A BULLYING TYPOLOGY? ESTIMATING THE PREVALENCE OF BULLYING WITH A LATENT CLASS CLUSTER MODEL Guy Notelaers, Leuven University, Research Center for Stress, Health and Well-being, Belgium and Federal Governement, Dep. of Labour, DIOVA-DIRACT Jeroen K. Vermunt, Tilburg University, Department of Methodology, Netherlands Hans De Witte, Leuven University, Research Center for Stress, Health and Well-being, Belgium Stale Einarsen, Bergen University, Department of Psychology, Norway Work, Stress and Health, 2006. The Sixth International Conference on Occupational Stress & Health. American Psychological Association. Miami .
1. Brief overview 2.Measuring bullying at work with the Negative Acts Questionnaire (Einarsen, et al., 1997) and identification of victims 3.Development of latent class cluster solution 4. Data & Measurements 5. Results - LCA : six clusters according to exposure to negative acts - Construct validity - Predictive validity 6. Conclusion
2. NAQ: measuring victims ( opera-tional criterium 1 act/week) How many times have you been subjected to the following acts during the last six months? never now once a once a and then month week -Someone withholding necessary information so that your work gets complicated -ridicule or insulting teasing -ordered to do work below your level of competence -being deprived of responsibility or work tasks -gossip or rumors about you -social exclusion from co-workers or work group activities -repeated offensive remarks about you or your private life -verbal abuse -hints or signals from others that you should quit your job -physical abuse or threats or physical abuse -repeated reminders about your blunders -silence or hostility as a response to your questions or attempts at conversations -devaluation of your work and efforts -neglect of your opinions or views -practical jokes -devaluation of your rights and opinions because of your age -Exploitation at work, such as running private errands -reactions from others that you work too hard
Problems with operational criterion Content low correspondence with self rating (Zapf, et al., 2003) high discrepancy between criteria and self rating in estimated number of victims 10-15% against 2-4% (Zapf, et al., 2003) criterion is only a snap shot that cannot capture the process of bullying, the stages in that process as implied by the definion of bullying Methodological Assumption that all negative acts have same properties, are equal to determine who is a victim do other answers reveal nothing else (are those answers really not significant, allowing them to be just ‘nulls’) Statistical a sumscore over all items implies that measurement is unidimensional ( design of the NAQ 4DIM (Einarsen & Raknes, 1997) or 2DIM (Einarsen & Hoel, 2001) recoding is reducing variance
3. Latent Class Cluster Analysis Is a systematic methodology to assign individuals to groups with whom they share properties... To create typologies.... Aim: to describe with as few latent classes as possible the relationships between indicators and latent trait Starting point: everybody resides in the same class Subsequently additional classes are added until a measurement model is attained that statistically fits the data (Bayesian Information Criterion)
4.1 Data Between 2003 and 2004, 6 175 questionnaires were collected from approximately 18 companies that were investigating stress and well-being at work The average age of the respondents is 41 years (std=10.7) . The average seniority is eleven years (std deviation = 10,3). 57% of the respondents completed a Dutch and 43% a French questionnaire. 48% private sector / 25% within the government or government institutions / 27 % health care institutions 9% blue collar, 31% white collar, 4% social worker, 13% nurses, 21% civil servant, 10% lower managerial and 11% managerial position 78% of the respondents had a fixed term contract, while 15% were on a temporary contract and 7% had other types of contract
4.2 Measurement instruments Negative Acts Questionnaire (Einarsen et al., 1997) Construct validity: Are you being bullied (10 companies) Have you been bullied during the last six months (2 companies) Legal definition (4 companies) Predictive validity (all companies) Strains from the VBBA (van Veldhoven et. al., 1994) Pleasure at work Worrying Recovery need Sleep quality
5.1.1Fit Measures from Latent Gold number of clusters BIC (LL) Npar L² red. L² df Bootstrap p-value Class. Err. 1 / nominal indicators 133571 48 56344 4294967247 2 / nominal indicators 119784 97 42135 0,25 4294967198 0,039 3 / nominal indicators 116320 146 38247 0,32 4294967149 0,075 4 / nominal indicators 114903 195 36407 0,35 4294967100 0,002 0,097 5 / nominal indicators 114243 244 35323 0,37 4294967051 0,008 0,11 6 / nominal indicators 113911 293 34567 0,386 4294967002 0,036 0,13 7 / nominal indicators 113790 342 34023 0,396 4294966953 0,064 0,14 8 / nominal indicators 113831 391 33640 0,4 4294966904 0,06 0,15 9 / nominal indicators 113929 440 33315 0,41 4294966855 0,038 6 / 3 local depend /nom ind 113107 320 33530 4294966975 0,112 0,17 5 / 3 local depend / nom ind 113272 271 34118 0,39 4294967024 0,07 0,138
5.1.2 Result: mean conditional probabilities not bullied at all not bullied neither / nor latent bullying work related bullying victim proportion 0,353 0,277 0,17 0,09 0,08 0,032 order of latent clusters 1 2 3 4 5 6 never 0,927 0,718 0,638 0,31 0,551 0,152 now and then 0,067 0,246 0,336 0,58 0,225 once a month 0,004 0,024 0,017 0,117 0,217 once a week or more 0,002 0,013 0,009 0,03 0,107 0,32
5.1.3 Results: closer look at meaning of clusters
5.2. Construct validity 1 (row%) subjective method clustermodel operational criterium Are you being bullied during work? not at all never in-between latent bully-ing work-related victim not a vic-tim vi-ctim total per-cen-tage 34,7 31,4 11,7 4,48 5,31 0,73 73,23 15,16 88,3 sometimes 0,7 1,43 1,83 3,22 1,33 1,56 5,687 4,323 10,1 often 0,03 0,17 0,4 0,23 0,399 0,831 1,23 always 0,1 0,033 0,333 0,37 total percent. 35,5 32,9 13,7 8,13 7,04 2,79 79,35 20,65 100
5.2. Construct validity 2 (row%) subjective method clustermodel operational-criterium Belgian legal defintion not at all never inbetween latent bullying work-related victim not a victim total per-centage no, never 41,9 9,86 21,3 3,05 0,9 76,8 0,36 77,1 yes, seldom 1,25 5,2 2,15 11,5 12,4 yes, sometimes 2,33 4,66 1,08 6,82 2,69 9,68 ja, +1x / week 0,54 0,18 0,72 total percent. 43,5 13,3 28,9 2,51 1,97 95,3 4,67 100
5.3. Predictive Validity (z-scores) Latent Cluster model Operational criterion Not at all bullied Not bullied In between Latent bullying Work related bullying Victim Not a victim of bullying Victim of bullying Pleasure at work 0,32 0,02a 0,07a -0,37 -0,62 -1,03 0,16 -0,53 Recovery need 0,31 -0,00a 0,01a -0,44b -0,45b -0,88 0,11 -0,36 Worrying 0,23 0,00a -0,35b -0,38b -0,72 0,09 -0,33 Sleep quality 0,29 0,06 -0,11 -0,45a -0,37a -1,01 The LC approach shows following test statistics: pleasure at work: F=125(5, 4934); p < .001; recovery need: F=98 (5, 4947); p < .001; worrying: F=59 (5, 5020); p < .001 and quality of sleep: F=98 (5, 909); p < .001. The F-statistics of the operational method are: pleasure at work: F=454 (1, 4929); p < .001; recovery need: F=191(1, 4943); p < .001; worrying: F=154 (1, 5015); p < .001and quality of sleep: F=203 (1, 4904); p < .001. Legend: Superscripts indicate pairs that are not significantly different. Consequently all other row wise comparisons are significant at a .0001 level.
6. Conclusion Latent class analysis permits a more precise measurement of bullying at work 3 non bullying cluster 3 bullying cluster with one victims cluster The six latent class solution has a higher level of predictive validity and a higher agreement with self ratings (construct validity) than the operational criterion