The Multivariate Wald-Wolfowitz Test for Recognizing Primary Teachers’ Burnout Syndrome with the Maslach Burnout Inventory Tool

The Maslach Burnout Inventory (MBI) is an instrument for measuring occupational burnout across different professions, leveraging 22 standardized questions. MBI considers three factors contributing to burnout, namely emotional exhaustion, depersonalization, and personal
achievements.
We explore the multivariate Wald-Wolfowitz test (WW-test), a nonparametric statistical test, as a novel means to analyze the answers to the 22 MBI questions, exploring the power of the non-Gaussian distribution hypothesis for the sample responses. As a research study, we apply and compare the proposed technique with an existing research paper studying burnout syndrome in primary education teachers in the Prefecture of Achaia, Greece.
We show that the WW-test not only offers similar analytical power compared to the established (statistical) tests (e.g., t-test and ANOVA) but also allows for deeper analysis and new insights, not previously identified in the literature.

Introduction

The World Health Organization (WHO) included burnout syndrome as an occupational phenomenon in its 11^th revision of the International Classification of Diseases (ICD-11). WHO defined it as “[..] a syndrome conceptualized as resulting from chronic workplace stress that has not been successfully managed. It is characterized by three dimensions: feelings of energy depletion or exhaustion; increased mental distance from one’s job, or feelings of negativism or cynicism related to one’s job; and reduced professional efficacy” (WHO, 2019).

In the context of the teaching profession, factors contributing to burnout include the lack of support from colleagues, heavy workloads, organizational culture, lack or ambiguity of duties, unsatisfactory salary, and work-life balance (Aloeet al., 2014; Brunstinget al., 2014; Halbesleben & Buckley, 2004; Jacksonet al., 1986; Kokkinos, 2007; Lee & Ashforth, 1996; Maslachet al., 2001; Melamedet al., 2006; Schaufeli & Bakker, 2004; Toker & Biron, 2012).

The Maslach Burnout Inventory (MBI) is an instrument to assess burnout feelings, initially designed for human services professionals, such as nurses, physicians, social workers, and police (Maslach & Jackson, 1981; Maslach, 1982; Maslachet al., 1997; Maslachet al., 2001; Maslachet al., 2018; Schaufeliet al., 2009). The original version, nowadays mentioned as MBI-HSS, was later adapted as MBI-HS (MP) for medical personnel, MBI-ES for educator surveys, MBI-GS for surveys not already covered by the previous, and MBI-GS (S) specifically for college and university students. In all cases, MBI comprises 22 standardized questions with adjusted wording. MBI is answered on a 7-point Likert scale, from 0 (Never) to 6 (Every day), reporting how often a person has specific feelings related to their work.

The 22 MBI questions are classified into three classes, reflecting the three burnout components (or dimensions according to WHO): emotional exhaustion (EE, nine questions), depersonalization (DP, five questions), and personal achievements (PA, eight questions). For each component, the answers are summed to derive an intensity level ranging from low (EE: 0-18, DP: 0-5; and PA: 40-48) to medium (EE: 19-26; DP: 6-9; and PA: 34-39), to high (EE: 27-54, DP: 10-30, and PA: 0-33). We note that higher EE and DP intensity imply burnout but lower PA implies burnout.

The established practice in the research literature is to analyze the answers to the MBI questionnaire against the demographics of the sample, such as gender and career stage. This often assumes a normal (Gaussian) distribution of the answers to simplify the analysis by applying traditional statistical techniques (e.g., the t-test).

The multivariate Wald-Wolfowitz test (WW-test) is a non-parametric test that assesses the commonality between two sets of multivariate observations (Friedman & Rafsky, 1979; Theoharatoset al., 2004; Zahn, 1971). This is an extension to the classical WW-test and compares two different tuples of vectorial observations by checking whether they form different branches in the overall minimal spanning tree (Cheriton & Tarjan, 1976; Gautam, 2025; Kruskal, 1956; Prim, 1957). The WW test calculates a value W that denotes the (dis)similarity of the two sets: The more negative the value is, the more dissimilar the two sets are (Theoharatoset al., 2004). Based on the W value, one can derive the p-value indicating the statistical significance of the null hypothesis (Theoharatoset al., 2004). The primary advantages of the WW-test are twofold: a) it makes no a priori assumption about the distribution and b) it is appropriate for small samples. Implementations and codes for the WW-test can be easily found in the international literature.

In this paper, we propose the WW-test approach for assessing the collected MBI answers as the “signatures” of the responders for burnout syndrome. In this sense, we consider each answered questionnaire both as a vector of 22 values (22 dimensions) and as three vectors, one for each burnout component with dimensions 9, 5, and 8 respectively.

The rest of this paper is structured as follows. First, we present our method regarding the collected data (answers to the MBI-ES questionnaire) for the specific research study. Then, we present the results of applying the WW test for the analysis of the answers. The discussion of our findings follows, concluding with possible future directions of work.

Materials and Methods

The research study is the burnout feeling of in-service primary education teachers in the Prefecture of Achaia, Greece. Primary education in Greece comprises a) kindergarten schools and b) elementary education schools. Attendance at kindergarten schools is compulsory for two years for all children being 4- and 5 years old. At the age of six, children enroll in elementary (also called “public”) schools, where attendance is compulsory for six more years. The Greek state provides public primary education for free. It is a requirement to hold a university-level degree in order to be employed as a teacher either in a kindergarten or elementary education school. Each type of school requires a different University degree (Eurydice, 2024). Achaia is located in Western Greece and features a large number of primary education schools, both small and large ones, in big cities and remote mountain villages alike, attracting teachers at all stages of their career and, thus, providing a representative case of country-wide demographics (Fotopoulou, 2013).

The teacher burnout feeling was assessed using an anonymous MBI-ES questionnaire administered in earlier work (Gkamari & Fotopoulou, 2024). There are five demographics of interest for our sample that may contribute to the occurrence of burnout: 1) gender; 2) additional tertiary education studies (e.g., an M.Sc. or Ph.D. degree); 3) years of in-service teaching experience; 4) type of school serving (kindergarten or elementary school); and 5) degree of satisfaction with the salary received.

Our research data comprise the n = 258 MBI-ES questionnaires filled by the participating kindergarten and elementary school teachers. We represent the 22 answers of each participant as a vector in the R^d space, where d = 22 for the whole burnout topic, d = 9 for the emotional exhaustion component, d = 5 for the depersonalization one, and d = 8 for the (reduced) personal achievements. In this sense, the answers to the 22 questions for each of the n = 258 questionnaires are mapped to a matrix comprising 258 rows and 22 columns, where every cell contains an integer value between 0 (Never) and 6 (Every day). Each row represents the answer of one teacher, i.e., their “signature” for burnout.

We implemented the WW test as a Matlab script leveraging on the built-in graphminspantree function for calculating the minimal spanning tree (MST). The script running time for all vectors (d = 22, 9, 5, and 8) of the sample (n = 258) was unnoticeable, comparable to running a classical t-test in SPSS. To the interested readers, we note that there are also readily available implementations for the WW-test in other programming languages, such as Python (e.g.,  https://shorturl.at/qa9kF and  https://shorturl.at/uLQkE).

Results

In the next, we examine how burnout correlates with the five chosen demographics presented in the previous section.

Salary Satisfaction and Burnout

Salary satisfaction is recorded with values between 1 and 5 in our sample (no 0 or 6 responses). We exclude those rows with value 5, being rare (n = 3), and end up with four sets of 63, 81, 83, and 28 teachers respectively for 1, 2, 3, and 4. For the sake of simplicity, the sets are named after the Likert scale value. We test the similarity of the 255 vectors using the WW-test.

We compare the four sets considering as the characteristic vector for each teacher all the 22 MBI-ES questions (overall burnout). Our objective is to examine whether the sets derive from the same population, i.e., they have the same distribution in the 22-dimensional space. Table I summarizes the results for both the W- and p-value.

Table I. WW-Test and p-Values for Salary Satisfaction (1–4) and Overall Burnout

Overall burnout (d = 22)
	2	3	4
	W/p-value	W/p-value	W/p-value
1	0.4017/NaN	−2.18/0.0146	−2.95/0.0016
2		−0.235/0.407	−4.07/0.0002

There is a statistically significant difference between sets 1 and 4 (W = −2.95, p = 0.0016): Those teachers ranking differently their salary satisfaction also differ in overall burnout levels. The same holds for sets 1 and 3 (W = −2.18, p = 0.0146) and sets 2 and 4 (W = −4.07, p = 0.0002). Hence, the more the income satisfaction (higher value), the less (lower value) the overall burnout.

As a next step, we compare the four sets based only on the “emotional exhaustion” (d = 9) component of burnout (respective mean values: 27.17; 30.22; 24.13; and 17.82). Table II summarizes the results for both the W- and p-value. There is a strong dissimilarity and statistically significant difference between sets 1 and 4 (W = −3.703, p = 0.0001); sets 2 and 3 (W = −2.5873, p = 0.0048); and sets 2 and 4 (W = −3.14, p = 0.0008). Hence, it can be confirmed that emotional exhaustion decreases as salary satisfaction increases. This is similarly reported in (Gkamari & Fotopoulou, 2024).

Table II. WW-Test and p-Values for Salary Satisfaction and Emotional Exhaustion Burnout Component

Emotional exhaustion (d = 9)
	2	3	4
	W/p-value	W/p-value	W/p-value
1	−0.1032/0.4589	−1.5159/0.0648	−3.703/0.0001
2		−2.5873/0.0048	−3.14/0.0008

Then, we compare the sets based on the “depersonalization” (d = 5) component (respective mean values: 7.64; 8.72; 7.96; 5.28). Table III summarizes the results for both the W- and p-value. Again, we note a strong dissimilarity and statistically significant difference between sets 1 and 4 (W = −2.344, p = 0.0095) and sets 2 and 4 (W = −4.016, p = 0.00003). The mean value for set 4 (5.28) is much lower compared to the other three. Therefore, we can assume a strong negative (reverse) relationship between salary satisfaction and depersonalization. The more the salary satisfaction, the less the depersonalization. This contrasts with Gkamari and Fotopoulou (2024), where no statistical significance was identified using Spearman’s rank correlation coefficient test.

Table III. WW-Test and p-Values for Salary Satisfaction and Depersonalization Burnout Component

Depersonalization (d = 5)
	2	3	4
	W/p-value	W/p-value	W/p-value
1	−1.605/0.054	−0.647/0.258	−2.344/0.0095
2		0.87/NaN	−4.016/0.00003

Last, we compare the four sets based on the “personal achievements” (d = 8) burnout component (respective mean values: 29.8; 31.0; 31.61; and 34.25). Table IV summarizes the results for both the W- and p-value. The W-values are negative, yet not as big (in absolute value) as the previous cases. There is a weak dissimilarity but not a statistically significant difference this time, as all p-values are higher than 0.05. Hence, we derive that salary satisfaction and personal achievements are not correlated. This is similarly reported in (Gkamari & Fotopoulou, 2024).

Table IV. WW-Test and p-Values for Salary Satisfaction and Personal Achievements Burnout Component

Personal achievements (d = 8)
	2	3	4
	W/p-value	W/p-value	W/p-value
1	−1.26/0.103	−1.515/0.0648	−0.8/0.211
2		−0.3915/0.3477	−1.229/0.1095

Years of Teaching Experience and Burnout

We classify the answers in three buckets based on years of teaching experience: bucket 1 contains those having up to 10 years; bucket 2 contains those having between 11 and 20 years; and bucket 3 those having between 21 and 30 years. Similarly to the previous section, we compare the overall burnout and its three components.

Table V summarizes the results for both the W- and p-value. Neither overall burnout nor its components show any significant dissimilarity or statistically significant difference when it comes to years of teaching experience except in two marginal cases: When comparing bucket 1 with bucket 3 for the emotional exhaustion (W = −1.66, p = 0.048) and personal achievements (W = −1.6641, p = 0.048). This means that burnout equally affects teachers at all stages of their careers rather than only in their early days in service. These findings are in alignment with (Gkamari & Fotopoulou, 2024).

Table V. WW-Test and p-Values for Years of Teaching Experience

Overall burnout (d = 22)
	11–20 years	21–30 years
	W/p-value	W/p-value
1–10 years	−0.278/0.3904	−1.4919/0.0679
11–20 years		−0.2497/0.4014
Emotional exhaustion (d = 9)
	11–20 years	21–30 years
	W/p-value	W/p-value
1–10 years	−0.278/0.390	−1.66/0.048
11–20 years		−0.605/0.273
Depersonalization (d = 5)
	11–20 years	21–30 years
	W/p-value	W/p-value
1–10 years	0.0585/NaN	−0.4603/0.3227
11–20 years		−0.4325/0.3327
Personal achievements (d = 8)
	11–20 years	21–30 years
	W/p-value	W/p-value
1–10 years	−0.4464/0.327	−1.6641/0.048
11–20 years		−0.605/0.2726

Additional Degrees and Burnout

We dichotomize the sample into those having the mandatory degree only (n = 92) and those having an additional tertiary education degree (n = 166), i.e., a second degree (n = 34), an M.Sc. (n = 119), or a Ph.D. (n = 13). Table VI summarizes the results for both the W- and p-value.

Table VI. WW-Test and p-Values for Additional Degrees

	W/p-value	Mean values
Overall burnout (d = 22)	−3.6020/0.0016
Emotional exhaustion (d = 9)	−2.4988/0.006	25.8675 vs. 26.2418
Depersonalization (d = 5)	−3.0979/0.01	7.9157 vs. 7.6484
Personal achievements (d = 8)	0.3116/NaN	32.1325 vs. 28.8791

There is a noticeable dissimilarity and statistically significant difference between the two groups for overall burnout (W = −3.60, p = 0.0016), emotional exhaustion (W = −2.50, p = 0.006), and depersonalization (W = −3.10, p = 0.01) components, despite the mean values being quite close. In contrast, the distance between the mean values for personal achievement components, albeit large in absolute value, is neither statistically significant nor attributed to a different distribution (W > 0, indicating similarity). Overall, this means that the possession of additional degrees is a differentiation factor when it comes to burnout. These findings highlight the analytical power of the WW-test; the ANOVA test used by Gkamari and Fotopoulou (2024) did not identify any statistically significant difference in any of the three burnout components or the overall burnout.

School Type and Burnout

Participants worked either in kindergartens (n = 87) or elementary education schools (n = 167). Four respondents did not answer this question, and the respective rows are excluded from the analysis below. Table VII summarizes the results for both the W- and p-value.

Table VII. WW-Test and p-Values for the Type of School (E: Elementary, K: Kindergarten)

	W/p-value	Mean values
Overall burnout (d = 22)	−2.7472/0.009
Emotional exhaustion (d = 9)	−2.508/0.006	E = 27.14 / K = 23.977
Depersonalization (d = 5)	−2.950/0.0016	E = 8.682 / K = 6.218
Personal achievements (d = 8)	−0.692/0.2442	E = 29.748 / K = 33.16

We observe a pattern similar to the previous case of additional degrees. The emotional exhaustion (W = −2.51, p = 0.006) and depersonalization (W = −2.95, p = 0.0016) components exhibit noticeable dissimilarity and statistically significant differences. These two components further contribute an alike result for overall burnout (W = −2.75, p = 0.009). Again, personal achievements are not a differentiating factor and their contribution to overall burnout is not strong, despite the large difference in mean values. Our findings differ from those in (Gkamari & Fotopoulou, 2024), where all three components exhibit statistically significant differences; here only emotional exhaustion and depersonalization do.

Gender and Burnout

Last but not least, we dichotomized the sample by the gender demographic: There are n = 67 male and n = 190 female participants. One respondent did not answer this question and the respective row is excluded from the analysis below. Table VIII summarizes the results for both the W- and p-value.

Table VIII. WW-Test and p-Values for Gender (M = male, F = female)

	W/p-value	Mean values
Overall burnout (d = 22)	−0.7291/0.2330
Emotional exhaustion (d = 9)	−1.8613/0.0313	M = 32.46 / F= 23.928
Depersonalization (d = 5)	1.44/NaN	M = 11.55 / F = 6.60
Personal achievements (d = 8)	−2. 391/0.0084	M = 26.47 / F = 32.19

Emotional exhaustion (W = −1.86, p = 0.0313) and personal achievements (W = −2.39, p = 0.0084) exhibit noticeable dissimilarity and statistically significant differences between male and female participants, which is also reflected in the respective mean values. The depersonalization component does not exhibit a similar behavior, despite the clear difference in mean values. The overall contribution of emotional exhaustion and personal achievements is not very strong (W = −0.73). It appears that depersonalization overshadows the other two components when it comes to overall burnout.

When it comes to emotional exhaustion and personal achievement, there is an agreement with earlier works of Andreou (2019), Gkamari and Fotopoulou (2024), Kamtsios and Lolis (2016); and Kosyvaset al. (2023) and contrast Tzima (2014) that reports no gender differentiation. However, when it comes to depersonalization, the WW-test indicates similarity, in agreement with Tzima (2014), while the other works indicate statistically significant differences between male and female participants.

Discussion

We explored the applicability of the multivariate WW-test in measuring burnout using the MBI instrument. To assess its analytical power, we compared it against an earlier work (Gkamari & Fotopoulou, 2024) that analyzed MBI-ES answers using classical statistical tools. Table IX summarizes whether the results here differ from those in the previous work. In the majority of the cases, the WW test offered the same results as the earlier work. There are a few notable exceptions that are worth discussing further below.

Table IX. Overlap of Findings with (Gkamari & Fotopoulou, 2024)

	EE	DP	PA
Gender	Same	Different	Same
School type	Same	Same	Different
Additional degrees	Different	Different	Different
In-service years	Same	Same	Same
Salary satisfaction	Same	Different	Same

When it comes to gender, depersonalization appears to be indifferent between males and females, despite the overall difference in mean values and intensity (medium vs. high) in the specific sample. As there is no consensus in the research literature on whether DP is gender-specific or not, this finding contributes to the ongoing discussion and calls for further studies in the future.

When it comes to additional degrees, the WW test identifies significant differences between those having and those not having, while earlier work did not identify such a correlation. This emphasizes the power of the WW test: Despite the close overall mean values, it is still able to distinguish the two classes as having different behavior. Hence, it is an interesting line of future work to further investigate how the additional degrees may impact burnout feelings.

WW-test reaffirmed that years-in-service do not impact the burnout feeling, nor EE, DP, or PA. This is a worrying finding in that burnout equally affects teachers at all stages of their careers rather than only in their early days in service. Hence, support mechanisms should be developed for everyone, not only newcomers to the profession.

When it comes to salary satisfaction, the WW test can identify a previously unreported correlation with DP: those ranking high, i.e., 4 out of 5, their salary satisfaction tend to have low DP feelings, while the rest tend to have medium ones. This finding calls for a deeper and larger future study with an increased number of participants to confirm it and explore means to improve salary satisfaction. This is further justified as the mean value (5.28) for those high-ranking in our sample lies in the borderline of low (range 0–5) and medium (6–9).

Conclusions

We proposed the multivariate WW-test as a method for analyzing MBI responses for assessing the burnout syndrome. The proposed vector-processing approach considers the 22 MBI-ES questionnaire responses as the signature of each teacher. The essential advantages of the WW test are its simplicity and analytical power even for small samples, which is a common pattern in this kind of studies. The initial findings are promising and offer new directions of work. We also aim to apply the WW test in further studies, complementing established statistical analysis techniques.