Abstract
1. Introduction
A key principle of the life-course paradigm is the notion of linked lives, that lives are lived interdependently, with sociohistorical influences expressed through networks of shared relationships (Elder, Johnson, and Crosnoe 2003). Although this principle is central to network and life-course analysis and has received increasing attention in the conventional regression framework (e.g., the actor-partner interdependence model, the multilevel p 2 model; for an overview, see Card, Selig, and Little [2011]; Kenny, Kashy, and Cook [2006]), it has been less developed in sequence analysis. This is unfortunate, because, since its inception in sociology in the 1980s, sequence analysis has been a new way of understanding and capturing individual lives over time (Abbott 1995; Abbott and Tsay 2000). Robette, Bry, and Lelièvre (this volume, pp. 1–44; hereafter RBL) make a significant contribution by developing a method for simultaneously detecting life-course patterns for those whose lives are linked (e.g., mothers and daughters in the illustrative case). RBL’s innovative globally interdependent multiple sequence analysis (GIMSA) has considerable potential to address important questions that are of central interest to many life-course scholars (and sociologists more generally) about lives lived in tandem and over time.
In this comment, we focus on the contributions of GIMSA and also broach three issues, all of which are substantive but have methodological implications: temporality, analytic weight, and variability. We also briefly discuss issues related to validity and estimation uncertainty.
2. Contributions
Not only life-course scholarship but sociology itself is to a large degree about discovering identifiable patterned social relationships over time. But the “over time” component has been little theorized or empirically assessed. Consider, for example, the large body of stratification research linking parents’ occupational status (at no set age or when the child is 16 years old) with the subsequent occupational status of the adult child (often at no set age). Or consider studies of parental poverty or unemployment when a child is a preschooler or adolescent. This flies in the face of the dynamic and complex life paths of both generations and the relationships between them. GIMSA as a method provides the means of capturing the dynamic patterning of lives, but it also encourages its users to theorize these patterns and intergenerational transmissions both over and in time, considering and measuring their multidimensionality and the degree of contemporaneous of their interdependence while modeling them in a parsimonious way.
GIMSA is thus an important addition to the toolkit of sequence analysis that raises issues about the clocks and calendars of lives and relationships over time. For example, what matters across generations? Is it the ordering of events, their durations, or their timing (in either the parents’ or the children’s biographies)? What is the role of social change and historical time? RBL point out that part-time work was not widespread in France until the 1980s, yet it was prominent in some daughters’ lives. They note that “in effect, parents’ sequences and those of their children can be of very different nature, reflecting the norms of their time, and our focus is on the relationship between sequences as wholes rather than on their synchronization in terms of age.”
3. Temporality, Analytic Weight, and Variability
3.1. Temporality
A key component rarely fleshed out in the linked-lives principle has to do with temporality. One person’s behavior could affect another person instantaneously or within a short time frame (“local interdependence”), or it could be that it is the whole history of one person that really matters (“global interdependence”). GIMSA seems to have been developed with the latter logic in mind. This obviously corresponds well with the value of sequence analysis as a holistic approach. Nevertheless, it does not appear to allow the examination, for example, of how a given transition in one person’s life is tied to temporal patterns in another’s life. And yet understanding these unfolding processes as turning points, contingencies, and so on, is exactly what the linked-lives formulation asks us to do.
Not requiring synchronization is a key advantage of GIMSA, which provides more latitude in addressing sequences involving different generations. However, it opens up other questions, such as the choice of the time dimension. RBL purposefully chose two different time clocks (mothers’ from ages 14 to 60, daughters’ from the completion of education to 15 years later) to demonstrate the power of GIMSA. Future users will also need to make decisions regarding which temporal dimensions to use. Are results sensitive to alternative time scales? If so, how should one decide which to use? Are there formal tests to discriminate alternative typologies?
A related issue is the vague temporal order, exacerbated by the lack of transparency in partial least square (PLS). Even in the example provided, there should be some overlap between the two life-courses. For example, assume that one daughter finishes her education when her mother is 45 years old. The mother’s sequence after age 50, therefore, is not expected to affect her daughter’s first 5 years (i.e., completion of education to 5 years later). In other words, not every year should “count” when identifying shared patterns.
3.2. Analytic Weight
It is not entirely clear about the analytic weight, or the relative status, of the two sequences. Lives could be linked in a variety of ways; for example, directionality could be one way or mutual. This is a distinction that needs to be clarified in life-course analysis. In this case, given the ages analyzed, the focus is unidirectional, from mothers to daughters. According to RBL, “the daughter’s situation at a given age is unlikely to be linked to her mother’s situation at the same age, but rather it is the mother’s whole employment history that is considered as part of the daughter’s social background.”
But this will not always be the case when the years considered overlap for both members of a dyad. In addition, the canonic PLS treats mother and daughter as symmetric, and what GIMSA draws on is shared information. It is hard to see the social meanings (how lives are linked) underlying the criteria used by PLS to detect common structures.
3.3. Variability
We are pleased that RBL point out that the sequence of one dimension of the dyad could be more heterogeneous than the other. This observation is not merely a methodological nuisance but has a substantive basis. Changing life-courses are exhibiting dual trends of individualization and standardization (relevant when studying parent-child dyads); women typically have more variable life-courses than men (relevant for couple dyads). Therefore, choices of weights could be consequential and should be considered as a routine robustness test, just as RBL do in their analysis.
4. Validity and Uncertainty
RBL clearly and succinctly lay out the processes of implementing GIMSA step by step. They also conduct several robustness tests to examine sensitivity. Questions remain, however, regarding validity and the comparative advantage of this method. For example, how well does it perform as compared with multichannel sequence analysis (Gauthier et al. 2010), at least for contemporaneous sequences such as those of couples? Future studies could be conducted to systematically evaluate the behavior of GIMSA under alternative scenarios.
We have a related comment about uncertainty. This is not a limitation unique to GIMSA but is shared by almost all classification methods. Given the multiple stages involved in GIMSA—optimal matching, multidimensional scaling, PLS, and clustering—how to quantify uncertainty seems particularly relevant. Uncertainty from one stage can easily become magnified in the next. Understanding the magnitude of this problem is important for future users.
5. Conclusion
We see a wide range of application potential for GIMSA that would stimulate theory building and capture identifiable patterned dynamics in relationships. This would be valuable for tracking the career trajectories of dual-earner couples as well as corresponding life trajectories of parents and children. Such analysis could shed light on within-couple inequalities as well as the intergenerational transmission of advantage and disadvantage, even in times of rapid social change. There has been a surge of interest in studies of couples’ conjoint careers (e.g., Moen 2003) and of multiple generations (e.g., Mare 2011). GIMSA seems to be a good candidate for understanding these intricate relationships, at least as an exploratory tool. Similarly, it could be used to study how multiple dimensions of lives—work, family, residence, health—interrelate with one another. (RBL briefly note that this extension requires a replacement of PLS with multiple factor analysis. As we commented, however, temporality and analytic weights could be even more complicated and critical for s≥3 cases.) Nevertheless, GIMSA is an important and useful first step.