How do you know your causal argument is true? What are the main formal properties by which your argument might be tested?
There are many criteria which can be used to assess the validity of an argument of causality. Some of these criteria are, of course, of much greater importance than the others. Here I will discuss four of the main criteria necessary for determining whether or not a causal argument is in fact valid or not. These four criteria are: plenitude, representativeness, transparency, and replicability.
The plenitude of an argument is a description of the number of cases and examples of the relationship of causality that one is wishing to prove. Because there is no way to determine whether something is caused by something else in the absence of a comparison with an example of when it did not happen, there is a definite need for at least one case of causation. Without one example there is absolutely no empirical support for a supposed assertion of causation, rendering the argument moot. Thus, at least one example is necessary. Plenitude follows from this same logic. If I can prove once that some stimulus has caused a change in a given system, then that argument of causality has some validity. Following from that, the more times that it can be proven that X impacts a situation and causes Y, the more validity is lent to the hypothesis.
Thus a large sample size is imperative to proving a causal argument true. After all, an event occurring once or twice only can possibly be brushed aside as a fluke or the result of some random occurrence either previously unknown or unaccounted for. But, if a causal relationship can be established in many differing instances, the argument will be given much greater credence as the opposite of the previous statement comes into effect: the abnormalities and counter-instances being of a much smaller number will be written off as the result of a fluke or unaccounted for variable. This large sample size can also have the effects of clarifying what is being looked for or explained, as well as providing a basis for determining what (within the confines of the observations) is normal and what is abnormal. Thus, plenitude (or a large number of cases or examples) is a necessity for confirming an argument of causality.
Representativeness is a measure of how well the sample groups can be used to generalize to the general population. This is incredibly important within the context of establishing a causal argument as relationships at a close, distinctly local level are hard to approximate within the greater context of entire population. For example, determining that certain actions on the part of members of a local PTA (Parent Teacher Association) will create distinct repercussions among the schools would be very hard to generalize and apply within the confines of the US capital among the Senators and Representatives of Congress working for educational reform, to say the least of their general application to legislative branches of government the world over. Despite the fact that all of these bodies are made up of citizens of countries trying their best to accomplish their civic duties and work for the betterment of the education of their children, the relationship between them is sufficiently different as to figure that the local PTA’s causal effects are hardly representative of Congress in general.
Representativeness lies in the use of a causal argument being discovered in one case that can be applied to as yet unstudied cases to predict the likely outcome. Determining the level of validity of a causal argument in the broader sense than simply that referenced in the distinct cases studied is necessary in determining the validity of the causal argument in general. In order for a causal argument to be useful to the intellectual community it is necessary to establish the level of its relative representativeness and applicability to a broader range of problems and comparisons. Without this representativeness, there can be no assurance that the causal argument being made will apply anywhere outside of the specific cases in which it was observed.
The criterion of transparency refers to the ability to trace a clear and distinct path of causality from cause X to result Y. This can be a multistep or single step path. The point is that the path of causality can be traced directly from occurrence X through all intervening steps to occurrence Y. Determining this causal path removes any and all doubts about the exact relationship between X and Y. If this path cannot be distinctly proven and followed, then the results of the relationship can still be questioned as it is not exactly known if X in reality causes Y or if there is some other unseen factor coming into play. Determining this path of causality lends a great deal of credence to any argument of causality as it can be used to both prove a relationship between X and Y as well as show the way they are connected chronologically.
The replicability of a research design or experiment is very important to the final resulting determination of a causal argument. This decisive factor has to do with the ability of others to replicate and achieve the same (or extremely similar) results of a given experiment or research set. An experiment designed to show a causal argument must be replicable and yield the same results each time, no matter who does the experiment or where (giving due diligence to factors and variables that might change depending on conditions necessary to the experiment). This is extremely important in the process of verifying and determining a causal relationship because it allows for a significant amount of cases to be studied to give greater credence to the relationship (related closely to plenitude as discussed above) as well as overcoming problems such as measurement error and possible falsification of findings. Again, as with plenitude, something that only happens once is suspect because there could be any number of mitigating factors affecting the results of the experiment and the determination of the causal relationship. Thus, without a good sense of replicability, any experiment or research design that is intended to validate a causal argument can effectively be disregarded.