You are on page 1of 12

Teaching and Teacher Education 32 (2013) 75e86

Contents lists available at SciVerse ScienceDirect

Teaching and Teacher Education

journal homepage:

Getting personal with teacher burnout: A longitudinal study on the development

of burnout using a person-based approach
Daniel Hultell*, Bo Melin, J. Petter Gustavsson
Karolinska Institutet, Department of Clinical Neuroscience, Division of Psychology, 171 77, Stockholm, Sweden

h i g h l i g h t s

< When using a variable-based approach burnout appeared to be stable over time.
< The person-based approach however identied seven different burnout trajectories.
< The person-based approach presents a multifaceted view of burnout development.
< Findings show that there is reason to question the stable nature of burnout.
< The majority of the beginning teachers coped well with the transition.

a r t i c l e i n f o a b s t r a c t

Article history: Studies have suggested that teachers burnout levels are stable over time. This might be because lon-
Received 17 February 2012 gitudinal studies on burnout have mainly used a variable-based approach. The purpose of this study was
Received in revised form to determine if a person-based approach could provide a more multifaceted perspective to the devel-
6 December 2012
opment of teacher burnout. 816 beginning teachers were tracked over the rst three years of their
Accepted 14 January 2013
employment. At group level, burnout levels were moderately low and stable over time. However, un-
derlying these levels were seven trajectories, of which six changed signicantly over time. Changes in
burnout trajectories were associated with concurrent changes in burnout-related variables.
Teacher burnout
2013 Elsevier Ltd. All rights reserved.
Cluster analysis
PATH study

1. Introduction a sink or swim experience (Johnson & Birkeland, 2003). In addi-

tion to entering employment and the associated challenges, for
The transition from higher education to employment is a major many beginning teachers, this is also a period during which other
life event and a challenge. It involves testing the knowledge ac- life events occur that characterize early adulthood (e.g., moving in
quired during higher education in a real-life work setting, and to a new home, marriage, and starting a family). It is thus an intense
nding out whether expectations of teaching are realized or not. period characterized by many changes and many are also faced
Furthermore, there is a process of organizational socialization, with with the new situation of balancing work and family life. Thus,
the aim of successful adaptation to the work role and work tasks, as although there are many positive changes occurring during this
well as gaining social acceptance from new colleagues and super- period it is also a period of many trials and stress. Research has
visor. Many experience a reality shock during this transition shown that this transition period is hard for teachers resulting in
(Gavish & Friedman, 2010) and this period has been likened to strain that may eventually lead to turnover (e.g., Ingersoll, 2012)
or have health impairing consequences such as burnout
(e.g., Goddard, OBrien, & Goddard, 2006).

* Corresponding author. Karolinska Institutet, Department of Clinical Neuro- 1.1. Burnout

science, Division of Psychology, 171 77 Stockholm, Sweden. Tel.: 468 524 838 06;
fax: 468 30 72 98.
E-mail addresses: (D. Hultell), (B. Melin), The concept of burnout was developed in the 1970s in studies of (J.P. Gustavsson). individuals with early career problems within the human service

0742-051X/$ e see front matter 2013 Elsevier Ltd. All rights reserved.
76 D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86

professions (Cherniss, 1980a; Freudenberger, 1975; Kramer, 1974). longitudinal studies, and, as stressed by Collins (2006), the study
The predominant view of burnout is that suggested by Maslach design should be guided by the theoretical model of change and
et al. (Maslach & Jackson, 1986; Maslach, Jackson, & Leiter, 1996) not by logistics.
who dene burnout as a work-related construct that consists of Burnout has previously been found to be a stable construct over
three dimensions: emotional exhaustion (depletion of energy), time (e.g., Brouwers & Tomic, 2000; Hakanen, Schaufeli, & Ahola,
depersonalization (detachment from work and colleagues at work), 2008; Schaufeli & Enzmann, 1998; Taris, Le Blanc, Schaufeli, &
and reduced personal accomplishment (feelings of inadequacy and Schreurs, 2005) and the magnitudes of the stability coefcients of
inefcacy) (Maslach, Schaufeli, & Leiter, 2001). The Maslach Burn- burnout are more in line with those found for psychological traits
out Inventory (MBI) (Maslach & Jackson, 1986; Maslach et al., 1996) (Roberts & DelVecchio, 2000). This is surprising, considering that
is the most frequently used instrument to measure burnout burnout is dened as a psychological state and, therefore, would be
(Schaufeli & Enzmann, 1998). The reduced personal accomplish- expected to vary over time, an issue which has been highlighted
ment dimension has, however, been found to be problematic and within the eld of research on burnout (e.g., Shirom, 2005). High
has the weakest empirical support of the three suggested burnout stability scores, however, do not mean that changes in levels do not
dimensions (e.g., Friedman, 1993; Lee & Ashforth, 1996; Shirom, occur, but rather they are indications of high rank-order stability. In
2011). Consequently, a core of burnout has emerged consisting of a recent review of 35 studies reporting the stability of burnout and
the two dimensions of emotional exhaustion and depersonaliza- 37 studies reporting the magnitude of changes in the mean level of
tion. Based on this core of burnout, Demerouti et al. have presented burnout over time, it was found that, after taking the effect of time
an alternative burnout measure called the Oldenburg Burnout In- in to account, the rank-order correlation per year was .66 and the
ventory (OLBI) that is based on two dimensions: exhaustion and change in mean levels expressed as effect size (Cohens d) per year
disengagement (Demerouti, Bakker, Nachreiner, & Schaufeli, 2001). was .13 (Hultell, 2011). These ndings show that burnout is a stable
Exhaustion is viewed as the result of intense physical, emotional, construct in regards to both rank-order and changes in levels, and
and cognitive strain. Disengagement refers to distancing oneself this was the case for both teachers and for individuals employed in
from work, and developing negative attitudes toward the work other occupations (Hultell, 2011). The stable nature of burnout
object, the work content, and to work in general. An additional found in these studies could be a consequence of the measure-
denition of burnout, based on the conservation of resources (COR) ments and methods of analysis. Neither the MBI nor the OLBI
theory (Hobfoll, 1989), has been suggested by Shirom and Melamed include a time reference and many of the items in these in-
and views burnout as the depletion of an individuals energetic struments are actually trait-like in character, something that could
coping resources to deal with occupational stressors (Shirom & be a reason for the apparently stable nature of burnout. Moreover,
Melamed, 2006). Burnout is conceptualized in this theory as the longitudinal studies on burnout that are available in the liter-
physical fatigue (lack of energy to perform daily work tasks), ature almost exclusively use a variable-based approach and thus
emotional exhaustion (lack of energy to engage in relationships have focused on statistical relationships between variables across
with people at work), and cognitive weariness (slow thinking individuals at the group level. Although this approach is useful
processes and low mental agility), and is assessed using the when development is linear and at group level, this might not al-
Shirom-Melamed Burnout Measure (SMBM) (Melamed, Shirom, ways be the case. Given the range of possible interactions between
Toker, Berliner, & Shapira, 2006). Although there are conceptual the individual and their environment, it seems likely that devel-
differences between these three views of burnout, they share many opment of psychological states will be characterized by complex
features. All instruments can be seen as assessments of the core paths as opposed to simple, uniform trends. In these cases the
dimensions of burnout (emotional exhaustion and depersonaliza- variable-based approach does not capture the individual differ-
tion in the MBI, exhaustion and disengagement in the OLBI, and ences in development in an accurate way. In such situations,
physical fatigue and emotional exhaustion in the SMBM). Fur- a person-based approach could be more useful because it has the
thermore, all denitions view burnout as a work-related psycho- advantage of being able to identify non-linear developmental pat-
logical state, i.e., it is context-dependent and varies with time. terns by clustering individuals within the sample who share similar
Although burnout is dened as a psychological state, it is com- patterns of development (Bergman, Magnusson, & El-Khouri,
monly agreed upon that it develops gradually over time, and the 2003). When reviewing the literature, three studies were identi-
process of developing burnout has been described as an erosion of ed in which a person-based approach was used when studying the
engagement (Maslach & Leiter, 1997; Maslach et al., 2001) and as development of burnout: (1) Mkikangas, Feldt, Kinnunen, and
a spiral of loss (Hobfoll & Shirom, 2000; Shirom, 2010). Tolvanen (2011), (2) Boersma and Lindblom (2009), and (3)
To gain better understanding of the burnout process it is Rudman and Gustavsson (2011). The rst two studies, however,
therefore necessary to use a longitudinal study design. However, only used data from two waves of measurement and thus suffered
when reviewing studies on burnout it is apparent that most are from the limitation that change over time could potentially be
cross-sectional, and the vast majority of the longitudinal studies confounded with measurement error. The third study (Rudman &
available are mainly limited to data from two waves of measure- Gustavsson, 2011) was based on data from three waves of meas-
ment. Relying on data from two waves limits the possibility of urement and individual levels of burnout at each wave were clus-
making causal inferences and studying development over time tered to identify trajectories. Eight trajectories were identied
because true change might be confounded with measurement error showing that underlying the stable group-level changes were
(Singer & Willett, 2003). When studying changes over time it is diverging developmental patterns (Rudman & Gustavsson, 2011).
therefore necessary to use data from three or more waves of These ndings thus indicate that substantial information about the
measurement. Moreover, when studying change it is preferable to development of burnout may be lost when using the variable-based
have a well-dened starting point. This makes it possible to study approach.
development more accurately by limiting the potential for con-
founding of explanatory factors. A common problem with many 1.2. Teacher development and burnout in beginning teachers
longitudinal studies on burnout is that the study participants have
varying work experience, which results in varying starting points Although the development of burnout is in no way restricted to
for the development of the phenomenon being studied. The use of the initial period of employment it has been found that this period
a sound temporal design is, therefore, of special importance in is decisive for many in regards to the development of work-related
D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86 77

attitudes and behaviors; and it was especially critical for the students, and dealing with individual differences), assessing stu-
development of burnout (Cherniss, 1980a). During this critical dents work and progress, and relations with parents (e.g., Friedman,
period the teachers are also assumed to develop their professional 2000; Kyriacou, 2001; Veenman, 1984). These aspects are associated
skills. According to the theory of teacher development (Ingersoll & with their perceived ability to successfully perform their job and
Strong, 2011), teacher development starts with pre-service prepa- relate to the construct self-efcacy. Self-efcacy is dened as be-
ration (i.e., during teacher education). The next stage is induction liefs in ones capabilities to organize and execute the courses of
which the teachers should receive when entering employment. action required to produce given attainments (Bandura, 1997, p. 3)
Induction is assumed to result in less turnover which allow the and affects an individuals behavior, motivation, and perseverance in
teachers to improve their classroom practices. Improved classroom achieving their goals. Self-efcacy beliefs are domain specic, and
practices are assumed to ultimately result in improved student professional self-efcacy in teachers is referred to as teacher self-
achievements. Teacher development is assumed to take place efcacy (TSE). TSE generally concerns classroom practices and
during a vulnerable period in regards to development of burnout. It common aspects covered in assessments of TSE are organizing and
therefore seems reasonable to assume that development of burn- planning, student engagement, instruction, and classroom man-
out both can affect and be affected by the different parts of the agement. The content of TSE instruments vary, however, because
theoretical model of teacher development. they are usually adapted to t the context of the teachers being
Considering that burnout is a work-related construct, most studied. TSE has been found to be related to burnout, and teachers
studies have focused on factors affecting burnout during employ- who experience low levels of TSE are more likely to develop burnout
ment. However, there are studies showing that factors during (e.g., Evers, Tomic, & Brouwers, 2004; Schwarzer & Hallum, 2008;
higher education can affect future work-related well-being. It has Skaalvik & Skaalvik, 2010). The importance of TSE in relation to
been found that achievement strategies and self-esteem during burnout is also supported by the qualitative studies on early career
education affect future burnout (Salmela-Aro & Nurmi, 2007; burnout by Cherniss (1980a, 1980b, 1995). One of the crucial factors
Salmela-Aro, Tolvanen, & Nurmi, 2009). In addition, Gavish and in the development of burnout for newcomers was a crisis of
Friedman (2010) found that beginning teachers had high levels of competence that refers to feelings of doubt and insecurity
burnout and suggested that this was likely to be due to the fact that regarding ones ability to perform work tasks despite having a for-
many already started to experience burnout during their education. mal education (Cherniss, 1980a). The crisis of competence corre-
The education process is designed to provide student teachers with sponds closely to self-efcacy (Cherniss, 1993) and TSE can be seen
the knowledge and skills they need to manage their work as as the operationalization of the crisis of competence in beginning
teachers. Education that fails to adequately prepare the student teachers.
teachers has been found to increase the risk of burnout early in the
teachers careers (Gold, Roth, Wright, & Michael, 1991). Thus, it is 1.3. The present study
relevant to take factors related to the educational period in to
consideration when studying burnout in beginning teachers. There is a need for longitudinal studies on teacher burnout using
The next stage of teacher development concerns induction. In- data from more than three waves of measurement. Furthermore,
duction refers to supportive programs developed for beginning burnout has been found to be stable over time despite being
teachers to help them with the transition from higher education to dened as a psychological state which could be a consequence of
employment. The content of the programs vary but can include the methodological approach used in previous studies. The initial
a reduced schedule, being assigned a mentor, seminars on period of employment has been found to be of particular interest
instructional strategies with other beginning teachers and more because it has been found to be a period often associated with
experienced teachers. These aspects have also been identied as strain, and it has also been found to be a formative time for the
alleviating of stress for beginning teachers (Gavish & Friedman, development of work-related attitudes and behaviors (Cherniss,
2010; Gilbert, 2005). In a recent review on the effects of induc- 1980a). The main purpose of this study was therefore to examine
tion it was concluded that induction had a positive impact on whether the use of a person-based approach could identify pat-
turnover, classroom practices and student achievement (Ingersoll & terns of intra-individual change in burnout during the rst three
Strong, 2011), thus supporting the assumed relationships in the years of employment for beginning teachers. Given that only one
theory of teacher development. Although it is assumed that in- study was found in which a similar study design was used, the
duction will ultimately lead to improved student achievement the hypotheses in this study are tentative. Based on previous ndings
rst goal is to make sure that teachers remain in the profession regarding the stability of burnout (e.g., Hultell, 2011; Schaufeli &
which will allow them to develop their skills. Hence, the purpose of Enzmann, 1998) and on the trajectories identied by Rudman and
teacher induction is to make the transition to employment easier Gustavsson (2011), six different types of burnout trajectories were
for beginning teachers with the initial goal of reducing turnover. expected to be found:
Turnover is a well-recognized problem for beginning teachers who
often leave their profession early in their careers (Guarino, 1. A trajectory for which levels of burnout increased over time.
Santibanez, & Daley, 2006; Ingersoll & Strong, 2011; Strong, 2005) 2. A trajectory for which levels of burnout decreased over time.
and it was recently reported that turnover rates during the initial 3. A trajectory characterized by an initial increase in burnout
period of employment are as high as 40e50% (Ingersoll, 2012). followed by a decrease.
Burnout has been assumed to be an antecedent to turnover, and 4. A trajectory characterized by an initial decrease in burnout
research has shown that burnout is positively related with turnover followed by an increase.
intentions (Friedman, 1993; Gold et al., 1991; Lee & Ashforth, 1996; 5. A trajectory characterized by low and stable burnout levels.
Schaufeli & Enzmann, 1998). Given the link between burnout and 6. A trajectory characterized by high and stable burnout levels.
turnover intention, it seems likely that teacher induction should
also have a buffering effect on the development of burnout. It was also of interest to determine how the development of
The next stage in teacher development concerns classroom burnout was related to concurrent changes in variables relevant to
practices. When reviewing studies focusing on stressors of teachers, burnout in teachers and determine if the burnout trajectories could
it is apparent that the most common stressors are related to class- be predicted by individual differences. This is of relevance because it
room practices (e.g., discipline, working with poorly motivated helps to explain differences between developmental patterns and
78 D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86

adds validity to the ndings. Because the present study focused on after the teachers had graduated and entered employment (T1, T2,
the transition period from education to employment it was decided and T3). Considering that work demands vary during the semester
to include explanatory variables that are central in the theory of it is likely that these variations will inuence burnout assessments.
teacher development since this theory also focus on this particular This means that observed changes in burnout could be an effect of
period. This framework also allowed for inclusion of variables viewed the time of measurement. Questionnaires were therefore sent out
as both predictors and consequences of burnout. The variables cho- at about the same time at each data collection during employment
sen in this study regarding concurrent changes were teacher self- to reduce this potential bias. Criteria for inclusion were that the
efcacy (TSE), turnover intention, and self-rated health (SRH). respondents had participated in all three waves of measurement
Although SRH is not a part in the teacher development model it was during employment and had answered more than half of the items
included due to the established link between burnout and health in the burnout assessment scale. A total of 816 teachers met the
(e.g., Guglielmi & Tatrow, 1998; Moya-Albiol, Serrano, & Salvador, criteria for inclusion and thus constituted the sample for the study.
2010; Nixon, Mazzola, Bauer, Krueger, & Spector, 2011; Schaufeli & There were 697 (85%) females, and the mean age of the sample at
Enzmann, 1998; Vinokur, Pierce, & Lewandowski-Romps, 2009). T1 was 34.95 years (SD 7.87).
TSE was assumed to be a predictor of burnout whereas turnover An attrition analysis was performed to detect whether there was
intention and SRH were assumed to be consequences. It was expected any systematic dropout from the study over time. A repeated
that burnout would be positively related to turnover intention and measures logistic regression analysis was performed using General
negatively related to TSE and SRH. In addition to examining concur- Estimation Equations (GEE). The dependent variable was dropout
rent changes, it was also of interest to determine whether burnout (responding vs. non-responding at each data collection with in-
trajectories could be predicted by individual variables, educational dividuals who chose to terminate their participation in the study
variables, and organizational variables. Based on the assumed link being considered as non-responding), and the independent vari-
between pre-service training with future teacher development ables were time, sex, age, immigrant background, self-rated health
(Ingersoll & Strong, 2011) and previous research into educational (SRH), depression, and life satisfaction. Each independent variable
achievements in relation to future burnout (e.g., Fives, Hamman, & was tested for interaction with time and main effects and inter-
Olivarez, 2007; Gold et al., 1991; Salmela-Aro et al., 2009), it was action effects were tested using Wald c2 statistics. All the inde-
expected that indicators of educational success (e.g., achievement of pendent variables were assessed at baseline (T-2), and the
educational goals) would have a buffering effect on burnout. It was dependent variable was assessed in each of the following waves of
also expected that indicators of well-being (e.g., low levels of measurement. Although it would have been preferable to have
depression, high levels of life satisfaction) during higher education included burnout as a predictor in the analysis, data were not
would be negatively related to future burnout. Finally, it was expec- available at baseline. The results showed that there were no in-
ted that teachers who participated in an induction program would be teractions for any of the variables with time and that there were
less likely to develop of burnout. main effects of time (c2(3) 87.76, p < .001), sex (c2(1) 16.64,
p < .001), and SRH (c2(1) 7.12 p < .001). These odd ratios (OR)
2. Methodology showed that the response rate deteriorated over time (OR .45,
p < .001) and that there was a lower response rate for males
2.1. Procedure and sample (OR 1.29, p .017) and for participants with lower levels of SRH
(OR 1.36, p .012).
The data used in this study originated from a nationwide lon-
gitudinal study performed in Sweden called the Prospective Anal- 2.2. Measurement
ysis of Teachers Health (PATH). Data was collected using
questionnaires and there were ve data collections: two during The outcome variable was burnout and this was assessed using
higher education and three during the initial period of employ- the Scale of Work Engagement and Burnout (SWEBO) (Hultell &
ment. The sampling frame of the PATH study consisted of two Gustavsson, 2010a,b). The SWEBO consists of two subscales, one
subgroups. The rst group consisted of student teachers who were measuring burnout and one measuring work engagement. Because
studying to become preschool teachers or teachers for students burnout is dened as a psychological state (i.e., context-dependent
between the ages of 7e11 years (Younger Age). The second group and varying with time), the SWEBO is both context-specic (work)
consisted of student teachers who were studying to become and includes a time reference (two weeks). The burnout subscale
teachers for students between the ages of 12e18 years (Older Age). consists of the dimensions of exhaustion, disengagement, and
Criteria for inclusion in both groups were that the student teachers inattentiveness. The SWEBO is conceptually similar to previously
had three semesters left in their program, and that they attended discussed operationalizations of burnout. The exhaustion and dis-
a school with more than 80 students registered in the dened engagement scales assess the core dimensions of burnout and
program in the semester prior to the rst data collection. There inattentiveness is quite similar to the cognitive weariness dimen-
were 4067 student teachers from 21 schools who met the criteria sion of the SMBM. What is different from these other measures is the
for inclusion and who were included in the sampling frame. operationalization of the SWEBO. Neither the MBI nor the OLBI are
Questionnaires were sent to all student teachers included in the both context-specic and include a time reference. Although the
sampling frame along with a letter that described the study and SMBM includes both these aspects, the items that assess emotional
explained that participation was voluntary and that they were free exhaustion focus on perceived ability to engage with other persons,
to leave the study at any time. A total of 2853 (70%) responded and and there is a risk that these items, therefore, measure emotional
thus constituted the cohort of the study. Because the focus of the efcacy beliefs instead of emotional exhaustion. This is not the case
present study was the development of burnout after entering for the SWEBO because it only includes mood adjectives and thus
employment, the data collections during education are labeled T-2 focuses solely on affective aspects. Sample items of the burnout
and T-1, and the data collections during employment are labeled T1, subscale are: In the past two weeks at work I have felt lethargic
T2, and T3. For an overview of the data collection process, see Fig. 1. (exhaustion); In the past two weeks, in relation to my work I have
The study design was longitudinal and used data from four waves of felt a sense of meaninglessness (disengagement); In the past two
measurement. The data were collected when the teachers had weeks while I have been working I have felt unfocused (inatten-
about one semester left of their education (T-1), and then annually tiveness). Items were rated using a frequency response format
D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86 79



Fig. 1. Overview of the data collection process and responses at each wave of measurement.

(1 Not at all, 2 Some of the time, 3 Most of the time, 4 All of the burnout. TSE, turnover intention, and SRH, were assessed in parallel
time). The measurement model of the burnout subscale is hierar- with burnout (i.e., in waves T1, T2, and T3). SRH was assessed using
chical with a second-order factor (i.e., burnout) loading on three a single item and was rated using ve response alternatives ranging
rst-order factors (i.e., exhaustion, disengagement, and inatten- from bad to good (1 Bad, 2 Rather bad, 3 Neither good nor bad,
tiveness). The psychometric properties of the SWEBO have been 4 Rather good, 5 Good). Turnover intention was measured using
evaluated using conrmatory factor analysis (CFA) with satisfactory a scale consisting of three items (Cohen, 1998). The items were
results regarding reliability and factorial validity for the burnout rated using a response format ranging from 1 to 5 (1 Strongly
subscale (a .90; c2(df) 138.66(24), SRMR .051, RMSEA .046, disagree to 5 Strongly agree). TSE was measured with a scale that
and CFI .99) (Hultell & Gustavsson, 2010a,b). In addition, the consisted of 12 items assessing the overall dimension of TSE that
SWEBO has previously been used in a study predicting burnout and consisted of four sub-dimensions (Instructional strategies, Give
the results of that study further strengthen the validity of the in- support to individual students, Classroom management, and
strument (Hultell & Gustavsson, 2011). The SWEBO has also been Teacher-parent interaction). The TSE scale was homegrown and the
compared to the OLBI on three occasions and cross-sectional cor- items were constructed to target qualication requirements within
relations ranged from .76 to .77 (Hultell, 2011), showing that the two each sub-dimension that novice teachers are expected to hold after
instruments share more than half of their variance. The SWEBO is completing their education and when entering employment (SFS
a fairly newly developed instrument and as such there are no clin- 1993: 100). Items were rated along an 11-point continuum with
ically validated cut-off values for burnout cases. Considering the anchors at 0% (No, I cannot do that), 50% (Maybe I can do that), and
response format and the pathological nature of the adjectives 100% (I am certain I can do that). At the nal wave of measurement
included in the scale, the following categorization of overall mean (T3) one TSE item was missing in the questionnaire. However, the
values was used: 1.49 as low burnout (majority of responses are correlations between the 11-item version and the 12-item version
Not at all), 1.50e1.99 as moderately low (majority of responses vary were .996 at T1 and .995 at T2 indicating that the missing item did
between Not at all and Some of the time), 2.00e2.49 as moder- not bias the results. Descriptive statistics, sample items, and reli-
ately high (majority of responses vary between Some of the time ability coefcients (Cronbachs alpha) for these variables are pre-
and Most of the time), and 2.50 as high burnout (majority of re- sented in Table 1. Moreover, a range of variables, mainly assessed
sponses are Most of the time or All of the time). during the nal year of education (T-1), were included as pre-
Additional variables were included in the study to describe dictors. These variables included sex, age, depression, SRH, life
concurrent development in external variables to better understand satisfaction, achievement of educational goals, TSE, satisfaction
the differences between the various developmental patterns of with education, being pressured by studies, being pressured by
80 D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86

Table 1 Table 2
Number of items, range, source of measurement, descriptive statistics, and reliability Description of the categorization of demographics, the predictors assessed during
scores (Cronbachs alpha) of the variables included in the study used to cross- the nal year of education, and the two time-invariant organizational predictors.
validate the trajectories of the identied clusters.
Variables Categorization
Variable N Range Sample item M (SD) a Demographics
items Sex Male vs. Female
Burnout T1 9 1e4 In the past two weeks 1.55 (.50) .90 Age 29; 30e39; 40
at work I have felt
Health related
Burnout T2 9 1e4 e 1.61 (.48) .89
Depression No vs. Yes
Burnout T3 9 1e4 e 1.65 (.50) .91
Depression was categorized according to the
TSE T1 12 1e11 Can you analyze and 8.59 (1.33) .92
DSM-IV (i.e., 5 out of 9 symptoms present most
evaluate student learning
of the time during the last two weeks, of which
and development?
symptom 1 or 2 must be present). Assessed using
TSE T2 12 1e11 e 8.91 (1.17) .92
the Major Depression Inventory (Bech, Rasmussen,
TSE T3 11 1e11 e 9.06 (1.10) .91
Olsen, Noerholm, & Abildgaard, 2001).
SRH T1 1 1e5 How would you rate 4.32 (.76) e
SRH Low (Bad, Rather bad, Neither good nor bad) vs.
your general health
High (Rather good, Good)
Life satisfaction Low vs. High
SRH T2 1 1e5 e 4.29 (.79) e
The scale consisted of ve items with the response
SRH T3 1 1e5 e 4.29 (.77) e
categories: 1 Strongly disagree, 2 Somewhat
Turnover 3 1e5 I think a lot about 1.90 (1.18) .90
disagree, 3 Neither agree nor disagree,
intention T1 leaving my current
4 Somewhat agree, 5 Strongly agree. Mean
values were calculated for each individual and a
Turnover 3 1e5 e 1.95 (1.20) .91
cut-off value of 3.50 was used (<3.50 Low
intention T2
vs.  3.50 High). Assessed using the Satisfaction
Turnover 3 1e5 e 1.99 (1.23) .91
With Life Scale (Diener, Emmons, Larsen, & Grifn,
intention T3

Educational outcomes
Achievement of Low vs. High
occupational choice, and spillover from studies to private life. In- educational goals The scale consisted of six items with the
duction was also included as a predictor from employment. The response categories: 1 To a very low degree,
predictor variables were all categorized, and a description of the 2 To a low degree, 3 To a high degree, 4 To
categorization and source of measurement are presented in Table 2. a very high degree. Mean values were calculated
for each individual and a cut-off value of 2.50 was
used (<2.50 Low vs.  2.50 High). Assessed
2.3. Data analysis using items from the National Survey of Student
Engagement (Kuh et al., 2001).
A univariate repeated-measures analysis of variance (rANOVA) TSE Low vs. High
The scale consisted of 11 items with the response
was used to study within-group changes of burnout levels across
categories: 1 No, I cannot do that, 6 Maybe I
the rst three years of employment, and post-hoc pair-wise com- can do that, 11 I am certain I can do that. Mean
parisons between T1 and T2 and between T2 and T3 were also values were calculated for each individual and
analyzed. Test-retest correlations were used to assess stability. a cut-off value of 6.50 was used (<6.50 Low vs.
 6.50 High). Homegrown.
To identify and classify individuals with similar developmental
Satisfaction with Low (Very bad, Bad) vs. High (Good, Very good)
trajectories, a cluster analysis was performed using SLEIPNER 2.1 education
(Bergman & El-Khouri, 2002). The analysis was based on the data Pressured by Low (Not pressured at all, Not particularly
collected after the teachers had entered employment (T1, T2, and studies pressured) vs. High ( Somewhat pressured, Very
T3). Initially data were scanned for outliers that could potentially pressured)
Pressured by Low (Not pressured at all, N. particularly pressured)
distort the clustering procedure using the RESIDUE module of
occupational vs. High ( Somewhat pressured, V. pressured)
SLEIPNER. Two individuals were identied as outliers and were choice
removed from the original data set. Next, the cluster analysis was Spillover from No vs. yes
performed using Wards hierarchical method. This clustering studies to family
method was chosen based on its ability to adequately recover true Organizational
clustering and consistently replicate clustering (Breckenridge, variables
2000). Based on the recommendations of Bergman et al. (2003) Induction No vs. Yes
Three stoppage rules were applied when choosing the cluster so-
lution: (1) the explained error sums of squares (EESS) should not be
lower than .67, (2) the homogeneity coefcients for the respective clusters were used to perform a K-means cluster analysis on the
cluster should not exceed 1.0, and (3) the merging of two clusters total sample (including the cases removed from the cluster ana-
should be comprehensible and make theoretical sense. Because the lyses) to assign cluster membership for all participants. SPSS 18
clustering method chosen was hierarchical, this could result in (SPSS Inc, 2009) was used to perform the K-means cluster analysis.
some individuals ending up in clusters in which they did not A two-way rANOVA (cluster  time) was performed to deter-
optimally belong. To correct for this potential misplacing of in- mine if there was a difference in change over time between the
dividuals, the RELOCATE module of SLEIPNER was used to move identied clusters. This was done to conrm that the clusters
individuals from one cluster to another where they t better. To actually represented different trajectories. Next, univariate rANO-
validate the chosen cluster solution, the SIMULATE module of VAs were performed to study changes over time within each
SLEIPNER was used to determine whether the solution explained cluster. Two-way rANOVAs (cluster  time) were performed to
signicantly more variance compared to 20 randomly generated determine if there were differences in changes between the clus-
data sets (i.e., testing a null hypothesis of no relations in the data). ters for each explanatory variable. Univariate rANOVAs were per-
After validating the cluster solution, the centroids of the different formed for each cluster looking at changes over time for the
D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86 81

explanatory variables. Because the outcome variables and the Table 3

explanatory variables were not all measured on the same scale, Prevalence and incidence of burnout.

Z-scores were calculated and calibrated in relation to the mea- Wave Low Moderately Moderately High Incidence Cumulative
surements during the rst year of employment and are presented burnout low high burnout incidence
in Fig. 2. Chi-square analyses were performed to study whether T1 54.4% 23.8% 16.3% 5.5% 5.5%a 5.5%d
clustering could be predicted using individual variables, educa- T2 54.4% 27.6% 18.1% 6.0% 3.9%b 9.2%d
T3 44.6% 26.5% 22.9% 6.0% 3.5%c 12.4%d
tional variables, and organizational variables.
SPSS 18 (SPSS Inc, 2009) was used to analyze the descriptive N at risk 816.
N at risk 771.
statistics, the reliability of the included variables, to perform the c
N at risk 741.
attrition analysis, to perform rANOVAs, and to perform the c2 an- d
Based on complete sample (N 816).
alyses. A p-value of maximum .05 was used to judge statistical
signicance (Glass & Hopkins, 1995).
The cluster analysis resulted in a seven-cluster solution that met
the three criteria for a satisfactory cluster solution. When testing
3. Results
the null hypothesis of no relations in the data, the results showed
that the chosen cluster solution explained more variance than
The mean burnout level increased successively over the three
could be expected by chance (T(20) 41.49; p < .001). After having
years, peaking at 1.65 in the third year of employment (Table 1). The
relocated individuals from one cluster to another where they t
results showed that there were signicant changes over time
better, the EESS of the nal solution was .76 and the homogeneity
(F(2) 19.52; p < .001) although the changes were small (h2 .02).
coefcients for the clusters ranged from .04 to .36. Descriptive
The difference between the lowest and highest levels of burnout
statistics for the seven clusters (based on the K-means analysis
differed by .20 (Cohens d). Test-retest correlations showed
including all 816 participants) are presented in Table 4, and the
that one-year stability ranged between .58 (r12) and .62 (r23), and
burnout trajectories are graphically plotted in Fig. 2.
two-year stability was .53 (r13). Cross-sectional prevalence of in-
The results of the rANOVAs showed that there was a signicant
dividuals with high burnout levels ranged from 5.5% to 6% and the
interaction (time  cluster) effect, indicating that the seven clusters
cumulative incidence was 12.4% (Table 3).
reected different developmental patterns. Furthermore, the re-
sults also showed that the clusters reected different devel-
opmental patterns between the rst and second and between the
second and third waves of measurement. Post-hoc tests showed
that there were signicant changes over time for six of the seven
clusters. Furthermore, three of these six clusters (1, 3, and 4) also
changed signicantly between T1 and T2 and between T2 and T3,
and these clusters also exhibited the greatest changes over time (h2
ranged between .32 and .76). The other three clusters with signif-
icant changes over time (2, 5, and 7) only changed signicantly
between T1 and T2, however, and the changes were smaller (h2
ranged between .02 and .16). Table 5.

3.1. Description of the seven clusters

The standardized scores at T1, T2, and T3 for each cluster are
plotted in Fig. 2 and the results of the c2 analyses are presented in
Table 6. Each cluster is described below and when presented are
discussed in relation to the levels of the other clusters.

3.1.1. Cluster 1: increase

This cluster comprised 13% of the teachers and was charac-
terized by initially low levels of burnout that increased signicantly
over time. The increase in burnout was reected in increases in
turnover intentions as well as a decrease in SRH and TSE. Inter-
estingly, the c2 analyses showed that this cluster did not differ
signicantly in regards to any of the predictors.

Table 4
Results of the cluster analysis for burnout: size of respective cluster, means and
standard deviations for the different waves of measurement and homogeneity
coefcients for respective cluster.

Cluster N (%) T1 M (SD) T2 M (SD) T3 M (SD) Homogeneity

1 108 (13) 1.40 (.23) 1.64 (.30) 2.20 (.25) .14
2 89 (11) 2.23 (.32) 2.07 (.25) 2.02 (.33) .16
3 83 (10) 1.61 (.23) 2.14 (.32) 1.73 (.25) .15
4 71 (9) 2.10 (.28) 1.44 (.24) 1.70 (.27) .14
5 204 (25) 1.08 (.13) 1.16 (.16) 1.14 (.15) .04
6 44 (5) 2.50 (.44) 2.68 (.38) 2.70 (.48) .36
Fig. 2. Graphically plotted developmental patterns of burnout, TSE, turnover intentions,
7 217 (27) 1.40 (.21) 1.46 (.22) 1.45 (.20) .09
and SRH for respective cluster.
82 D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86

Table 5
Results of the rANOVAs analyzing change over time and whether the development differs among the clusters (Cluster  time interaction), the post-hoc tests concern
development over time for respective cluster regarding the different variables.

T1 T2 T3

F p h2 F p h2 F p h2
Burnout (Time) 19.52 <.001 .02 13.33 <.001 .02 7.66 .006 .01
Burnout (Cluster  time) 85.14 <.001 .39 86.79 <.001 .39 67.80 <.001 .34
Post-hoc (Time) 1 315.71 <.001 .75 49.33 <.001 .32 333.41 <.001 .76
2 9.73 <.001 .10 17.00 <.001 .16 .93 .338 .01
3 85.84 <.001 .51 124.74 <.001 .60 107.64 <.001 .57
4 111.65 <.001 .62 207.54 <.001 .75 32.75 <.001 .32
5 14.13 <.001 .07 25.69 <.001 .11 1.86 .174 .01
6 2.43 .097 .05 3.96 .053 .08 .04 .851 <.01
7 4.75 .009 .02 7.99 .005 .04 .29 .592 <.01
Turnover intention (Cluster  time) 6.86 <.001 .05 7.60 <.001 .05 7.21 <.001 .05
Post-hoc (Time) 1 16.68 <.001 .14 .01 .940 <.01 32.07 <.001 .23
2 .32 .718 <.01 .24 .624 <.01 .09 .770 <.01
3 13.99 <.001 .15 27.75 <.001 .25 8.78 .004 .10
4 8.13 .001 .10 15.91 <.001 .19 1.11 .296 .02
5 1.10 .329 .01 .63 .428 <.01 .68 .412 <.01
6 .14 .857 < .01 .14 .713 < .01 .34 .561 .01
7 .63 .533 < .01 .40 .529 <.01 1.38 .241 .01
TSE (Cluster  Time) 2.83 .001 .02 3.66 .001 .027 1.78 .10 .01
Post-hoc (Time) 1 7.05 .002 .06 10.48 .002 .092 .19 .66 <.01
2 9.40 <.001 .10 13.99 <.001 .14 1.41 .239 .02
3 15.38 <.001 .16 3.00 .087 .04 15.33 <.001 .16
4 26.37 <.001 .29 31.91 <.001 .33 .52 .472 <.01
5 11.98 <.001 .06 4.88 .028 .02 8.31 .004 .04
6 1.74 .182 .04 2.54 .118 .06 .02 .896 <.01
7 46.62 <.001 .18 41.13 <.001 .16 10.86 .001 .05
SRH (cluster  time) 2.53 .003 .019 3.75 .001 .028 1.53 .165 .011
Post-hoc (Time) 1 4.52 .012 .042 2.41 .124 .023 2.23 .139 .021
2 .34 .690 .004 .42 .518 .005 .82 .369 .009
3 3.80 .024 .045 7.34 .008 .083 3.28 .074 .039
4 4.72 .014 .064 9.72 .003 .123 1.80 .184 .025
5 .36 .695 .003 .59 .444 .003 .06 .814 <.001
6 .02 .982 <.001 .03 .872 .001 .02 .881 .001
7 .13 .882 .001 .23 .630 .001 .09 .761 <.001

3.1.2. Cluster 2: decrease younger (age  29) than those in other clusters. In regards to the
The second cluster comprised 11% of the teachers and was health-related indicators, the results showed that during higher
characterized by higher initial levels of burnout that decreased over education the members of this cluster had higher levels of
time. The decrease was, however, small and the high levels of depression and lower levels of life satisfaction. In regards to the
burnout were reected in high levels of turnover intentions along variables related to the individuals education, the results showed
with low levels of SRH and TSE. The results of the c2 analyses that members of this cluster had lower levels of TSE and experi-
showed that individuals in this cluster were, to a greater extent, enced more pressure related to their studies.

Table 6
Results of the c2 analyses used to predict cluster belonging. Bold numbers indicate adjusted residuals with z-values equal to or greater than 2.

Variables % c2 p

Cluster Total 1 2 3 4 5 6 7
Sex Female 85 87 82 87 69 90 89 86 20.63 .002
Age <29 39 46 54 41 35 25 48 41 48.43 <.001
30e39 30 31 21 39 42 31 27 27
>40 31 23 25 21 23 45 25 33

Health-related variables
Depression High 8 8 15 8 15 3 31 4 26.46 <.001
SRH High 88 85 85 82 80 97 74 90 29.75 <.001
Life satisfaction High 55 52 40 38 38 82 18 55 100.37 <.001

Educational outcomes
Achievement of educational goals High 70 76 67 68 58 79 56 68 18.15 .006
TSE High 94 94 87 95 93 99 86 95 20.15 .003
Satisfaction with education High 91 93 88 91 85 94 77 92 15.17 .019
Pressured by studies High 72 75 82 80 77 59 87 71 28.08 <.001
Pressured by occupational choice High 43 50 51 51 50 22 80 45 63.24 <.001
Spillover from studies to family High 67 66 71 78 69 58 77 69 13.99 .030

Organizational variables
Induction Yes 49 51 58 41 63 44 64 47 14.42 .025
D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86 83

3.1.3. Cluster 3: increase followed by decrease burnout during the initial period of employment for beginning
This cluster comprised 10% of the teachers and was charac- teachers. Furthermore, it was also of interest to study how con-
terized by initially low levels of burnout that increased between T1 current development for variables related to burnout differed be-
and the T2 and then recovered between T2 and T3. An identical tween clusters, and whether individual variables, educational
pattern was found for turnover intention, the opposite pattern was variables, and organizational variables could help explain cluster
found for SRH, and the levels of TSE increased over all three years. membership. When adopting a variable-based approach, the re-
The results of the c2 analyses showed that these individuals were sults indicated that burnout was a stable construct. Although there
older (age  40) and they had lower levels of life satisfaction during was a signicant increase in burnout over time, the changes were
higher education. small in magnitude. These ndings are in line with previous lon-
gitudinal studies on burnout (e.g., Brouwers & Tomic, 2000;
3.1.4. Cluster 4: decrease followed by increase Schaufeli & Enzmann, 1998; Shirom, Oliver, & Stein, 2009; Taris
This cluster comprised 9% of the teachers. They initially started et al., 2005), all indicating that the stability of burnout, despite
out fairly high but recovered between T1 and T2 only to increase being dened as a psychological state, is more in line with the
again between T2 and T3. Both of these changes over time were stability of a psychological trait. When a person-based approach is
signicant. Similar concurrent patterns were found for turnover adopted, however, the data tell a different story. The results showed
intention, and the opposite pattern was found for SRH and TSE. The that the clusters represented different trajectories, and that there
results of the c2 analyses showed that there were fewer females in were changes in burnout levels over time that evened out when the
this cluster and that there were more individuals between the ages variable-based approach was adopted. Based on previous results
of 30 and 39. They also exhibited signs of poor health during the describing the development of burnout it was expected to nd six
nal year of education (e.g., high levels of depression and low levels different types of trajectories. The expected trajectories were
of SRH and life satisfaction). In regards to the educational variables, identied, thus conrming the hypothesized developmental pat-
individuals within this cluster experienced lower levels of ach- terns. Even though an additional cluster with stable levels emerged,
ievement of educational goals. Interestingly, this was the only these individuals exhibited low levels of burnout and their devel-
cluster that differed concerning induction, and there were more opmental pattern thus corresponded to the expected group but
individuals in this cluster who received formal induction than in with low, stable burnout levels.
the other clusters. Although the trajectories of the clusters corresponded fairly well
to previously identied developmental patterns, they do not nec-
3.1.5. Cluster 5: stable low essarily represent typical burnout trajectories. It could therefore be
This cluster comprised 25% of the teachers and was charac- argued that the clusters identied were too sample-specic or too
terized by low levels of burnout and small changes over time. This context-dependent, and this might very well be the case. Although
was also reected in concurrent low levels of turnover intentions it is likely that some of these trajectories can be identied in other
combined with high levels of SRH and TSE. The results of the c2 teacher populations, the purpose of the study was not to identify
analyses showed that the individuals in this cluster were, to burnout trajectories that were generalizable to all teacher pop-
a greater extent, females and were older. In regards to mental and ulations. Rather, the aim of the study was to examine if the person-
physical health, the results showed that they were healthy in all based approach could be used to identify substantial patterns of
aspects (e.g., high levels of SRH and low levels of depression). In intra-individual change. The results show that the person-based
regards to educational outcomes, they experienced higher levels of approach offers a multifaceted perspective to the development of
TSE, achievement of educational goals, and lower levels of pressure burnout and that there is a signicant amount of individual varia-
related to studies and occupational choice. tion that may be lost when relying on a variable-based approach. It
thus appears that simply relying on a variable-based approach
3.1.6. Cluster 6: stable high when studying the development of burnout may be associated with
This cluster comprised 5% of the teachers and was characterized the risk of not identifying substantial developmental patterns as
by higher levels of burnout that did not change signicantly over well as drawing incorrect conclusions about change over time. The
time. This was reected in concurrent stable high levels of turnover identication of diverging change patterns allows for a better un-
intentions along with stable low levels of SRH and TSE. The high derstanding of change, and makes it possible to study which factors
levels of burnout were also reected in symptoms of poor mental affect change for that particular trend. For instance, an in-depth
and physical health during the nal year of higher education (e.g., analysis of cluster 1 will likely provide valuable insights about
high levels of depression and low levels of SRH). Furthermore, factors that drive the burnout process, whereas a closer examina-
during higher education these individuals experienced less TSE, tion cluster 3 might reveal which factors that are the causes of their
were less satised with their education, and experienced more recovery. Maybe teachers in cluster 1 were exposed to a poor
pressure related to both their studies and their occupational choice. psychosocial work climate (e.g., inadequate social support from
colleagues, heavy workload, difcult students, did not teach the
3.1.7. Cluster 7: stable moderate subjects they were trained for) that eventually became too much
This cluster comprised 27% of the teachers and was charac- and resulted in the chronic increase in burnout. Teachers in cluster
terized by initially moderate levels of burnout that changed little 3 on the other hand also appear to have had a rough start but
over time. Both TSE and SRH changed signicantly over time. The perhaps left their job and started teaching in a different school
results of the c2 analyses showed that this cluster only differed which was the reason for their recovery. Although these questions
concerning levels of depression and that individuals had, to are of great interest they are beyond the scope of this study.
a greater extent, lower levels of depression during higher Nonetheless, the results show great promise for future studies and
education. that a person-based approach provides researchers with a valuable
methodological tool for studying and understanding change over
4. Discussion time. This is the key nding of the study, and this is most likely the
case independent of the nationality of the study sample or the
The purpose of this study was to examine whether the use of cultural context. Thus, the person-based approach is a valuable
a person-based approach could identify developmental patterns of methodological resource regardless if one is interested in studying
84 D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86

development in burnout over time in Asia, Europe, or America. Previous research suggests that the initial period of employment
Furthermore, the ndings in the present study are not restricted to is especially challenging for beginning teachers (e.g., Cherniss,
the development of burnout. The person-based approach can just 1980a; Goddard et al., 2006). Although this was clearly the case
as well be used to improve the understanding of change in other for some, the overall ndings of the present study contradict this
relevant variables for beginning teachers such as TSE, turnover, or notion, and it was found that the majority of the participants
work engagement. experienced low levels of burnout. The results of the variable-based
The results also conrmed that the explanatory variables rela- analyses show that during the rst three years of employment the
ted to burnout changed in the hypothesized manner. Turnover beginning teachers were, on average, quite healthy and had mod-
intention was positively related to burnout, whereas SRH and TSE erately low, albeit increasing, levels of burnout. This was also
were inversely related to burnout. These ndings add validity to reected when examining the burnout levels of the clusters. The
the cluster solutions as well as conrm previous ndings on their two largest clusters (5 and 7), comprising just over half of the re-
relationship with burnout (e.g., Fernet, Guay, Senecal, & Austin, spondents, had low burnout levels (i.e., <1.50) across all three
2012; Friedman, 1993; Guglielmi & Tatrow, 1998; Lee & Ashforth, measurement waves, whereas only 5% of the participants (cluster
1996). The inverse relationship between burnout and TSE in- 6) had burnout high burnout levels (i.e., 2.50) across all three
dicates that teachers who do not feel condent in their ability to waves. Although the proportion of teachers with high burnout was
perform their job are more likely to develop burnout. These nd- small, burnout levels were above 2.00 (i.e., moderately high or high
ings correspond well to the ndings of Cherniss (1980a) regarding burnout) in 5 of the 7 clusters (48% of the beginning teachers) at
the negative impact of the crisis of competence on the develop- some time during the rst three years of employment. It should
ment of burnout. Self-efcacy beliefs are most malleable early in also be mentioned that the proportion of participants in the clus-
learning, and once established appear to remain stable unless ters with burnout levels exceeding 2.00 increased over time
persuasive and conicting evidence leads to re-evaluation (T1 25%, T2 26%, and T3 29%). Based on the total sample of
(Bandura, 1997). Considering that TSE is likely to be affected by individuals, the corresponding gures were 22% at T1, 24% at T2,
the teachers education, teachers who have not received sufcient and 29% at T3. In summary, the results showed that the majority of
knowledge and skills during their education to meet the qual- the beginning teachers had low levels of burnout, indicating that
ication requirements they are expected to hold after completing they coped well with the transition from education to employment.
their education are more likely to have less condence in their However, the results also showed that more than one in ten
ability and are thus more vulnerable to developing burnout. This experienced burnout at some point during this period, that about
notion is supported by the ndings in the present study in regards half of the teachers experienced moderately high burnout or high
to the impact of educational success (i.e., achievement of educa- burnout at some time, that there was an increasing trend in
tional goals, not being pressured by studies, and low levels of burnout levels, and that nearly one-third experienced moderately
spillover) on burnout, which has also been demonstrated in pre- high burnout or high burnout in the nal wave of measurement.
vious research (Gold et al., 1991). Given the stable nature of self- The problems experienced by many beginning teachers have
efcacy, ensuring the development of a strong sense of efcacy resulted in the development of induction programs aiming to make
during education and early in the professional career can be the transition from education to employment easier. Results of
expected to have longstanding positive effects, whereas failure to previous studies show that these programs have had positive ef-
do so might lead to longstanding detrimental effects (Tschannen- fects (Ingersoll & Strong, 2011; Smith & Ingersoll, 2004). However,
Moran, Hoy, & Hoy, 1998). when the results of the present study were examined, induction
Two things became apparent upon examination of the predic- appeared to have little effect on burnout. Only cluster 4 differed in
tive analyses. The rst was that indicators of mental and physical this aspect in that signicantly more of the members of this cluster
health as well as of educational success were related to future had received a formal induction. The shape of the trajectory could
levels of burnout. Individuals with good health and educational be an indication that the induction had an initial buffering effect on
success had lower initial levels of burnout whereas those who had burnout, but after the induction period was over the demands of
poor health and who experienced strain during their education employment caught up and burnout increased. It must be noted,
had higher initial levels of burnout. These results were in accord- however, that the only measure was whether the new teachers
ance with the hypothesized associations between the variables. received any induction or not, and that no qualitative aspects (e.g.,
The second insight was that although the predictors assessed type of activity, number of activities, or duration) of the induction
during the nal year of education could serve as indicators of were included. Including qualitative aspects of induction in the
future burnout, they did not clearly predict changes in the burnout assessment might have yielded different results. In an attempt to
trajectories. Instead they appeared to be related only to initial explain why teachers in cluster 4 received more induction they
levels of burnout. It could, however, be argued that these in- were compared to teachers in the other clusters. When comparing
dicators could predict the trajectories of clusters 5 (Stable low) and the characteristics of the clusters only two differences were evi-
6 (Stable high). Individuals within cluster 5 consistently showed dent. There were more males in cluster 4 and there were more
signs of good mental health, good physical health, and educational teachers that were between 30 and 39 years old. It seems likely that
success prior to entering employment, whereas the opposite pat- these were not the reasons to why they received more induction
tern was the case for cluster 6. The two clusters thus appear to and more plausible that this has to do with factors that were not
represent two groups of individuals: one non-vulnerable and one included in the study. Related to the strain experienced when
vulnerable. From a prevention perspective, the cluster that perhaps entering employment is the consistent nding that burnout is
was of greatest interest was cluster 1 (Increase). Considering that negatively related to age (Brewer & Shapard, 2004), indicating that
these individuals consistently increased in burnout, it would have younger teachers may be more vulnerable to burnout. Although
been interesting to have identied some potential early risk factors this nding is consistent, it has been highlighted previously that
of burnout. However, these individuals did not differ in any sig- age can be confounded with work experience (Maslach et al., 2001).
nicant manner, and were very typical. It thus appears as if In the present study, however, the participants all had the same
changes in burnout were mainly related to the concurrent situa- amount of work experience and age was still negatively related to
tion indicating that it might be the result of work-related or burnout, thus indicating that being older might have a protective
personal-life strain. effect against burnout.
D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86 85

4.1. Limitations of the study and recommendations for future Acknowledgments

The study received nancial support from the AFA insurance
The results of the attrition analysis showed that the response company and funding from Karolinska Institutet.
rate decreased over time, and that males and participants with
initially low levels of SRH had a lower response rate. It is thus References
necessary to take these issues into consideration when interpreting
the results. In regards to the effect of time on the response rate, Bandura, A. (1997). Self-efcacy: The exercise of control. New York: Freeman.
Bech, P., Rasmussen, N. A., Olsen, L. R., Noerholm, V., & Abildgaard, W. (2001). The
there are a variety of reasons why this could occur (e.g., paternity sensitivity and specicity of the Major Depression Inventory, using the Present
leave, sick leave, simply failing to respond), and it is therefore dif- State Examination as the index of diagnostic validity. Journal of Affective
cult to speculate what effect time had on the results. Because the Disorders, 66, 159e164.
Bergman, L. R., & El-Khouri, B. M. (2002). SLEIPNER e A statistical package for
proportion of males was so small in the sample and there is no
pattern-oriented analyses. Stockholm: Department of Psychology, Stockholm
strong relation between sex and burnout, it not likely that this had University.
any signicant effect on the results. The effect of SRH on attrition Bergman, L. R., Magnusson, D., & El-Khouri, B. (2003). Studying individual develop-
implies that individuals in the sample may be healthier compared ment in an interindividual context: A person-oriented approach. New Jersey:
Lawrence Erlbaum Associates.
to dropouts. Because burnout is negatively related to SRH there is Boersma, K., & Lindblom, K. (2009). Stability and change in burnout proles
chance that the burnout levels were underestimated. over time: a prospective study in the working population. Work & Stress, 23,
In the present study intra-individual trajectories of change were 264e283.
Breckenridge, J. (2000). Validating cluster analysis: consistent replication and
identied using cluster analysis. There are, however, alternative symmetry. Multivariate Behavioral Research, 35, 261e285.
person-based methods of analysis available, and one new method Brewer, E. W., & Shapard, L. (2004). Employee burnout: a meta-analysis of the
that is being more frequently used is latent growth mixture mod- relationship between age or years of experience. Human Resource Development
Review, 3, 102e123.
eling (LGMM). LGMM has many advantages compared to cluster Brouwers, A., & Tomic, W. (2000). A longitudinal study of teacher burnout and
analysis, and it could be argued that it would have been preferable perceived self-efcacy in classroom management. Teaching and Teacher Edu-
to have used this method of analysis rather than cluster analysis. cation, 16, 239e253.
Cherniss, C. (1980a). Professional burnout in human service organizations. New York:
However, when restricted to only three waves of measurement Praeger.
LGMM only allows for modeling linear trajectories, and because Cherniss, C. (1980b). Staff burnout: Job stress in the human services. Beverly Hills:
non-linear change was of interest it was decided to use cluster Sage Publications.
Cherniss, C. (1993). The role of professional self-efcacy in the etiology and amel-
analysis. Cluster analysis and LGMM are both person-based ap-
ioration of burnout. In W. B. Schaufeli, C. Maslach, & T. Marek (Eds.), Professional
proaches and are therefore more sensitive to identify different burnout: Recent developments in theory and research (pp. 135e150). Washington,
patterns of change compared to variable-based approaches. How- DC: Taylor & Francis.
ever, the subgroups identied by both methods are still based on Cherniss, C. (1995). Beyond burnout: Helping teachers, nurses, therapists, and lawyers
recover from stress and disillusionment. New York: Routledge.
aggregated levels and it is likely that there still are individual dif- Cohen, A. (1998). An examination of the relationship between work commitment
ferences within these subgroups. One approach to better under- and work outcomes among hospital nurses. Scandinavian Journal of Manage-
stand the effect of these differences is to add theoretically relevant ment, 14, 1e17.
Collins, L. M. (2006). Analysis of longitudinal data; the integration of theoretical
variables to the data analysis process. An additional alternative to model, temporal design, and statistical model. Annual Review of Psychology, 57,
gain better understanding of the burnout process is to use a qual- 505e528.
itative approach. In the same way as using a person-based approach Demerouti, E., Bakker, A. B., Nachreiner, F., & Schaufeli, W. B. (2001). The job
demands-resources model of burnout. Journal of Applied Psychology, 86,
can shine new light on variable-based results, using a qualitative 499e512.
approach might add a new perspective to the results of the person- Diener, E., Emmons, R. A., Larsen, R. J., & Grifn, S. (1985). The satisfaction with life
based results, especially if longitudinal data is available. scale. Journal of Personality Assessment, 49, 71e75.
Evers, W. J. G., Tomic, W., & Brouwers, A. (2004). Burnout among teachers: stu-
Although the SWEBO has been psychometrically validated,
dents and teachers perceptions compared. School Psychology International, 25,
a clinical validation of the scale is needed to make more accurate 131e148.
estimations of the prevalence of burnout and to allow for com- Fernet, C., Guay, F., Senecal, C., & Austin, S. (2012). Predicting intraindividual
changes in teacher burnout: the role of perceived school environment and
parisons with other more established burnout scales. To get a better
motivational factors. Teaching and Teacher Education, 28, 514e525.
sense of the effect of induction programs on burnout future studies Fives, H., Hamman, D., & Olivarez, A. (2007). Does burnout begin with student-
need to include qualitative dimensions of the induction programs teaching? Analyzing efcacy, burnout, and support during the student-
(e.g., content, structure, length, and participant satisfaction). One teaching semester. Teaching and Teacher Education, 23, 916e934.
Freudenberger, H. J. (1975). The staff burnout syndrome in alternative institutions.
aspect that was not included in the present study was behaviorally Psychotherapy: Theory, Research and Practice, 12, 73e82.
related outcomes of burnout (e.g., actual turnover and sick leave). Friedman, I. A. (1993). Burnout in teachers: the concept and its unique core
This issue needs to be addressed in future studies to better un- meaning. Educational and Psychological Measurement, 53, 1035e1044.
Friedman, I. A. (2000). Burnout in teachers: shattered dreams of impeccable pro-
derstand the consequences of burnout. fessional performance. Journal of Clinical Psychology, 56, 595e606.
Gavish, B., & Friedman, I. A. (2010). Novice teachers experience of teaching: a dy-
5. Conclusions namic aspect of burnout. Social Psychology of Education, 13, 141e167.
Gilbert, L. (2005). What helps beginning teachers. Educational Leadership, 62,
The present study both strengthens and expands upon previous Glass, G. V., & Hopkins, K. D. (1995). Statistical methods in education and psychology.
ndings. The results of the study conrm the previously found Boston: Allyn & Bacon.
Goddard, R., OBrien, P., & Goddard, M. (2006). Work environment predictors of
relationships between identied predictors of burnout. Of greater beginning teacher burnout. British Educational Research Journal, 32, 857e874.
interest is, however, the ndings supporting the usefulness of Gold, Y., Roth, R. A., Wright, C. R., & Michael, W. B. (1991). The relationship of scores
a person-based approach when studying change in burnout over on the Educators Survey, a modied version of the Maslach Burnout Inventory,
to three teaching-related variables for a sample of 132 beginning teachers.
time. The person-based approach offers a new non-linear per-
Educational and Psychological Measurement, 51, 429e438.
spective on the development of burnout compared to studies only Guarino, C. M., Santibanez, L., & Daley, G. A. (2006). Teacher Recruitment and
using variable-based methods. Capitalizing on this alternative retention: a review of the recent empirical literature. Review of Educational
method when studying change over time will hopefully lead to new Research, 76, 173e208.
Guglielmi, R., & Tatrow, K. (1998). Occupational stress, burnout, and health in
insights not only into the burnout process but also on development teachers: a methodological and theoretical analysis. Review of Educational
in other variables relevant to teaching. Research, 68, 61e99.
86 D. Hultell et al. / Teaching and Teacher Education 32 (2013) 75e86

Hakanen, J. J., Schaufeli, W. B., & Ahola, K. (2008). The Job Demands-Resources Nixon, A. E., Mazzola, J. J., Bauer, J., Krueger, J. R., & Spector, P. E. (2011). Can work
model: a three-year cross-lagged study of burnout, depression, commitment, make you sick? A meta-analysis of the relationships between job stressors and
and work engagement. Work & Stress, 22, 224e241. physical symptoms. Work & Stress, 25, 1e22.
Hobfoll, S. E., & Shirom, A. (2000). Conservation of resources theory: applications to Roberts, B., & DelVecchio, W. (2000). The rank-order consistency of personality
stress and management in the workplace. In R. T. Golembiewski (Ed.), Handbook traits from childhood to old age: a quantitative review of longitudinal studies.
of organization behavior (2nd ed.). (pp. 57e81) New York, New York: Marcel Psychological Bulletin, 126, 3e25.
Dekker. Rudman, A., & Gustavsson, J. P. (2011). Early-career burnout among new graduate
Hobfoll, S. E. (1989). Conservation of resources. A new attempt at conceptualizing nurses: a prospective observational study of intra-individual change trajec-
stress. American Psychologist, 44, 513e524. tories. International Journal of Nursing Studies, 48, 292e306.
Hultell, D., & Gustavsson, J. P. (2010a). A psychometric evaluation of the Scale of Salmela-Aro, K., & Nurmi, J.-E. (2007). Self-esteem during university studies pre-
Work Engagement and Burnout (SWEBO). Work, 37, 261e274. dicts career characteristics 10 years later. Journal of Vocational Behavior, 70,
Hultell, D., & Gustavsson, J. P. (2010b). The manual of the Scale of Work Engagement 463e477.
and Burnout (SWEBO) (No. B 2010:1). Stockholm: Karolinska Institutet. Salmela-Aro, K., Tolvanen, A., & Nurmi, J.-E. (2009). Achievement strategies during
Hultell, D., & Gustavsson, J. P. (2011). Factors affecting burnout and work engage- university studies predict early career burnout and engagement. Journal of
ment in teachers when entering employment. Work, 40, 85e98. Vocational Behavior, 75, 162e172.
Hultell, D. (2011). Lost in transition? A study of newly graduated teachers experiences Schaufeli, W. B., & Enzmann, D. (1998). The burnout companion to study and practice:
during the initial period of employment. Ph.D. thesis, Karolinska Institutet, A critical analysis. London: Taylor & Francis.
Stockholm. from Schwarzer, R., & Hallum, S. (2008). Perceived teacher self efcacy as a predictor of
Ingersoll, R. M., & Strong, M. (2011). The impact of induction and mentoring pro- job stress and burnout: mediation analyses. Applied Psychology, 57, 152e171.
grams for beginning teachers: a critical review of the research. Review of Edu- Shirom, A., & Melamed, S. (2006). A comparison of the construct validity of two
cational Research, 81, 201e233. burnout measures in two groups of professionals. International Journal of Stress
Ingersoll, R. M. (2012). Beginning teacher induction: what the data tell us. Phi Delta Management, 13, 176e200.
Kappan, 93, 47e51. Shirom, A., Oliver, A., & Stein, E. (2009). Teachers stressors and strains: a longitu-
Johnson, S. M., & Birkeland, S. E. (2003). Pursuing a "sense of success": new dinal study of their relationships. International Journal of Stress Management, 16,
teachers explain their career decisions. American Educational Research Journal, 312e332.
40, 581e617. Shirom, A. (2005). Reections on the study of burnout. Work & Stress, 19, 263e270.
Kramer, M. (1974). Reality shock: Why nurses leave nursing. Saint Louis: Mosby. Shirom, A. (2010). Job-related burnout: a review of major research foci and chal-
Kuh, G. D., Hayek, J. C., Carini, R. M., Ouimet, J. A., Gonyea, R. M., & Kennedy, J. lenges. In J. Campell Quick, & L. E. Tetrick (Eds.), Handbook of occupational health
(2001). NSSE technical and norms report. Bloomington: Indiana University, psychology (2nd ed.). (pp. 223e242) Washington DC: American Psychological
Center for Postsecondary Research and Planning. Association.
Kyriacou, C. (2001). Teacher stress: directions for future research. Educational Shirom, A. (2011). Job-related burnout: a review of major research foci and chal-
Review, 53, 27e35. lenges. In J. Campell Quick, & L. E. Tetrick (Eds.), Handbook of occupational health
Lee, R. T., & Ashforth, B. E. (1996). A meta-analytic examination of the corre- psychology (2nd ed.). (pp. 223e242) Washington DC: American Psychological
lates of the three dimensions of job burnout. Journal of Applied Psychology, Association.
81, 123e133. Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Modeling
Mkikangas, A., Hyvnen, K., Leskinen, E., Kinnunen, U., & Feldt, T. (2011). A person- change and event occurrence. New York: Oxford University Press.
centred approach to investigate the development trajectories of job-related Skaalvik, E., & Skaalvik, S. (2010). Teacher self-efcacy and teacher burnout: a study
affective well-being: a 10-year follow-up study. Journal of Occupational and of relations. Teaching and Teacher Education, 26, 1059e1069.
Organizational Psychology, 84, 327e346. Smith, T. M., & Ingersoll, R. M. (2004). What are the effects of induction and
Maslach, C., & Jackson, S. E. (1986). Maslach burnout Inventory manual (2nd ed.). Palo mentoring on beginning teacher turnover? American Educational Research
Alto, Califorina: Consulting Psychologists Press, Inc. Journal, 41, 681e714.
Maslach, C., & Leiter, M. P. (1997). The truth about burnout. San Francisco: Jossey- SPSS Inc. (2009). SPSS 18.0 for windows. Chicago: SPSS Inc.
Bass Publications. Strong, M. (2005). Teacher induction, mentoring, and retention: a summary of the
Maslach, C., Jackson, S. E., & Leiter, M. P. (1996). Maslach burnout Inventory manual research. The New Educator, 1, 181e198.
(3rd ed.). Palo Alto, CA: Consulting Psychologists Press. Taris, T. W., Le Blanc, P. M., Schaufeli, W. B., & Schreurs, P. J. G. (2005). Are there
Maslach, C., Schaufeli, W. B., & Leiter, M. P. (2001). Job burnout. Annual Review of causal relationships between the dimensions of the Maslach Burnout In-
Psychology, 52, 397e422. ventory? A review and two longitudinal tests. Work & Stress, 19, 238e255.
Melamed, S., Shirom, A., Toker, S., Berliner, S., & Shapira, I. (2006). Burnout and risk Tschannen-Moran, M., Hoy, A. W., & Hoy, W. K. (1998). Teacher efcacy: its meaning
of cardiovascular disease: evidence, possible causal paths, and promising and measure. Review of Educational Research, 68, 202e248.
research directions. Psychological Bulletin, 132, 327e353. Veenman, S. (1984). Perceived problems of beginning teachers. Review of Educa-
Moya-Albiol, L., Serrano, M. A., & Salvador, A. (2010). Burnout as an important tional Research, 54, 143e178.
factor in the psychophysiological responses to a work day in teachers. Stress Vinokur, A. D., Pierce, P. F., & Lewandowski-Romps, L. (2009). Disentangling the
and Health: Journal of the International Society for the Investigation of Stress, 26, relationships between job burnout and perceived health in a military sample.
382e393. Stress and Health, 25, 355e363.