Causality as validity: Some implications for the Social Sciences

1 Running head: Causality as Validity Article type: Short communication Causality as Validity: Some implications for the Social Sciences Rafael Moreno and Rafael J. Martínez University of Seville Address for correspondence: Department of Experimental Psychology. Calle Camilo J. Cela s/n 41018 Seville (Spain) Telephone: +34 954557670 Fax: +34 954551784 e−mail: [email protected], [email protected] The final publication is available at Springer via http://dx.doi.org/10.1007/s11135-006-9060-4 2 Causality as Validity: Some implications for the Social Sciences Abstract We analyse the concept of causality in the social sciences, whose development is insufficient and lesser than the methodology developed for its study. The nature of the causal process as the production of effects remains unclear and the relationships considered to be manifestations of that process cannot be taken for proof of its existence. Given these difficulties, we suggest that, aside from the inherited interpretations, the practice of the concept of causality makes reference to correctly specified relationships not confounded by others; characteristics identical to those which define validity. In that way, causality is equivalent to the validity of a relationship. Beyond merely re-understanding causality, this proposal permits the deduction that the temporal precedence of the cause is a necessary condition only for one type of causality, making it possible to consider other types, not admitted by the traditional notion, in which the cause is consequent or simultaneous to the variable to be explained. Examples and characteristics of these types of causality are presented and considered to be useful for the social sciences. Key Words: Causality, Validity, Temporal asymmetry, Simultaneous Relations, Nondirectional Causality, Causal Models, Social Sciences 3 1. Introduction Causality is a central theme in the methodology of the social and natural sciences, which corresponds to the relevant place it holds in our culture (Thompson, 1993). In that sense, experimental methodology holds the study of causal relations as its objective, and the possibilities for the evaluation of causal models for non-experimental data in the social sciences has increased greatly in recent years (e.g. Aickin, 2001; Bentler, 1985; Bollen, 1989; Bollen & Long, 1993; Dalhbäck, 2001; Hayduk, 1987; James et al., 1982; Jöreskog & Sörbom, 1988; Long, 1983; Mc Donald, 1985; Muthén, 1987; Seidel & Eicheler, 1990). This methodological development has been accompanied by a broad consensus about the nature and characteristics of causality: a) It is understood as a process of production or transmission of forces by which the cause brings about the effect; b) It is considered that that process is manifested in specific relationships and that it appears when certain variables, such as common causes and intervening variables, which may influence the components of the relationship under study, are controlled (Wright, 1934, 1960; Simon, 1954; Blalock, 1962; 1964; Duncan, 1966). Relationships which do not occur under these conditions are considered to be spurious or indirect, products of other variables, except in the case of omission of another variable – termed a suppressing variable– with similar effects, but of the opposite sign, on the variables of a causal relationship (Pedhazur, 1982; Bollen, 1989). Finally, c) It is understood that, as manifestations of a process of production, causal relationships are temporally asymmetrical or directional from the cause to the effect. However, such consensus certainly does not translate into a clear concept of the supposed process of production, which is therefore maintained as one of the frequently discussed themes in social science (Doreian, 2001; Notterman, 2004). As Kenny (1979: 4) affirmed, "It is difficult to convey this notion of causality formally, just as it is to formally 4 define space or time". The principle difficulties of each one of the indicated characteristics are the following, respectively: a) That link is not explained beyond the general declaration that the cause is the generating force of the effect. What exactly the nature of a causal force is and what causal forces in different areas, such as mechanical physics and social processes, have in common, are questions which are not answered, despite being basic for the postulation of the existence of that causal force; a force said to be "almost vitalistic," given its lack of definition (Kenny, 1979: 4); b) The supposition that the causal force has an existence which is made manifest in specific relationships has not been validated, given that the force is not defined independently of the studied relationships. These relationships may not be considered to be the manifestation or proof of the existence of a force which is itself exclusively defined by those relationships. It is simply a circular argument which cannot validate what it intends to, and ; c) Although the temporal asymmetry or direction of the causal relationships is clearly identified in terms of the precedence of a variable, the temporally asymmetric relationship is not a characteristic specific to causation, for example the succession “cockcrow - sunrise”. Thus, although the causal theme in the social sciences presents an important development of formal procedures, it has serious difficulty to define and validate the supposed existence of a causal force of production. As have indicated Glymour et al. (1989: 58), "mathematical representation is the easiest part; the hard part concerns what inferences are to be drawn from causal claims", which is due precisely to the poor definition of these claims. It appears that there exists a conceptual problem regarding the causal process – a problem compounded by the existing procedures for the study of relationships. To have procedures for the purpose of studying something which is poorly defined is a situation which should be reconsidered in order to recover the sense of the procedures and of the causal notion itself. It is a problem most likely linked to the insufficient attention paid by methodology, as has been 5 indicated by others: "A basic introduction to the concept of causal theories is hard to find, since most of the time the discussion is already put in mathematical form" (Saris & Stronkhorst, 1984: 22). Frequently, methodological work on causality attributes the insufficiencies in definition to lapses in the philosophy of science. That would show that the social sciences have neglected to undertake the necessary conceptual analysis, passing it on to philosophy. Nonetheless, given that that attitude has not been satisfactory, it would appear reasonable to approach the causal question from scientific methodology itself, analysing its specific bases and practices. 2. Causality as validity As a starting point, we consider that the existing methodological procedures and their continuous application express the concept of causality actually used in scientific practice – a concept which, nonetheless, appears to be masked by the usual and inherited interpretation of causality as a force of production. In our opinion, to adequately define the concept of causality does not insist that one questions the action of the supposed force, but that one questions the facts which are under investigation, and are considered to be reasonable, when researchers claim to be studying causality. Such as with any other concept, it may be true for causality that in identifying the situations in which a community considers the use of the concept to be adequate, we are demonstrating the meaning which that community gives it in practice (Wittgenstein, 1953), coinciding or not with the way that that community describes that practice itself. Therefore, to identify the current concept of causality in the social sciences demands the analysis of the facts used in the methodological bibliography as characteristics of causality, despite the interpretations which have been superimposed over them; analysing what they do, more than what it is said that they do. That should be the necessary and 6 sufficient material to meet the indicated objective. In the methodological bibliography of the social sciences, principally constituted of structural equation models (SEM), causality is referred to in terms of relationships (e.g. Bollen & Long, 1993; Dalhbäck, 2001; Hayduk, 1987; Jöreskog & Sörbom, 1988; Seidel & Eicheler, 1990). A large part of the relationships of many variables xj with other variables yi may be indicated by the additive linear model yi = γi1x1 + γi2x2+...+ γiqxq + ζi (1) where each ij is the expected change in yi for a one-unit change in x1 when the rest of the variables are maintained constant, and where ζi represents the random component of yi or the effects of variables such as errors of measurement or characteristics which are unknown or of little relevance and are therefore not included in the model as independent variables. The mentioned expression takes into account that the variables of each studied relationship may be related to other variables which they should not confound. Thus, when studying each relationship (yi, xj) in equations like (1), it is required that the variables contained in ζi, beyond merely having minimal effects and constant variance, are not related among themselves nor among the variables xj, meaning that E(ζi)=0, V(ζi)=σζi2, COV(ζiζi')=0 for all ii', and COV(ζixj)=0, for all i and j. This demands the inclusion of the maximum number of relevant variables, and demands that their control is carried out by operations previous to data collection and/or by statistical means, in experimental or non-experimental research. This is what Steyer et al. (2000) have indicated in terms of causal unbiasedness and unconfoundedness. Additionally, when representative inclusion and control of variables are not sufficient, assumptions or partial restrictions must be made, evaluating the possible relevance of each one of them by successive replications. Some conclusions may be extracted from what has been expressed. While the 7 traditional notion of causality is understood as a process of production of facts, which is manifested in controlled relationships between variables, in practice regulated by the scientific methodology of SEM, it occurs that: 1) The meaning of causality is exclusively referred to as adequately specified or represented relationships, not confounded by other variables. 2) These two characteristics also define the validity of a relationship, (Kerlinger, 1973; Shadish et al, 2002). In consequence, to identify a relationship as causal is simply, in practice, to identify it as valid. 3) Causality then is formed by the relationships subject to study and by those others which must not confound the first, which means that these and other relationships are equally as necessary and that neither is sufficient to establish causality. Aside from this conjunction, it is true that "correlation does not prove causation" and that "a lack of correlation does not disprove causation" (Bollen, 1989: 52). Finally, 4) the notion sets the causal theme in identifiable terms and rescues it from interpretations with problems of definition or verification; the adequately specified and not confounded relationships should be recognised as necessary and sufficient material to constitute the meaning of causality, instead of being underestimated as simple manifestations of supposed underlying processes never defined independently with respect to those relationships. To assume the previously enumerated ideas permits a different and perhaps more significant way of thinking about the two characteristics of the traditional notion: Causal force, manifested in relationships. Instead of saying that a relationship reflects causal force studied through specified and not confounded relationships, it may be said that this relationship is termed causal. This last point describes the circumstances in which the concept of causality is applied in scientific practice, different from the supposition of a force based on data considered to be manifestations of that very force. The causal theme therefore becomes a problem for the way it is planted rather than for not having been solved. The conception of 8 causality as the transmission of a hidden force would, by definition, be a superfluous problem which could be abandoned. 3. No necessity of temporal asymmetry in causal relationships The temporal asymmetry or directionality of relationships, understood as the precedence of one variable, is an inevitable characteristic of the assumption of causality as a productive force. The supposed transmission of that force between variables may only occur if the cause precedes the effect. In consequence, simultaneous relationships are usually ruled out, even considering that in relationships of such appearance - such as some from non-recursive models - the interval between cause and effect is smaller than the observation interval. Within this framework, temporal asymmetry or precedence is not questioned, as it is assumed to be a necessary condition for the relationship and any other possibility would be absurd. Keeping the precedence of the cause constant, the covariation between cause and effect is studied, but not if that covariation depends on the cause preceding the effect. The same occurs in the case of reciprocal causality, in which the cause precedent at time t is the effect variable at a posterior time t+k, which assumes bidirectionality or successive temporal asymmetry. The basis differs from the notion of causality as validity understood in terms of unbiasedness and unconfoundedness. This notion does not include reference to temporal asymmetry of causal relationships, which may indicate that this characteristic is not a necessary condition for all effects. This means that the cause or necessary condition for a valid relationship must not necessarily be precedent in time, and may have several consequences. One may be that if temporal asymmetry or direction is not a necessary condition for causal relationships, then the order of occurrence of determined variables is one thing, and quite another is if that precedence is necessary for that relationship. For that 9 reason, the temporal asymmetry of a determined causal relationship must not be understood as the simple precedence of one variable, but as the conditionality one relationship may have with respect to the temporal precedence of some of its variables. Also, if the temporal precedence of one variable is not a condition for all causal relationships, that means that there may be valid causal relationships without this precedence of the variable considered to be the cause, occurring simultaneous to the effect or following it. In other words, in accordance with the notion of causality as validity used in methodological practice, one may speak of three types of causal or valid relationships in function of the temporal location of the cause or necessary condition: a) Directional causality with preceding cause; b) Directional causality with consequent or posterior cause; and, c) Non-directional causality, with cause simultaneous to the effect. Given the novelty of the latter two types, it is convenient to explore their characteristics and present possible examples. Abundant examples may be found of directional causal relationships with consequent cause in the psychological literature of operant conditioning, with wide generalization across species. In any of its variations, the phenomenon is described as the increment of those responses of the organism which are followed by determined events or reinforcement. The posterior reinforcer is considered the cause of the increase of the behaviour, and the key criteria to distinguish between behaviour directed by its consequences and that directed by previous stimuli which occur as pavlovian conditioning. Another example of this directional causality with consequent cause is found in the explanations of intentional action or anticipation-guided ends or goals, such as that found in Action Psychology (e.g. Harré et al, 1985). In reference to non-directional causal relationships, some primary examples may be extracted from the work of Kuhn (1977) on causality in physics. Having objectives quite 16 Note. 1. The proposed types may represent the four aristotelian types of causality. The relationship of the precedent cause appears to correspond to causality of the efficient type, those of consequent cause to final causality and those non-directional causal relationships to the formal and material types, in which a variable is explained by being validly related to a representation or model of its structure or of its component elements, respectively.