Mining for Association between life courses

We explore three kinds of association between life courses. First, there is case-based association, which relies on the average contingency between states in different domains. This kind of association turns out not to be very practical: measures based on entropy of state contingencies are insensitive to order of states. However, a new measure based on contingency of subsequences turns out to be a numerical anomaly. Second we investigate global association between dissimilarities or distances in either domain and apply Escouffier’s RV and Mantel’s coefficient to life-course data. Unfortunately, such coefficients do not pick up associations since they presume monotonicity of distances across domains. Finally, we focus on local association, exploiting neighborhoods. It turns out that this approach leads to a coefficient that has a nice interpretation in terms of predictability and is numerically tractable and seems well behaved at the full range of associations.