Static versus dynamic analysis of the interaction between female labour-force participation and fertility

Numerous studies have found a negative relationship between female labour-force participation and fertility. In theory, there could be three explanations of this finding: (i) causality runs from labour-force participation to fertility, (ii) causality runs from fertility to labour-force participation, (iii) causality runs both ways. Alternatively, the relationship may not be a causal one. In practice, empirical studies covering a wide range of Western countries at different times, and utilizing a great variety of methods and techniques, have shown all four possibilities to be plausible. This may be because outcomes differ from country to country for socio-cultural reasons, or from period to period for historical ones. If so, applying various methodologies to data for one country at a particular point in time should yield consistent results that all point in one direction only. If they did not outcomes would appear to be method-dependent. The single data set used in this study refers to the Netherlands in 1984 (ORIN project). The relationship between fertility and labour-force participation in this data set is investigated by means of three methodologies, ranging from ‘static’ to ‘dynamic’, i.e., differing according to the degree in which they take the temporal aspects of the decision-making process underlying this relationship into account: simultaneous logit analysis, Granger analysis and Markov analysis. Each main approach is applied in two different ways or on two different subgroups, for a total of six applications. In spite of diverging operationalizations of the basic variables, it turns out that four of these six analyses favour the inference that fertility decisions do have an impact on labour force participation decisions but not the other way around, whereas the other two confirm earlier findings (from data sets collected during the 1970s) that the relationship is reciprocal. Substantively, this might indicate that the pattern of covariance is changing. But ‘static’ simultaneous logit analysis is the only method to consistently point at this causal unidirectionality, while outcomes from Granger and Markov analysis depend on the modality applied. Methodologically, this means that the issue of method-dependency, at least in this area, remains largely unresolved.