46 MAIRE NI BHROLCHAIN is exogenous.Divorce and the outcomes with which as the outcome,although significant coefficients it is associated almost certainly have determinants in were found in the individual analysis,the sibling- common.'Divorce effects'authors are in the position based analysis revealed no significant coefficients of ascribing outcomes that are for the most part of on non-intact families.Indeed,the results suggest unknown origin to a cause -divorce that is also a positive effect associated with the presence of a largely of unknown origin.Present-day conventions stepfather in early adolescence.Such findings are in the social sciences condone this faulty logic. suggestive,but would clearly need to be repli- cated-not an easy matter with existing data. 'Utopian'Conditions and Technical Ingenuity Further strategies are considered in a later section. One response to the methodological comments made here is that although they may be justified in Why is Causal Inference Overdone? principle,they are utopian in practice,and that social science cannot expect to meet such require- Excessive confidence in causal inference in social ments.An initial answer to this is that we must cut science appears currently to be the rule rather than our coat to suit our cloth-if we do not have evi- the exception.Textbooks on social research methods dence sufficient to justify causal inference,then do little to discourage the practice.Many treat the such inference is unscientific.In any case,the sug-subject in only a negative sense,offering the maxim gestion that social science cannot be scientifically that 'correlation does not imply causation.Little rigorous appears a rather pessimistic view(cf.Lie-guidance is offered on how causal inference might berson,1988:395).Primarily lacking at present are actually proceed.21 Present-day social science lacks data resources rich enough to examine the many the methodology,scientific conventions,and fund competing hypotheses in this area.Many of the fea-of experience to provide guidance in matters of cau- tures of family functioning on which data are sparse sal inference (McKim and Turner,1997).Causal at present are in principle measurable,and so acces- claims are not that often scrutinized in social science sible to observation.As one scientist remarked in and so authors have not been accustomed to defend- connection with over-enthusiastic interpretations ing such claims.And so,the 'divorce effects' of evidence in another area:'We need to be rolling literature suffers from an inferential malady that up our sleeves and collecting new data,and be more afflicts social science as a whole. rigorous about our analysis of existing data'(geneti- Demography and sociology abound with causal cist Ryk Ward,quoted in Lewin,1998:26).That issues similar to the question about the effects of par- better data and imaginative design can greatly alter ental divorce.Examples include:the effects of perspectives is evident in recent work on sibship size education on fertility (would the fertility of those and intellectual development,long interpreted as a women who are in fact educated have been lower if causal relationship(Guo and VanWey,1999).For the they had been less well educated?);the effects of the present,the findings of an ingenious study of out-availability ofbirth controlon fertility(would the fer- comes associated with family structure are of tility of those women to whom birth control is particular interest(Yeung efal.,1995).One approach available have been higher if birth control had not to the problem of missing or hard-to-measure vari-been accessible to them?);the effects of cohabitation ables on family environment is to use siblings as on maritalstability(would those couples who cohabit controls-apparently impossible in the case of before marriage have had a lower risk of marital divorce,since all children in a family experience par-breakdown if they had married first?).In all cases, ental divorce.Nevertheless Yeungetal.overcame this the analogy with a natural science experiment is difficulty by using the PSID(Panel Study of Income inadequate not only because experiment is not pos- Dynamics)to define family-structure stage specifi-sible but precisely because,as Lieberson (1985) cally.They compared outcomes for children whose discusses,individual choice and the social process parents were separated at a particular stage of child-are inherently selective.It is therefore unrealistic in hood with those of older siblings who had been in ways that are critical to the validity of causalinference an intact family at that stage.With school completion to imagine these processes in experimental fashion. 
                   2
 3 4                  "       >  >      "   8              =#  /
     <   1                  /                         +                      >         G  /             !                        # J  %&''A =&-* 8 "                      6                              +    "                  A 2?                            4 #   I" ?     J  %&&'A 9E* 5                    "   0                 #K   ;? %&&&* F                               #,    %&&-* 1                       >                    < ,    G      8!
#8    !
*           5                                       ?             G                 G      !          3                               >           F              > 
 <   ; ?                         5 "               6    /       3    2       4 J     3          9% 8    "                         #6  5  %&&.* )         0                    +   2 3 4     3       R      
                  3         A  3          #                             @*(  3              #                               @*(  3           #               "    "       @* !                             J #%&'-*               !