You are on page 1of 20

8-------------------------------The Validity of the Maslach Burnout Inventory in Three National Samples

DIRK ENZMANN, WILMAR SCHAUFELI AND NOELLE GIRAULT

In a recent article on "occupational morbidity and burnout. lessons and warnings for HIV/ AIDS carers" Miller (1991) posed ten questions havmg priority to be answered. The first is: "Does burnout really exist in HIV care?" (Miller, 1991, p. 447). Prerequisites for giving an answer to this question are a clear and consistently used definition of burnout and its operanonalization given by a well developed measure. A multitude of publications exist that stress the risk of burning out in AIDS health care. Many of them merely rely on vague descnptions ofthis "phenomenon" (e.g. Bolle, 1988; Howze, 1988; Roozenburg, 1988, Bor, 1989; Lelsourdars, 1989; Voelcker, 1991, Broadhead, 1992). The vast majority of those researchers, however, who study the issue of burnout in AIDS health care by means of standardized questionnaires, employ the Maslach Burnout Inventory (MBI; Maslach and Jackson, 1981, 1986), which lS - together with the Burnout Measure (BM; Pmes, Aronson and Kafry, 1981, Pines and Aronson, 1988) - the most frequently used mstrument in research on burnout (cf. Schaufeli, Enzmann and Girault, 1993). In doing so they adopt the multidimensional defirution of burnout according to the subscales of the MBI.
"Burnout IS a syndrome of emotional exhaustion, depersonalization, and reduced personal accomplishment that can occur among individuals who do 'people work' of some kind The Emotional Exhaustion subscale assesses feelings of being emotionally overextended and exhausted by one's work The Depersonalization subscale measures an unfeeling and Impersonal response towards recipients of one's service, care, treatment, or mstruction The Personal Accomplishment subscale assesses feelings of competence and successful achievement In one's work With people" (Maslach and Jackson, 1986, pp 1-2)

132

Dirk Enzmann et al.

The demand for consistency in the definition of burnout and the prevailing use of a specific instrument puts one on the horns of a dilemma. On the one hand it provides comparability between studies and cross-validation of findings, on the other hand It makes it difficult to Improve the psychometric properties of the measure or to adopt it to new theoretical developments. Thus, if the internationally prevailing use of the MBI is to have a favourable effect on the research on burnout, it is especially important to ascertain whether the MBI is psychometrically valid in different translations and populations. In order to answer the first priority question (does burnout really exist m AIDS health care?) and to compare the findings internationally, at least three steps of research are necessary: the reliability and vahdity of the measure has to be ascertained (1) internationally and (2) in the particular samples of AIDS health carers compared with other human services professionals, (3) the burnout scores in AIDS health care have to be compared with scores in other occupational fields and to other studies in the field of AIDS The importance of the first Issue is not limited to international research on burnout in AIDS health care, but also extends to research on burnout in other fields of human services Whereas the data required to answer the second question are few and far between, there IS sufficient matenal to answer the first. Pertaining to the thud issue, only very few studies on burnout m AIDS health care are pubhshed. In this stage of research, it would be too early to give evidence on the true prevalence on burnout in the field of AIDS (cf Enzmann, Chapter ?). The purpose of this chapter ISto address the first Issue and to present findings about the reliability and cross-national validity of the MBI in three European samples of human services providers from the Netherlands, Germany, and France The fmdmgs serve to mdicate whether it is possible to translate the MBI into different languages, while preserving ItS psychometnc properties. Three questions will be addressed: (1) Are the three MBI scales reliable indicators of burnout", (2) Is the MBI a valid measure of burnout?; and 0) Is the factorial structure of the MBI invariant across nations? These questions pertain to interItem reliability, construct validity, and factorial validity, respectively. Before considenng these questions, we will describe bnefly the three samples and the measures that were used for validation purposes

SAMPLES All three samples consisted of human services professionals. The Dutch sample is relatively homogeneous and consists of nurses who work m general hospitals (20%), mental hospitals (25%), nursmg homes (18%), community health (23%), mstitutions for the mentally handicapped (11%), and other health care mstitu-

Internal consistency coefficients for the BM were very high in the Dutch and German samples (both Cronbach's a were . The remaining 17% was employed in other wards where the work is particularly stressful. either In hospitals (20%. such as paediatrics. WIth several additional measures. followed by nurses 02%) and personnel officers (21 %). Most subjects were female (73%). translated versions of the MBIwere employed. The German sample (n = 266) included general hospital nurses (24%). Sixtyone percent of the sample was male. The mean age was 34.such as occupational therapists and counsellors (11%).93). .The Validity oj the Maslacb Burnout Inventory 133 tions (3%). r. and a remaining group of various professionals.e intensive care units and oncology wards) or in mobile emergency teams (63%. The nurses' mean age was 33 2 years (Sd 6 5). 27% were male. and Study D (n = 351).9 years (Sd 8. Most professionals in the total sample were physicians (46%). and youth centres). psychologists (9%). 23% was younger than 25. and 39% was male. agencies for family care. or with cancer patients 03%) (they represent the first respondents of the ABBA-project. Data were collected independently in four studies: Study A (n = 68). The majority of the sample (61%) was female. Study C (n = 118). The health professionals worked m particularly demanding jobs. MEASURES In the three national samples. StudyA included 121 professionals from public Institutions (e g general and mental hospitals. This was considered more appropriate in the medical context in which the inventory was employed. see Kleiber and Enzmann m Chapter ?). About half (52%) of the sample was between 25 and 35 years of age.8) ThIS sample was combined from two studies. geriatric nurses (9%)' social workers (21%).e ambulances or fire bngade). the BM was completed as an alternative burnout instrument. and 39% was female. In the Dutch as well as In the German sample A. educators (8%). For the current chapter these samples were combined into one composite sample (n = 659). i. With the elderly (27%). physicians 05%). The French sample (n = 267) consisted of health professionals (790/0) and personnel officers (21%). The term "recipient" was replaced by "patient" in the Dutch and French versions. and 25% was older than 35. A subs ample of 58 physicians and nurses who worked in ambulance emergency teams were studied in more detail. Study B (n = 122). Study B included 145 professionals who work in different public Institutions with HIV-mfected and AIDS patients (60%).

(2) those who reacted in a passive way With depressive moods. The nurses' level of reactivitywas. withdrawal from clients and coworkers (a = . and looking for social support from colleagues (n = 26).73) was measured with a self-constructed thirteen-item invemory (sample item: "I don't bust a gut for clients who are unwilling to cooperate"). respectively. 1979).~ssessed by employing the corresponding subscale of the Strelau Temperament Inventory (a = . 1972).83) and mental distress (a = . In sample B four additional measures were included. In sample A the level of depression (a = . Disillusionment (a = .87) was assessed by an inventory developed by Mohr (1986). over involvement m administrative duties). Rush. even when extra activities are no longer required. withdrawal behaviours towards patients and colleagues were assessed (e. Irritability refers to a particular kind of fatigue that is caused by prolonged activity. 1970) and by a scale developed by Warr and Jackson (1982).88) was assessed by Beck's Depression Inventory (BDI. 1983).82) and selfesteem (a = . "Vragenlijst Organisatie Stress" (VaS. Finally. On the basis of their answers.68) (Strelau.60) was measured with a self-constructed seven-item inventory (sample item: "Only those persons who are willing to make personal sacrifices for the task should choose a job similar to mine"). asking for a time-out for recuperation. Reactivity refers to the intensity of an individual's reaction IO both external and internal stimuli. emotional and behavioural reactions that occurred after stressful penods at work were assessed by employing a freeresponse format Respondents were classified by the third author into two a priori categories: (1) those who showed active reactions such as crymg. bursts of anger. not all measures were employed in all of the subsamples. and the individual feels tense. Bergers. Marcelissen and de Wolff. anxiety. the individual activates his or her additional resources. Job satisfaction (a = . The individual is forced to remain active. and unable to relax. respondents were categorized . Level of organizational commitment (a = .90) was assessed by the Orgaruzational Commitment Questionnaire (Mowday. irritability (a = . 1986). resources are depleted. compulsive thinking. Beck.73) were measured by the N-scale of Eysenck's Personality Questionnaire (Eysenck. nervous. several additional measures were obtained. In the Dutch sample). excitation. Steers and Porter. "escaping" in rest rooms. Neuroticism (a = . In an attempt to meet the demands.80) was measured by a self-constructed six-item inventory (sample item: "I like being at my working place")' Finally. despite being fatigued. and withdrawal (n = 32) Moreover. Shaw and Emery. In the French subs ample (n = 58) . selfreported physical stress symptoms (a = .g cynicism. Psychosomatic reactions (a = 88) were measured with a scale developed by Mohr (1986) in a sample of industrial blue collar workers. For both German samples.82) were assessed by the corresponding scales of the Dutch adaptation of the Organizational Stress Questionnaire. inhibition.134 Dirk Enzmann et al Because the national samples were combined from several studies. As a consequence.

INTERNAL CONSISTENCY AND INTER-SCALE CORRELATIONS Table 1 shows the internal consistency coefficients. 1978. German. and those who intended to contmue m their Jobs (n = 32). PA = personal accomplishment = depersonahzation. Again two groups were considered: those who intended to quit (n = 26). . ex 82 88 71 . the respondents in this subsample were asked whether or not they intended to leave their jobs in the mobile emergency teams within one year.The Validity of the Maslach Burnout Inventory 135 into two groups: those who withdrew from their patients and colleagues (n = 23). the internal consistencies of the MBI scales are satisfactory in every sample. In Table 2 the correlations are presented between the MBI scales In each TABLE 2 correlations in three national German (n = MBI inter-scale Dutch (n samples French (n = 535) 266) = 267) DP EE DP DP -043'" EE 052'" -0 25'" DP -0 15' DP EE 027'" -010 PA 0. 245). ex 86 . ex 90 . "p< 01. scale ex Emotional exhaustion Depersonahzation Personal accomphshment German (n = 266) French (n = 267) cor.58'" -040'" -017" Notes 'p < 05. p.84 79 ex 84 71 81 cor.83 84 ex 80 78 66 cor.88 72 77 According to the commonly employed criterion of a > . Because the value of Cronbach's a depends on the number of items included in the scale. Table 1 also presents the corrected a-value which is standardized for a test-length often Items.70 (Nunnally. according to the Spearman-Brown formula TABLE 1 (Cronbach's ex) of the MBI scales and French samples Dutch (n = 659) Internal consistency 10 the Dutch. EE = emotional exhaustion. "'p ~ 001. and those who did not (n = 35). particularly when the test-length is standardized (corrected a-coefficient). Finally.

the scales are substantively and almost evenly correlated.The inter-scale correlations differ slightly between the samples from the three countries. and discriminant validity (i. whereas correlations in the German and particularly in the French. physical.136 Dirk Enzmann et al sample. This was to be expected. to what extent does a similar factorial structure emerge in each sample/).e. The validity of the MBI is not questioned by these findings. CONSTRUCT VALIDITY In this section we discuss the construct validity of the MBI in each national sample separately. and Kafry (1981) define burnout as emotional. however. to what extent do different burnout invemories measure the same construct/). Two aspects are considered: convergent validity (i. DP accomplishment. EE = emotional exhaustion. because different additional measures have been included. samples are much lower However. the pattern of correlations is quite similar across the three samples: emotional exhaustion and depersonalization are fairly strongly and positively related. and mental exhaustion. the factorial validity of the MBI will be studied cross-nationally (i. Dutch Sample In order to study the convergent validity of the MBI. The correlations With depersonahzation and personal accomplishment are modestly positive and weakly negative. correlations were computed with the total BM score (see Table 3) TABLE 3 burnout and related concepts Relationships (Pearson's r) between (Dutch sample BM 059'" 078'" n = 659) EE Physical symptoms Mental distress BM 052'" 068'·' 075'" DP 026'" 043'" 043'" PA -018" -038'" -030'" Notes "p< 01. respectively. It appears that the BM score is highly positively correlated with emotional exhaustion. BM = burnout measure = depersonalization. because burnout is conceptualized multidimensionally by .e. to what extent can burnout be differentiated from related constructs?).e. In the next section. In the Dutch sample. "'p~ PA = personal 001. since Pines. whereas personal accomplishment is more weakly and negatively related to both other scales. Aronson.

two-factor analyses with scale means were performed in subsamples in which other relevant variables had been assessed.e. 1983). Stout and Williams. the discriminant validity of the emotional exhaustion scale with self-reported mental distress is relatively poor: 45% of the variance of both variables is shared. correlations of the MBI scales with mental distress are similar to those with the BM. personal accomplishment loads on the second component. Most remarkably. and physical stress symptoms. Table 4 shows the results of the principal components analyses after subsequent varimax rotation. (i. It follows that depersonalization. together with personality characteristics such as reactivity and self-esteem. Although the results of both factor analyses are not identical. In contrast. As can be seen from Table 5. Thus. without exception the correlations between burnout and mental distress are stronger than with physical stress symptoms. in both samples no burnout component emerged on which the BM and all scales of the MBI load exclusively. mental distress. This finding confirms the results from other similar factor analyses (Brookings et al. correlations with self-reported physical stress symptoms and mental distress). emotional . Physical stress symptoms are moderately correlated with emotional exhaustion and with the BM. neuroticism. therefore. Typically. Instead. some important similarities can be observed. Similar relationships between the two burnout instruments have been found in other studies (Corcoran. can be reasonably discriminated from self-reported mental distress and physical stress symptoms. However. but weakly with depersonalization and personal accomplishment. Table 3 also yields information about the discriminant validity of the MBI. 1988). German Sample Different sets of measures were used in both German subsamples. depersonalization. mental distress is less substantially related to both other MBI. Accordingly. validity is examined separately for each subsample In order to study the convergent validity of the MBI in Sample A. the three scales were correlated with the BM. In order to evaluate the discriminant validity of the MBI more thoroughly. MBIemotional exhaustion and the BM show a considerable overlap with these general indicators of ill-health. 1985. it was expected that correlations of the MBI with physical symptoms and mental distress would be weaker than with the BM. The discrimmant vahdity of the BM is even poorer: 60% of its variance is shared with mental distress. Dignam and West. scales. On the other hand. and particularly personal accomplishment. The first component is dominated by strains such as (emotionali exhaustion.The Validuy of tbe Maslacb Burnout Inventory 137 Maslach and Jackson (1986). 1986.

a) Not included In sample C Not included in sample D BM = burnout measure. EE = emotional exhaustion.46 0. depression.1% 0.46 -0.83 078 Sample D (n = 351) II -043 -0. EE = emotional exhaustion DP = depersonalization. As in the Dutch sample. --'p::.138 Dirk Enzmann et al. 001. TABLE 4 Factor analysis of scale means (Dutch subsamples) Scales Sample C (n = 188) II BM EE DP PA Physical symptoms Mental distress Reactivity Self-esteem') Neuroticism" Explained variance 0.81 075 035 -037 0.75 407% 245% Notes.66---057--- Notes "p < 05. PA = personal accomplishment Only factor-loadmgs ~ 1. Furthermore. TABLE 5 (Pearson's r) between burnout and related concepts (German sample A n =121) Relationships EE Irntabihty Depression Job satisfaction BM 048--040---052--069--- DP 029-015-047--023-- PA -036---041--038---041--- BM 068--0. irritability was expected to correlate more strongly WIth emotional exhaustion than with the other scales of the MBI.30 I are presented bl exhaustion correlated highly positively with the BM score. --p< 01.79 35.73 -037 0. Only the relationship between the BM and MBI depersonalization is somewhat weaker in the German sample.79 0. The three MBI scales were expected to correlate more strongly with the BM than with these three variables.34 084 0. and job satisfaction. the remaining scales of the MBI correlate only modestly or weakly WIth the BM. PA = personal accomplishment. BM = burnout measure The discriminant validity of the MBI was Investigated by correlating its scales with irritability. The results. presented In Table .77 0.52 0.1% 30.62 069 -0.67 089 -066 0. DP = depersonalization.

respectively. However.The Validity of the Maslacb Burnout Inventory 139 5. the BM. whereas personal accomplishment and depersonalization are most strongly related to disillusionment and withdrawal. and depression). Additional findings concerning the discriminant validity of the MBI come from Sample B Broadly speaking. respectively. these scales lack discriminant validity with such related concepts as irritability and depression. all correlations in Table 7 are moderate. Hence. Contrary to the expectation that all burnout scales would load on one component and the related concepts on other components. at the same time. These moderate relationships underscore the conceptual validity of the MBI scales. the BM. Emotional exhaustion reflects the person's level of generic stress and is therefore likely to be related to psychosomatic strains . only partly confirm these expectations. Job satisfaction loads substantially on the second-as well as the third component. this is not considered a negative result. because the BM correlates highly (and even more strongly than all the MBI scales) with irritability.2% 047 179% PA exhaustion. The remairung two components are dominated by depersonalization and personal accomplishment. DP = depersonahzation personal accomplishment Only factor-loadings ~ I 30 I are presented = A better understanding of the construct validity of the MBI can be obtained by a factor analysis of the scale means of the MBI. depression. the results of this analysis indicate the convergent validity of emotional exhaustion and the BM. and job satisfaction. However. irritability. depersonalization correlates more highly with job satisfaction than with the BM. Emotional exhaustion shows the highest correlations with psychosomatic symptoms and with commitment. the first principal component includes all strains (emotional exhaustion. As in the Dutch sample. it is concluded that the BM has a considerably lower discriminant validity than the MBI. TABLE 6 Factor analysis of scale means (German sample A n Scales BM EE DP PA Depression Irntabihty Job satisfaction Explained variance Notes EE = emotional II = 121) III 087 068 050 093 093 085 077 -038 389% -058 21. In particular. and the three related concepts (see Table 6). a result much like that found in the Dutch sample.

disillusionment. a factor analysis with the scores of the scales was performed (see Table 8). •••p::. DP = depersonalization.p< 01. EE = emotional exhaustion.140 Dirk Enzmann et at Relanonships TABLE 7 (Pearson's r) between burnout and related concepts (German sample B n =108) EE Psychosomatic reactions commitment DP 021* 031" -020' 044··· PA -040'" -048'" 029" -030'· Disillusionment Orgaruzational Withdrawal 054'" 0. which is obviously related negatively to disillusionment Again. '.40'" -040'" 039"· Notes "p < 05. The results from this analysis confirm the previous interpretations. 001. feelings of personal accomphshment result from a positive evaluation of one's work with recipients. and lack of commitment. DP = depersonahzation PA = personal accomplishment Only factor-loadings :? I 30 I are presented Depersonalization is considered to be a particular coping response that is dominated by distancing and withdrawing from contacts with recipients Finally. the lack of accomplishment is not restncted to the personal domain but also affects the organizational and professional domains The second component is dominated by depersonalization and withdrawal This confirms that withdrawal from coworkers and recipients on a . PA = personal accomplishment TABLE 8 Factor analysis of scale means (German Scales EE DP PA DISillusionment Organizational commitment Withdrawal Psychosomatic reacuons Explained vanance sample B n II = 108) III 060 087 -071 081 -0. This indicates that reduced personal accomplishment is accompanied by negative reactions towards the organization (diminished commitment) and by a loss of confidence in the effectiveness of one's professional role Accordingly.68 037 2650/0 035 073 254% 064 -031 088 201% Notes EE = emotional exhaustion. The first component is dominated by lack of personal accomplishment.

as a result of assessing different variables for each scale. Unfortunately. French. neither an alternative burnout inventory nor other instruments that measure related concepts were included in the French study. Girault. was sigruficantly higher than those of their colleagues who reacted in a more active fashion (tt'. which Illustrate the inadequacy of passive ways of coping. The mean emotional exhaustion score of the health professionals who reacted in a passive way to stress. levels of depersonahzation were significantly higher for professionals who tended to Withdraw from their patients and their colleagues.25. The expectation that emotional exhaustion would be associated With passive reactions to stress was confirmed. P < . structural equations modelling was employed. Moreover. and to the intention to leave the job (d. to withdrawal.e. the third component consists mainly of both strains that were included in the analysis. French Sample Unfortunately. the construct validity of the MBI was examined by relating the MBI scores to emotional and behavioural reactions. only two studies have employed a similar procedure (I. p < .The Validity of the Maslach Burnout Inventory 141 behavioural level goes along with negative attitudes towards recipients Finally.66.e. p < . than for those who did not (t(56)= 1. 1981).6. and German samples.88. This result confirms Maslach's (1982) original mterpretation of depersonalization as a way of mental distancing Finally. emotional exhaustion and psychosomatic symptoms. 1981-1986) can be reproduced In the Dutch. In order to test the fit of the original theoretical model. 1981. the validity of the MBI could not be studied analogously to the Dutch and German samples.001). i. 1989). confirmatory factor analysis). particularly with respect to the emotional component of burnout. As expected. feelings of professional inefficacy and demotivation are associated with the desire to leave the Job It is concluded that the three MBI dimensions were successfully validated in the French sample. respondents who indicated that they would like to quit their jobs had significantly lower personal accomplishment scores than those who intended to continue their jobs (t(56)= 5.05). In the French subsarnple (n = 58). Therefore.01). Pines and Kafry. THE CROSS-NATIONAL VAlIDITY OF THE MBI The question to be addressed in this section is the extent to which the origmal three-factorial structure of the MBI (Maslach and Jackson. have also been reported in other studies (Gillespie. using the LISRELVII computer program (Ioreskog and Sorbom. = 3. in one study seven items were added to the original . Similar results. as expected. 1989} Thus far.

so that our results can only be compared the findings of Gold et al. Because m the present study the sizes of the national samples differ considerably. According to this approach. n: df: mcrernental (Type 2) X2 X2 of target model X2 of null model degree of freedom of target model. They can be ordered according to their degree of restnction For instance. 1990). Equation 1).: A one-factor model that assumes that all Mfsl-rtems load on one single factor. This index does not provide information about the absolute fit of a particular model. the tradinonally employed g~odness-of-fit index (X2) strongly depends on sample size. a special strategy was followed. an oblique factormodel that allows the three dimensions of the MBI to correlate is less restrictive than an orthogonal or uncorrelated model. emotional exhaustion. MBI (Lee and Ashforth. Balla and McDonald. Mo corresponds to the hypothesis that there are Just as many uncorrelated factors as there are Items The factorial models of the MBI with which Mo IScompared are less restnctive. I e. At the basis of these model comparisons lies the most restrictive model: the socalled null-model (Mo) In the present case.142 Dirk Enzmann et al.: (1989). a so-called Type 2 incremental fit index (X2_J2) IS computed. three models were tested. It assesses the fit relative to a (nested) model in that particular sample. M. In order to investigate the cross-national factonal validity of the MBI. Each comparison of the null-model with one of these models results In an incremental fit index (X2-12) which can be compared with the corresponding fitindex of the other samples of different size Accordingly. with The Procedure of Fitting Models In structural equations modelling. 1986).t)/(n - dD (Equation 1) X2-12: t. 1988) (d. depersonalization. as well as between different samples. the goodness-of-fit of different models within one sample. and personal . which is scarcely affected by sample size (Marsh. Mz: The three-factor model used by Maslach and Jackson (1981. x2-I2 with = (n . which assumes Implicitly that the MBI items load on three uncorrelated factors. can be assessed. rather.

001. and the RMR (Root Mean Square Residual). and therefore also In a different factor-Interpretation) My A three-factor oblique model In which the three factors of M2are allowed to be correlated. It IS remarkable that the fit of Mz in the Amencan sample of Gold et ai. respectively. Low values of X2and RMR. Gold et al adapted the MBI slightly in order to make the instrument suitable for their student sample. However. 1989). (989) (X2-12= 771) is poorer compared to the fit obtained In the three European samples. An oblique rotation would probably have resulted In different factor-loadings. Table 9 shows (aside from the X2) additional fit-indices that are provided by LISREL. and the French samples. given that the MBI was constructed Via an inductive approach using an American population. in the Gold et al sample. The best relative fit of the three factorial models is found in the German sample. . one would not have expected the even worse fit of M. like the GFI (Goodness-of-Fit-Index). Perhaps these changes had a negative influence on the factorial structure of the instrument Moreover. The X2-12-values (which are least affected by sample size) are shown in Table 10. This is most probably caused by the large sample sizes. Indicating a rather poor absolute fit. unweighted items-scores are used for computing the scale-scores. followed by the Dutch. the American. (The X2-I2-values of the sample of Gold et al . (989) have been computed by the authors. Their factor-interpretation is based upon loadings that resulted from orthogonal rotation (The three scales are correlated.The Validity of the Maslach Burnout Inventory 143 accomplishment.) In each of the national samples the three-factor oblique model (M) fits the data Significantly better than the orthogonal model (M2). High values are indicative of a good model fit. Absolute and Relative Model Fit The results of the analyses are presented In Table 9 (model tests) and in Table 10 (model comparisons). given that the Amencan study used the untranslated English version of the MBI. because instead of factor-scores. the AGFI (Adjusted Goodness-of-Frt-Index). In each of the national samples.and AGFI-values are indicative of a good model fit (Ioreskog and Sorbom. while considenng a hypothesized factor structure. and high GFI. the probability levels of the X2statistic are less than. Such a confirmatory factor analytic approach IS particularly well suited to the study of complex constructs such as burnout because it has the capacity to analyze correlations among factors simultaneously.

the Moprovrdes information as to whether or not an item loads on the "right" factor. Improving the Model Fit LISREL provides information about which items in each national sample are responsible for the rather poor fit of the original three-factorial orthogonal model (M).669 . M] = one-factor model.870 421 . Mz M3 X2 447303 143859 1143. we computed a corresponding value (. d) Corrected by the authors. Mo = null model. (The Mo for a particular Item indicates the increase of the x2-statistic if that item would be "allowed" to load on another particular factor.~X2-12) which can be interpreted similarly to the X2_J2 statistic The Mos for Item twelve on the . onginal 54344 i It is important to note that researchers from other countries who want to study the cross-national validity of their MBI versions can compare the obtarned fit-indices with the values of Tables 9 and 10.078 293 122 171 095 German (n = 266) French (n=267) American" (n = 147) Notes pS 001 for all x'-statltstlcs.710 RMR 287 096 189 097 245 .83 50899 41678 161569 95128 51769 49546 162862 67447c) 53444d) 48326 df 231 209 209 206 231 209 209 206 231 209 209 206 231 209 209 206 GFI 345 736 832 . M3 = three-factor oblique model a Gold et at (989).) In order to compare the Mos across samples. M2 = three-factor orthogonal model. b) Corrected by the authors.144 Dirk Enzmann et al. TABLE 9 Test of MBI factonal model 10 three national samples Sample Dutch (n = 535) Model Mo M1 M2 M3 Mo M] M2 M3 Mo M. M2 M3 Mo M.848 864 542 713 856 860 336 636 747 764 AGFI 282 680 797 840 366 599 816 833 498 653 826 829 273b) 559 693 . ongmal 0 730. c) Corrected by the authors.91 196128 957. In other words.94 814.121 141 071 . ongmal 76447. Inspection and subsequent testing of the so-called modification indices (Mo) reveals that in each European sample a significant Improvement of fit would occur if Mz is re-estimated allowing item 12 ("I feel very energetic") to load additionally on emotional exhaustion.192 117 098 .

In the French sample.031.020. the other (minor) adaptations are country-specific. Mo = null model.880 051 472 781 . French. . M2 = three-factor orthogonal model.014 . The corresponding X2-I2-values are: . item sixteen ("Working with people directly puts too much stress on me". = and Mo = 38.795 . emotional exhaustion) tends to load on depersonalization (M. M3 = three-factor oblique model Gold et al (989). personal accomplishment) tend to load on the depersonalization factor (Mo= 30. ilX2_J2-I2 = . because "not feeling energetic" can be considered as just another symptom of exhaustion.008).12 32903 1003.0. = 12. and . and German samples respectively.020.012) Thus.2. all values have been computed by the authors a) x emotional exhaustion factor are 132. 29. The remaining significant modification indices seem to be rather specific for each of the national samples.-M3 Mo-M. a. Mo-M2 Mo-M3 M. A significant factor loading of item twelve on emotional exhaustion clearly makes sense psychologically. Item ten ("I've become more callous toward people since I took this job".0.1. In the Dutch sample. Mo-M2 Mo-M3 M2-M3 Mo-M. In the German sample. except for one common adaptation of the three-factor orthogonal model that leads to an improved fit (item twelve). 35.I " The Validity oj the Maslach Burnout Inventory 145 ! TABLE10 Compansons of MBI factonal models 10 three national samples Sample Dutch (n = 535) Companson Mo-M.3. items twelve and eighteen ("I feel exhilarated after working closely with my recipients". ilX2_J2= .1X2_J2 . = 22.672 771 805 034 German (n = 266) French (n=267) American" (n = 147) Notes pS 001 for all 2-stahtstlcs. ilX2I2 = .009.7. M. Mo-M2 Mo-M3 M2-M3 Mo-Mo Mo-M2 Mo-M3 M2-M3 X2 df 22 22 25 3 22 22 25 3 22 22 25 3 22 22 25 3 XQ2 303444 332909 3658. .2 for the Dutch. pepersonalization) tends to load on emotional exhaustion (M.007.45 145229 154450 9221 66441 109800 112023 2223 95415 109418 114536 5118 712 781 857 077 573 829 . = one-factor model. . respectively).

(1989) cannot be explained by their somewhat atypical sample (college students) and rephrased items. forthcoming). (In the French sample. GENERAL DISCUSSION AND CONCLUSIONS Particularly In the Dutch and German samples. Can the deviations from the theoretical model (M) that we found in the European samples be attributed to cultural differences with the American normative sample in which this theoretical model was developed? In other words. However. Enzmann. is the MBI cross-nationally invalid. or are the deviations independent of cross-national differences? If the poor fit of the model m the study of Gold et al. Second. a rigorous test of this hypothesis has yet to be done (Schaufeli et al. Golembiewski and his colleagues have also shown the congruence of MBI dimensions across . an interesting question emerges. (2) both models capture a substantial aspect of the data (model comparisons)... then their results point in the direction of factorial mvalidity of the original model. From the current analyses we conclude that the factorial validity of the MBI is not as convincingly demonstrated as one would have been expected from the literature. certain results are remarkably stable across the samples: (1) significant misfits occur for the orthogonal as well as for the oblique model (model tests). In sum. A test of the assumption that the misspecifications in the model are of a theoretical rather than of a cultural nature is in progress. we only compared three national samples. Girault and Maslach.146 Dirk Enzmann et al. or merely differences In translation can be held responsible for these somewhat disappointing results. The fact that the factorial model did not fit perfectly is most probably not caused by cultural bias of some sort. the samples consist of different occupational groups. but by misspecifications in the original model. (3) the oblique three-factor model fits the data significantly better than the Other models.) Moreover. However. the samples are rather small (particularly the German and French samples) Finally. Is the MDI theoretically or cross-nationally invalid? At this point. cultural influences. forthcoming). it seems unlikely that sampling effects. we showed that the three-dimensional structure of the MBI is invariant across samples of three different countries. very few analyses could be performed because almost no additional measures were included. via a reanalysis of the American normative sample (Schaufeli. It is clear that further analyses of the factonal validity of the MBI are needed that apply linear structure modelling techniques to representative samples from different countries. the MBI positive psychometric results were found. there are some limitations pertaining to our analyses First. In spite of these obvious limitations.

Depersonalization is particularly linked to withdrawal from patrents and colleagues. such as hardiness (Kobasa. In their study. personal accomplishment is strongly related to favourable personality features. This example Illustrates a most important point: the discriminant validity of the MBI can only be discussed fruitfully within a conceptual framework that specifies the relationships of the MBI dimensions with each other and with various other vanables (cf. social workers who maintained a sense of accomplishment did not intend to quit despite high levels of exhaustion. the psychometric resulrs of the MBI are similar across nations. control and commitment. neuroticism. they employed a slightly modified version of the MBI and used a less rigorous and more subjective method of testing. However. Thus. and interpretations about. This hints at a conceptual relatedness of personal accomplishment WIth psychological coping resources. these resources playa stress-buffering role because they concern a person's sense of mastery. irritability. '" Seven conclusions can be drawn from our cross-national validity study of the MBI. and mental and physical stress symptoms It has been argued by Watson and Pennebaker (1989) that subjective measures of health and well-being reflect. our results also point to the poor discrirrunant vahdiry of the emotional exhaustion scale of the MBI and of the BM scale Feelings of exhaustion overlap considerably with similar strains like depression. the individual's internal sensations. This view is supported empirically by Koeske and Koeske (1989). 1993). 1993) Generally. and to positive attrtudes towards the organization and one's professional role. In our study. First. Cherruss. On the other hand. the convergent validity of the emotional exhaustion scale is demonstrated convincingly: about one half of ItS variance is shared WIth the BM. a common underlying factor of "negative affectivity" might explain the relatively poor discriminant validity of the MBI emotional exhaustion scale and of the BM scale Positive indications were found for the discnmmant vahdiry of both of the other MBI scales. 1982) and self-efficacy (cf. The correlations of both other MBI scales with the BM are somewhat less substantive.The Validity oj the Maslacb Burnout Inventory 147 different nations (Golembiewski. psychosomatic reactions. 1993). Second. Leiter. This agrees with Maslach's (982) interpretation of depersonalization as being a way of coping WIth emotional exhaustion by mental distancing (see also Leiter. . Scherb and Boudreau. and most importantly. the perceptions of. 1993). to a large extent. who showed that feelings of personal accomplishment moderate the relationship between emotional exhaustion and a particular stress reaction: the intention to quit. whereas those who were exhausted and also had low feelings of accomplishment did mtend to quit.

g. but would also strengthen the validity of this scale (see . 1990) Sixth. Fourth. the implicit orthogonal model from the test manual that assumed the factors to be uncorrelated shows a relatively poor fit (cf Gold et ai. This onedimensional indicator of burnout only taps its most generic dimension (i e exhaustion) and is also substantially related to similar strains. Thus. self-esteem). for instance. Lee and Ashforth. irritabihty. whereas personal accomplishment is related to favourable personality characteristics (e. Obviously. DIRECTIONS FOR FUTURE RESEARCH The MBI can be Improved in a number of ways. Thus. results from the three national samples are highly consistent That is. clearly related to other strains such as depression. 1989. the discriminant validity of the BM is also poor. emotional exhaustion is correlated substantially with related concepts. and mental distress. The threefactor oblique model that allows the three factors to be correlated fits best to the data of each sample. as far as the cross-national validation of the factor-structure of the MBI is concerned. Fifth. with the most complex factor loadmgs Adding a couple of Items.148 Dirk Enzmann et al. the depersonalization and personal accomplishment scales appear to be discriminantly valid. Moreover. and feelings of disillusionment. neuroticism. Depersonalization is the shortest (and therefore least reliable) scale of the MBI. that is. Additional psychometnc development of the depersonahzation dimension seems necessary. the factorial structure of the MBI is similar across nations. preferably about the behavioural elemem of depersonalization. emotional exhaustion must be considered a generic psychological stress reaction. so the discriminant validity of this scale is rather poor. Finally. This is not very surpnsing when the content of this particular Item is considered ("I feel energetic"). psychosomatic reactions. organizational commitment. indications were found that the misspeciftcauons that cause the Imperfect fit of the original factorial model of the MBI most probably reflect theoretical flaws rather than cultural bias resulting from. inadequate translation or cross-national differences in perception or expression of symptoms. Third. would increase not only the internal consistency. physical stress symptoms. burnout (as measured by all three of the MBI scales) ISa multidimensional construct that cannot be merely reduced to feelings of exhaustion. our results show that m the original model one particular item loads on the "wrong" factor: additionally on emotional exhaustion instead of only on personal accomplishment. Depersonalization is related to withdrawal.

W B (1989) The dimensionality of a modified form of the Maslach Burnout Inventory for uruverstty students m a teacher-trammg program Educational and Psychological Measurement. whereas those of the personal accomplishment scale are phrased positively It cannot be completely ruled out that this fact explains the substantial correlation between the first two scales (see Tables 3. 5. 30) reveals that item twelve is the weakest and most complex Item m the accomplishment scale. Scherb. 549-561 Golembiewski. 1-16 Girault. 184-187 Broadhead. Department of Psychology Bolle. G ( 1988) SOCial support in the workplace Tests of SIX theoretical models AmencanJournal of Communuy Psychology. Similar results have been obtained in other exploratory factonal validity studies (Koeske and Koeske. R S (1992). Rush. DC Taylor & FranCIS Corcoran. 49. G P A. B . & de Wolff. J B.g. D F (1981) Correlates for active and passive types of burnout Journal of Social Sennce Research. 23. C J1993) The role of professional self-efficacy In the etiology and amehoration of burnout In W. H J (1970) The structure of human personality London Methuen Gillespie. and 7) and the particular behaviour of the latter scale (e. AT. The occupational risks of AIDS outreach workers In AS Trebach. 1. & Emery. 1986. R A (993) Burnout 10 cross-national settings . B F. R (1989) AIDS counselling AIDS Care. N (989) Burnout Emergence et strategies d'adaption [Burnout Emergence and strategies of adaptation) Dissertation Paris Uruversite Rene Descartes Gold. Y . Inspection of the MBI manual (Maslach and Jackson. Marcelissen. J L (1988) Supporting the deliverers of care Strategies to support nurses and prevent burnout Nursing Climes of North Amenca. & T Marek (Eds ). REFERENCES Beck. its high correlations with favourable personality charactenstics) Finally. & Boudreau. The Netherlands. 107-112 Dignam. K (Eds ). one item ("I feel energetic") should be deleted from the MBI since It not only loads on the intended personal accomplishment dimension. C Maslach. A (1985) Self-reported Job burnout among female human service professionals Journal of Occupational Bebaoiour. 1. J T. DC Drug Policy Foundation Brookings. Schaufeh. Moreover.The Validity of the Maslach Burnout Inventory 149 Table 8). F H. Ch J (986). the Items of the emotional exhaustion and depersonalization scales are phrased negatively. K. & McEvoy. S. P A. G (1972) Cogrutnie therapy of depression New York.701-724 Eysenck. G. Brown.143-150 Chermss. an equal balance of positively and negatively worded items is strongly recommended In the present version. 1989. Handletding Vragenhjst Orgarusattestress [Test-manual Organizational Stress Questionnairel University of Nijmegen. Mor and Laliberte.B.. Bachelor. 843-850 Bor. & Zeese. C E . 1984). Strategies for Change New Dtrecttons In Drug Policy (pp 195-202) Washington. which also loads on the exhaustion factor.4(2). 16. Bolton. Shaw. bur also on emotional exhaustion. Guilford Bergers. & Michael. R T. Professional Burnout Recent Developments In Theory and Research 035-149) Washmgton. 6. K J (986) Measunng burnour A rehabihty and convergent validity study Journal of Social Bebatnor and Personality. & West. p. A J .

S C. M. & Pennebaker. Washmgton. 131. & Aronson. 99-113 Maslach. & Ashforth. J M (1983) Companson of rwo measures of burnout Psychological Reports. NJ: Erlbaum Koeske. 234-254 .743-747 Leiter. B E (1990). v66rkomen en voork6men [Burnout In the Services for the HIV-Infected. D (989). T (988). & Koeske.50-53 Watson. Enzmann. Dealing With stress and burnout. D ( 1991) Occupational morbidity and burnout Lessons and warnmgs for HIV/AIDS carers Iniernauonal Reuteto of Psychiatry. and distress Explonng [he central role of negative affecnviry Psychological Retneui. and Kafry. 25.144 Lebourdais. (982) The hardy personality. & Maslach. actnnty New York Academic Press Voelcker. AM.). P (993) Burnout as developmental process Consideration of models In W B Schaufeh. 274-283 Mowday. C. 14. C & Jackson. & WIlliams. DC Taylor & FranCIS Maslach. R M. J W (989) Health complaints. Girault. Prevalence and Prevention) (Nationale Comrmssie AIDS-Besrnjdmg. D (981) Burnout From Tedium to Personal Growth New York Free Press Roozenburg. 439-449 Mohr. J C (1978) Psychometric theory New York McGraw-HIli Pmes. The AIDS Caregtuer's Handbook (pp.Toward a social psychology of stress and health. E (1988) Career Burnout Causes and Cures New York Free Press PInes. In T. 224-247 Nunnally. B Schaufeh. Ill. & G Sanders. L (1984) Burnout among hospice staff Health and SOCial Work. Oktober 1988. D. C (982) Burnout A SOCial psychological analysis In J. N . W B. S E (1981) The measurement of experienced burnout journal of Occupational Bebainour. & Porter. V . stress. In J Suls. 140. C Maslach & T Marek (Eds ). C (forthcoming) The factonal ualidity of the Maslach Burnout Inventory A cross-national confirmatory analysts Stout. Professional Burnout Recent Developments in Theory and Research (pp 237-250) Washmgton. W Jones (Ed). No 578. & Jackson. New York St Martin's Press joreskog. E . C. 138-147). L W (1979) The measurement of orgaruzational commitment Journal of Vocational Bebauior. Aronson. RD.Lrueruentionsttns 30-53). G F . AM. R T. Enzrnann.150 Dirk Enzmann et at Genenc and model-speciftc perspectives In W. E (1989) Hopelessness and helplessness Treatmg the doctors who treat AIDS patients Canadian Medical ASSOCiatIOnjou rnal. J (1983) Temperament. 9. & T Marek (Eds ). WashIngton Taylor & FranCIS Schaufeh. Edison (Ed. 2. #6815/HRI 88) Schaufeli. llSREL VII User's reference guide Mooresville: SCientific Software Kobasa. CA Consulting Psychologists Press Miller. D . The Burnout Syndrome Current Research. H (1988) Burnout tn de bulpoerierung by Hlv-infectie. Social psycbology of bealtb and Illness(pp 333). London House Press Maslach. Park Ridge. S E (1986) Maslacb Burnout Inventory Manual (2nd edn ) Palo Alto. D .). 96. KG. On the meaning of Maslach's three drmensions of burnout journal of Applied Psychology. N (993) Measurement of burnout A review In W B Schaufeh. 440-443 Lee. 75. J K. J (1991) AIDS groups address caregiver burnout The Advocate.. Professional Burnout Recent Developments In Theory and Research (pp 199-215). G (1986) Die Erfassung psycbtscher Befindensbeeintracbtigungen bel Industnearbettern [The assessment of psychological distress m blue-collar workers) Frankfurt/Main Lang Mor. 283-289 Strelau. Steers. E. C Maslach. Maslach & T Marek (Eds. Hillsdale. (1989) Construct validity of the Maslach Burnout Inventory A Critical review and reconceptualization journal of Applied Bebainoral SCience. & Girault. Tbeoty. & Sorbom. 53. DC Taylor & FranCIS Howze. Professional Burnout Recent Developments in Theory and Research (pp 218-236). W B . R T. & Laliberte. 3. personality.