50 MAIRE NI BHROLCHAIN tools of this kind be mentally available to us,we need of large-scale surveys are not often in the public to reduce our intellectual dependence on regression domain.There may be disincentives to publish equations and coefficients as the means of testing information about the deficiencies of a data-set hypotheses.In particular,the design of investiga- since it might damage the reputation and funding tions needs a great deal more attention,as prospects of a study to do so.To overcome this pro- Freedman argues.In general,as Lieberson suggests, blem a convention could be established that research social science needs to expand ideas about the types groups independent of those responsible for major of causal mechanism linking phenomena of interest. data-sets should be funded to evaluate and publish For example,are cause and effect similar in scale?- quality assessments of those datasets. has a particular spectacular effect resulted from a Social demographers have available to them a vast spectacular cause or is the cause,perhaps,relatively fund of expertise and lore in the work of minor in scale?Or has a particular sudden cause demographic colleagues involved in the core demo- resulted in a sudden effect,or has it taken effect by a graphic activity of estimation,especially in Third long,slow process,only identifiable in the relatively World settings.Such measurement preoccupations long term? seem somehow to have been forgotten in social In epidemiology,as in other sciences,establishing demography,though a couple a decades ago they causality is considered a demanding task.Investiga- were prominent.Similar concern about measure- tors expect to have causal claims carefully ment and description have been voiced by others. scrutinized.If standards of causal inference in social Both Goldthorpe(this volume)and Abbott (1998) science are to approach those in epidemiology,we advocate that description be recognized anew as of have to accept that the same robustly expressed scep-key importance and Freedman(1985)has long urged ticism will be the order of the day.We also have to the same view.Abbott remarks that 'Sociology will accept that causal processes will probably be identifi-not be taken seriously again as a general science of able less often,and with less certainty,than in the social life until it gets serious about description.We epidemiological domain,because of the biological will know that sociology and social demography are nature of epidemiology's subject matter and the getting serious about description when applied greater access to intervention in the form of e.g.ran- debate centres on it-when,for example,state- domized clinical trials.I am not suggesting that ments such as that the extent to which fertility has epidemiology is the only source of practical exam-declined in Zambia is hotly debated'are a common- ple,but that it is a close neighbour that has found place.We would do well to remember that more systematic ways of dealing with inferential measurement is of enormous importance in the nat- problems that are similar to our own. ural sciences.Millikan's oil-drop experiment-one of the most famous in the history of science -was designed simply to measure the charge on an Data Quality,Description,and Measurement electron.Probably the greatest debate and measure- The need for greater attention to data quality was ment effort in social science fields currently is in noted earlier.It is crucial when issues become con- policy-related research-the measurement of troversial.Basic conventions we might adopt are unemployment,of deprivation,of inequalities in that research reports should always specify the level health,and the like.This is no doubt because such of missing information associated with every vari-topics can be of acute political significance.But the able used,any procedures used to impute values divorce issue is also a controversial one,and surely during the editing of the data,numbers in the sam-the response to the uncertainties surrounding it ple oforigin and in the sample analysed,the identity should be increased investment in the quality of of the raters evaluating outcome measures,and the our information like.The need for blinding to ensure bias-free mea- sures in some circumstances was noted earlier.The Culture of Restraint and Modesty quality of large data sources used repeatedly in secondary analysis needs particular attention.It is a Causal inference is a hard-won commodity in any matter of concern that detailed studies of the quality science.With our present(lack of)causal knowledge   "                        G          !                   F   !   J                 "     F      3    @ >         3                      @ 1                 3    " 3              @ !                " !            0 !                                    ?                                     4  /                   0   !                                                ' 9 '  "  5                  !          7                                                                            " 5                    5                       !                              5                                    5                      /                                           "                   5 ?                                              7  K  #   *  + #%&&'*       0    "    F #%&'-*       + "  2    "                   4 ?  "                     >             2            S     4    ?                      6"4   >             >             8          3             >                   " 5                  7                                       
  7   "  )            ?     #" *   "