Foundational consequences. Question: why should we bravely defend causality? Still work on definition, rationale, methods? Is it just a fashionable topic for academics? NO goals of social sciences, cognitive and practical. Both need causation.
I start from the scientific practice. Reading the scientific literature, here are three ‘morals’ I drew – the whole work is then devoted to substantiate those claims. Survey of the scientific literature, in particular of causal analysis in several disciplines in the social sciences. E.g.: demography, econometrics, epidemiology … (yes, it is debatable whether epidemiology is a social science – I take social sciences in a very broad sense, and insofar society, well, distribution and causes of disease in society, is concerned, epidemiology, to my eyes, falls into the social sciences) Morals: Causal relations are not characterized in terms of physical processes. Instead, statistical characterization. Use of statistical notions. Of course, problem of correlation is not causation, statistical probabilistic characterization. Causal relations are not sharply deterministic. Not an ontological commitment to inderminism, just a problem of epistemic acces. Role of causal context. Use of background knowledge. Formulation of causal hypotheses in this context, H-D methodology to confirm. Social scientists seem to be primarily interested in testing variations among variable. This might be trivial to practising scientists, I’ll give arguments why it is not.
Briefly present the context of Caldwell study on child mortality and mother’s education. 1979 article in Pop Studies. developing countries, in part rural & urban area in Nigeria; objective= understand factors that influence child mortality; former studies: focus on sanitary, medical, social, political factors. evidence gathered in other studies. In analyzing impact of public health service, Caldwell notices that many socio-ec factors provide little expl of mortality rates. INSTEAD, mother’s education is of surprising importance A table that show how cm rate varies depending on certain factors, namely: mother’s place of residence, mother’s education, husband occupation and education, type of marriage. Briefly explain methodology employed Analysis and comparison of contingency tables. Those tables show how frequencies of some factors vary depending on other factors. Discuss quotation Caldwell’s causal reasoning
Briefly explain the meaning of the structural equation. What is Y, what X, what . The linear relation, equality is not algebraic equality … interpretation: variations in X accompany variations in Y. quantify those variations. So, variation is the primary concept used in this equation representing a causal relation. since variation is not a causal concept, what guarantees the causal intepretation wait and see the methodological consequences.
PT in philosophy have been developed in slightly different ways by different authors. Present Suppes (i) pioneer but more accessible than I.J. Good (ii) bcz in a way or another every current theory refers to Suppes’ (iii) bcz we just need to grasp the basic intuition behind and not deal with all criticisms, objections etc 2 temporally distinct events, event C preceded event E; we can attach probabilities to them, and a cause is an event such that P(E|C)> P(E); Suppes aware of possibility that events lower (not increase) prob (negative causes, preventatives) P(E|C) < P(E). In general P(E|C) P(E) compare marginal and conditional probability check whether the marginal probability differs from conditional the cause is responsible for the variation.
Theory of causal explanation. Causes explain because they make effects happen. Causal generalizations are change-relating or variation-relating of course, problem of distinguishing causal from spurious change-relating relations have to show a certain invariability as prescribed by invariance condition in structural models. Shall see that invariance is not bottom-line concept.
Foundations. I show that forefathers of quantitative analysis in the social sciences, and also Mill (renowned methodologist of the experimental method) rely on the notion of variation – without fully conceptualizing its importance for causality Start from Quetelet. objective of his work is to study the causes that operate on the development of man, in particular the goal is to measure their influence and their mode of reciprocal action. explaining, quite informally, how he intends to detect and measure the causes that influence his average man . Basically, what he describes is the comparative method as Mill will elaborate in detail. Quetelet’s causal reasoning is impregnated of the variation rationale because his search for causes of human development starts with the observation of a wide variability in the mortality tables he calculated . And then looks for factors that are responsible of those variations. Causal factors are detected by testing variations in mortality tables.
In the System of logic the experimental inquiry is seen as the solution to problem of what process for ascertaining what phenomena are related to each other as causes and effects. We have, says Mill, to follow the Baconian rule of varying the circumstances, and for this purpose we may have recourse to observation and experiment. In this general idea of varying the circumstances the variation rationale is clearly already at work. The experimental inquiry is basically composed of four methods: 1. Method of agreement (comparing together different instances in which the phenomenon occurs), 2. method of difference (comparing instances in which the phenomenon does occur with instances in other respects similar in which it does not), joint method of agreement and difference (that consists in a double employment of the method of agreement, each proof being independent of the other, and corroborating it), 3. method of residues (subducting from any given phenomenon all the portions which can be assigned to known causes, the remainder will be the effect of the antecedents which had been overlooked or of which the effect was as yet an unknown quantity), 4. method of concomitant variation is particularly useful in case none of the preceding methods (agreement, difference, residues) is able to effect a variation of circumstances. For instance in presence of permanent causes or indestructible natural agents which is impossible to either to exclude or to isolate, which we can neither hinder from being present nor contrive that they shall be present alone. OSS: that the experimental method inapplicable in the social sciences. Interestingly, Durkheim goes against this view (les Règles de la methode sociologique, ch VI), in particular he maintains that the method of concomitant variation is fruitfully used in sociology.
suicide as a social phenomenon ; searches for the social causes of suicide, namely for the factors depending on which the social rate of suicide varies . study the variability across time of the suicide rate (= ratio between number of voluntary deaths on the overall population). This variability is quite insignificant across time within the same population, but it is instead considerable across different societies. - The rationale of variation permeates Durkheim’s causal reasoning about suicide as a social phenomenon, because by examining how the suicide ratio varies across societies he aims at detecting the social factors this variation depends on. The variation rationale in Le suicide is extrapolated from his argumentation along the essay. However, the rationale is definitively explicit in Les règles de la méthode sociologique . - The comparative method is, according to Durkheim the only way to make sociology scientific. However, not all procedures of the comparative method will do. Only the Millian method of concomitant variations will. Durkheim has, nonetheless a determinist conception of the causal relation, for he believes in the principle “same cause, same effect”; the comparative method is scientific, i.e. conformed to the principle of causality, only if comparisons are analyzed under the supposition that to the same effect always corresponds the same cause (1912 : 157). In Durkheim a probabilistic characterization of causal relations is totally absent, and if one of the two phenomena occurs without the other, this might be due to the effect of a third phenomenon operating against the cause or to the fact that the cause is present under a different forms. Durkheim of course is aware of the fact that a concomitant variation might be due to a third phenomenon acting as a common cause or as an intervening factor between the first and the second. He concludes that the results of the method of concomitant variation are therefore to be interpreted. If a direct link from the cause to the effect is not self evident, then the mechanism responsible for the concomitant variation has to be unveiled in order to rule out a case of common cause Interestingly, in Durkheim we can also find an “invariance condition” ante litteram . What we have to compare is not isolated variations but series of variations, regularly constituted (1912 : 165).
Suppose we are interested in unemployment. We might be interested: in whether the same characteristic, say unemployment rate, varies across time (taxon 1) – e.g. in two successive surveys, or across individuals (taxon 2) – e.g. individuals in the survey may show radically different employment histories, or across characteristics (taxon 3) – e.g. unemployment rate may be different according to different levels of education. we can model counterfactual variations (taxon 4), for instance the individual probability of finding a job given certain characteristics, or, in experimental studies, we can check whether variations hold between the test and control group. Finally (taxon5), variations can be merely observed – when we deal with observational data, or can be the result of interventions – if we can manipulate and operate directly on data.
Here we come back to the issue anticipated before: if variation is not per se a causal concept, what guarantees the causal interpretation? Main idea: distinguish associational models – causal models. What associational models do, what they are constituted of. What causal models do, what they have more than associational models. Explain features.