The Effect of Social Support on Prenatal Smoking in Romania

Full text

1
The Effect of Social Support on Prenatal Smoking in Romania
Economics 399
Katherine LeMasters
Smoking is one of few controllable risk factors for pregnancy and it is associated with poor
health outcomes for mothers and children. While prenatal smoking is now low in many places,
Eastern European countries continue to see high prenatal smoking rates. My study explores the
effect of women’s social support on prenatal smoking decisions, a relationship that has not been
explored in the context of Romania. To do this I use probit models, which allow me to determine
the likelihood that a woman is a current or recent smoker, depending on her level of social
support. I find that some facets of social support are more protective against smoking than
others, such as the number of people that a woman can count on for recognition of her talents
and skills. I also find that having a medium level of social support (e.g., three or four people to
count on for support) is just as, if not more effective, than having a high level of social support
(e.g., five people or more to count on for social support). I then assess whether the effects of
social support on smoking behaviors can be explained by a decline in stress. I find that stress is
not a statistically significant predictor of reduced smoking and that social support remains
statistically significant when stress is included in the model. While social support is directly
linked to a lower probability of women smoking six months before pregnancy, I find little
evidence to suggest that it encourages women who smoke within the six months before
pregnancy or during her pregnancy to quit. While I cannot reach any causal conclusions, I do
conclude that Romania should look further into the social support systems that pregnant women
have when assessing how to improve prenatal smoking behaviors.

2
Table of Contents
I. Introduction……………………………...……………………………………………... 3
II. Historical Context……………………………...………………………………............ 5
III. Literature Review........................................……………………………...…………... 8
IV. Data, Descriptive Statistics, and Correlations…………...……………………............ 11
V. Empirical Model and Results………………...………………………………….......... 17
VI. Concerns……………………………...……………………………………………..... 26
VII. Conclusions and Implications……………………………...……………………....... 28
IX. Tables and Figures……………………………...…………………………................. 30
X. Appendix……………………………...…………………………………..................... 38
XI. Bibliography……………………………...…………………………………............... 40

3
I. Introduction
Smoking during pregnancy is one of few controllable risk factors for poor birth outcomes and
poor childhood health, yet it remains prevalent in many societies (Holtrop, Meghea, Raffo,
Biery, Chartkoff, Roman, 2010). This health behavior is associated with pregnancy
complications, low birth weight, and preterm birth, as well as other poor health outcomes
(Lumley, Chamberlain, Dowswell, Oliver, Oakley, Watson, 2009). Postpartum maternal smoking
also exposes children to secondhand smoke (SHS), which, in turn, further deteriorates their own
health and increases their chances of smoking later in life (Been, Nurmatov, Cox, Nawrot, Van
Schayck, & Sheikh, 2014; Lumley et al., 2009; Jackson & Henriksen, 1997).
1
In the long run,
children born to women who smoked during pregnancy and/or postpartum have higher rates of
respiratory, physical, and behavioral issues (Holtrop et al., 2010).
Many high-income countries, specifically the United States (US) and those in Western
Europe, have implemented a variety of programs to reduce prenatal smoking with focuses on
health education, mental health, partner involvement, reduced nicotine addiction, and more
(Lumley et al., 2009). In the US, 23 percent of women smoked three months before pregnancy in
2010 with 11 percent continuing to smoke during pregnancy (Center for Disease Control, 2010).
However, many countries in Central and Eastern Europe (CEE) do not have targeted
interventions for pregnant smokers and continue to have high smoking rates today. In 2008,
women’s smoking rates in Romania reached a high of 41 percent, with 15 percent of women
continuing to smoke while pregnant (Meghea, Rus, Rus, Holtrop, & Roman, 2012). In a more
recent study, 24 percent of women sampled smoked six months before pregnancy and 15 percent
of all women continued to smoke while pregnant (Meghea et al., 2012). So, while smoking rates
1
The postpartum period begins when the child is born and ends when the child is one year old.

4
prior to pregnancy decreased, smoking rates during pregnancy did not, which may indicate that
women addicted to nicotine or those that previously smoked a high number of cigarettes per day
were not reached.
2
Most health outcomes related to smoking are not measured in Romania, but infant mortality
rates (IMR) are, as continued smoking during pregnancy increases the chances of infant
mortality (Kleinman, Piere, Madans, Land, & Schramm, 1987; Rantakallio, 1978). In Romania,
the IMR steadily decreased from 64.1 deaths per 1,000 live births in 1968 to 31.2 deaths per
1,000 live births in 1985 when it started to increase again, as this was a time when health
facilities offered poor quality care and Nicolae Ceaușescu’s regime was arguably at its most
repressive point, which I expand on in the next section (The World Bank Group, 2014A). IMR
then decreased again in 1990 and has continued to decrease at a slow rate, reaching 10.5 deaths
per 1,000 live births in 2013 (The World Bank Group, 2014A). While this decrease may seem
like a success, Romania’s IMR is much higher than many neighboring countries’. The Czech
Republic, Hungry, and Ukraine have IMRs of 2.9, 5.2, and 8.6, respectively (The World Bank
Group, 2014A). Additionally, the US’s IMR is 5.9, making it almost half that of Romania’s (The
World Bank Group, 2014A). While IMR is related to many variables other than smoking, these
statistics do serve as a useful frame of reference to show how smoking during pregnancy affects
infants.
There is no consensus in Romania or most of CEE as to causes of prenatal smoking and there
is a scarcity of current studies comprehensively characterizing pregnant women and their
smoking behaviors and exposure to smoke in this region (Meghea, Rus, & Dirle, 2010; Lopez,
2
It is interesting that while women’s smoking rates in Romania are much higher than those in
the US, a similar percentage continue smoking while pregnant. This finding is further indicative
that methods to decrease smoking do not often target pregnant women. However, I do not have
time to expand on a cross-country analysis here.

5
Collishaw, & Piha, 1994; Thomas, Fayter, Misso, Ogilvie, Petticrew, Sowden, & Worthy, 2008;
Gilman, Breslau, Subramanian, Hisman, & Koenen, 2008).
3
This paper attempts to fill this void
by assessing the role of social support on pregnant women’s smoking behaviors in Romania. I
use probit models to assess whether social support is correlated with women’s decisions to
smoke six months prior to getting pregnant, and, conditional on smoking six months before
pregnancy, whether social support is correlated with quitting within the six months before
pregnancy or during pregnancy. I find that women with more social support (specifically women
with more people to count on for guidance, friendship, help, comfort, recognition of talents and
skills, and emotional support) are less likely to smoke six months prior to pregnancy. However, I
find that social support does not influence women’s decisions to quit within the six months
before pregnancy or during her pregnancy. Having medium levels of social support (e.g., three or
four people to count on for support) is just as effective at influencing smoking behaviors than
having high levels of social support (e.g., five or more people to count on for support). I also
assess whether social support affects smoking decisions indirectly by decreasing a woman’s
stress levels. I find that the coefficients on social support are virtually unchanged when I control
for stress, which suggests that the effects of social support cannot simply be explained by a
reduction in stress.
II. Historical Context
First, I must look at Romania’s historical context to fully comprehend why its smoking rates
are as high as they were in many other countries three decades ago (Krstev, Marinkovic, Simic,
3
There is little consensus as to the causes of smoking during pregnancy in the US and other
Western countries as well. However, one study attributes increases in women’s smoking to
increased social acceptance, liberalization of norms for women’s behavior, rebelliousness, and
potential weight loss (Waldron, 1991). These findings are not generalizable across societies, so I
do not assume that these factors weigh heavily in Romania.

6
Kocev, & Bondy, 2012).
4
Romania was ruled by a communist government starting in 1947 until
the Romanian people revolted in December of 1989. They overthrew Ceaușescu’s regime in
what is known as the December Revolution and replaced this regime with a semi-presidential
republic. Before the revolution, all social institutions, including healthcare, were publically
owned, were of poor quality, and were highly centralized. The government infringed upon many
of women’s sexual and reproductive health (SRH) rights, did not offer health education during or
after pregnancy, and outlawed abortion (Hord, David, Donnay, & Wolf, 1991). So, there was no
information regarding the harmful effects of smoking, little incentive for women to receive
prenatal care, and a fear of the formal medical system.
Since the revolution, most health policy has focused on rebuilding the health system’s
infrastructure (i.e., free access to medical care in state-owned institutions) rather than on
changing health behaviors (i.e., increasing breastfeeding), so most care is geared towards
curative physical health in urban areas rather than preventative education that is transferrable
throughout the country (Hord et al., 1991).
5
Romania has no central protocol for health-related
information given during prenatal care, and while most women are told that smoking is bad for
their child’s health during prenatal appointments, they are neither given guidance on how to quit
nor are told about how grave the consequences from continued smoking can be.
4
For a comparison of smoking rates, the US states had an adult smoking rate of 42.4 percent in
1965, 30.1 percent in 1985, 24.7 percent in 1995, 20.9 percent in 2005, and 19.0 percent in 2011
(Centers for Disease Control and Prevention, 2013). The smoking rate in Romania as of 2011
was 26.7 percent for all adults but was over 37 percent for males (Irimie, 2011).
5
One consequence of the healthcare system’s focus being on urban areas is that the IMR is
double for rural areas and minorities (i.e., Roma) (UNICEF, 2014). This is important in my later
empirical tests, as location is a covariate in my regression analyses. While I do have observations
for those of Roma ethnicity (a marginalized minority), there are too few observations to conduct
quantitative analysis on them, but I do include their descriptive statistics in Tables 1, 2, and 4.

7
Early efforts to decrease smoking lacked conviction (e.g., the Tobacco Control Program), as
the health system was uninvolved, the price of cigarettes did not increase, and no actors had a
vested interested in lowering smoking rates. In 2005, the Ministry of Health (MOH) began the
national program, ‘Stop Smoking,’ and the World Health Organization (WHO) introduced the
Framework Convention on Tobacco Control (FCTC), which showed smoking in a bad light but
did not cater to pregnant women and subsequently caused no decrease in Romania’s smoking
rates (Webb, 2013).
Additionally, Romania’s newly liberalized economy allowed multinational tobacco
companies to enter the country starting in 1989, gain political power, and advertise smoking as a
sign of feminism, all of which increased social pressure for women to smoke (David, Esson,
Perucic, & Fitzpatrick, 2010). The combination of the health sector’s lack of preventative
healthcare and the country’s lack of tobacco control have increased Romania’s already high
smoking rates. This increase is most pronounced for women, as female tobacco usage increased
from 11 to 25 percent from 1989-2000 and as smoking rates are even higher for women of
childbearing age (Irimie, 2011).
6
In one study conducted in 2008, smoking among women of
childbearing age was 41 percent with 15 percent of all women smoking while pregnant (Meghea,
2012).
It was not until smoking became more expensive in 2010 through an increase in the excise
tax along with an increased exchange rate with the Euro and an economic crisis that smoking
rates began to decrease for all socioeconomic groups (Thomas et al., 2010). Yet, smoking rates
in Romania remain around 30 percent (smoking rates for women ages 25-44 remain around 24
6
Specifically, between 1995-2000, women’s smoking rates increased from 15.2 percent to 25.0
percent while men’s smoking rates increased from 42.7 percent to 48.0 percent (Irimie, 2011).
So, while men smoke more than women, women experienced a much larger percentage increase
in smoking rates after communism fell.

8
percent), as high as those of the West three decades ago (Irimie, 2011; Krstev et al., 2012).
Additionally, the Romanian government was one of only four countries to vote against the
Tobacco Products Directive negotiations in the European Union (EU) in 2013, indicating that
Romania still lacks the political clout to end smoking (Joossens & Raw, 2014).
III. Literature Review
An important risk factor affecting smoking during pregnancy is the lack of a spouse, friends,
and family that provide psychological and material resources to women (Cohen & Wills, 1985;
Meghea et al., 2010; Connor & McIntyre 1999; Harley & Eskenazi, 2006; Elsenbruch, Benson,
Rucke, Rose, Dudenhausen, Pincus-Knackstedt, Klapp, & Arck, 2007). Social support is defined
as the presence of psychological resources, particularly social stability and social participation,
which provide emotional and instrumental support to women (Glazier, Elgar, Goel, & Holzapfel,
2004). Supportive social relationships promote healthy behavioral change, so a lack of social
support likely perpetuates unhealthy behaviors (Cohen, Mermelstein, Kamarck, & Hoberman,
1985). This study uses the Social Network Support Scale to measure social support, which has
been modified from the Lubben Social Network Scale.
7
I thus focus on how social support can
potentially serve as a mediating factor for women to smoke during pregnancy.
Types of Social Support
The literature considers two types of social support: structural and functional support.
Structural support is defined as a numerical measure of social relationships and social network
size (e.g., the quantity of support) while functional social support is defined as an individual’s
own perception of the support available or support received (e.g., the quality of support) (Harley
7
Other metrics that are proven to measure social support are the Functional Social Support
Questionnaire and the Norbeck Social Support Questionnaire (Harley & Eskenazi, 2006;
Schaffer & Lia-Hoagberg, 1997).

9
& Eskenazi, 2006). Some studies focus on the effect of the number of social relationships on
women’s smoking decisions. For example, Meghea et al. (2010) finds that women with no one to
help with the baby have a higher probability of continued smoking during pregnancy than those
that do. However, other studies find that a high number of social relationships is not indicative of
support, claiming that a structural measure does not suffice and that we need to look at the
quality of social support, not the quantity of social support that a woman has (Berkmann, Glass,
Brissette, & Seeman, 2000).
For example, Harley and Eskenazi’s study (2006) finds that married women are less likely to
continue smoking during pregnancy. If women have a supportive husband and are less likely to
smoke, they have functional support (e.g., they have a high quality of social support), but if there
is an effect solely from being married, the women have structural support (e.g., the have a higher
quantity of social support than unmarried women). When the authors measured both marital
status and the husband’s social support, they found that a husband’s social support does not
influence the wife’s smoking rates but being married does, indicating that the husband provides
only structural support (Harley & Eskenazi, 2006). By separating out marriage and social support
the authors were able to extract quality of support from quantity of support and found that being
married provides a high quantity, not quality of support.
Studies that measure functional support specifically found a strong, negative relationship
between social support and smoking behaviors, meaning that women with high levels of social
support typically smoke less (Orr, Blazer, & Orr, 2011; Bradstreet, Higgins, Heil, Badger,
Skelly, Lynch, & Trayah, 2012; Harley & Eskenazi, 2006). However, all of these studies have
been conducted in the US, emphasizing the need for similar studies to be done in an Eastern

10
European context as social support may function differently here.
8
Orr’s study (2011) found that
women without functional support did not seriously consider quitting or attempting to quit.
Bradstreet’s study (2012) found that those with more non-smoking friends and family had lower
levels of social discounting (e.g., they decreased their generosity hyperbolically within a social
network) and Harley and Eskenazi’s study (2006) found that those with increasing functional
social support were less likely to smoke during pregnancy.
9
Thus, the more functional support
that pregnant smokers had, the more likely they were to consider and attempt quitting, have
lower social discounting (e.g., a greater likelihood of quitting), and actually quit.
10
Social Support Mechanisms
In addition to my primary question regarding social support’s affect on prenatal smoking, I
analyze the different mechanisms through which social support could affect smoking rates. It is
largely accepted that smoking is associated with high stress levels, so some studies claim that
social support is protective in regards to smoking behaviors by buffering the impact that life
stress has on a mother’s emotional well-being (Glazier et al., 2004; Schaffer et al., 1997; Meghea
et al., 2010; Goedhart, Van der Wall, Cuijpers, & Bonsel, 2008). These studies claim that when a
woman has high social support levels, she has lower stress levels, which, in turn, give her a
8
For example, Romanians typically report a significantly greater number of family members
they feel close to than other Europeans or Americans, so it is possible that they rely on social
support more for their mental health than those in other countries do (Coleman, Carare, Petrov,
Forbes, Saigal, Spreadbury, Yap, & Kendrick, 2011).
9
Social discounting is a significant predictor of smoking status, and is defined as the degree to
which one discounts generosity within a social network (Bradstreet et al., 2012). It was used to
assess the amount of hypothetical money a person is willing to give up to share with individuals
in their social network from the person emotionally closest to them to a mere acquaintance
(Bradstreet et al., 2012). The amount women are willing to forgo to share decreased
hyperbolically as a function of social distance and smokers had steeper social discounting
functions (e.g., less generosity) than quitters or nonsmokers (Bradstreet et al., 2012).
10
Unfortunately, both of these studies and my study are not able to isolate whether or not those
providing functional social support to women are smokers or not themselves.

11
lower likelihood of smoking during pregnancy. This stance is further supported by a study
involving alcohol in which social support increased women’s self-efficacy and self worth, which
then lowered stress levels and ultimately lowered alcohol-related hospital admission rates (Booth
et al., 1992).
11
However, other studies claim that social support directly influences health behaviors
regardless of stress levels (Harley & Eskenazi, 2006). This hypothesis has led to more mixed
results. While some studies have controlled for mental health and other measures of wellbeing
and find that social support still influences smoking directly, others find that social support only
decreases smoking indirectly through stress or even increases smoking depending on the type of
social support measured (Schaffer et al., 1997; Aaronson, 1989).
Thus, while there is consensus that social support influences smoking decisions during
pregnancy, there is not consensus as to the pathways through which social support functions.
These potential mechanisms are important to consider and I address them later in the paper.
Also, it is important to remember that pathways between social support and smoking may differ
by context and, as most of these studies have been done in the US, may not be transferrable to
Romania.
IV. Data, Descriptive Statistics, and Correlations
Data
The primary data source for this paper is Advancing Maternal and Child Health in Romania:
an integrated assessment of the determinants of pregnancy outcomes, also known as the MAIA
project. This study was conducted at Babes-Bolyai University’s Center for Health Policy and
11
Smoking and drinking are often considered to be complementary but, because smoking is
much more prevalent in Romania, I focus on this. Additionally, rates of alcohol consumption in
my data source are incredibly low, so I cannot analyze them further.

12
Public Health in Cluj, Romania in the department of Community and Behavioral Health from
2012-2014. The timing of this research is ideal because it comes at a time when women in
Romania are smoking at high rates but are beginning to see and respond to incentives to quit.
The MAIA questionnaire, given to pregnant women 18 years and older, has 1,395
observations at the individual level and contains smoking, demographic and socioeconomic,
pregnancy, and mental health indicators. The survey was conducted at five public hospitals in the
Cluj region of Romania, was completely voluntary, and contains self-reported data. For the
smoking variable, women were asked if they were a smoker six months prior to pregnancy.
12
If
the women answered ‘yes,’ they were asked if they were current or recent smokers. Options
included: ‘smoking as much as before,’ ‘smoking a reduced number of cigarettes,’ ‘quit after
learning about the current pregnancy,’ and ‘quit before learning about the current pregnancy.’
Women were then asked about their levels of SHS exposure on a daily basis and whether or not
they lived with a smoker, both of which allow me to assess how risk factors may differ between
direct and indirect smoke exposure.
Basic demographic and socioeconomic questions measure age, education level, ethnicity,
urban or rural residence, marital status, and income while pregnancy indicators measure the
wantedness of pregnancy and whether or not the woman has other children. I measure social
support by the Social Network Support Scale and stress levels by the Perceived Stress Scale
(PSS). The Social Network Support Scale targets functional social support, or, the ‘peer effect’
(Hoxby, 2000). This scale has six sub questions that address the number of people that women
can count on for specific facets of social support (measured as: zero, one or two, three or four,
12
This is an important indicator, because even if women did not smoke six months prior to
pregnancy, they may have smoked earlier in life and may still have health problems after giving
birth depending on how long they smoked for and they may have a high risk of relapsing after
pregnancy (Meghea et al., 2012; Flemming, Graham, Heirs, Fox, & Sowden, 2012).

13
and five or more people): advice and guidance, friendship, comfort and welfare, help and
accountability, recognition of talents and skills, and closeness and emotional stability.
13
While
the Social Network Support Scale is measured by conglomerating these six facets of social
support into a scale that ranges between zero and 18, I am able to break the scale down into its
six components, which I describe later.
Descriptive Statistics
In all descriptive statistics tables (Tables 1-4), the results for the entire MAIA survey are in
the left panel and results for the subsample I use are in the right panel.
14
Both panels are
displayed in order to see whether or not my sample is representative of the survey. My
subsample (N=1,119) includes all observations that contain data for every variable in my
regressions and draws only from the MAIA survey.
15
In regards to outcome variables (as shown
in Table 3), approximately 30 percent of women smoked six months prior to pregnancy, and 14
percent of all women smoked during pregnancy (45 percent of women smoking six months prior
to pregnancy continue to smoke). Of the 30 percent that smoked six months prior to pregnancy,
five percent continue to smoke the same amount, 41 percent smoke less than they did before, 45
13
While the Social Network Support Scale does measure the number of people that women can
count on, it measures the number that provide them with a high quality of support, thus
measuring functional rather than only structural support.
14
As I stated before, for both panels in Tables 1, 2, and 4 I show the descriptive statistics for the
Roma specifically. The Roma are a very marginalized minority in Romania and the way in which
their social support influences their smoking decisions may be different than how the majority
population’s social support does so. While there are not enough Roma observations in my
sample to conduct analyses on them (N=28), I insert some of their descriptive statistics as a point
of reference.
15
My subsample includes all observations with entries for smoking, social support, education
level, marital status, income, and SHS exposure. I do include descriptive statistics for age,
unwanted pregnancy, and having other children, but my subsample is not based on these
indicators because doing so restricts my sample size and there is not strong enough theory to
justify including them. Later, when I introduce stress into the model, my subsample decreases to
N=1,024.

14
percent quit after learning about their current pregnancy, and nine percent quit before learning
about their current pregnancy (e.g., with the six months before they were pregnant). In my
subsample the statistics for smoking are similar, with about 29 percent smoking six months
before pregnancy and 44 percent of prior smokers continuing to smoke during pregnancy. About
50 percent of women in the entire survey are not exposed to SHS daily (which is the same as in
the subsample) while the other 50 percent are. Of those that are exposed to SHS on a daily basis,
40 percent are exposed for less than half an hour a day and the remaining 60 percent are exposed
for over half an hour a day. Fifty-nine percent of women do not live with a smoker while the
remaining 41 percent live with at least one. This variable is similar in the subsample, with about
60 percent of women living with at least one smoker.
As shown in Table 1, the average age of women is 30 years old (Romanian women’s average
age at first birth is 26), 55 percent of women have had education beyond high school (this
seemingly high percentage of women with a post high school education is common for formerly
communist countries), 81 percent are ethnically Romanian, 34 percent are rural (the country is 47
percent rural), most have what is considered to be an average Romanian income (their income
convergence is common in formerly communist countries), and over 85 percent are married
(Central Intelligence Agency, 2011). As seen in Table 2, 73 percent of women wanted their
pregnancy at this time, and over 80 percent have at least one other child. All of these percentages
are for the entire survey and are comparable to those in the subsample.
Social support levels (my independent variables of interest) are split into their respective
sub-questions, with the majority of women in the survey and the subsample having either
medium or high levels of support. As a reminder, women listed how many people they could
count on for each of these facets of social support (zero, one or two, three or four, or five or

15
more) and those that answered either nobody or one to two were categorized as having low social
support, those that answered either three to four were categorizing as having medium social
support, and those that answered five or more were categorized as having high social support.
Shown in Table 4, about 35 percent of women have medium levels of advice and guidance while
another 43 percent have high levels, over 70 percent have medium-high levels of friendship,
about 30 percent have a medium sense of comfort and welfare while another 50 percent have a
high sense, about 37 percent have medium levels of help while 43 percent have high levels,
about 28 percent have medium recognition of their talents and skills and 53 percent have high
recognition, and about 35 percent have medium levels of closeness and emotional stability while
41 percent have high levels.
16
So, for all facets of social support, the majority of women said that
they had either medium or high levels of social support with only about 20 or 30 percent of
women claiming that they have low levels of social support.
Conditional Probabilities
Conditional probabilities between social support, smoking status, and various individual
characteristics (e.g., marital status and having other children) are presented in Tables 5 and 6.
Table 5 reports the probability that a woman is married, has other children, and is either a current
smoker, has reduced her smoking, or is a former smoker (also referred to as a prior smoker or a
quitter), given her levels of social support. For a switch from low to high social support there is a
seven percentage point increase in being married, a four percentage point increase in having
other kids, a four percentage point decrease in being a current smoker, a three percentage point
decrease in having reduced the number of cigarettes smoked per day, and a seven percentage
16
I cannot split friendship into medium and high levels due to data coding issues. All women
with three, four, or five or more people to rely on for friendship are coded as having only three or
four people to rely on.

16
point decrease in having been a prior smoker, all of which are similar to the correlations within
the subsample.
17
Table 6 reports the probability of a woman being married, having other children, and having
a high level of social support given her smoking decisions. Forty-eight percent of current
smokers have high social support, while 53 percent of those that have reduced smoking do, 54
percent of those that quit either in pregnancy or within six months of becoming pregnant do, and
61 percent of those that did not smoke six months before pregnancy do. So, as smoking
behaviors improve, social support increases. This relationship is similar in the subsample with 52
percent of smokers, 53 of reduced smokers, 55 percent of quitters, and 62 percent of nonsmokers
having high social support. There is no clear association between smoking decisions and having
other children. Sixty-one percent of current smokers are married, 64 percent of reduced smokers
are married, 83 percent of quitters are married, and 91 percent of nonsmokers are married, so
there is a strong association between reduced smoking and being married. Again, this
relationship is similar in the subsample.
Correlations
I then look at the correlation coefficients for the social support scales, which are reported in
Table 7. Most facets of the scale are highly correlated (between .40 and .60), as I would expect
them to be given that they are all elements of a woman’s social support. Comfort and guidance,
help and guidance, help and comfort, emotional support and comfort, and emotional support and
help are all very highly correlated, with coefficients of .60 or greater. Because the regressions
each target a different facet of social support, these coefficients are not too concerning, but they
17
Low social support is defined as women ranking between zero and 12 on the social support
scale and high social support is defined as women ranking between 13 and 18 on the social
support scale. As I described earlier, I break down this scale into facets of social support in my
main analysis, but, for the sake of simplicity, I use the scale to look at conditional probabilities.

17
may indicate that comfort and help (because they are very highly correlated with other elements)
may not be the strongest predictors of social support.
18
Regardless, as these variables are not
completely multicollinear, it is important to split up the social support scale into its components
because they may measure different facets of social support.
V. Empirical Model and Results
Modeling Social Support
My study uses probit regression in order to look at a binary dependent variable constrained
between zero and one (i.e., smoking within the six months leading up to pregnancy versus not
smoking six months before pregnancy) and then estimates the maximum likelihood that a woman
will switch categories from zero (e.g., not smoking) to one (e.g., smoking). I report the marginal
effect, which is the probability for an infinitesimal change in each continuous independent
variable and for a discrete change in the probability for dummy variables. I also use linear
probability models to see whether or not there is a difference between the two models, and find
similar results using both specifications.
19
For simplicity, I focus my discussion on the probit
results.
Not Smoking Six Months Before Pregnancy: Without Controls
First, I use a probit regression with the following model:
(1) Nonsmokeri = Φ (α + β1SocialSupporti + εi),
where Nonsmokeri indicates that the woman did not smoke six months prior to pregnancy. I first
run the regression without controls to see how social support is correlated to a woman’s decision
18
While I did run probit regressions with different combinations of the social support sub-
questions as my variables of interest, no results were particularly informative. So, for the sake of
simplicity, I do not describe them further.
19
Please see the Appendix for both the results from (Table 1) and discussion of the linear
probability model that measures how social support affects women’s decisions to smoke six
months before pregnancy.

18
to have smoked six months before pregnancy. I run the regression separately for the number of
people that provide the woman with guidance, friendship, comfort, help, recognize their talents
and skills, and provide the woman with emotional support. First, these categories are measured
on a scale as low, medium, and high with women that have zero, one, or two people to count on
for these facets of social support being low, women that have three to four people to count on for
these facets of social support being medium, and women that have five or more people to count
on for these facets of social support being high. Probit results are reported in Table 8 Panel A in
the row ‘No Controls.’
Women with medium levels of guidance (rather than low levels of guidance) are four
percentage points more likely to have not smoked six months before pregnancy, women with
medium-high friendship are six percentage points more likely to not have smoked, women with
medium comfort are four percentage points more likely, women with medium help are five
percentage points more likely, women with medium recognition of talent are six percentage
points more likely, and women with medium emotional support are five percentage points more
likely. Seventy percent of women did not smoke six months before pregnancy so women having
medium levels of social support increase their likelihood of not having smoked at least six
months before pregnancy from about 0.70 to approximately 0.75. Because I measure facets of
social support on a scale, the effect of moving from medium to high levels of social support is
double the effect of from moving from low to medium levels of social support. So, women with
high levels of guidance are then four percentage points more likely to have not smoked six
months before pregnancy than those with medium levels of social support and are then eight
percentage points more likely to have not smoked six months before pregnancy than those with
low levels of social support. This relationship holds true for other facets of social support as well.

19
Not Smoking Six Months Before Pregnancy: With Controls
While the baseline regression is a useful frame of reference, it is important to control for
individual characteristics that may influence a woman’s decision to smoke during pregnancy
other than the amount of social support that they have. The new regression is as follows:
(2) Nonsmokeri = Φ (α + β1SocialSupporti + X/β2 + εi),
where X includes individual characteristics such as education (bachelor’s degree or higher versus
high school or lower), urban residence status, marital status, income (making the Romanian Leu
equivalent of $465 or more versus making the equivalent of less than $465), and SHS exposure
(being exposed to SHS daily versus not being exposed daily). Finally, ε indicates the error term,
which captures unobserved characteristics. In regards to the controls, I expect that someone who
is more educated, living in a rural environment, married, wealthier, and not exposed to SHS daily
will smoke less than their counterparts. Ebrahim, Floyd, Merritt, Decoufle, and Holtzman (2000)
find that women that are more educated and are married are less likely to smoke both while
pregnant and while not pregnant in the US and Fukuda, Nakamura, and Takano (2004) find that
people that are wealthy are less likely to smoke in Japan.
Additionally, Irimie (2011) finds that in Romania more people in urban areas smoke and that
more smokers are exposed to SHS than nonsmokers. Studies also find that younger people are
more likely to smoke, women with unwanted pregnancies are more likely to continue smoking
during their pregnancy, women with other children are more likely to continue smoking during
pregnancy, and women in their first pregnancy are more likely to relapse if they smoked before
pregnancy (Fukuda et al., 2004; Hellerstedt, Pirie, Lando, Curry, McBride, Grothauy, & Nelson,
1998; Cnattingius, 2004; Connor & McIntyre, 1999).

20
Probit results from regression (2) are located in Table 8 Panel A in the second row. Here,
being recognized for talents and skills is statistically significantly associated with not having
smoked six months prior to pregnancy but all other sub questions of social support are not
significant. Specifically, women with three or four people that recognize them for their talents
and skills are three percentage points more likely to have not smoked six months prior to
pregnancy than are those with zero, one, or two people that recognize them for their talents and
skills. This means that women with five or more people that recognize them for their talents and
skills are then six percentage points more likely to have not smoked six months prior to
pregnancy than those with zero, one, or two people to recognize them for their talents and skills.
Not Smoking Six Months Before Pregnancy: Nonlinear Effects
There may be differential effects depending on whether women have three to four people to
count on for social support or five or more people to count on for social support. By previously
measuring social support on a scale (zero, one, two), I forced the effect of having high support to
be double that of medium support. I rerun the regressions with and without controls and allow
for medium and high levels of social support to have different effects. The coefficients for
medium and high levels of all facets of social support are now large and statistically significant.
However, because I am not able to separate out medium and high levels of social support for
friendship, the results for friendship are the same as those from above. First I run the regressions
without controls to establish initial relationships, which is shown in Table 8 Panel B in the first
row. For most facets of social support, having medium support has a stronger effect than having
high support, indicating that having a fifth person to rely on for support may not be important.
Additionally, this fifth person may even indicate that a woman has too many people providing
her with support (e.g., too high a quantity) so that quality of support decreases because the

21
coefficient from medium to high support actually decreases rather than stays the same for some
facets of social support.
Women with either medium or high guidance are 11 percentage points more likely to have
not smoked six months prior to pregnancy, and this relationship is similar for the different facets
of social support. Thus, there is no consensus as to whether medium or high social support is
more important. But, the coefficients on both are over twice as strong as the coefficients when I
combined medium and high levels of support in Panel A and they are different from one another,
so I separate medium and high levels of support for the remainder of my analysis.
When I introduce controls (e.g., educational status, urban or rural residence, marital status,
income, SHS exposure) into this regression, social support coefficients mostly remained
significant and all coefficients on medium levels of social support remained significant,
indicating that having three or four people to rely on for support may be the most accurate social
support measure. These results are in Table 8 Panel B in the second row. Having medium
guidance makes women nine percentage points more likely to have not smoked six months prior
to pregnancy while having high guidance makes women seven percentage points more likely,
medium comfort makes women eight percentage points more likely, medium help makes women
11 percentage points more likely, medium and high talent both make women nine percentage
points more likely, and medium emotional support makes women nine percentage points more
likely. The remaining indicators lose their statistical significance when I introduce controls.
Because friendship loses statistical significance, it may be that having many friends is only a
measure of structural support (quantity), not functional support (quality), which is why it is not
statistically significant. The number of friends a woman has may only get at the quantity of
social support a woman has but indicators such as the number of people providing women with

22
emotional support may get at the quality of support more directly. However, because I cannot
separate medium friendship from high friendship, I cannot be sure that this is why, as most facets
of social support were only significant once I separated medium and high effects.
As I said earlier, it is possible that high levels of social support may not be as important as
medium levels of social support, which holds true when I introduce control variables. This could
mean one of two things: (1) having a fifth friend to count on does not matter, or (2) having a fifth
friend proxies for a woman having a high quantity of support but not a high quality of support.
20
Not Smoking Six Months Before Pregnancy: Stratified Controls
The prior regressions control for many potential covariates, but it is possible that these
covariates have differential effects. For example, the relationship between social support and
smoking may operate differently for women with high education versus those with low
education. To assess whether or not these effects are different, I run each regression with
controls but stratify each regression by the level of the controls, the results of which are in Table
9. Table 9 Panel A shows high and low education, with women with high education typically
having a stronger association between social support and smoking rates than women with low
education. For women with high education, having medium levels of guidance makes them 13
percentage points more likely to have not smoked six months prior to pregnancy but for women
with low education the effect is not statistically significant. This means that women with high
education may benefit from having higher levels of social support but women with low education
may not. For most facets of social support, this relationship remained stronger for women with
20
To test for whether or not medium and high levels of social support are statically different
from one another, I run t-tests for all facets of social support, which are reported in the Appendix
in Table 2. I find that only medium and high levels of comfort are statistically significantly
different from one another, which may explain why many coefficients for medium and high
support are similar or the same.

23
high education, but for some, having either high or low education is statistically significant, and
for a couple (high help and high talent), only low education is statistically significant. Because I
earlier established that medium levels of social support are robust to more specifications, I focus
most on the relationship between controls and medium levels of support.
When I separate urban and rural areas (Table 9 Panel B), I find that for those living in an
urban area, the relationship between social support and smoking status is strong but for those
living in a rural area, the relationship is weaker. This may mean that for people living in an urban
area, social support is more protective against smoking than for people living in a rural area.
However, for those living in a rural area, measures of comfort, help, and talent are important.
When I separate women who are married from those who are not (Table 9 Panel C), I find
that social support is more statistically significant for smoking when women are married. When
women are not married, social support may not be as important. What is important here is that
social support remains statistically significant when I restrict my sample to married women. This
indicates that structural (quantity) support from being married exists as opposed to functional
(quality) support, which I spoke about earlier.
For most measures of social support, having low income makes social support buffer against
smoking but having high income does not (Table 9 Panel D). For example, for women with low
income, having medium help makes them 22 percentage points more likely to not have smoked
six months prior to pregnancy but for women with high income there is no statistically
significant effect. This may mean that for people with low income, having social support is more
helpful than for women with high income.
Finally, I separate out women that are exposed to SHS and those that are not (Table 9 Panel
E). For women that are exposed to SHS daily, having medium guidance results in a 17

24
percentage point increase in the likelihood that the woman did not smoke six months prior to
pregnancy. For women that were not exposed to SHS regularly, having medium comfort results
in a 14 percentage point increase in the likelihood that the woman did not smoke six months
prior to pregnancy. So, the relationship between social support and smoking does change
depending on the type of social support studied.
Social Support, Smoking, & Stress
Because there is a debate as to the mechanisms that social support functions through, I
introduce stress into the probit model. By including stress, I am able to test whether social
support directly influences women’s smoking decisions or whether it only indirectly influences
women’s smoking decisions by buffering women’s stress levels. If the variable for social support
remains statistically significant, then there is a direct effect. If the variable for social support
loses statistical significance and the variable for stress level is significant, then there is only an
indirect effect. I split stress into high and low stress by using the mean response level for the
PSS, as there is no established threshold by which to split the PSS.
21
By splitting the PSS at the
mean, the actual categories for high and low stress are uninteresting but they are still useful for
my analysis. The equation used here is the following and the results are reported in Table 10:
(3) Nonsmokeri = Φ (α + β1SocialSupporti + β2StressLeveli + X/β3 + εi)
When controlling for stress but no other covariates, women with medium levels of guidance
are 12 percentage points more likely to have not smoked six months prior to pregnancy than
those with low guidance. Additionally, those with high stress are six percentage points more
likely to have smoked six months prior to pregnancy than those with low stress. These results
21
The PSS ranges from zero to 40 and the mean response for this study is 14.35 and the median
response is 14. Because the mean and median are so similar, I split the PSS at the mean level
with low stress being zero to 14 and high stress being 15 to 40.

25
may indicate that both social support and stress directly influence women’s smoking decisions.
However, when I introduce covariates into the model (e.g., education, residence, marital status,
income, and SHS exposure), the statistical significance on stress goes away but the statistical
significance on social support stays the same. This indicates that the effect of stress on smoking
is due to these other covariates and not to stress itself. Most importantly, because the social
support indicator (e.g., guidance) remains statistically significant and the strength of the
coefficient remains high, I conclude that social support directly effects women’s smoking
decisions. I see the same relationship for other facets of social suppport, with stress losing its
significance when adding controls but social support remaining significant.
Quitting Smoking Either During Pregnancy or Within Six Months of Pregnancy
Next, I assess the effects of social support on a woman’s decision to quit smoking within six
months of becoming pregnant or during her pregnancy, conditional on her having smoked six
months before pregnancy. I estimate equation (2) again, but this time I use Quitter as my
dependent variable, making it equation (4):
(4) Quitteri = Φ (α + β1SocialSupporti + X/β2 + εi),
where Quitter indicates that a woman quit smoking either during pregnancy or within the six
month time period before becoming pregnant. This allows me to assess whether social support
predicts whether or not a woman smoking six months prior to pregnancy quits before or during
pregnancy. Here my sample size is much smaller, only capturing women that are either currently
smoking or have quit recently (N=326).
The results from these regressions are in Table 11. First, I run the regression with linear
effects and find that once introducing controls, no facets of social support are statistically
significant. Then, I run the regressions with nonlinear effects to see if a different relationship

26
exists when stratifying medium and high levels of social support. The relationship is much
weaker here than when assessing the likelihood of not smoking six months before pregnancy.
Here, women that have medium-high friendship, medium or high talent, or high emotional
support are more likely to quit smoking than to continue smoking but there are no statistically
significant results for other facets of social support.
When I control for education, living in an urban or rural area, marital status, income, and
exposure to SHS, the significance for all facets of social support go away. However, the strength
of the coefficients for some facets of social support is still fairly high. For example, those with
medium levels of recognition of talent and skills are 10 percentage points more likely to quit than
those with low levels. Medium and high levels of friendship and emotional support have similar
results, but medium and high amounts of guidance, comfort, and help are not strongly associated
with quitting smoking. Because there is no statistical significance here and the strength of the
indicators varies greatly, I conclude that social support does contribute to women’s decision to
not smoke six months before pregnancy but does not contribute to women’s decision to quit
during pregnancy or within six months of getting pregnant rather than continuing to smoke when
controlling for education, residence, marital status, income, and exposure to SHS.
VI. Concerns
In the results above, I find that social support most influences women’s decisions to smoke
six months before pregnancy. However, one should be careful when interpreting these
coefficients as causal effects of social support. Reverse causality is a large issue, as smoking
rates may determine the amount and type of social support a woman has rather than the other

27
way around.
22
I also worry about omitted variables, especially those related to women’s
personalities, as these variables likely impact both the features of women’s social networks and
how she responds to health information about smoking.
There are large issues with self-reporting, as women’s responses about their smoking status
are likely biased downward and women’s responses about the number of people they can rely on
for social support are likely biased upwards. Additionally, the women were interviewed at
different stages in their pregnancy, which may affect women’s answers regarding smoking.
However, prior studies have found that a variable for the week of pregnancy at interview is not
statistically significant (Meghea et al., 2012).
Because the data only captures women who are 18 years and older, it misses many teen
pregnancies, which make up approximately three percent of live births in Romania (The World
Bank Group, 2014B). These teen pregnancies are disproportionately Roma, so discarding them
makes the sample less representative of all people in Romania but likely does not impact the
relationship between social support and smoking for ethnically Romanian women. Abortion rates
are incredibly high in Romania, as abortion was outlawed during communism when Ceaușescu
advocated for a pronatalist state and is now used at a very high rate (David & Baban, 1996). The
year after communism fell there were over 3,000 abortions for every 1,000 live births and there
are still almost 480 for every 1,000 live births (Hord et al., 1991; UNFPA, 2013). This means
that the MAIA data does not capture all pregnancies because many unwanted pregnancies are
terminated.
22
Splitting the social support scale into its respective sub-questions may help to correct for this.
For example, the question targeting the number of people the woman can rely on for guidance
isolates the ‘wise peer effect’; there is likely less of an impact of smoking on the number of wise
people in your life as opposed to the number of friends in your life. However, I cannot be certain
that splitting up the scale corrects for endogeneity, so it is still a large concern.

28
VII. Conclusions and Implications
When assessing if and how social support effects women’s smoking decisions during
pregnancy in Romania, I find that social support levels are significant contributors to these
decisions. Social support effects whether or not a woman has smoked six months before
pregnancy more so than whether or not a woman has quit within those six months or during her
pregnancy. By introducing stress into the model, I am able to conclude that social support
directly impacts women’s smoking behaviors and does not only impact smoking through
lowering women’s stress levels.
I also find that social support effects these women’s decisions differentially by the type of
social support looked at. For example, the number of people that a woman says recognize her
talents and skills has a different effect than the number of people that a woman says provide her
with help and accountability. It is thus important to stratify the types of social support looked at
rather than creating an index. Finally, once a woman can count on three or four people for these
types of social support, the fifth person may not be of benefit, as women having medium levels
of support (three or four people) often having a stronger effect than women having high levels of
support (five or more people).
While I cannot draw causal conclusions here, I do find that social support is an important
mechanism through which women’s continued smoking during pregnancy can be mitigated.
While smoking cessation programs are likely beneficial, most Romanian women that report
quitting smoking say they have done so without professional guidance (Irimie, 2011). So, by
increasing the focus put on social support, women’s prenatal smoking rates in Romania could
significantly fall.

29
Tables and Figures

30

31

32

33
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
N = 1,119
23
‘Linear Effects’ indicate that the facets of social support are measured on a scale in one variable (e.g., zero is low social support,
one is medium social support, and two is high social support).
24
‘Non-Linear Effects’ indicate that the facets of social support are split between low, medium, and high using two variables. Those in
the low category (which is the omitted variable) rely on zero, one, or two people for social support, those in the medium category rely
on three or four people, and those in the high category rely on five people.
Table 8: Social Support and the Probability of Not Smoking Six Months Before Pregnancy
Guidance
Friendship
Comfort
Help
Talent
Emotional
Panel A: Linear Effects23
No
Controls
.04**
(.02)
.06*
(.03)
.04**
(.02)
.05***
(.02)
.06***
(.02)
.05***
(.02)
Controls
.02
(.02)
.03
(.03)
.01
(.02)
.01
(.02)
.03*
(.02)
.02
(.02)
Panel B: Non-Linear Effects24
Med
High
Med/High
Friendship
Med
High
Med
High
Med
High
Med
High
Guidance
Comfort
Help
Talent
Emotional
No
Controls
.11***
(.04)
.11***
(.04)
.06*
(.03)
.12***
(.04)
.11***
(.04)
.15***
(.03)
.12***
(.03)
.13***
(.04)
.14***
(.04)
.15***
(.03)
.11***
(.03)
Controls
.09**
(.04)
.07*
(.04)
.03
(.03)
.08**
(.04)
.04
(.04)
.11***
(.04)
.05
(.04)
.09**
(.04)
.09**
(.04)
.09***
(.04)
.05
(.03)
Table 7: Correlations for Social Support Scales
Guidance
Friendship
Comfort
Help
Talent
Emotional
Guidance
1.00
Friendship
.50
1.00
Comfort
.60
.53
1.00
Help
.60
.49
.63
1.00
Talent
.51
.42
.55
.55
1.00
Emotional
.57
.47
.64
.64
.58
1.00

34
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
N = 1,119
Table 9: Social Support and the Probability of Not Smoking Six Months Before Pregnancy with Stratified Controls
Med
High
Med/High
Friendship
Med
High
Med
High
Med
High
Med
High
Guidance
Comfort
Help
Talent
Emotional
Panel A: Education
High Education
.13***
(.05)
.09*
(.05)
.00
(.04)
.16***
(.05)
.10*
(.06)
.14***
(.05)
.07
(.05)
.11**
(.05)
.07
(.05)
.13***
(.04)
.08*
(.04)
Low Education
.05
(.06)
.08
(.05)
.07
(.05)
.04
(.06)
.09
(.05)
.09*
(.05)
.13**
(.05)
.10*
(.05)
.16***
(.05)
.14***
(.05)
.11**
(.05)
Panel B: Residence
Urban
.12***
(.04)
.10**
(.04)
.04
(.04)
.16***
(.04)
.12**
(.05)
.14***
(.04)
.09**
(.04)
.10**
(.05)
.10**
(.05)
.19***
(.04)
.12***
(.04)
Rural
.09
(.06)
.10
(.06)
.08
(.05)
.04
(.07)
.11*
(.06)
.14**
(.06)
.16***
(.06)
.15**
(.06)
.18***
(.06)
.06
(.06)
.08
(.06)
Panel C: Marital Status
Married
.06
(.04)
.05
(.04)
.02
(.03)
.08**
(.04)
.07*
(.04)
.11***
(.04)
.07*
(.04)
.09**
(.04)
.11***
(.04)
.11***
(.03)
.08**
(.03)
Not Married
.22**
(.10)
.16
(.10)
.17**
(.08)
.14
(.11)
.15
(.10)
.21**
(.10)
.16
(.10)
.13
(.11)
.11
(.10)
.24**
(.10)
.04
(.10)
Panel D: Income
High Income
.09*
(.05)
.05
(.05)
.02
(.04)
.11**
(.05)
.07
(.05)
.06
(.05)
.04
(.05)
.11**
(.04)
.10**
(.05)
.11**
(.04)
.06
(.04)
Low Income
.12**
(.06)
.17***
(.05)
.12**
(.05)
.10*
(.06)
.17***
(.05)
.22***
(.05)
.20***
(.05)
.13**
(.06)
.19***
(.05)
.20***
(.05)
.18*
(.05)
Panel E: Secondhand Smoke Exposure
Exposed to SHS
.17***
(.06)
.11*
(.06)
.07
(.05)
.05
(.06)
-.01
(.06)
.20**
(.05)
.06
(.05)
.18***
(.06)
.15***
(.05)
.10*
(.05)
.06
(.05)
Not Exposed to
SHS
.06
(.04)
.08**
(.04)
.03
(.04)
.14***
(.04)
.17***
(.05)
.13***
(.04)
.11***
(.04)
.06
(.04)
.08**
(.05)
.13***
(.04)
.11***
(.04)

35
Table 10: Social Support and the Probability of Not Smoking Six Months Before Pregnancy with Stress Controls
Med
High
Med/High
Friendship
Med
High
Med
High
Med
High
Med
High
Guidance
Comfort
Help
Talent
Emotional
No
Controls
.12***
(.04)
.10***
(.04)
.06*
(.03)
.13***
(.04)
.10***
(.04)
.14***
(.04)
.11***
(.03)
.14***
(.03)
.12***
(.04)
.16***
(.03)
.10***
(.03)
Controls
.10*
(.04)
.06*
(.04)
.03
(.03)
.08***
(.04)
.03
(.04)
.10***
(.04)
.04
(.04)
.11**
(.04)
.08*
(.04)
.09***
(.04)
.04
(.04)
Stress
No
Controls
-.06**
(.03)
-.07**
(.03)
-.06**
(.03)
-.06**
(.03)
-.05*
(.03)
-.06**
(.03)
Controls
-.04
(.03)
-.04
(.03)
-.04
(.03)
-.04
(.03)
-.03
(.03
-.03
(.03)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
N = 1,024

36
Table 11: Social Support and the Probability of Quitting Smoking During Pregnancy or Six Months Before Pregnancy
Guidance
Friendship
Comfort
Help
Talent
Emotional
Panel A: Linear Effects
No
Controls
.00
(.03)
.16***
(.06)
.05
(.03)
.06*
(.03)
.05
(.03)
.06
(.03)
Controls
-.04
(.04)
.10
(.07)
.01
(.04)
.00
(.04)
-.00
(.04)
.03
(.04)
Panel B: Nonlinear Effects
Med
High
Med/High
Friendship
Med
High
Med
High
Med
High
Med
High
Guidance
Comfort
Help
Talent
Emotional
No
Controls
-.00
(.07)
.01
(.07)
.16***
(.06)
.02
(.08)
.09
(.07)
.00
(.07)
.11
(.07)
.12*
(.07)
.11*
(.07)
.10
(.07)
.11*
(.06)
Controls
-.09
(.09)
-.09
(.08)
.10
(.07)
-.01
(.09)
.02
(.08)
-.06
(.08)
-.00
(.08)
.07
(.08)
.01
(.08)
.06
(.08)
.06
(.07)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
N = 326

37
Appendix
I. Linear Probability Model
I now run regression (2) using a linear probability model, which then becomes regression (5)
due to the new model specification:
(5) Nonsmokeri = α + β1SocialSupporti + X/β2 + εi
Because the dependent variable is between zero and one, a probit model should be a better fit,
but running linear probability allows me to clarify whether or not the probit model is sufficient.
Here I use equation (5) and report my findings in Table 1.
I find that using a linear probability model produces similar results to using a probit model,
indicating that the probit model is sufficient. In the linear probability model, when controls are
included, women with medium guidance are nine percentage points more likely to have not
smoked six months prior to pregnancy versus those with low guidance. Without controls, the R-
squared values range from between 0.4 percent and 1.9 percent, and with controls the R-squared
values range from between 12 percent and 13 percent. This means that with a linear probability
model, controlling for a woman’s level of social support, educational attainment, marital status,
residence, income, and exposure to SHS explains about 13 percent of the variation in her
smoking decisions. In the probit model the results are exactly the same, as women with medium
guidance are nine percentage points more likely to have not smoked as well. This holds true for
other facets of social support, thus indicating that the probit model is sufficient.

38
Table 1: Social Support and the Probability of Not Smoking Six Months Before Pregnancy Pregnancy Using A Linear
Probability Model
Med
High
Med/High
Friendship
Med
High
Med
High
Med
High
Med
High
Guidance
Comfort
Help
Talent
Emotional
No Control
.12***
(.04)
.11***
(.04)
.06*
(.03)
.14***
(.04)
.12***
(.04)
.16***
(.04)
.13***
(.03)
.15***
(.04)
.15***
(.04)
.16***
(.04)
.13***
(.03)
Controls
.09**
(.04)
.07*
(.04)
.03
(.03)
.07**
(.04)
.04
(.04)
.11***
(.04)
.05
(.04)
.10**
(.04)
.09**
(.04)
.09***
(.04)
.06
(.03)
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
N = 1,119
II. T-Tests
Table 2: T-Tests
F
Prob > F
Medium and High Guidance
1.28
0.26
Medium and High Friendship
N/A
N/A
Medium and High Comfort
1.48
2.22
Medium and High Help
3.99
0.05
Medium and High Talent and Skills
0.10
0.75
Medium and High Emotional Support
1.73
0.19

39
Bibliography
1. Aaronson, L. S. (1989). Perceived and received support: Effects on health behavior during
pregnancy. Nursing Research, 38(1), 4-9.
2. Been, J. V., Nurmatov, U. B., Cox, B., Nawrot, T. S., van Schayck, C. P., & Sheikh, A.
(2014). Effect of smoke-free legislation on perinatal and child health: a systematic review
and meta-analysis. The Lancet.
3. Berkman, L. F., Glass, T., Brissette, I., & Seeman, T. E. (2000). From social integration to
health: Durkheim in the new millennium. Social science & medicine, 51(6), 843-857.
4. Bradstreet, M. P., Higgins, S. T., Heil, S. H., Badger, G. J., Skelly, J. M., Lynch, M. E., &
Trayah, M. C. (2012). Social discounting and cigarette smoking during pregnancy. Journal of
behavioral decision making, 25(5), 502-511.
5. Centers for Disease Control and Prevention. (2013). “Trends in Current Cigarette Smoking
Among High School Students and Adults, United States, 1965–2011.” CDC. Web.
6. Centers for Disease Control and Prevention. (2010). “Trends in Smoking Before, During, and
After Pregnancy—Pregnancy Risk Assessment Monitoring System, United States, 40 Sites,
2000–2010.” Morbidity and Mortality Weekly Report 2013.
7. Central Intelligence Agency (2011). The World Factbook.
8. Cnattingius, S. (2004). The epidemiology of smoking during pregnancy: smoking prevalence,
maternal characteristics, and pregnancy outcomes. Nicotine & Tobacco Research, 6(Suppl 2),
S125-S140.
9. Cohen, S., Mermelstein, R., Kamarck, T., & Hoberman, H. M. (1985). Measuring the
functional components of social support. In Social support: Theory, research and applications
(pp. 73-94). Springer Netherlands.
10. Cohen, S., & Wills, T. A. (1985). Stress, social support, and the buffering hypothesis.
Psychological bulletin, 98(2), 310.
11. Connor, S., McIntyre, L. (1999). The sociodemographic predictors of smoking cessation
among pregnant women in Canada. Canadian Journal of Public Health. 90(5). 352-355.
12. David, H. P., & Baban, A. (1996). Women's health and reproductive rights: Romanian
experience. Patient Education and Counseling, 28(3), 235-245.
13. David, A., Esson, K., Perucic, A. M., & Fitzpatrick, C. (2010). Tobacco use: equity and
social determinants. Equity, social determinants and public health programmes.
14. Ebrahim, S. H., Floyd, R. L., Merritt II, R. K., Decoufle, P., & Holtzman, D. (2000). Trends
in pregnancy-related smoking rates in the United States, 1987-1996. Jama, 283(3), 361-366.
15. Elsenbruch, S., Benson, S., Rucke, M., Rose, M., Dudenhausen, J., Pincus-Knackstedt, M.,
Klapp, B., Arck, P. (2006). Social support during pregnancy: effects on maternal depressive
symptoms, smoking, and pregnancy outcome. Human Reproduction. 22(3). 869-877.
16. Flemming, K., Graham, H., Heirs, M., Fox, D., Sowden, A. (2012). Smoking in pregnancy: a
systematic review of qualitative research of women who commence pregnancy as smokers.
Journal of Advanced Nursing, 69(5), 1023-1036.
17. Fukuda, Y., Nakamura, K., & Takano, T. (2005). Socioeconomic pattern of smoking in
Japan: income inequality and gender and age differences. Annals of epidemiology, 15(5),
365-372.
18. Gilman, S. E., Breslau, J., Subramanian, S. V., Hitsman, B., & Koenen, K. C. (2008). Social
factors, psychopathology, and maternal smoking during pregnancy. American journal of
public health, 98(3), 448.

40
19. Glazier, R. H., Elgar, F. J., Goel, V., & Holzapfel, S. (2004). Stress, social support, and
emotional distress in a community sample of pregnant women. Journal of Psychosomatic
Obstetrics & Gynecology, 25(3-4), 247-255.
20. Goedhart, G., Van der Wal, M., Cuijpers, P., Bonsel, G. (2009). Psychosocial problems and
continued smoking during pregnancy. Elsevier. 34, 403-406.
21. Harley, K., & Eskenazi, B. (2006). Time in the United States, social support and health
behaviors during pregnancy among women of Mexican descent. Social science & medicine,
62(12), 3048-3061.
22. Hellerstedt, W. L., Pirie, P. L., Lando, H. A., Curry, S. J., McBride, C. M., Grothaus, L. C.,
& Nelson, J. C. (1998). Differences in preconceptional and prenatal behaviors in women with
intended and unintended pregnancies. American Journal of Public Health, 88(4), 663-666.
23. Holtrop, J. S., Meghea, C., Raffo, J. E., Biery, L., Chartkoff, S. B., & Roman, L. (2010).
Smoking among pregnant women with Medicaid insurance: are mental health factors
related?. Maternal and child health journal, 14(6), 971-977.
24. Hord, C., David, H. P., Donnay, F., & Wolf, M. (1991). Reproductive health in Romania:
reversing the Ceausescu legacy. Studies in family planning, 231-240.
25. Hoxby, C. (2000). Peer effects in the classroom: Learning from gender and race variation
(No. w7867). National Bureau of Economic Research.
26. Irimie, S. (2011). Global Adult Tobacco Survey. Ministry of Health Romania. Bucharest,
27. Romania.
28. Jackson, C., & Henriksen, L. (1997). Do as I say: parent smoking, antismoking socialization,
and smoking onset among children. Addictive behaviors, 22(1), 107-114.
29. Kleinman, J. C., Piere, M. B., Madans, J. H., Land, G. H., & Schramm, W. F. (1988). The
effects of maternal smoking on fetal and infant mortality. American Journal of
Epidemiology, 127(2), 274-282.
30. Krstev, S., Marinkovic, J., Simic, S., Kocev, N., Bondy, S. (2012). Prevalence and predictors
of smoking and quitting during pregnancy in Serbia: results of a nationally representative
study. International Journal of Public Health. 57, 875-883.
31. Lopez, A. D., Collishaw, N. E., & Piha, T. (1994). A descriptive model of the cigarette
epidemic in developed countries. Tobacco control, 3(3), 242.
32. Lumley, J., Chamberlain, C., Dowswell, T., Oliver, S., Oakley, L., & Watson, L. (2009).
Interventions for promoting smoking cessation during pregnancy. Cochrane Database Syst
Rev, 3(3).
33. Meghea, C. I., Rus, D., & Dirle, I. A. (2010). Characteristics and health behaviors of
pregnant women in Romania. GINECO RO, 6(3), 166-171.
34. Meghea, C. I., Rus, D., Rus, I. A., Holtrop, J. S., & Roman, L. (2012). Smoking during
pregnancy and associated risk factors in a sample of Romanian women. The European
Journal of Public Health, 22(2), 229-233.
35. Michell, M. P. L. (2000). Smoke rings: social network analysis of friendship groups,
smoking and drug-taking. Drugs: Education, Prevention, and Policy, 7(1), 21-37.
36. Orr, S. T., Blazer, D. G., & Orr, C. A. (2012). Maternal prenatal depressive symptoms,
nicotine addiction, and smoking-related knowledge, attitudes, beliefs, and behaviors.
Maternal and child health journal, 16(5), 973-978.
37. Rantakallio, P. (1978). Relationship of maternal smoking to morbidity and mortality of the
child up to the age of five. Acta Paediatrica, 67(5), 621-631.

41
38. Schaffer, M. A., & Lia‐Hoagberg, B. (1997). Effects of Social Support on Prenatal Care and
Health Behaviors of Low‐Income Women. Journal of Obstetric, Gynecologic, & Neonatal
Nursing, 26(4), 433-440.
39. The World Bank Group (2014A). Adolescent Fertility Rate. Data. Web
40. The World Bank Group (2014B). "Health Nutrition and Population Statistics." World
DataBank. Web.
41. Thomas, S., Fayter, D., Misso, K., Ogilvie, D., Petticrew, M., Sowden, A., ... & Worthy, G.
(2008). Population tobacco control interventions and their effects on social inequalities in
smoking: systematic review. Tobacco Control, 17(4), 230-237.
42. Unicef. (2014) "Romania: Country Profile." Unicef. Web.
43. United Nations Population Fund (2013). Adolescent Pregnancy in Eastern Europe and
Central Asia. Eastern Europe and Central Asia Regional Office.
44. Waldron, I. (1991). Patterns and causes of gender differences in smoking. Social science &
medicine, 32(9), 989-1005.
45. Webb, Douglas. (2013). Tobacco Control for Health and Development. [Issue brief]. HIV,
Health, and Development, United Nations Development Program, New York, NY.