The role of pregnancy health problems on maternal smoking behaviours

ABSTRACT

To explore the effects of pregnancy health problems (PHPs) on smoking behaviours during and after pregnancy (‘smoking-inducing’ effect), we estimate a two-period model that jointly determines prenatal and postnatal smoking decisions, taking into consideration the presence of PHPs. While PHPs are likely to reduce prenatal (except for heavy smokers) and postnatal smoking propensity, we still observe considerable postnatal relapse in the sample, which can be attributed to smoking addiction, as well as information asymmetries and maternal stress associated with PHPs. Thus, we advocate for smoking cessation policies and programmes throughout and beyond pregnancy to avoid potential intertemporal substitution between prenatal and postnatal cigarette consumption.

KEYWORDS:

• Smoking intensity
• pregnancy health problems
• addictive behaviours
• information asymmetry
• stress

JEL CLASSIFICATION:

• I12
• I18
• I10

Acknowledgments

We are grateful to The Centre for Longitudinal Studies, UCL Institute of Education for the use of the data and to the UK Data Archive and UK Data Service for making them available. However, they bear no responsibility for the analysis or interpretation of these data. All errors are our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ According to the Centers for Disease Control and Prevention, during the past 50 years more than 100,000 infants who had been exposed to cigarettes died of Sudden Infant Death Syndrome or other smoking-caused health conditions (DHHS 2014).

² Previous studies show that health shocks, in general, positively affect health-promoting behaviours such as less substance use, and that higher health risks can alter adult smoking decisions, rather than specifically studying the population of pregnant women (Sundmacher 2012; Marti and Richards 2017).

³ The model excludes the pre-pregnancy smoking decision ($t_0$), since the dataset has little information on pre-pregnancy characteristics.

⁴ Smoking may or may not be valued because it generates different utilities for women at different times. For example, it may directly enhance the utility from cigarette consumption for long-term smokers, but it decreases the utility of women who choose not to smoke, as smoking imposes a burden on them.

⁵ For more information on the dataset, the sampling process and the timing and content of the survey, see Hansen (2012).

⁶ These quartiles are defined based on the distribution of average daily cigarette consumption during pregnancy in the sample. The estimation results remain qualitatively the same when we adopt the same definition as in Bradford (2003) – 0–10 cigarettes per day as light smoker, 10–20 as moderate smoker, and 20 or more cigarettes as heavy smoker – or when we use tertiles of the cigarette consumption distribution to define light, moderate, and heavy smokers.

⁷ Many pregnancy-related health conditions are caused by pregnancy hormones (CDC 2017), and we thus consider them as relatively exogenous.

⁸ The psychology literature shows that personality traits are relatively stable over the lifetime (Almlund et al. 2011).

⁹ Results available upon request from the authors.

¹⁰ Second-hand smoking effects could also be captured by controlling for the presence of clean indoor air laws which would increase the costs of smoking in a private setting. However, for the time period in our sample (early 2000 to early 2002) the amendments to the Smoke Control Areas in England, Wales and Scotland – as well as any previous laws that were in effect across the different countries in the U.K. (i.e. Clean Air Act 1993) – will be reflected in the region fixed effects. See the official U.K. website http://www.legislation.gov.uk for further details on smoking legislation.

¹¹ As indicated at the bottom of Table 4, the reported correlations among the four prenatal smoker types within the period, as well as those between the two periods, are all statistically significant at the 1% level. This is consistent with our two-period model, which accounts for observed and unobserved traits that affect both prenatal and postnatal smoking decisions, taking into consideration within- and across-period error correlations.

¹² The estimated coefficients for the effects of each trimester prenatal care on smoking behaviours used to calculate these probabilities are given in Appendix Table A1.

¹³ Such a bias may be strongest for non-chronic PHPs such as gestational diabetes, vomiting or urinary track infections. We estimate the effects of specific PHPs on maternal smoking decisions, but do not find a clear pattern across chronic and non-chronic PHPs (Appendix Table A2). Focusing on mothers who experience non-chronic PHPs (diabetes, vomiting, UTIs, and other illness) compared to more chronic PHPs (early or late bleeding, anaemia, raised blood pressure) – an exploratory factor analysis indicates which conditions reflect the same underlying trait – we do not find statistically significant differences in postpartum smoking between the two types of PHPs (Appendix Table A3). However, chronic conditions negatively affect moderate smoking, and positively affect heavy smoking, compared to non-chronic PHPs, suggesting that PHPs can be endogenous.

In the same vein, we sort PHPs into two groups, based on whether or not a PHP is easily detectable: PHPs that can be detected only after a medical diagnosis, which we call ‘unobservable’ PHPs, and ‘observable’ PHPs otherwise. In Appendix Table A4, we show that, compared to those with observable PHPs, women with unobservable PHPs decrease the probability of postnatal smoking by 2.4 p.p., and are 3.3 p.p less likely to become a heavy postnatal smoker. In addition, the difference in the estimated marginal effects between the two different types of PHPs is statistically significant.

¹⁴ Appendix Table A5 shows that mothers with and without PHPs differ in terms of observed characteristics.

¹⁵ Neither the Lewbel nor the BMI instruments are redundant according to the $LM-$redundancy test (${\chi }^2=95.69$). We also use maternal pre-pregnancy weight (with and without the Lewbel instruments), and the validity of the instruments is similar.

¹⁶ If we re-estimate Tables 4–6 accounting for the potential endogeneity of PHPs, we still find the same pattern, with the estimated coefficients being higher by 5–10% (results available upon request).

¹⁷ Due to data limitations, we cannot estimate a maternal fixed-effect model. To alleviate concerns that mothers may respond to these questions in reference to a different (current) pregnancy, we repeat the analysis only for mothers who are not currently pregnant ($N=14,941$) (Appendix Table A6, Panel A), only mothers who have received prenatal care ($N=15,099$) (Appendix Table A6, Panel B), and only first-time mothers ($N=7,171$), using sampling weights. For all these cases, the results are equivalent to the ones from the main analysis. As additional robustness checks, we estimate the main model when disaggregating by education level (college versus non-college graduates), by marital status (married versus unmarried), and by poverty status (above versus below poverty). All results remain qualitatively the same. These results are available upon request.

¹⁸ The age of a mother, especially for new mothers, may also play a role in acquiring information relating to pregnancy. For example, an adult mother may receive a different set of information for her planned pregnancy relative to a teenage mother who experiences an accidental/unplanned pregnancy. To consider the age effect on information asymmetry, we have conducted a series of analyses including interaction terms among age, firstborn, and PHPs, but do not find any statistical significant relationship between these factors on the propensity of maternal smoking. These results are available from the authors upon request.