With 703 million older people worldwide, and the rapid increase in global average life expectancy (from 64.6 years (1990-1995) to 72.3 years (2015-2020)), there is significant pressure on governments to protect the health of ageing populations (SNIPH, 2006; WHO, 2018). Within western developed nations this has led to a strong focus on goal-orientated concepts related to positive health in later life, such as healthy, positive and active ageing (Beard et al., 2016; Dizon, Wiles, & Peiris-John, 2019). The emergence of these ideas has given rise to discourse around "...developing and maintaining the functional ability that enables well-being in older age" (WHO, 2015), and the necessary resources and capacity to support older people to pursue what is meaningful to them. These more positive aspects of health amount to a sense of 'ontological security', where the individual strives for certainty and the belief that they can be active participants within society (Giddens, 1991). However, such constructs are not easily demarcated for a heterogeneous older population and have attracted criticism for being individualistic (Dizon

Although there are a number of approaches to constructing a measure of multidimensional social exclusion in later life, theoretical and methodological challenges exist around the aggregation and weighting of constituent indicators. This is in addition to a reliance on secondary data sources that were not designed to collect information on social exclusion. In this paper, we address these challenges by comparing a range of existing and novel approaches to constructing a composite measure and assess their performance in explaining social exclusion in later life. We focus on three widely used approaches (sum-of-scores with an applied threshold; principal component analysis; normalisation with linear aggregation), and three novel supervised machine-learning based approaches (least absolute shrinkage and selection operator; classification and regression tree; random forest). Using an older age social exclusion conceptual framework, these approaches are applied empirically with data from Wave 1 of The Irish Longitudinal Study on Ageing (TILDA). The performances of the approaches are assessed using variables that are causally related to social exclusion.

The measurement of the complex, multidimensional and dynamic concept of old-age social exclusion has been constrained due to theoretical and methodological challenges as well as a reliance on secondary data sources not designed to collect social exclusion indicators. Limitations in measuring social exclusion in later life hinder the expansion of our empirical and conceptual understanding of social exclusion. In this paper, we seek to address these limitations by developing a composite measure of old-age social exclusion using three methods: 1) normalisation through re-scaling with linear aggregation, 2) a sum-of-scores approach with an applied threshold and, 3) classification and regression trees (CART), a machine learning approach. Using the conceptual framework of old-age exclusion presented by Walsh et al., (2017), these three approaches are applied empirically with data from Wave 1 of The Irish Longitudinal Study on Ageing (TILDA). The measures are assessed in terms of their ability to explain a validated measure of psychological well-being. Results suggest that despite the challenges associated with secondary data and measurement techniques that implicitly measure social exclusion, the newly proposed composite measure computed using CART performed better than the other two measures which are more prevalent in the literature.

No abstract

No abstract