The National Survey on Drug Use and Health (NSDUH) was used to examine the smoking behavior of youth and young adults from 2002 to 2011. NSDUH is an annual cross-sectional, national- and state-based survey on the use of tobacco products, alcohol, illicit drugs, and mental health status among the U.S. civilian, non-institutionalized population aged 12 or older (SAMHSA, 2013a). The survey has been conducted since 1971 and collects data quarterly through face-to-face interviews. NSDUH employs a state-based design with an independent probability sample within each state and the District of Columbia. This study used data from the 2002–2011 NSDUH. Data prior to 2002 were not used because of the methodological changes introduced to the survey design in 2002, which affected respondent reporting of substance use (Kennet & Gfroerer, 2005). Annual sample sizes are around 3600 for each of the eight states with the largest populations and around 900 for each of the remaining states and the District of Columbia. Youth and young adults were oversampled to achieve roughly equal numbers of respondents aged 12 to 17, 18 to 25, and 26 or older each year. This corresponds to annual sample sizes of about 1200 for each age group in large states and 300 per age group in small states. The percentage of eligible households that completed the initial screening was 90.7% in 2002 and declined to 87.0% in 2011. Interview response rates among youth and young adults were 90.0% and 85.2%, respectively, in 2002, and declined somewhat to 85.0% and 80.5%, respectively, in 2011. The 2002–2011 study sample contained data on 226,315 youths and 226,228 young adults.

Cigarette smoking was assessed using five NSDUH indicators. The first indicator was initiation of cigarette smoking which is defined as having smoked the first lifetime cigarette during the 12 months before the survey interview. The measure is not defined for those who had a first lifetime cigarette more than 12 months before the survey interview. The second indicator was past-month cigarette smoking which is defined as having smoked all or part of a cigarette during the 30 days before the interview. This measure is defined for all respondents. The third indicator was smoking cessation which is defined as ever having smoked daily and smoked during the period 13 to 24 months ago, but not having smoked during the past 12 months. This measure is defined only for those who reported having been a daily smoker (i.e., having smoked 30 consecutive days) at some point in their lifetime and who reported smoking during the 13 to 24 months before the survey interview. The fourth indicator was a quantity measure among past-month smokers, in terms of the number of days smoked during the past month. The fifth indicator was the number of cigarettes smoked among past-month smokers on a typical day during the past month.

In the statistical models for youth and young adults we controlled for the following set of respondent characteristics: sex, race/ethnicity, age, presence of parents in the household, birthplace (U.S. or not), household income, county type (e.g., metropolitan), religious service attendance, importance of religion, school enrollment, privacy during the survey interview, attitudes toward risky behavior, frequency of seatbelt use, and having moved 2 or more times during the past 5 years. In the models for youth, we also included two control variables measuring the perception of parents' attitude toward smoking one or more packs of cigarettes per day and perceived risk of smoking. In the models for young adults we included control variables for pregnancy, education, employment status, and health insurance status.

Several state-level covariates were used to account for differences in tobacco control policies and antismoking sentiment across states. State excise taxes on a pack of 20 cigarettes were obtained from the Tax Burden on Tobacco (Orzechowski & Walker, 2012), which contains annual averages that were converted to quarterly numbers through linear interpolation. Quarterly indices for the presence of a smoke-free law in bars, restaurants, and private workplaces as covariates were also included. These were collected from the ImpacTeen project (ImpacTeen, 2013) and CDC's State Tobacco Activities Tracking and Evaluation (STATE) system (CDC, 2013). The values that each index assumed were “no restrictions,” “some restrictions,” and “complete smoking ban.”

An annual measure of the lack of compliance with laws prohibiting the sale of tobacco products to individuals younger than 18 was also included. Following the Synar Amendment of 1992, states were required to enact and enforce such laws and conduct annual, unannounced inspections of tobacco sales outlets (SAMHSA, 2011a; US DHHS, 2012b). The variable used in the analysis is the percentage of retailers that violate youth tobacco access laws, as reported by each state (SAMHSA, 2013b). The violation rate is expected to have limited explanatory power, because sale restrictions do not apply to young adults, and youth also rely on social sources (e.g., friends and family) to obtain cigarettes (Altman, Wheelis, McFarlane, Lee, & Fortmann, 1999; DiFranza & Coleman, 2001; Forster et al., 1998). However, we included the variable as a proxy for antismoking sentiment in a state.

We included quarterly per capita tobacco control funding, obtained from CDC's STATE system. To account for potential delayed effects of this funding, we calculated a discounted cumulative amount of funding for each quarter, which included the current quarter and the previous seven quarters, where past funding amounts were discounted at a quarterly rate of 25% (Farrelly, Pechacek, Thomas, & Nelson, 2008). The final state-level covariate used in the model is a NSDUH estimate of smoking prevalence by quarter. This allowed us to control for the general downward trend in smoking that may have affected smoking among youth and young adults independently of the tax change. Additionally, this variable helped account for differences between states in general attitudes toward smoking. We used the smoking prevalence among 18- to 25-year-olds in the models for youth and the smoking prevalence among 26- to 49-year-olds in the models for young adults.

Before the analysis could proceed, appropriate values of the state-level variables had to be assigned to or linked with each individual record in NSDUH. Because the smoking measures relate to different time periods (e.g., past month, past 12 months), the assignment mechanism varied across the different outcome analyses. For the initiation indicator, we distinguished between those who initiated and those who did not. For those who initiated smoking, the month and year of initiation were known, so that we could assign values of the state-level covariates from the corresponding quarter. For individuals who did not initiate smoking, we calculated a “look-back date” of 45 days prior to the survey interview and assigned values of state-level variables from the quarter that contained the look-back date. Thus, for those not initiating, values were chosen from either the interview quarter or one quarter prior.

A similar approach was followed for the cessation outcome. For respondents who quit smoking 13 to 24 months before the interview date, the month and quarter of quitting were known and we assigned values of the state-level variables from the corresponding quarter. For respondents who did not quit smoking, we calculated a look-back date of 410 days (=365 + 45) before the survey interview and assigned values of the state-level variables from the quarter that contained the look-back date. Finally, the past-month smoking indicator and smoking intensity outcomes (days and cigarettes per day) refer to the 30 days before the survey interview. For these outcomes, we assigned to these individuals the values of the state-level covariates from the quarter that contained the majority of that 30-day period. Thus, in these analyses we assigned values of state-level covariates from either the interview quarter or one quarter prior (if the interview date fell within 15 days of the start of a quarter).

Data were analyzed using SAS-Callable SUDAAN®(RTI International, 2012) to account for NSDUH's complex survey design and sampling weights. Smoking behavior was analyzed separately for youth and young adults, and the male, female, black and white populations within those age groups to examine potential differences in the response to the increase in the federal excise tax. Before estimation, all explanatory variables measured in dollars were adjusted for inflation using the Bureau of Labor Statistics' Consumer Price Index (U.S. Department of Labor, 2013). To adjust for missing data, an ‘unknown’ category was added for unordered, categorical variables. Missing values of ordered or continuous variables were imputed (SAMHSA, 2011b). Regression coefficients and odds ratios (ORs) were estimated using Generalized Estimating Equations with Taylor linearization to calculate standard errors and confidence intervals (CIs).

The statistical models used to estimate the impact of the April 1, 2009, federal tax increase on each of the examined outcomes were pre-post regression models. They included as covariates the individual and state-level characteristics discussed before, as well as a post-period indicator (equal to 0 before April 1, 2009, and equal to 1 thereafter). The value of the indicator, corresponding to the reference period of the smoking outcome, was linked to the NSDUH record following the same procedure described above. For the binary outcomes (smoking initiation, current smoking, and cessation), we used a logistic regression model. For the smoking intensity outcomes (smoking days and cigarettes per day) of current smokers, we used a linear regression model. The analysis for smoking days was based on the sample of all current smokers, whereas the analysis for cigarettes per day was conducted separately for daily and nondaily smokers.

Table 1 contains the average rates of smoking initiation, past-month smoking and smoking cessation before and after the April 1, 2009 federal tax increase. Rates are shown separately for youth aged 12–17, young adults aged 18–25, and four subgroups. Table 1 also shows, for current smokers, the number of days smoked during the past month and the number of cigarettes consumed per day of smoking.

Smoking initiation rates among youth and young adults were lower after the April 1 2009 tax increase compared to before the federal tax increase. The same was true for past-month cigarette use. The cessation rate among youth was lower after the tax increase, compared to before, whereas for young adults the rate was slightly higher. Among youth and young adult current smokers, the number of days smoked (during the past month) and the number of cigarettes smoked per day were lower after the tax increase, but compared to the averages before the tax increase, the reductions were marginal.

Table 2 shows the odds ratios (ORs) for the post-period indicator, estimated from the logistic regression models for smoking initiation, past-month cigarette use and smoking cessation. The OR represents the odds of the outcome occurring after April 1, 2009 relative to the odds of it occurring before April 1, 2009. ORs less than one indicate that, the outcome was less likely after compared to before the federal tax increase, controlling for individual and state-level characteristics.

For youth, the tax increase was associated with a reduction in the odds of smoking initiation (OR = 0.83, p < 0.0001). A similar result was found for the male, female, and white youth populations, but not for black youth. Initiation among young adults overall was not affected (OR = 0.91, p = 0.0532), but the odds of initiating were lower for white young adults after the tax increase (OR = 0.88, p = 0.0268).

The odds of past-month cigarette use were lower after the tax increase for the youth population overall (OR = 0.83, p < 0.0001) and three out of four subpopulations (male, female, and white, ORs = 0.75–0.92, p < 0.05). For young adults, the odds of past-month cigarette use were also lower after the tax increase (OR=0.92, p < 0.0001). The same was true for two out of four subpopulations of young adults (female and white, ORs = 0.87–0.89, p < 0.0001). For smoking cessation, none of the estimated ORs were statistically significant.

Table 3 shows the coefficients of the post-period indicator, estimated from the linear regression models for the number of days smoked during the past 30 days and the average number of cigarettes consumed per day of smoking, among current daily and nondaily smokers. Negative estimates indicate a reduction in the outcome following the April 1, 2009 tax increase.

Among the overall population of youth smokers, the average number of days smoked during the past month fell by 0.97 days (p = 0.0001) after the tax increase. Statistically significant reductions were also found for male, female and white youth. Among the overall population of young adults, the average number of days smoked during the past month fell by 0.84 days (p < 0.0001). The reductions were statistically significant for all subpopulations, ranging from 0.78 days for female young adults (p = 0.0002) to 0.95 for black young adults (p = 0.0164).

Youth daily smokers smoked on average 1.02 cigarettes per day less after the federal tax increase (p = 0.0011). Statistically significant reductions were also found for male, female and white youth daily smokers. In the overall population of youth nondaily smokers, 0.25 fewer cigarettes (p = 0.0114) were smoked per day after the tax increase. Similar results were found for female and white youth who were nondaily smokers.

Among young adult daily smokers, the average number of cigarettes smoked per day decreased by 0.92 (p < 0.0001) after the federal tax increase. Statistically significant reductions were also found for daily smokers in the male, female and white youth populations. Young adult nondaily smokers also smoked fewer cigarettes per day (−0.45, p < 0.0001), with similar reductions for young adult nondaily smokers in the male, female and white subpopulations.