Beyond Transitions: Applying Optimal Matching Analysis to Life Course Research

(2008) - Peter Martin, Ingrid Schoon, Andy Ross

Journal: International Journal of Social Research Methodology
Link:: http://www.tandfonline.com/doi/abs/10.1080/13645570701622025
DOI:: 10.1080/13645570701622025
Links::
Tags:: #paper #NCDS #SequenceAnalysis #LifeCourse #BCS
Cite Key:: [@Martin2008]

Abstract

Life course researchers have increasingly explored optimal matching analysis (OMA) as a tool for the analysis of sequences, such as sections of people’s status biographies. OMA is usually employed in combination with cluster analysis (CA) to create classifications of sequences. In this article, we introduce an analytic strategy that allows assessing the classification’s internal validity. Using ideal typical sequence representations, we test different cluster algorithms and are able to optimise the fit to the data. An application analyses economic activity sequences collected for two British birth cohorts born in 1958 and 1970, investigating historical changes in passages to adulthood. The results suggest that passages into adulthood have become more diverse since the 1970s. The analytic strategy produced a classification with better fit than straightforward CA.

Notes

“OMA is usually employed in combination with cluster analysis (CA) to create classifications of sequences.” (Martin et al., 2008, p. 179)

“An application analyses economic activity sequences collected for two British birth cohorts born in 1958 and 1970, investigating historical changes in passages to adulthood. The results suggest that passages into adulthood have become more diverse since the 1970s” (Martin et al., 2008, p. 179)

“In his seminal text, Elder (1985) defines transitions as ‘changes in state that are more or less abrupt’ (pp. 31–32).” (Martin et al., 2008, p. 180)

“Although the methods of investigation vary, a quantitative analysis of survey data has often used a series of ‘marker states’ (such as financial independence, marriage, full-time employment and living outside the parents’ home) to signal the completed transition to adult status (Berrington, 2001; Osgood, Ruth, Eccles, Jacobs, & Barbe, 2005; Shanahan, 2000).” (Martin et al., 2008, p. 180)

“In operational terms, we can define a passage as a sequence. Whilst a transition denotes a single status change, ‘sequence’ refers to a temporally ordered series of status positions; in other words, a sequence embodies a section of a person’s status biography (Levy, 1977; Pollock, Antcliff, & Ralph, 2002; Sackmann & Wingens, 2001).” (Martin et al., 2008, p. 180)

“OMA proceeds by calculating the dissimilarity between two given sequences.” (Martin et al., 2008, p. 181)

“Dissimilarity is defined as the minimum ‘cost’ of transforming one sequence into the other. The cost, in turn, is a function of the number of insertions, deletions and substitutions needed for the transformation. Calculating the transformation costs for every possible pair in the sample yields a dissimilarity, or distance, matrix. The distance matrix is then subjected to further processing; usually, researchers have used cluster analysis (CA) to classify the sequences (an alternative to CA is multidimensional scaling; see Halpin & Chan, 1998, for an application)” (Martin et al., 2008, p. 181)

“Cluster analysis is a convenient method for creating classifications and has been used successfully in many scientific disciplines (Everitt, Landau, & Leese, 2001). However, it has significant weaknesses. These include the subjectivity of cluster enumeration, the existence of different clustering algorithms that produce substantially different classifications (Everitt et al., 2001), and multiple solutions in data with ties (Morgan & Ray, 1995). We shall expand on these problems below.” (Martin et al., 2008, p. 181)

“he post-war years’ extraordinary economic growth, coupled with a mood of progress-optimism (Hobsbawm, 1994), provided life chances that made these trajectories seem both attainable and desirable for many people” (Martin et al., 2008, p. 182)

“Several authors have argued that standardised trajectories into work and family have been broken by changing gender relations, expansion of the education system, a decoupling of educational qualifications and professions, and the increased risk of youth unemployment (Buchmann, 1989; Dewilde, 2003; Furlong & Cartmel, 1997; Hareven, 2000).” (Martin et al., 2008, p. 182) Broken comparatively? Sounds a bit off to me

“first, authors have doubted whether post-war transitions were as straightforward as the argument suggests (Goodwin & O’Connor, 2005). Second, scholars rightly point to a lack of rigorous empirical evidence (Shanahan, 2000).” (Martin et al., 2008, p. 182)

“October is a useful month insofar as it is relatively free from seasonal variation, and marks the beginning of the academic year, so that full-time students will be recorded adequately (Anyadike-Danes & McVicar, 2003; Rindfuss, 1991). We differentiate six activity states: full-time employment (F), part-time employment (P), full-time education (E), government training (T), unemployed seeking work (U) and out of the labour force (O). These six1 activities constitute the state space: they are the basic elements from which our sequences are constructed.” (Martin et al., 2008, p. 183)

“However, complete sequences were only available for 20,399 individuals (55%). We employed three procedures aimed at increasing the analysis sample size and at avoiding severe non-response bias; these procedures included replacement of missing entries, utilisation of incomplete sequences and a weighting scheme. They are summarised in Appendix A.” (Martin et al., 2008, p. 183)

“Cluster enumeration always bares a subjective element (Aldenderfer & Blashfield, 1984; Everitt et al., 2001). This might be in the nature of classifying: classifications are human constructions, and the categories we use to classify are usually merely convenient pigeon holes that order a chaotic world.” (Martin et al., 2008, p. 185)

“Ta b le 2 Classification of Economic Activity Sequences” (Martin et al., 2008, p. 188)

“Moreover, almost all NCDS men (96%) can be classified into one of the six largest groups (Nos. 1–6). In contrast, the six largest BCS70 groups (Nos. 1–4, 6 and 8) account for only 90% of men.” (Martin et al., 2008, p. 189)

“lthough the results are slightly less clear among women, they also suggest an increase in diversity.” (Martin et al., 2008, p. 189)

“Ta b le 3 Percentages of Group Membership by Gender and Cohort (Weighted Sample)” (Martin et al., 2008, p. 189)

“there is an increase in the number of distinct pathways into full-time employment: the paths via government training, and via an intermediate period of ‘return to education’ are ‘new’. Also, a small but increased minority of ‘late starters’ has a delayed entry into full-time work for reasons other than fulltime education or training.” (Martin et al., 2008, p. 190)

“This speculation reveals a weakness of our data: at any one time point, they only record a single activity state, although in reality people can occupy a combination of statuses.” (Martin et al., 2008, p. 190)

“It is important to note that non-normative and unusual sequences do not in themselves indicate disorderly passages. For example, the only group that would seem to fit the ‘yo-yo’ movement in and out of adult roles described by the European Group for Integrated Social Research (2001) are those who return to education.” (Martin et al., 2008, p. 191)

“Our conclusions are limited by the focus on employment careers only. For a thorough analysis of passages into adulthood, other status indicators (relating to partnership, parenthood and housing) would need to be taken into account.” (Martin et al., 2008, p. 191)