1. Tailoring interventions for those in most need: Grouping employees by burnout and engagement Carolyn Timms and Paula Brough Griffith University
2. Rationale Most engagement and burnout research is based on aggregated data and uses statistical techniques that are out of reach of ordinary practitioners and management personnel. An important limitation of our continued dependence on quantitative research could be that individual perspectives may be lost or overlooked in broad sweeping strokes provided by statistical analysis.
3. Aims To group respondents according to their specific burnout and engagement responses To identify the resulting group similarities of workload, job control, reward, community, fairness and values (Areas of Work-life Survey, AWS). Overall aim: to illustrate the work perceptions of the most at-risk workers, with a view to tailoring interventions specifically for the employees in most need.
4. Method Two independent samples of employed workers in Australia who responded to a survey. Sample one (n = 953) were employees of an education union Sample two (n = 260) was a mixed group of employees representing a number of organizations.
5. Demographics Sample two Sample one N= 953 All members of an education union 701 (74%) female & 252 (26%) male Mean age 45 to 49 year age banding Mean hours worked =46 hpw (SD = 12.88) N=260 Various professional groups 155 (60%) female & 105 (40%) male Mean age 40-44 year age banding Mean hours worked 45 hpw (SD = 12.55)
6. K-means cluster analysis Previously used in medical research to identify individuals in most need of tailored interventions Used in molecular biology for gene expression data analysis Current research used a predetermined number of five clusters to represent five points in the continuum between the most engaged and the most burned out workers.
7. Used to define clusters The Oldenburg burnout Inventory (OLBI) Exhaustion & Disengagement The Utrecht work engagement scale (UWES) Absorption, Vigour & Dedication The research sought to determine people who were most alike on these variables to group them into clusters. Standardised – z scores
13. Differences between clusters and the AWS According to Clatworthy et al. (2005) it is important to use some variables that have not been used in defining the clusters in order to demonstrate that the clustering is valid. All AWS variables demonstrated significance between clusters (MANOVA analysis).
14. Used to confirm differences between clusters The Areas of Worklife Survey (AWS) Workload, reward, control, community, fairness and values. Used to investigate differences between the groups identified in the cluster analysis.
17. Findings Five groups that differed from each other on burnout and engagement were found. These groups demonstrated different patterns of response in regard to the AWS, which provided an indication of their workplace experience.
18. Findings People can experience aspects of engagement (e.g. absorption) and burnout (e.g. exhaustion) simultaneously. Further investigation may find some interesting interactions between these variables. The research has demonstrated how the needs of workers differ and suggests that this method of analysis may provide some suggestions as to how more effective interventions may be tailored.
19. Examples of possible interventions using Cluster Analysis Identification of a group in the workplace such as the ‘under-pressure group’ (who reported moderate engagement scores as well as exhaustion) suggests that interventions addressing workload and work-life balance issues may be more effective strategies for improving the intrinsic motivation of these workers.
20. Examples of possible interventions using Cluster Analysis #2 Identification of large numbers of people whose responses are similar to the un-engaged group may suggest job enrichment schemes might be effective strategies. Identification of people who are burned out within an organisation suggests focused management education and training as well as improved organisational communication.
21. Conclusion The five groups identified in the current study have differing needs. This demonstrates that ‘one size fits all’ approaches to organisational interventions is not necessarily the most effective approach. Approaches that target these groups are therefore strongly recommended in the interest of improving individual and organisational outcomes.
Editor's Notes
Good afternoon everyone. This research uses similarities among people at a group level using K-means cluster analysis. The measures used to identify the groups were burnout and work engagement. The purpose for doing this was to demonstrate that it is possible to locate groups within a population and thereby target organisational intervention to be most effective.
This gap between research and practise has been identified by several researchers and includes aspects such as 1. Lack of knowledge among practitioners of sophisticated statistical techniques (Saari and Judge, 2004).2. Unfortunately what ends up happening is that practitioners then do no utilize the most salient of research findings and outdated knowledge is perpetuated. For example burnout may be viewed as the weakness of an individual, leading to individual intervention, rather than interventions that focus on the environment in which they are working.One important disadvantage of statistical techniques using aggregated data is that such an approach overlooks the insights to be gained from specific individual perspectives – which is often what practitioners can most easily relate to.
Survey
The majority of respondents were either professionals (n = 115, 44%) representing occupations such as journalism, nursing or engineering; or associate professionals (n = 81, 31%) drawn from occupations such as insurance broking, enrolled nursing and information technology programming. The majority of the remaining respondents (n = 46, 18%) worked in various clerical occupations. Respondents from traditional blue collar occupations such as the trades and manual work represented 7% (n = 18) of survey respondents.
Predetermined number of clusters this approach was called McLachlan (1992, p. 32) as a “mixture-likelihood based approach to clustering. According to McLachlan it is of considerable utility because it assumes a well-defined model where data can be classified according to existing theory.
The Oldenburg Burnout Inventory (OLBI; Demerouti, Bakker, Vardakou & Kantas, 2003) Two components: exhaustion and disengagement (measured on a 4 point scale)The Utrecht Work Engagement Scale (UWES; Schaufeli & Bakker, 2003) was employed to assess individual levels of work engagement. The UWES consists of three subscales: dedication (e.g., “I am proud of the work that I do”), vigour (e.g., “At my job I feel strong and vigorous”) and absorption (e.g., “Time flies when I am working”)Measured on a 7 point scale. We used standardised scores because (1) the UWES and OLBI measures had differing scales and (2)because some measures of similarity are sensitive to differences in the variance of variables
This is a graph of the groups identified in the cluster analysis with sample 1. We used standardised scores because (1) the UWES and OLBI measures had differing scales and (2)because some measures of similarity are sensitive to differences in the variance of variablesGroup 1 we called the empowered group because they indicated that they were very engaged in their work and were not burned out.Group 2. we called the under pressure group – because right up there with their engagement variables of absorption was exhaustion.Group 3. we called the Unengaged group – this sounds more ambivalent than dis-engaged.Groups 4 & 5 different degrees of burnout – represented 35% of sample 1.
Comparison of Sample 1 and Sample 2. Basically this just reveals that the structure of the groups remains reasonable stable between the two samples. E.g. # The empowered group# the two burnout groups Most differences occurred with the unengaged group.
Comparison of Sample 1 and Sample 2. Basically this just reveals that the structure of the groups remains reasonable stable between the two samples. E.g. # The empowered group# the two burnout groups Most differences occurred with the unengaged group.
Comparison of Sample 1 and Sample 2. Basically this just reveals that the structure of the groups remains reasonable stable between the two samples. E.g. # The empowered group# the two burnout groups Most variation between samples occurred with the unengaged group.
Work environment. Six aspects of the work environment were assessed by six subscales from the Areas of Work-life Survey (AWS; Leiter & Maslach, 2006). All responses are recorded on a five point scale from 1 (strongly disagree) to 5 (strongly agree).