fraction of minorities Blacks or Hispanics in Massachusetts and the high fraction in Arizona. These two
groups tend to have low smoking rates but poor birth outcomes. Pretreatment smok- ing and low birth-weight rates were 15 and 7.7 percent in Illinois, respectively, and 20
and 8 percent respectively in Michigan.
How these four states spent the revenues generated from the tax hikes differed. Arizona, Illinois and Michigan raised cigarette taxes primarily to generate state tax
revenues to pay for nontobacco-related programs. More than 60 percent of the rev- enue collected from Arizona’s 1994 40-cent tax hike went toward medical care for
the uninsured and poor. The remainder of the revenue went toward antitobacco pro- grams and health research on prevention and treatment of tobacco-related illnesses
The Arizona Republican, February 16, 1995, p. A5. Illinois’ 1993 14-cent cigarette tax hike was passed to ease budgetary woes over funding for health services for the
poor St. Louis Post Dispatch, July 13, 1993, p. 1A. Michigan’s 1994 50-cent tax increase was passed to raise revenues to replace a portion of those lost when resi-
dents voted to eliminate the property tax New York Times, March 14, 1994, p. A15. Massachusetts’ 1992 25-cent state cigarette excise tax increase funded the
Massachusetts Tobacco Control Program that included activities such as antismok- ing media campaigns, enforcement of local antismoking laws, and educational
programs targeted primarily at teenagers, and pregnant women The Boston Globe, December 5, 1993, p. 4, and December 19, 1993, p. 50. Revenue from Massa-
chusetts’ tax hike also went to health programs unrelated to smoking The Boston Globe
, March 5, 1993, p. 10. In the next section, we describe a model that estimates the “treatment effect” asso-
ciated with each state’s major tax hike. In the case of Arizona, Illinois, and Michigan, the text above suggests that we can reasonably attribute all of the change in con-
sumption to the higher cigarette tax. In the case of Massachusetts, however, the treat- ment included a tax hike followed by a large antismoking media campaign. In a
companion paper, Lien 2001 examines in detail the impact of the Massachusetts antismoking campaign on tobacco use among pregnant women and finds no addi-
tional change in smoking after the start of the media campaign. This suggests that the drop in smoking we show for Massachusetts is driven by the tax change and not the
media campaign.
III. Econometric Model