Strategy for testing a proposition that expresses a necessary condition

condition A is known to be absent or the dependent concept B is known to be present. If the values of the concepts are unknown, any population could be selected from the domain. It is tested in this popu- lation whether the frequency of occurrences of instances with the val- ues A absentB present is zero as expected if the proposition is true in the population or is very small according to a “pragmatic determin- ist” criterion, as discussed in Box 8. The hypothesis is rejected if the proportion of instances with the values A absentB present is larger than zero or larger than the proportion specified. The same argument about inefficiency of the survey as discussed above for the use of the survey for testing a sufficient condition applies here as well.

4.4.3 Strategy for testing a proposition that expresses a deterministic relation

A proposition that expresses a deterministic relation implies that for each single instance in the domain the proposition is true according to the theory. This means that the proposition can be tested in a single instance. The preferred strategy for testing a deterministic relation is the experiment. In such an experiment it must be demonstrated that each change in the value of the independent concept results in a predicted change in the value of the dependent concept. Depending on whether condition A can be administered in different dosages, the experiment could either be cross-sectional in which different values of A are administered to different groups or longitudinal in which the value of A is, for instance, gradually increased over time. The hypothesis is confirmed if the effect B increases according to the prediction. If an experiment is not feasible, the longitudinal single case study or the comparative case study is the second-best strategy. In the longitu- dinal single case study one instance is selected for measurement of both the independent and the dependent concept over time. It is assessed for each measurement point separately whether the value of dependent concept corresponds to the expected value. In the com- parative case study, two or more instances are selected each with a different value of the independent concept and the value of the dependent concept is observed, or one instance is selected for meas- urement of both the independent and the dependent concept over time. It is assessed for each measurement point separately whether the value of the dependent variable corresponds to the expected value. The survey might be used to test a deterministic relation as well. For a test in a survey, any population can be selected from the domain. The statistical analysis could compute for each pair of instances in the sample or in the population whether an observed difference in the values of the dependent concept B in the two instances of the pair corresponds in the way predicted by the proposition with the differ- ence in the values of the independent concept A. It is tested in this population whether the frequency of occurrences of pairs of instances in which B does not follow A in the predicted direction is zero as expected if the proposition is true in the population or is very small according to a “pragmatic determinist” criterion, as discussed in Box 8. The hypothesis is rejected if the proportion of instances with the values A absentB present is larger than zero or larger than the proportion specified. The same argument about inefficiency of the survey as dis- cussed above for the use of the survey for testing a sufficient or a nec- essary condition applies here as well.

4.4.4 Strategy for testing a proposition that expresses a probabilistic relation

The experiment is the preferred research strategy for testing a prob- abilistic relation. The effect of an independent concept causal factor A is investigated by comparing the change in value of a dependent concept effect B in an experimental group which was exposed to the causal factor A with the change in value of B in a control group which was in the same condition as the experimental group but with- out the independent concept A causal factor. Different experimen- tal conditions, with different values of A, might be created and the range of values of B in each of these conditions is measured. Differences in the values of B between the different experimental groups are analysed, usually statistically, in order to draw a conclusion about how the values of B co-vary probabilistically i.e. on average with the values of A. If such an experiment is not feasible, the survey is the next best strat- egy for testing a probabilistic relation. In a survey, the co-variation between the values of two or more concepts is observed in a group of real life non-experimental instances. These are usually cross- sectional measurements i.e. at one point in time, but sometimes it is possible to design a prospective and longitudinal survey, allowing the