DIVORCE EFFECTS'AND CAUSALITY IN THE SOCIAL SCIENCES 47 One area in which detailed attempts have been not.The question would be an epidemiological made to find evidence for selective vs causative pro-one and,on the evidence,divorce would be consid- cesses is in the relation of marital status to health and ered no more than a risk factor-a correlate -of mortality.It has long been known that mortality is leukaemia.The entire question would revolve higher among the unmarried than among the mar-around whether the association was causal or not. ried.While some authors have interpreted the The full panoply of epidemiology's investigative relation as reflecting a positive influence of marriage methods -case-control studies,cohort studies, on health,more quantitative specialists have recog- and guiding principles in relation to causation(of nized that the mortality differential by marital status which more in the next section)-would be could arise in one of two ways,or perhaps from a brought into play.At its best,the search for evidence combination of the two:a protective effect of mar- would proceed with a certain intellectual tough- riage and a selection effect (cf.Livi-Bacci,1985).ness.There is no reason why the same rigorous Unmarried people may be less healthy because standards of evidence should not be brought to good health is a criterion in choosing a marriage bear on the study of sociological and demographic partner and so the less healthy will marry less outcomes.Experiences such as school-leaving,early often.Also,poor health may raise divorce risk and/ partnership,extra-marital pregnancy,and the like or may reduce the chances of remarriage.On the are probably as easy to measure as most medical out- other hand,marriage might promote good health. comes and so could,in principle,be investigated in It could do so by increasing social contact,limiting similar ways.Epidemiologists nevertheless have an risk-taking behaviour,providing additional health enormous advantage in the relative constancy of care or encouraging take-up,and other such human biology across time and place,and so bor- mechanisms(Goldman,19930). rowing some types of study design from A counterfactual is involved here similar to that in epidemiology might be more complex to imple- the 'divorce effects'issue.Nevertheless,the apparent ment.This means of course that the social scientist impossibility of answering such questions has been will usually have much less secure grounds for infer- met with some resourcefulness in the associated ring causes than will the epidemiologist demographic research.Competing explanations Nevertheless,causal relationships are inferred much have been approached via the longitudinal tracking more readily in social science than in epidemiology. of differentials,international comparisons,analysis of change through time,identification of causes of death,district-level analyses,and detailed analysis Some Solutions of cultural practices in relation to marital search and matching (Livi-Bacci,1985;Kisker,1990;Hu Routine unjustified causal inference is a problem in and Goldman,1990;Goldman,1993a,1993b;Gold- social science not just because it is unfounded.It manefal.,1995).And while the issue may not yet have also misleads us about our access to causal know- a definitive answer,the question has been clarified ledge and so prevents us from seeking defensible and the area of uncertainty around it reduced some- ways of approaching causal inference.We cannot what. be sure at present,in my view,that strong causal inference is possible in social demography and A Thought-Experiment sociology,outside very particular phenomena.One day we may be able to tell whether it can be done and Suppose the question at issue was whether parental later,perhaps,we will come to know how.In the divorce resulted in an increased risk of,say,leuk-meantime,we could certainly develop sounder aemia in the children of the marriage up to the age investigative practices. of,say,25.Would evidence of the kind widely A key element of any solution-and one pro- accepted now as causal in relation to educational posed and discussed in different ways by Freedman and socio-demographic outcomes be accepted as (1985,1997),Cox (1992),and Goldthorpe (this indicating that parental divorce caures leukaemia in volume)-is that we look to theory,with its atten- their children?We can be fairly sure that it would dant specification of the mechanisms through1                                          !    "                ?            B     B                0      3                        A     3        3 # J7 %&'-*                                  +       " N         1            !             " "            "        #K %&&=* +             2 3 4   <                            )           "  3                                             #J7 %&'-( " %&&:( $  K %&&:( K %&&= %&&=( K  %&&-* +                                  %  6                        "    "              9- ?     "                               "   @ ?          5                        "   >    >   " 5                     5     4      >                   #       * >         +                     5                                             "                                                                       5                              <                    I  /             /       !           "         "        ?                                1                     "  !                 + "      >         3   F #%&'- %&&.* ) #%&&9*  K  #   * >    "