630 likes | 652 Views
This study explores the differences in prosocial behavior among identical twins, focusing on the impact of religion and education. Analyzing correlations and causal relationships, it investigates why these factors influence giving and volunteering behaviors.
E N D
Differences in Prosociality Among Identical Twins: Religion Matters, Education Does Not René Bekkers Department of Sociology Vrije Universiteit (VU) Amsterdam ECSR Conference, Lausanne September 12, 2019
Usually… • We seek evidence on social influences. • But a good counterfactual is lacking. • So we do the best we can with statistical controls & other tricks we’ve learned from econometricians. • Still our estimates are far from perfect because there are many unobserved confounding variables.
Suppose… • …you could run a regression in which the ceteris paribus = holding everything else constant assumption is true. • …you could add an almost perfect clone for each observation, varying just 1 characteristic at a time. • …you could eliminate genetic effects and isolate purely environmental effects.
Jacques en Fred Kueter, identicaltwinparticipants in a study on Alzheimer’sdiseasein Amsterdam, https://www.vumc.nl/nieuws/nieuwsdetail/tweelingen-studie-bij-eenvandaag-1.htm
Three questions • How alike are twins in the United States with respect to prosocial behavior? • Are differences among twins in giving and volunteering related to differences in education and religion? • If so, what explains these relationships?
Ubiquitouscorrelates of philanthropy • Religion: • Affiliation (yes>no) • Denomination (Protestant>Catholic) • Participation (churchattendance) • Education: • Level achieved The variancebetweenfields of study is small
Where do the correlationsoriginate? The more general research questions: • Why are religion and educationcorrelatedwith prosocial behavior? • To whatextent are these relationships the result of environmentalinfluences? • Are these relationshipscausal?
The ideal experiment wouldrandomizeeducation VWO = highersecondaryeducation VMBO = lowervocationaleducation
The uniqueenvironmentalinfluence of education Note that shared environmental influences are also excluded by design in this analysis
Whatmoleculargeneticists do • Genome Wide Association Studies (GWAS): identify ‘candidategenes’ thatcouldexplainvariance in someoutcomevariable. • Typically, individualgeneslike OXTR and DRD4 explaintinyfractions of variance (<1%). • Typically, all single nucleotide polymorphisms (SNPs) combinedexplainlessvariance (16% of education) thanestimated in biometric models (35%) – ‘missing heritability’ problem.
Geneticists analyzed 2,165,398 single nucleotide polymorphisms and found NONE telllingus who is a friendly person https://twitter.com/renebekkers/status/806922437250011136
Relatedliterature • The “twin fixed effects model” has been used in economics to estimate the influence of schooling on income since the 1970s (Behrman & Taubman, 1976; Ashenfelter & Kreuger, 1994; Ashenfelter & Rouse, 1998; Isacsson, 1999; Miller, Mulvey & Martin, 1995; Bonjour et al., 2003).
Environment mediation model Notethatthis is a uniqueenvironment mediation model
Open Science • Download the data at ICPSR, run the code at https://osf.io/ujhpm/, and obtain the results reported in the preprint posted at https://osf.io/yjbd4/
The MIDUS data • Survey on Midlife in the United Statessponsoredby the McArthur Foundation. • The RDD sample selection procedure includedtwinscreeningquestions. • Only English-speaking respondents aged 25-74 living in the US who were physically and mentally able to complete the interview were allowed to participate.
ACE model results Entries computed with Falconer’s formulas, A = 2 (rMZ - rDZ); C = 2 rDZ - rMZ; and E = 1 – rMZ.
The highereducatedgive more These differences are massive: amountsdonated in thehighestcategoriesare seventoninetimes the amountdonated in the lowestcategory. Data werewinsorized at the 99th percentile.
The highereducatedvolunteer more Again, large differences – up to 2 times
The religiousgive more Religiousgiving is included in thisfigure. Excludingdonations to religion, the differences are smaller, but stillsizeable.
The religiousvolunteer more Thisfigureincludesvolunteering for religiousorganizations.
Twobasicregression models • Betweeneffects model: ignores the twin pair structure, replicatesbivariate analyses. Includesgenetic + shared and uniqueenvironmentaleffects (ACE). • Within MZ twinfixedeffects model: does the highereducated / more religioustwin of an MZ pair give and volunteer more than the lesseducated / religiousco-twin? Includesonlyuniqueenvironmentaleffects.
Regressors • Church attendance (times per year) • Religious affiliation: none (reference), Catholic, Protestant, Other (0-1) • Level of education (1-12) • Strength of religiosity (z-standardized)
Education and volunteering *** *** *** p <.001
Churchattendance and giving ** ** ** * **
Strength of religiosity and giving (*) (*)p <.10
Religiosity and volunteering *** *** ** ** (*) (*) ***p <.001; ** p < .01; (*) p < .10
Conclusions • The associationbetween the level of education and giving and volunteering is due to genetic and / or shared environmentaleffects. • The associationbetweenreligiosity and charitablegiving is duetouniqueenvironmentaleffects, but it is limited to churchcontributions. • Religiositynurturesvolunteering, alsobeyondreligiousorganizations.
Causalinference? • Notsofast. Perhapsvolunteeringnurturesreligiosity, andcausality runs theother way. • Or perhapsanomitted (shared?) environmental effect nurturesvolunteering and religiosity. We cannotinfercausalityfrom the twinfixedeffects model.
Overcontrolling? • Perhaps we have eliminated effects of education through religiosity? • There could be a suppressor effect if education reduces religiosity. • If so, we may see a positive relationship between prosocial behavior and the level of education if we do not control for religiosity.
Which questions Would you try to answer using data from twins? ……..using this model?
Futuredirections • Replicatethisfindingusing data fromother samples of twins, in the US andbeyond, e.g. Twinlife in Germany. • Add wave 2 and 3 measures. • Examineotherdependent variables usingthismethod: health, trust, subjective well being, prosocial values...
René Bekkers Department of Sociology Vrije Universiteit (VU) Amsterdam r.bekkers@vu.nl Blog: http://renebekkers.wordpress.com Twitter: @renebekkers
The current version of this paper, all Supplementary Information and the code that produces the results are available at https://osf.io/ujhpm/