Impact of Smoking Cessation on Household Food Security

Abstract

Purpose:

Smokers can spend a substantial amount on cigarettes, potentially constraining their ability to purchase food. We tested the association of smoking cessation and household food security.

Methods:

Using the Current Population Survey (2001 to 2019), we longitudinally linked the Tobacco Use Supplement and the Food Security Supplement (n=71,278). Among adult smokers (n=13,144), we used modified Poisson regression to model household food insecurity as a function of quit status (continuing smokers versus recent quitters), adjusting for sex, age, household size, children in the household, and other household smokers. We also used multinomial logistic regression to examine more detailed food security status (high, marginal, low, very low).

Results:

The adjusted probability of household food insecurity at follow-up was 11% (95% CI: 8.7%-13%) for recent quitters and 20% (95% CI: 19%-21%) for continuing smokers. Continuing smokers had a lower adjusted probability of high food security (69% vs 80%) and a higher adjusted probability of marginal (11% vs 9.8%), low (12% vs 7%), and very low food security (7.8% vs 3.6%) compared to recent quitters.

Conclusions:

Cigarette cessation is associated with lower risk of household food insecurity. Therefore, promoting tobacco cessation alongside food assistance and poverty reduction policies may help alleviate food insecurity.

INTRODUCTION

Food insecurity is a major problem in the United States—in 2020, 10.5% of households did not have enough food throughout the year to support active, healthy living.¹ Since money is critical for obtaining enough and adequate (safe, nutritionally sufficient, and culturally acceptable) food in our food system, food insecurity is inequitably distributed by household resources. Approximately 29% of households living below 185% of the federal poverty threshold experienced food insecurity in 2020 compared with just 5% of households living at or above that level.¹ Food insecurity is also socially patterned due to systemic racism which affects social determinants of health such as socioeconomic status and neighborhood resources and, in turn, leads to increased food insecurity among racial/ethnic minority groups.²

Similarly, cigarette smoking is distributed inequitably due to complex structural factors including increased tobacco marketing in low-income or Black neighborhoods,³ unequal access to cessation resources across class and race lines,^4,5 and increased exposure to financial strain and chronic stressors that can lead to more smoking.^6-9 For instance, 32% of adults with a GED were current smokers in 2020 compared to only 6% of adults with a bachelor’s degree.¹⁰ Similarly, 20% of adults with a household income less than $35,000 smoked cigarettes compared with 6% of adults earning more than $100,000.¹⁰ Despite being commonly consumed by people who struggle financially, cigarettes are expensive. The national average price for a pack of cigarettes in 2021 was $8.00,¹¹ which translates to approximately $240 per month for a household that smokes one pack per day (the median cigarettes per day for smokers was 11 in 2022,¹² and smoking households often have 2+ smokers). Thus, households that contain smokers may have less money to spend on food.

Indeed, a growing body of evidence suggests that tobacco use and food insecurity commonly co-occur.^13-15 However, the nature and direction of this relationship is debated and may be complex. Drawing on 19 studies, Kim-Mozeleski and Pandey recently developed a conceptual model that posits a bidirectional association between tobacco use and food insecurity.¹³ Tobacco use may contribute to food insecurity because tobacco spending could expend household funds that otherwise would have been available for food.¹³ Additionally, tobacco use contributes to declining health, which, in turn, may lead to food insecurity by limiting earnings and increasing out-of-pocket healthcare spending.^13,16,17 However, food insecurity also leads to stress, negative emotions, and psychological distress.¹³ Tobacco users often use nicotine to cope with stress,¹⁸ which can make quitting challenging.^19,20 Additionally, nicotine is known to suppress appetite;²¹ people who experience food insecurity may use tobacco to dampen hunger.¹³ Therefore, food insecurity may also contribute to cigarette smoking and other tobacco use through psychological distress and appetite suppression pathways.¹³

Prior studies’ cross-sectional designs have limited the ability to draw conclusions about the nature of the relationship between cigarette use and food insecurity. Out of the 19 studies published between 2008 and 2018,¹³ 16 had cross-sectional designs^22-37 while only three were longitudinal.^38-40 However, none of these studies posed the question of whether reductions in cigarette use are linked to food security. Instead, the longitudinal studies were designed to examine how food insecurity status influenced cigarette smoking and did not assess changes in cigarette smoking status as an exposure.^38,39

More recent longitudinal studies explicitly examine change in food security status or change in smoking status but offer inconclusive findings.^41-43 Sheira et al. found that changes in food insecurity are associated with increased odds of smoking and that food insecurity is associated with higher smoking intensity.⁴³ Similarly, Kim-Mozeleski et al. (2019) found that becoming food insecure is associated with a higher likelihood of starting cigarette smoking and a lower likelihood of cessation.⁴² However, the separation between measures was long (12 years) and it is unclear when these transitions occurred.⁴² Conversely, Bergmans et al. found that becoming food insecure increases odds of cigarette cessation.⁴¹ Descriptive epidemiology shows that food insecurity has increased more among current cigarette smokers compared to the general population (54% vs 30% increase).⁴⁴ This raises the question of whether tobacco use could be putting households at risk for food insecurity. However, only one study has examined whether smoking increases risk of food insecurity. Kim-Mozeleski et al (2018) found that cigarette smoking status significantly predicts food insecurity severity after one year when controlling for baseline food insecurity severity.⁴⁰

Using longitudinal data from the Current Population Survey (CPS), we (1) investigated the association between cigarette cessation and household food security and (2) compared this association for various demographic groups. We hypothesize that, compared to individuals who quit smoking, individuals who continue to smoke cigarettes have higher risk of experiencing household food insecurity in the subsequent year. Further, we expect that the association between continuing smoking and household food insecurity is consistent across sex, household composition, household income, and racial/ethnic self-classification.

METHODS

Data

We used data from the CPS Tobacco Use Supplement (TUS) and the Food Security Supplement (FSS). The CPS is a monthly survey of approximately 60,000 households administered by the US Census Bureau and the Bureau of Labor Statistics.⁴⁵ The “basic monthly survey” covers demographic characteristics and employment information. Respondents also receive supplemental surveys each month. Methods to longitudinally link households in the CPS across a 16-month period have been developed by IPUMS.⁴⁶

The TUS collects information about tobacco use from non-institutionalized civilians.⁴⁷ The TUS has been administered 19 times between 2001 and 2019, with data collection occurring in various months and with multiple supplements in some years and none in other years. All CPS household members aged 15 years and older (18+ after 2010) are eligible for the TUS. Most interviews are administered via self-interview, but proxy responses are given by other household members in some cases (see the Methods Appendix for details).⁴⁷

The FSS is a household-level supplement administered annually in December that gathers information on use of food assistance programs. The FSS includes the 18-item module to assess household food security developed by the United States Department of Agriculture (USDA).⁴⁸ Questions are answered by a single, knowledgeable respondent about the household.⁴⁹

Study Design & Sample

Figure 1 shows our study design. Using IPUMS-CPS data from 2001-2019, we linked individuals who completed the TUS (Time 1) forward to their responses to the FSS (Time 2). For consistency, we linked only supplements that were 10-13 months apart. We validated the linked records, dropping individuals who did not have matching age, sex, and racial self-classification.

Figure 1. Study design.

Figure 1 shows how key measures were derived from our study design. Using IPUMS-CPS data from 2001-2019, we linked household members who completed the Tobacco Use Supplement (TUS) forward to their responses to the Food Security Supplement (FSS). All linked supplements were 10-13 months apart. We used recall questions from the TUS at Time 1 to establish whether self-reported former smokers at Time 1 had quit within the last year. We classified individuals at Time 1 as never smokers, non-recent quitters (former smokers who quit more than one year prior), recent quitters (former smokers who quit within the prior year), and current smokers. We assessed household food security status at Time 2 based on a twelve-month reference period. Covariates, including sex, age, household size, whether the household has any children, and number of current smokers in the household were assessed at Time 1.

We then excluded individuals who were under 18, did not complete the FSS, or were missing food security information. We also excluded proxy responses to the TUS as questions needed to establish the timeline of smoking cessation are only administered to self-respondents, and we excluded those with insufficient smoking information. We then randomly selected one participant per household to include in our final sample (n=71,278) since the FSS is a household-level supplement. Appendix Figure 1 shows a flow chart of inclusion criteria.

Measures: Exposure

We defined non-smokers as individuals who smoked less than 100 tobacco cigarettes in their lifetime or who recalled quitting smoking cigarettes more than one year prior to the TUS. We defined smokers as individuals who smoked cigarettes one year prior to the TUS (Time 0). Within our smoking sample, we defined recent quitters as former cigarette smokers at the TUS (Time 1) and who quit smoking within the past year (between Time 0 and Time 1). We defined continuing smokers as current cigarette smokers at the TUS (Time 1).

Measures: Outcomes

Using the 18-item USDA 12-month household food security survey module from the FSS at Time 2,⁴⁸ we classified households as food secure (experiencing high or marginal food security) vs food insecure (experiencing low or very low food security). We then used a more detailed measure, classifying households as experiencing high, marginal, low, or very low food security.

Measures: Covariables

We also used demographic information from the basic monthly survey administered alongside the TUS at Time 1. The races provided as response options in the CPS changed multiple times during the study period. For consistency across years, we aggregated options into White, Black, Asian/Hawaiian/Pacific Islander, and American Indian/Aleut/Eskimo. Hispanic ethnicity was assessed through a separate question. We used the aggregated race responses and the ethnicity variable to describe racial/ethnic self-classification, allowing people to be classified as multiple races/ethnicities. We also report participant age, education (less than high school; high school or equivalent; some college or associate’s; bachelor’s degree or higher), family income (under $20,000; $20,000-$29,999; $30,000-$39,000; $40,000-$49,999; $50,000-$59,999; $60,000-$74,999; $75,000+; not reported), household size, whether the household has any children, and the total number of household smokers. Additionally, we report household poverty status (below 185% poverty; above 185% poverty or income not reported) at Time 2. These measures were used to characterize our sample. Additionally, we considered sex, age, household size, whether the household has any children, and the number of current household smokers as potential confounders.

Statistical Analysis

We first descriptively compared detailed food security status for non-smokers, recent quitters, and continuing smokers. Within the smoking sample (n=13,144), we used modified Poisson regression to generate the risk of a household experiencing food insecurity among continuing smokers compared to recent quitters. We adjusted for sex, age, household size, whether the household had any children, number of current household smokers, and panel timing. We then calculated the predicted probabilities of experiencing food insecurity for recent quitters and continuing smokers using the “margins” command in Stata which calculates probabilities that are standardized to the distribution of covariates of participants in our sample. We conducted this analysis in the smoking sample overall before stratifying by household characteristics including presence of children, poverty status, and respondent sex. We were also interested in assessing whether the association between quitting smoking and household food insecurity differs for people subject to more racism. However, our data does not include measures of interpersonal discrimination or structural racism, so we stratify by respondent racial/ethnic self-classification (condensed to a dichotomous classification of non-Hispanic white vs Hispanic or non-white due to small sample sizes) as an imperfect proxy for experiences of racism. We also formally tested for interaction via interaction terms.

We then used multinomial logistic regression to investigate the association between continuing smoking and the detailed food security outcome measure, adjusting for the same factors listed above. As before, we estimated predicted probabilities of each level of household food security (high, marginal, low, very low) separately for recent quitters and continuing smokers. We also repeated the multinomial logistic regression in stratified subsamples.

All analyses were weighted using custom, longitudinal, household-level weights (see the Methods Appendix for details).

RESULTS

Most of the 71,278 adults in our sample were non-smokers (82%), while 2% were recent quitters and 16% were continuing smokers. Compared to non-smokers, recent quitters and continuing smokers were more likely to be younger, male, have less education, and live below 185% of the federal poverty level (Table 1). Food insecurity (defined as low or very low household food security) was highest among continuing smokers (19%) followed by recent quitters (13%) and non-smokers (8.5%) (Figure 2)

Table 1.

Sample characteristics

		SMOKING SAMPLEa (n=13,144)
	Non- Smokersb (n=58,134)	Recent Quittersc (n=1,422)	Continuing Smokers d (n=11,722)	FULL SAMPLE (n=71,278)
	N (%) or mean (sd)e
Demographics at TUS
Age, years	49.7 (17.1)	43.0 (15.6)	45.4 (15.0)	48.9 (16.8)
Sex
Male	23,622 (41%)	662 (48%)	5,353 (47%)	29,637 (42%)
Female	34,512 (59%)	760 (52%)	6,369 (53%)	41,641 (58%)
Racial/ethnic self-classification (check all that apply)f
White	50,385 (83%)	1,292 (88%)	10,220 (84%)	61,897 (83%)
Black	5,197 (12%)	99 (10%)	1,084 (13%)	6,380 (12%)
Asian/Hawaiian/Pacific Islander	2,285 (4.9%)	27 (2.0%)	292 (2.8%)	2,604 (4.5%)
American Indian/Aleut/Eskimo	835 (1.4%)	27 (1.7%)	327 (2.5%)	1,189 (1.6%)
Hispanic	4,955 (12%)	90 (8.4%)	635 (7.0%)	5,680 (11%)
Socioeconomic Status
Education at TUS
Less than high school	6,082 (11%)	152 (11%)	1,875 (17%)	8,109 (12%)
High school or equivalent	15,927 (26%)	477 (34%)	4,675 (39%)	21,079 (29%)
Some college or associate’s	15,727 (27%)	484 (34%)	3,555 (30%)	19,766 (28%)
Bachelor’s degree or higher	20,398 (36%)	309 (21%)	1,617 (14%)	22,324 (32%)
Employment at TUS
Employed	35,343 (62%)	926 (66%)	7,306 (62%)	43,575 (62%)
Unemployed	1,515 (2.9%)	60 (5.0%)	672 (6.6%)	2,247 (3.5%)
Not in labor force	21,276 (35%)	436 (29%)	3,744 (31%)	25,456 (35%)
Family income at TUS
Under $20,000	9,008 (16%)	277 (21%)	2,968 (26%)	12,253 (17%)
$20,000 to $29,999	6,371 (11%)	175 (12%)	1,653 (14%)	8,199 (11%)
$30,000 to $39,999	6,422 (11%)	181 (12%)	1,517 (13%)	8,120 (11%)
$40,000 to $49,999	5,048 (8.6%)	131 (9.4%)	1,081 (8.7%)	6,260 (8.6%)
$50,000 to $59,999	4,851 (8.1%)	145 (9.8%)	988 (8.2%)	5,984 (8.2%)
$60,000 to $74,999	5,856 (9.7%)	140 (9.6%)	1,014 (8.9%)	7,010 (10%)
$75,000 and over	17,178 (30%)	305 (22%)	1,851 (16%)	19,334 (28%)
Not reported	3,400 (5.8%)	68 (4.8%)	650 (5.8%)	4,118 (5.8%)
Household poverty status at FSS
Above 185% poverty or not reported	44,044 (77%)	992 (71%)	7,444 (64%)	52,480 (75%)
Below 185% poverty	14,090 (23%)	430 (29%)	4,278 (36%)	18,798 (25%)
Household Composition at TUS
Respondent’s household size	2.6 (1.4)	2.5 (1.3)	2.5 (1.5)	2.5 (1.4)
Number of children in household	0.7 (1.1)	0.7 (1.0)	0.7 (1.1)	0.7 (1.1)
Total current smokers in household	0.06 (0.25)	0.1 (0.4)	1.3 (0.5)	0.2 (0.5)
Household food security at FSS
Detailed food security status
High food security	49,834 (85%)	1,096 (77%)	8,328 (70%)	59,258 (82%)
Marginal food security	3,792 (7.0%)	136 (10%)	1,256 (11%)	5,184 (7.6%)
Low food security	3,075 (5.8%)	115 (8.3%)	1,264 (12%)	4,454 (6.8%)
Very low food security	1,433 (2.7%)	75 (4.8%)	874 (7.5%)	2,382 (3.5%)
Condensed food security status
Food secure (high/marginal)	53,626 (92%)	1,232 (87%)	9,584 (81%)	64,442 (90%)
Food insecure (low/very low)	4,508 (8.5%)	190 (13%)	2,138 (19%)	6,836 (10%)
Panel Timing g
Nov 2001 (TUS) to Dec 2002 (FSS)	6,417 (8.9%)	158 (8.2%)	1,802 (13%)	8,377 (9.5%)
Feb 2002 (TUS) to Dec 2002 (FSS)	3,431 (4.7%)	99 (6.1%)	887 (6.6%)	4,417 (5.1%)
Feb 2003 (TUS) to Dec 2003 (FSS)	3,472 (5.1%)	94 (5.6%)	837 (6.2%)	4,403 (5.3%)
Nov 2003 (TUS) to Dec 2004 (FSS)	8,815 (15%)	192 (13%)	1,983 (17%)	10,990 (15%)
Jan 2007 (TUS) to Dec 2007 (FSS)	9,917 (16%)	294 (19%)	2,206 (18%)	12,417 (16%)
Jan 2011 (TUS) to Dec 2011 (FSS)	9,656 (16%)	226 (16%)	1,786 (16%)	11,668 (16%)
Jan 2015 (TUS) to Dec 2015 (FSS)	8,812 (17%)	200 (16%)	1,276 (12%)	10,288 (16%)
Jan 2019 (TUS) to Dec 2019 (FSS)	7,614 (18%)	159 (16%)	945 (11%)	8,718 (17%)

TUS = Tobacco Use Supplement; FSS = Food Security Supplement; sd = standard deviation

Data = Current Population Survey (CPS), 2001-2019

^aThe smoking sample consists of recent quitters and continuing smokers. All individuals in this sample were current smokers at Time 0, one year prior to the TUS.

^bNon-smokers include never smokers and former smokers who quit more than one year prior to the TUS.

^cRecent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (i.e., between Time 0 and Time 1)

^dContinuing smokers are current smokers at the TUS (Time 1) and one year prior (Time 0).

^eProportions, means, and standard deviations are all sample weighted using custom, household-level longitudinal weights. Corresponding counts are unweighted.

^fParticipants can be classified as multiple races/ethnicities. Therefore, the proportions in the racial/ethnic classification categories add up to more than 100%.

^gPanels were created by linking responses for individuals who completed the TUS and subsequently completed the FSS 10-13 months later.

Figure 2. Detailed food security status by smoking category (n=71,278).

Figure 2 shows detailed household food security status by smoking category for the full sample of smokers and non-smokers (n=71,278) using IPUMS-CPS data from 2001-2019, Non-smokers include never smokers and former smokers who quit more than one year prior to the Tobacco Use Supplement (TUS) at Time 1. Recent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (between Time 0 and Time 1). Continuing smokers are current smokers at the TUS (Time 1). Detailed household food security status was assessed through the FSS (Time 2) approximately 10-13 months after participants completed the TUS. All estimates are sample weighted using custom household-level longitudinal weights.

Among the smoking sample (n=13,144), continuing smokers had 1.85 times the risk of household food insecurity (95% CI: 1.51-2.27) compared with recent quitters after adjusting for individual and household characteristics (Table 2). The predicted probability of food insecurity was 11% (95% CI: 8.7%-13%) for recent quitters and 20% (95% CI: 19%-21%) for continuing smokers. Similar associations were observed in all subsamples (Table 2).

Table 2.

Modified Poisson Regression: Relative risk of food insecurity for continuing smokers compared to recent quitters for smoking sample^a overall and stratified by household characteristics

	OUTCOME = HOUSEHOLD FOOD INSECURITY
	Predicted Probability (95% CI)d,e		RR (95% CI)d,e	Interaction
Sub-sample	Recent quitterb	Continuing smokerc	Ref=recent quitter	p-value
overall (n=13,144)	11% (8.7%-13%)	20% (19%-21%)	1.85 (1.51-2.27)	N/A
Households with children (n=5,045)	12% (8.9%-16%)	24% (22%-25%)	1.91 (1.42-2.58)	0.753
Households without children (n=8,099)	10% (7.5%-13%)	17% (16%-19%)	1.70 (1.26-2.28)
Below 185% poverty (n=4,708)	26% (20%-32%)	36% (35%-38%)	1.40 (1.12-1.76)	0.405
Above 185% poverty (n=8,436)	5.7% (3.7%-7.6%)	10% (9.4%-11%)	1.83 (1.26-2.66)
Male respondent (n=6,015)	9.1% (6.2%-12%)	17% (15%-18%)	1.82 (1.29-2.55)	0.408
Female respondent (n=7,129)	12% (9.5%-15%)	23% (21%-24%)	1.83 (1.42-2.36)
Non-Hispanic white respondent (n=10,653)	10% (8.2%-13%)	17% (16%-18%)	1.63 (1.30-2.06)	0.644
Non-white respondent (n=2,491)	14% (8.5%-19%)	28% (26%-31%)	2.05 (1.36-3.10)

TUS = Tobacco Use Supplement; FSS = Food Security Supplement; sd = standard deviation

Data = Current Population Survey (CPS), 2001-2019

^aThe smoking sample consists of recent quitters and continuing smokers. All individuals in this sample were current smokers at Time 0, one year prior to the TUS.

^bRecent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (i.e., between Time 0 and Time 1)

^cContinuing smokers are current smokers at the TUS (Time 1) and one year prior (Time 0).

^dWe used modified Poisson regression to model food insecurity as a function of smoking status (continuing smokers vs recent quitters), adjusting for sex, age, household size, whether the household has any children, number of current smokers in the household, and panel timing. We then calculated predicted probabilities of food insecurity separately for recent quitters and continuing smokers.

^eAll estimates are sample weighted using custom, household-level longitudinal weights.

Our multinomial logistic regression showed that continuing smokers had a lower predicted probability of high food security (69% vs 80%) and a higher probability of marginal (11% vs 9.8%), low (12% vs 7%), and very low food security (7.8% vs 3.6%) compared to recent quitters (Figure 3). If a person was a continuing smoker (compared to a recent quitter), they were 1.28 (95% CI: 0.97-1.69), 2.04 (95% CI: 1.50-2.77), and 2.53 (95% CI: 1.72-3.70) times as likely to live in a household experiencing marginal, low, and very low food security respectively compared to high food security after adjusting for individual and household characteristics. Again, this pattern held across all stratified analyses (Appendix Table 1).

Figure 3. Predicted probabilities of detailed household food security status by smoking status.

Figure 3 shows the predicted probabilities of each level of household food security for continuing smokers and recent quitters for individuals in the smoking sample (n=13,144). All individuals in this sample were current smokers at Time 0, one year prior to the TUS. Recent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (i.e., between Time 0 and Time 1). Continuing smokers were current smokers at the TUS (Time 1) and one year prior (Time 0). We used multinomial logistic regression to model detailed household food security status as a function of smoking status (continuing smokers vs recent quitters), adjusting for sex, age, household size, whether the household has any children, number of current smokers in the household, and panel timing. The abbreviation RRR refers to the relative risk ratios from the multinomial logistic regression. We then calculated predicted probabilities of each level of food security (high, marginal, low, very low) separately for recent quitters and continuing smokers. All estimates are sample weighted using custom, household-level longitudinal weights. The predicted probability of high food security (not shown above) is 69% for continuing smokers and 80% for recent quitters.

DISCUSSION

We found that the risk of experiencing household food insecurity at follow-up was nearly twice as high among individuals who continued to smoke cigarettes compared with smokers who recently quit. Further, the relative risk associated with continuing smoking was largest when considering very low food security, suggesting that, if the association we observe is causal, quitting smoking may have the biggest impact on food security for households with the greatest unmet food needs. These patterns were consistently observed across households with and without children; households at or below 185% of the federal poverty line; male and female respondents; and non-Hispanic white and Hispanic or non-white respondents.

Complex structural factors lead to a higher prevalence of both cigarette use and food insecurity among low-income populations. Tobacco marketing disproportionately targets low-income individuals; for example, SNAP-authorized (the Supplemental Nutrition Assistance Program) stores have 3 times the odds of displaying interior tobacco advertisements.⁵⁰ Additionally, low-income smokers may have less access to cessation resources, lower social support for quitting, and higher levels of dependence on tobacco.⁵¹ Similarly, there are structural issues within the food system (e.g. the transition to a reliance on ultraprocessed foods⁵² and the unequal neighborhood distribution of quality food retailers^53,54) that make it difficult for households with limited financial resources to access healthy food.

Because both tobacco use and food insecurity are so closely tied to socioeconomic disadvantage, financial strain is likely a key explanation for the association between cigarette use and food insecurity.²⁹ Households with smokers may spend a substantial amount of money on tobacco, potentially limiting funds available to purchase food. Lower income households are more likely to spend a larger proportion of their total expenditure on cigarettes.⁵⁵ This can lead to smoking-induced deprivation; 28% of smokers in the US responded that they have spent money on cigarettes that they knew would be better spent on household essentials like food.⁵⁶ For context, households participating in SNAP receive approximately the same amount, on average, in monthly benefits ($240) as the average US household that purchases one pack per day spends on cigarettes.^11,57 Given that SNAP participation reduces household food insecurity,⁵⁸ the financial savings from quitting smoking could also go far towards reducing risk of food insecurity.

If the relationship between quitting smoking and improved food security is causal, promoting tobacco cessation could potentially be an effective way to alleviate food insecurity among cigarette smokers and among non-smokers who live in smoking households. Government-sponsored food assistance programs, including SNAP and the Special Supplemental Nutrition Program for Women, Infants and Children (WIC), should provide information about cigarette cessation and help connect participants who smoke to appropriate cessation interventions and treatments. Additionally, government funding for non-profit and charitable food organizations (e.g., food pantries and home meal delivery programs) should incentivize organizations to provide cessation resources. Expanding existing tobacco control policies (e.g., regulating nicotine content in cigarettes; banning flavored products; increasing tobacco taxes) and increasing funding for tobacco prevention may also help alleviate food insecurity. Finally, implementing stronger poverty reduction policies will reduce both tobacco use and food insecurity.

A key strength of this analysis is our use of the CPS to longitudinally investigate how a change in smoking status is associated with subsequent household food security. By linking the TUS and the FSS, we were able to leverage a large, longitudinal sample that is considered the gold standard source of information on food security in the US. Another strength is the use of multiple linked panels of data across two decades which shows that the relationship between continuing smoking and food insecurity is an enduring feature of addiction and household economics in the US rather than a short-term trend.

However, we acknowledge several limitations. First, while our data is longitudinal, we only have one measure of household food security (at Time 2). Thus, we were unable to incorporate a baseline assessment of food security prior to or concurrent with our assessment of cigarette cessation. Relatedly, we cannot rule out the possibility that improvements in food security could have contributed to smoking cessation and acknowledge that the relationship is likely bidirectional. However, regardless of the exact nature and direction of the relationship between cigarette smoking and food insecurity, quitting smoking could free up financial resources for food insecure individuals that could be used to acquire food. Second, we use responses from a single time point to assess change in smoking status, relying on participants to report both their current smoking status and recall how recently they quit smoking (if applicable). However, self-reported smoking history questions have high validity and reliability^59,60 and are commonly used in tobacco research. Third, we were unable to include proxy responses to the tobacco questions. If individuals who could not be reached for a self-interview differed from included individuals in ways that were associated with tobacco use and food security, this could bias our results. For example, if people who were more likely to both be continuing smokers and experience household food insecurity were disproportionately excluded from our analysis, we may underestimate the association between continuing smoking and food insecurity. However, the prevalence of food insecurity is similar for TUS proxy respondents (8%) and self-respondents (9%) so we expect any resulting bias to be small. Finally, because our measure of food security is based on a 12-month recall period, recall bias may be present. Further, this measure is unable to capture whether households experience episodic or persistent food insecurity. The results should be interpreted with these limitations in mind.

CONCLUSION

Households with smokers can spend a substantial amount of money on tobacco, potentially constraining resources available to purchase food. Risk of food insecurity is lower among households where one smoking adult has quit smoking cigarettes than among households where people continue to smoke. Therefore, promoting cigarette cessation could potentially be an effective way to reduce food insecurity. Food assistance programs and other interventions designed to reduce food insecurity should help connect participants who smoke cigarettes to cessation resources.

ABBREVIATIONS & ACRONYMS

CPS

Current Population Survey

FSS

Food Security Supplement

SNAP

Supplemental Nutrition Assistance Program

TUS

Tobacco Use Supplement

USDA

United States Department of Agriculture

WIC

Special Supplemental Nutrition Program for Women, Infants, and Children

Methods Appendix

CPS Data

The Current Population Survey (CPS) is a longitudinal survey that follows a 4-8-4 monthly rotating panel design. After being selected into the sample, household members are surveyed each month for four consecutive months, not surveyed for eight subsequent months, then the same households are surveyed once again monthly for four additional consecutive months.⁴⁵ As a result, household members are surveyed eight times over a period of 16 months. However, due to the technical obstacles associated with linking CPS files across months, analyses of CPS data traditionally have not taken advantage of the longitudinal design.⁴⁶

Researchers at the University of Minnesota’s Institute for Social Research and Data Innovation have recently addressed some of the challenges associated with linking CPS data.⁴⁶ Integrated, harmonized, and fully linkable data is now available through IPUMS at the University of Minnesota.⁶¹ Using IPUMS-CPS data allows users to extract longitudinal data on household and individuals from basic and supplemental surveys across the entire 16-month panel.

Proxy Responses in the Tobacco Use Supplement

All CPS household members aged 15 years and older (changed to 18 years or older in samples from 2010 forward) who completed the CPS core items are eligible to complete the TUS. Prior to 2014, attempts were made to self-interview all eligible individuals, and proxy responses were given by other household members after four failed callback attempts. Between 2014-2015, two to three eligible individuals per household were randomly selected for self-interviews, and proxy responses were collected for the remaining household members. Beginning in 2018, supplement-eligible responses not selected for self-interview were not given a proxy-response interview. Proxy-response interviews continued to be given to those randomly selected for self-interview after four failed callback attempts.⁴⁷

Sample Weighting

The Bureau of Labor Statistics provides cross-sectional, individual- and household-level sample weights for the CPS that account for sampling procedures and non-response. However, longitudinal weights are not available for linked analyses. Following guidelines established by IPUMS,⁶² we created custom, longitudinal weights for this analysis using Iterative Proportional Fitting (also called “raking”) with the “ipfraking” package in Stata.^63,64 Using the cross-sectional household-level FSS supplement weights as a base, we first created a set of population counts for the individuals at Time 2 who were eligible to have responded to the TUS at Time 1. We then adjusted the weights for those who actually linked so that the population count of the linked samples is equivalent to the population count of those who were eligible to link. Population totals were based on three combinations of individual-level variables in the Time 2 data: (1) Hispanic ethnicity, age, and sex totals; (2) race, age, and sex totals; and (3) state, age, and sex totals. We used the resulting raked weight for all primary analyses.

In supplementary analyses, we compared these results to estimates generated using no sample weights and using the cross-sectional, household-level base weights from the FSS and found comparable results (Appendix Table 2, Appendix Table 3).

Appendix Figure 1.

Flow chart showing linkage steps and inclusion criteria

Appendix Table 1.

Multinomial logistic regression: Association between continuing smoking (versus quitting smoking) and detailed household food security status (smoking sample,^a n=13,144)

		OUTCOME = DETAILED HOUSEHOLD FOOD SECURITY
Sub-Sample	Smoking Statusb,c	High Food Security	Marginal Food Security	Low Food Security	Very Low Food Security
		RRR (95% CI)d,e
Overall (n=13,144)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.28 (0.97-1.69)	2.04 (1.50-2.77)	2.53 (1.72-3.70)
Households with children (n=5,045)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.03 (0.70-1.52)	2.00 (1.29-3.10)	3.17 (1.71-5.90)
Households without children (n=8,099)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.57 (1.03-2.39)	2.03 (1.29-3.18)	1.82 (1.13-2.95)
Below 185% poverty (n=4,708)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	0.92 (0.61-1.38)	1.54 (1.01-2.34)	1.76 (1.08-2.87)
Above 185% poverty (n=8,436)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.30 (0.85-1.98)	1.85 (1.13-3.04)	2.41 (1.20-4.86)
Male respondent (n=6,015)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.44 (0.91-2.27)	2.00 (1.23-3.26)	2.19 (1.20-4.02)
Female respondent (n=7,129)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.20 (0.84-1.71)	1.99 (1.34-2.96)	2.68 (1.64-4.38)
Non-Hispanic white respondent (n=10,653)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.30 (0.94-1.79)	1.68 (1.20-2.36)	2.20 (1.44-3.35)
Non-white respondent (n=2,491)	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	0.98 (0.55-1.74)	2.49 (1.28-4.84)	2.77 (1.22-6.30)
		Predicted Probability (95% CI)d,e
Overall (n=13,144)	Recent quitter	80% (77%-82%)	9.8% (7.6%-12%)	7.0% (5.2%-8.8%)	3.6% (2.4%-4.8%)
	Continuing smoker	69% (68%-70%)	11% (10%-12%)	12% (11%-13%)	7.8% (7.1%-8.4%)
Households with children (n=5,045)	Recent quitter	73% (68%-78%)	14% (10%-19%)	9.2% (5.9%-13%)	3.0% (1.4%-4.5%)
	Continuing smoker	64% (62%-65%)	13% (12%-14%)	16% (14%-17%)	7.9% (6.7%-9.0%)
Households without children (n=8,099)	Recent quitter	83% (79%-86%)	7.0% (4.5%-9.4%)	5.5% (3.4%-7.6%)	4.7% (2.8%-6.7%)
	Continuing smoker	73% (72%-74%)	9.6% (8.7%-11%)	9.8% (8.8%-11%)	7.5% (6.7%-8.3%)
Below 185% poverty (n=4,708)	Recent quitter	53% (46%-60%)	22% (16%-27%)	16% (11%-20%)	10% (6.4%-14%)
	Continuing smoker	46% (44%-48%)	17% (16%-19%)	21% (19%-22%)	16% (14%-17%)
Above 185% poverty (n=8,436)	Recent quitter	88% (86%-91%)	5.9% (3.8%-8.0%)	4.2% (2.4%-6.0%)	1.4% (0.005%-2.3%)
	Continuing smoker	83% (81%-84%)	7.1% (6.4%-7.9%)	7.2% (6.3%-8.0%)	3.2% (2.6%-3.8%)
Male respondent (n=6,015)	Recent quitter	83% (79%-87%)	7.6% (4.7%-11%)	5.8% (3.4%-8.2%)	3.3% (1.5%-5.0%)
	Continuing smoker	74% (72%-75%)	9.7% (8.7%-11%)	10% (9.1%-11%)	6.3% (5.4%-7.2%)
Female respondent (n=7,129)	Recent quitter	76% (72%-80%)	12% (8.4%-15%)	8.2% (5.6%-11%)	4.0% (2.3%-5.7%)
	Continuing smoker	66% (64%-67%)	12% (11%-13%)	14% (13%-15%)	9.0% (8.0%-10%)
Non-Hispanic white respondent (n=10,653)	Recent quitter	81% (78%-84%)	8.4% (6.2%-11%)	7.0% (5.0%-9.0%)	3.4% (2.1%-4.7%)
	Continuing smoker	73% (72%-74%)	9.8% (9.0%-11%)	10% (9.6%-11%)	6.6% (6.0%-7.3%)
Non-white respondent (n=2,491)	Recent quitter	70% (61%-78%)	17% (9.9%-24%)	8.4% (3.8%-13%)	4.9% (1.6%-8.2%)
	Continuing smoker	58% (56%-61%)	14% (12%-16%)	17% (15%-19%)	11% (9.2%-13%)

TUS = Tobacco Use Supplement; FSS = Food Security Supplement; sd = standard deviation

Data = Current Population Survey (CPS), 2001-2019

^aThe smoking sample consists of recent quitters and continuing smokers. All individuals in this sample were current smokers at Time 0, one year prior to the TUS.

^bRecent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (i.e., between Time 0 and Time 1)

^cContinuing smokers are current smokers at the TUS (Time 1) and one year prior (Time 0).

^dWe used multinomial logistic regression to model detailed household food security status as a function of smoking status (continuing smokers vs recent quitters), adjusting for sex, age, household size, whether the household has any children, number of current smokers in the household, and panel timing. We then calculated predicted probabilities of each level of food security (high, marginal, low, very low) separately for recent quitters and continuing smokers.

^eAll estimates are sample weighted using custom, household-level longitudinal weights.

Appendix Table 2.

Comparison of Sample Weights for Modified Poisson Regression: Relative risk of food insecurity for continuing smokers compared to recent quitters for smoking sample^a (n=13,144)

	OUTCOME = HOUSEHOLD FOOD INSECURITY
	Predicted Probability (95% CI)d,e		RR (95% CI)d,e
Weights	Recent quitterb	Continuing smokerc	Ref= recent quitter
(1) Unweighted	11% (9.5%-13%)	19% (18%-19%)	1.62 (1.35-1.94)
(2) Cross-sectional, household level weights from the FSS	11% (8.9%-13%)	19% (19%-20%)	1.78 (1.45-2.18)
(3) Custom, longitudinal, household-level weights	11% (8.7%-13%)	20% (19%-21%)	1.85 (1.51-2.27)

TUS = Tobacco Use Supplement; FSS = Food Security Supplement; sd = standard deviation

Data = Current Population Survey (CPS), 2001-2019

^aThe smoking sample consists of recent quitters and continuing smokers. All individuals in this sample were current smokers at Time 0, one year prior to the TUS.

^bRecent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (i.e., between Time 0 and Time 1)

^cContinuing smokers are current smokers at the TUS (Time 1) and one year prior (Time 0).

^dWe used modified Poisson regression to model food insecurity as a function of smoking status (continuing smokers vs recent quitters), adjusting for sex, age, household size, whether the household has any children, number of current smokers in the household, and panel timing. We then calculated predicted probabilities of food insecurity separately for recent quitters and continuing smokers.

Appendix Table 3.

Comparison of Weights for Multinomial logistic regression: Association between continuing smoking (versus quitting smoking) and detailed household food security status (smoking sample,^a n=13,144)

		OUTCOME = DETAILED HOUSEHOLD FOOD SECURITY
Weights	Smoking Statusb,c	High Food Security	Marginal Food Security	Low Food Security	Very Low Food Security
		RRR (95% CI)d,e
(1) Unweighted	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.26 (1.00-1.60)	1.77 (1.38-2.28)	2.00 (1.47-2.71)
(2) Cross-sectional, household level weights from the FSS	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.32 (1.00-1.73)	1.97 (1.45-2.67)	2.35 (1.61-3.44)
(3) Custom, longitudinal, household-level weights	Recent quitter	-	-	-	-
	Continuing smoker	(base outcome)	1.28 (0.97-1.69)	2.04 (1.50-2.77)	2.53 (1.72-3.70)
		Predicted Probability (95% CI)d,e
(1) Unweighted	Recent quitter	79% (77%-82%)	9.6% (7.7%-11%)	7.0% (5.6%-8.5%)	4.4% (3.2%-5.5%)
	Continuing smoker	71% (70%-72%)	11% (10%-11%)	11% (10%-12%)	7.6% (7.1%-8.1%)
(2) Cross-sectional, household level weights from the FSS	Recent quitter	80% (77%-83%)	9.3% (7.2%-11%)	6.9% (5.2%-8.7%)	3.8% (2.6%-5.1%)
	Continuing smoker	70% (69%-71%)	11% (9.9%-11%)	12% (11%-12%)	7.8% (7.1%-8.4%)
(3) Custom, longitudinal, household-level weights	Recent quitter	80% (77%-82%)	9.8% (7.6%-12%)	7.0% (5.2%-8.8%)	3.6% (2.4%-4.8%)
	Continuing smoker	69% (68%-70%)	11% (10%-12%)	12% (11%-13%)	7.8% (7.1%-8.4%)

TUS = Tobacco Use Supplement; FSS = Food Security Supplement; sd = standard deviation

Data = Current Population Survey (CPS), 2001-2019

^aThe smoking sample consists of recent quitters and continuing smokers. All individuals in this sample were current smokers at Time 0, one year prior to the TUS.

^bRecent quitters are former smokers at the TUS (Time 1) who reported quitting within the last year (i.e., between Time 0 and Time 1)

^cContinuing smokers are current smokers at the TUS (Time 1) and one year prior (Time 0).

^dWe used multinomial logistic regression to model detailed household food security status as a function of smoking status (continuing smokers vs recent quitters), adjusting for sex, age, household size, whether the household has any children, number of current smokers in the household, and panel timing. We then calculated predicted probabilities of each level of food security (high, marginal, low, very low) separately for recent quitters and continuing smokers.

Data Availability

IPUMS-CPS data were downloaded from https://cps.ipums.org/cps/. All analyses were completed using Stata, Version 17. Code for longitudinally linking the supplements, creating sample weights, and analyzing the data are available upon request to the corresponding author (Kaitlyn M. Berry, berry590@umn.edu).