ACSPRI Conferences, RC33 Eighth International Conference on Social Science Methodology

Font Size:  Small  Medium  Large

An application of statistical learning to characterisation of 44 international healthcare accreditation organisations

Max Moldovan, Charles Shaw, Wendy Nicklin, Ileana Grgic, Triona Fortune, Stuart Whittaker, Nicholas Nechval, Marjorie Pawsey, David Greenfield, Reece Hinchcliff, Virginia Mumford, Johanna Westbrook, Jeffrey Braithwaite

Building: Law Building
Room: Breakout 2 - Law Building, Room 026
Date: 2012-07-10 01:30 PM – 03:00 PM
Last modified: 2011-12-19


We collected the responses to a questionnaire distributed to 44 international healthcare accreditation organisations and containing 165 unique questions. The set of questions corresponded to 10 thematic subgroups covering policy and governance specifics, funding details, the scope of services provided, among others. We further defined a concept of organisational vitality and collected the related vitality scores for each organisation based on expert opinion. The resulting series was used as a supervisory variable in the generalised linear regression modelling type of analysis.

The analysis identified two organisational attributes from the broad Policy and Governance subgroup as most statistically associated with organisational vitality. We observed potentially confounded associations with an economic prosperity factor (represented by the purchasing power parity adjusted gross domestic product per capita series), a population size factor and a political legacy factor (i.e. whether or not an organisation is from a country in the post-communist bloc). When controlled for the most influential Policy and Governance attribute and the economic prosperity factor, being an organisation from a post-communist bloc country increased the chances of falling into the less vital group by about 8 times. We do not establish any conclusions on causality at this stage, but have noticed this empirical evidence.

Realising that the two Policy and Governance attributes were the most statistically noticeable in the supervised statistical learning type analysis, we further analysed this subgroup using one of unsupervised learning techniques. In particular, the hierarchical clustering technique was applied to 10 categorical Policy and Governance attributes coded as asymmetric binomial variables. We computed the dissimilarity matrix based on Gower’s (1971) similarity coefficient because this method is capable of accommodating categorical as well as continuous types of data. The choice was motivated by the intention to extend the same type of analysis to exogenous variables, such as country population sizes or political system descriptions. This analytical approach has identified three clusters broadly grouping countries according to their vitality scores. We believe the groups correspond to the three distinct Policy and Governance styles that can be at least in part responsible for organisational success. If confirmed, the finding can have a wide range of implications to the establishment, funding and continuous improvement of international healthcare accreditation organisations.

Gower, J. C. (1971) A general coefficient of similarity and some of its properties. Biometrics 27, 857–874.