7/25/2019 33_HPU272_Friedman.pdf
1/22
Meharry Medical College Journal o Health Care or the Poor and Underserved 27 (2016): 869890.
Te Intergenerational ransfer of EducationCredentials and Health: Evidence from the 2008
General Social Survey-National Death Index
Esther M. Friedman, PhD
Peter Muennig, MD, MPH
Abstract:Background. Te educational attainment o a parent is a powerul predictor ochildrens wellbeing, but little is known about why this is the case. Methods. We used the2008 General Social Survey to explore actors that may explain the relationship between ones
athers education and ones own mortality. Tese include (1) intellectual traits, (2) materialwellbeing, (3) psychological characteristics, (4) personality characteristics, and (5) socialcapital. Results. Te education credentials o ones ather are signicantly associated withones risk o death. Te strongest mediators are own educational attainment, amily income,home ownership, and subjective socioeconomic status. o a lesser extent, respondentshappiness with riends and work and social bonding were also pathways. Conclusions.A athers educational attainment appears to inuence his childrens health, and may doso not only by improving the childs material circumstances but also through his or hereducational attainment and other psychological and social characteristics.
Key words:Intergenerational; socioeconomic status; survival analysis; General Social Survey;National Death Index.
he Servicemens Readjustment Act o 1944 ollowing World War II greatly increasedthe number o Americans with higher education.1Such improvements in educa-tional attainment are strongly linked to a variety o positive lie outcomes, including
a longer, healthier lie,26 greater job satisaction, and better work conditions.7Tere
is also a direct effect o education on cognitive traits and non-cognitive traits (e.g.,
learning to navigate social relationships with peers).8
Te benets o schooling are not only related to the wellbeing o individuals but also
have implications across generations. Tese benetsthe material as well as the cogni-
tive and socialmay all be transerred to children. It is well established, or instance,
that more educated parents have more educated children.9 In addition, children o
parents with high educational attainment tend to be exposed to more vocabulary, more
ORIGINAL PAPER
ESTHER M. FRIEDMANis a Behavioral and Social Scientist at RAND and Professor at the PardeeRAND Graduate School. PETER MUENNIGis with the Mailman School of Public Health at ColumbiaUniversity, New York. Authors listed in alphabetical order, but contributed equally to the paper. Teauthors would like to acknowledge Jaeseung Kim for invaluable help with variable construction andmodel development in the early stages of this papers development. Please address correspondence to:Esther M. Friedman, RAND, 1776 Main St, Santa Monica, CA 90401 or email: friedman@rand .org.
33_HPU272_Friedman indd 86933_HPU272_Friedman.indd 869 4/11/16 2:27 PM4/11/16 2:27 PM
7/25/2019 33_HPU272_Friedman.pdf
2/22
870 Intergenerational education and mortality
advanced math skills, and receive greater positive affi rmation than children o parents
with less educational attainment.10,11 Tis may explain why the inormation, values,
ideas, and belies that students acquire while in school produce intergenerational health
benets.1216
Tat children acquire a variety o cognitive and non-cognitive traits romwell-schooled parents raises the intriguing idea that a short-term social policy such the
Servicemens Readjustment Act might permanently and undamentally alter the collec-
tive ideas, belies, values, and cognitive traits o society as a whole or generations. Tis
most clearly occurs by educating less advantaged people who would otherwise have
not had the opportunity to achieve these levels o schooling and who transmit these
changes to the next generation in their amilies. Such a policy could have signicant
implications or the health o at least two generations. It is also plausible that such a
policy could weave knowledge and norms into the collective social abric or even
more generations to come.
In this paper, we use a unique dataset rst to establish that the health benets o
higher educational attainment are passed on to children, and second to examine how
this might occur. With respect to the ormer question, there is already a body o worksuggesting that parental education is related to offsprings health and mortality.46Much
less is known about the latter question, however. Possibly, parents who attended college
have acquired traits that later inuenced their childrens own educational attainment,
income, home ownership, subjective perceptions, and social capitaltraits that have all
been linked to health and longevity.1720Such traits might include knowledge, personality
characteristics, IQ, or even social/environmental characteristics such as perceived trust
in others (which is one oundation upon which social capital is built).21,22We explore
the intergenerational transer o a number o characteristics and resources, including:
(a) intellectual traits (i.e., verbal IQ), (b) material wellbeing (represented here by home
ownership), (c) psychological characteristics (e.g., happiness), (d) personality charac-
teristics (e.g., optimism), and (e) social capital (e.g., ties with amily and riends). We
also consider the mediating role o the respondents own education and income and
sel-rated health.
Methods
Data. Our analysis was perormed using the 2008 General Social Survey-National
Death Index (2008 GSS-NDI) dataset, which links the 19782002 waves o the GSS to
NDI data through 2008.23Te 2008 GSS-NDI provides three decades o data that can
be weighted to be representative o the U.S. non-institutionalized civilian population.
It includes a total o 32,830 people, o whom 9,271 were deceased as o 2008. Te
mortality linkage has been extensively tested and validated.23,24
For this analysis, we excluded the Black and Hispanic oversamples in the 1982 and1987 waves (or a total n=32,173); removed subjects missing data on education, age,
gender, race, and survey year (n=120); removed subjects with missing inormation on
income (n=3,144). We also excluded subjects who do not report inormation on their
athers educational attainments. Tis includes subjects who report that this is not ap-
plicable (n=4,423), those who report that they do not know their athers educational
attainment (n=1,674), and those with no response (n=173). Tis lef us with 22,759
33_HPU272_Friedman indd 87033_HPU272_Friedman.indd 870 4/11/16 2:27 PM4/11/16 2:27 PM
7/25/2019 33_HPU272_Friedman.pdf
3/22
871Friedman and Muennig
individuals ollowed over 361,807 person-years, and capturing 5,990 deaths over
this time period. Te sample sizes change a good deal in each analysis because some
explanatory variables were not obtained in particular waves o the GSS. However these
omissions were systematic and should not affect the representativeness o our sample.All analyses are weighted using the GSS-provided sampling weights.
Measures. Our principal outcome o interest was mortality hazards. Our primary
independent variable is the athers highest degree. Tis variable is grouped into ve
categories, less than high school, high school diploma, junior college degree, bachelors
degree, graduate degree, respectively. Because the number o observations or the junior
college group is extremely small, we combined junior college and high school into one
single group or the analyses.
Te decade o interview included two categories: whether individuals were surveyed
in 19801990 or in 19902008. We also broke down the second group urther, but
results did not differ or the two recent decades. We controlled or age, race, and gender
to capture xed socio-demographic characteristics (as conounders). We chose not to
control or actors that plausibly link the mechanism o interest, such as smoking ordrinking, in the happiness-mortality hazards pathway.
We dened non-xed characteristics as potential explanatory variables. Specically,
we explored ve domains o items captured in the GSS:
(1) Verbal IQ. IQ was measured in the GSS through a 10-item vocabulary test based
on the Gallup-Torndike verbal intelligence test, with a total score ranging rom 010
words correct. Larger numbers indicate more words were correct.
(2) Socioeconomic Status(SES). We included a number o measures o SES: whether
respondent owned his or her house or rented it; income in quintiles; and respondents
own educational attainments (based on degree attainment inormation in the GSS).
(3) Relative SES.Respondents relative SES was calculated in terms o responses to
three items rom the GSS: relative income, measured in response to a question com-
pared with American amilies in general, would you say your amily income is ar below
average, below average, average, above average, or ar above average? (with response
categories ranging rom ar above to ar below average); respondents satisaction with
his/her nancial situation based on a response to We are interested in how people are
getting along nancially these days. So ar as you and your amily are concerned, would
you say that you are pretty well satised with your present nancial situation, more or
less satised, or not satised at all? with response categories o pretty well satised,
more or less satised, not at all satised ; and subjective class identication identied
in response to a question asking about I you were asked to use one o our names
or your social class, which would you say you belong in: the lower class, the working
class, the middle class, or the upper class? with response categories ranging rom lower
to upper class. Items were averaged together and the overall index ranged rom 1 to 4.(4) Psychological wellbeing.Psychological wellbeing included measures o sel-reported
overall happiness (I you were to consider your lie in general, how happy or unhappy
would you say you are, on the whole?); reported marital happiness (aking things
all together, how would you describe your marriage? Would you say that your mar-
riage is very happy, pretty happy, or not too happy?; and satisaction with ones job
(On the whole, how satised are you with the work you dowould you say you are
33_HPU272_Friedman indd 87133_HPU272_Friedman.indd 871 4/11/16 2:27 PM4/11/16 2:27 PM
7/25/2019 33_HPU272_Friedman.pdf
4/22
872 Intergenerational education and mortality
very satised, moderately satised, a little dissatised, or very dissatised?), and with
riends and non-work hobbies.
(5) Social characteristics.Social characteristics included responses to the ollowing
measures o social capital: Would you say that most o the time people try to be helpul,or that they are mostly just looking out or themselves? ; Generally speaking, would
you say that most people can be trusted or that you cant be too careul in dealing with
people?; and social support, which included responses to how ofen the respondent
spends time with riends, relatives, and amily or go to bars/taverns, as well as religious
activity, which may capture another side o social capital/connectedness.
(6) Optimism. Optimism was measured in the GSS through a question asking:
Te way things are in America, people like me and my amily have a good chance o
improving our standard o livingdo you agree or disagree?.
Parental education. Parental education could have been coded in a variety o different
ways: athers education, mothers education, maximum education, or some average.
We chose to ocus primarily on athers education because or older cohorts within
developed nations, the athers highest degree may be a better predictor o amilialsocioeconomic status than the mothers highest degree.25In the U.S., or instance, in the
1940s through the late 1960s ew women obtained college or university degrees.25As a
sensitivity analysis, we also replicated all analyses using mothers education instead o
athers education and provide these results in the Appendix tables (ables A-2 to A-5).
As described above, the GSS includes a variety o scales measuring different social
and psychological domains. We took the items rom the GSS and perormed explana-
tory actor analyses to reduce the dimensionality o domains 3, 4, and 5 enumerated
above, which contained multiple survey items. Te analysis identied housing tenure
and subjective perception o socioeconomic status as distinct domains (satisaction
with nancial situation and subjective assessment o nancial situation relative to
average, actor loadings ranging rom 0.53 to 0.61). Within the psychological domains,
two actors were identied: existential satisaction (overall happiness, happiness with
marriage, and satisaction with job, actor loading ranging rom 0.33 to 0.59) and sat-
isaction with leisure time (riends and hobbies, actor loadings ranging rom 0.41 to
0.42). Within the social capital domain, we have our distinct actors, bridging social
capital (trust in others, eeling that people look out or themselves, actor loadings
ranging rom 0.48 to 0.49) requency o contact with riends (actor loadings ranging
rom 0.41to 0.43); requency o contract with amily (actor loadings ranging rom 0.63
to 0.67); and requency o involvement in religious activities (actor loadings ranging
rom 0.61 to 0.72). Within each domain, items were averaged together or one score.
Specic items in each scale are described in more detail in Appendix able A-1 along
with descriptive statistics.
Finally, we also explored whether sel-rated health played a meditational role, sincehaving athers with higher education degrees might improve respondents health.
Similarly, we also included amily income and respondents education to see their
meditational effects.
Statistical analyses. We use discrete time hazard models to calculate the hazard
ratios (HR) or those parents with the highest degree levels relative to those with less
than a high school diploma. Tese models estimate the proportion o the sample that
33_HPU272_Friedman indd 87233_HPU272_Friedman.indd 872 4/11/16 2:27 PM4/11/16 2:27 PM
7/25/2019 33_HPU272_Friedman.pdf
5/22
873Friedman and Muennig
experiences the event (in this case, death) during a specic time period (beginning in
1978, with sampling conducted through 2002 and mortality ollow-up through 2008).26
Our dependent variable is the vital status o the individual within a given year, which
is dichotomous in nature (alive/dead). We chose the most parsimonious unctionalorm or our models, and the quadratic orm provided the best model t. We calculate
hazard rates using the complementary log-log link.27
o test the proportional hazard assumption, we rst examined the interaction o
duration o survival and the athers highest degree, which was not signicant. Ten,
we plotted the log-log survival curves or each group o the athers highest degree,
and these were proportional. Tis suggests that we employed the appropriate survival
analysis, the cloglog, which yields a HR.
o test the inuence o our constructed domains on the parental education-mortality
hazards relationship, we employed a traditional mediation approach that combined item
responses as a continuous variable.28,29First, we checked the relationship between the
athers highest degree and the explanatory variables under study. I this relationship is
signicant, we next examined whether adding explanatory variables reduced the totaleffect o the athers education on mortality by measuring changes in the hazard ratio
(HR). Potential psychosocial and material explanatory variables were added one at a
time to the baseline regression, and were tested in separate models. Te GSS-NDI was
approved by the Columbia University Institutional Review Board.
Results
able 1 shows the descriptive characteristics o the sample by the athers educational
attainment. Respondents with less-educated athers tend to be older, non-White, less
well off, and with less educational attainment. College graduation rates were 13% or
individuals whose athers had less than a high school education, 26% o respondents
whose ather completed high school, 54% o respondents whose athers have a bachelors
degree, and 61% o respondents whose athers obtained graduate degrees themselves.
Similarly, subjective SES (opinion o amily income, satisaction with nancial situation,
and social class) differed by the respondents athers education. Respondents with the
least-educated athers, or example, have an average subjective SES score o 2.43 (out
o 4) compared with 2.75 or those respondents whose ather completed a graduate
degree. wenty-eight percent o individuals whose athers do not complete high school
are in air or poor health compared to 14% o those whose athers completed high
school, and 10% o those whose athers completed college or have a graduate degree.
Individuals with less-educated athers are also less likely to be homeowners and are
less satised with their marriages, amilies, and jobs. Tey also have less social capital
measured as requency o contact with amily and riends or measured as social bond-ing or trust. Te means and standard deviations or the individual items that constitute
the constructed scales we use are provided in Appendix able A-1.
able 2 depicts the adjusted HRs associated with each additional degree garnered
by the respondents ather. We present ve models. Te rst model depicts the HR or
each level o parental education or the total sample with controls or the respondents
age, gender, race, and survey decade. Te second model is similar to Model 1 only
33_HPU272_Friedman indd 87333_HPU272_Friedman.indd 873 4/11/16 2:27 PM4/11/16 2:27 PM
7/25/2019 33_HPU272_Friedman.pdf
6/22
Table1.
DESCRIPTIVE
CHARACTERISTICSOFTH
ESAMPLEBYFATHERSED
UCATION.2008GENERALS
OCIAL
SURVEYNAT
IONALDEATHINDEXa
FathersEducation