The concept of job crafting can be defined as ‘self-initiated change behaviours that employees engage in with the aim to align their jobs with their own preferences, motives, and passions (Wrzesniewski and Dutton, 2001; see also Berg et al., 2010)’ (Tims and Bakker, 2010, p. 173). There are different conceptualizations of job crafting. We will focus here on the conceptualization based on the job demands-resources model (JD-R; Bakker and Demerouti, 2017), differentiating between increasing structural resources, increasing social resources, increasing challenging demands, and decreasing hindering demands (Tims and Bakker, 2010; Tims et al., 2012).

The JD-R was adapted to the higher education context (Clements and Kamau, 2018; Lesener et al., 2020). Students cope with comparable demands like time-, and performance pressure and access similar resources like support and autonomy as employees. Likewise, the concept of job crafting has been applied to the study context (Ferreira, 2020; Körner et al., 2021). Increasing structural study resources can include, for example, asking lecturers for more latitude; increasing social study resources can be achieved by proactively asking lecturers for feedback; and increasing challenging study demands is about getting involved in new and interesting projects. On the other hand, reducing hindering study demands corresponds to postponing or avoiding too mentally or emotionally demanding tasks. In a study among first-year students in South Africa, an increase in structural resources was positively related to study engagement (Ferreira, 2020), but an increase in social resources was not associated with increased engagement in this sample. In a weekly diary study in a sample of higher education students enrolled at a German university, study resources were shown to be indirectly related to an increase in structural and social resources via increased study engagement (Körner et al., 2021).

Data collected by Mäkikangas (2018) from Finnish rehabilitation center employees using latent cluster analysis revealed two distinct profiles of job crafting behaviour. The vast majority of employees (94%) representing the actively crafting group showed slightly more crafting in all dimensions. The second group comprised the so-called passive job crafters (6%), who mainly engaged in reducing hindering demands. The group that showed more approach crafting behaviour scored higher in work engagement. Similar findings on the distinctions between the two domains of approach (promotion) and avoidance (prevention) have been reported in meta-analyses (Rudolph et al., 2017; Bruning and Campion, 2018; Lichtenthaler and Fischbach, 2019; Lazazzara et al., 2020). There were also divergent empirical results related to job crafting dimensions, as proposed by Tims and Bakker (2010). Whereas positive outcomes were reported for increasing structural and social resources, as well as for increasing challenging demands, reducing hindering demands was shown to be negatively related to engagement and satisfaction and positively related to emotional exhaustion (Lichtenthaler and Fischbach, 2019; Zhang and Parker, 2019).

The COVID-19 pandemic changed the way of studying in universities in various ways. The spatial conditions shifted from the university to home, making building and cultivating social support or friendships more complex. Asynchronous teaching formats, delivered by videos and task assignments on demand, resulted in the increased need for self-management (Hoss et al., 2021), and also might have increased opportunities for students to proactively seek more structural resources, and challenging demands.

Our study focused on approach crafting, seeking further differentiation within this category. Two aspects guided our decision: (1) based on the findings reported by Mäkikangas (2018), only a marginal share of respondents (6%) were categorized as avoidance crafters, (2) meta-analytic findings (Rudolph et al., 2017; Lichtenthaler and Fischbach, 2019) showed that avoidance crafting could not be seen as a beneficial strategy (albeit causality for the negative relationship of avoidance crafting with burnout is still an open question). Approach crafting showed consistently positive relationships with motivation, and well-being, and hence should be the focus of interventions aiming at increasing the use of crafting strategies.

To capture the holistic interplay between the different facets of study crafting, we looked for profiles used as potential predictors for different health and motivational outcome criteria. Students may pursue other crafting dimensions more or less compared to employees, and combinations of crafting strategies might well show differential effects for students. In the first step, we looked for homogeneous subsamples of students that could be characterized by similar combinations of study crafting strategies. As LPA is an explorative endeavour, we formulated an open research question:

In a second step, we aim to investigate potential differences between the study crafting profiles concerning health and well-being indicators. Building on previous research on job crafting and the JD-R (Bakker and Demerouti, 2017), we used emotional exhaustion and study engagement as outcome variables. Since student health is of overarching importance, we furthermore included general well-being. Whereas emotional exhaustion and engagement are rather domain-specific aspects, general well-being represents an extension to other life domains.

To recruit participants, we contacted all about 31,000 students of the Johannes Gutenberg-University of Mainz, Germany, by e-mail via a central mailing list. The questionnaire was made available online in June 2020. Except for a few courses, teaching was delivered online in a mixture of asynchronous (‘on demand’), and synchronous online courses. Students voluntarily participated in the survey and were informed about the study’s objectives. As an incentive, six vouchers for 20 euros were raffled. The students could choose between a regional help initiative for gastronomy named ‘Mainz help’ or an amazon voucher. The study was performed following the Ethical Principles and Guidelines for the Protection of Human Subjects of Research by the American Psychological Association (APA). Ethical approval of the study was obtained (2020-JGU-psychEK-S008).

In total, 3,066 university students answered the online questionnaire. After excluding missing data, a final sample of 2,882 students could be retained. More female than male students responded, with a few participants classifying themselves as diverse. On average, participants were 23.4 (SD = 4.4) years old. The majority of participants were enrolled in a bachelor’s degree program, about one quarter were enrolled in a master’s degree program, and about one quarter studied in programs requiring a state examination. Only very few students were enrolled in expiring qualification programs, such as magister and diploma. The focus of the students’ study ranged across disciplines, including natural and social sciences, as well as humanities, theology, medicine, music, language, and education. Table 1 provides more details of the study sample.

The data in this study were first prepared using IBM SPSS Statistics 27. The following analyses were performed using Mplus 7.3 (Muthén and Muthén, 1998–2012). LPAs were performed to look for homogenous subgroup patterns of study crafting behaviour (cf., Wang and Hanges, 2011). An LPA accounts for uncertainty about an individual’s true profile and, thus, an individual only contributes to the profile-specific means depending on their profile probability (Magidson and Vermunt, 2002). Since the assumption of the deterministic assignment of individuals to profiles is dropped, it is, therefore, also referred to as a probabilistic cluster analysis procedure (Bacher and Vermunt, 2010).

The profile analysis was performed according to the automatic BCH method (Asparouhov and Muthén, 2014; Bakk and Vermunt, 2015). As often recommended, the model was considered independently of its statistical relationship to covariates and criteria for this purpose (Vermunt, 2010; Asparouhov and Muthén, 2014; Bakk and Vermunt, 2015). In mixed distribution models, if the criteria are directly included in the calculation, the profiles may be determined not only by the original latent profile indicators, but also by the criteria and, thus, be biased or altered in number (Huang et al., 2010). All models (profile solutions) were computed with 1,000 initial value sets to reduce the risk of a local likelihood maximum, and each of these sets was processed with up to 500 iterations (Geiser, 2011).

The Akaike Information Criterion (AIC; Akaike, 1987), the Bayesian Information Criterion (BIC; Schwartz, 1978) and the sample-adjusted Bayesian Information Criterion (SABIC; Sclove, 1987) were used as information criteria to select the model with the most appropriate profile number. Lower values indicate a better model fit. In addition, the model-comparative Lo-Mendell-Rubin test (LMR; Lo et al., 2001; Asparouhov and Muthén, 2012) was used to select the model. The BIC and LMR tests together are recommended for choosing a model (Nylund et al., 2007), so these two were chosen as primary indicators. The entropy of the model was retained as an indicator of the quality of the classification (Celeux and Soromenho, 1996). The entropy measures the probability that any person from the sample is assigned to the best matching profile.

Based on Merians et al. (2019), the selected profile solution was validated using double cross-validation. For this purpose, the overall dataset was randomly split into two halves, and all profile solutions of the overall dataset were replicated with the first half. Using the previously described information criteria and the LMR test, the profile solutions were compared to exclude, as far as possible, a random distribution of the profile solutions based on the sample. Furthermore, the final profile solution from the first half, as well as the matching profile solution from the second half, were checked for their visual correspondence with the selected profile solution of the overall data set.

Following this, a criterion-related validity check of the latent profiles (Bacher and Vermunt, 2010)—the difference between the latent profiles in the criteria emotional exhaustion, study engagement, and well-being—was conducted. For this purpose, the output file of the final profile analysis of the total data set was used. In the automatic BCH method, the measurement model for the latent profile variable is specified during profile analysis, and the criteria are specified as such. The profile analysis is then computed independently of the criteria, but the output file displays the profile differences concerning the criteria. The results of the chi-square estimates and p values of each Wald test were used to test for differences between profiles in the outcome variables.

Five latent profile models with different numbers of profiles ranging from two to six, were estimated. Regarding the choice of the best fitting profile number, Table 3 summarises the information criteria of AIC (Akaike, 1987), BIC (Schwartz, 1978), and SABIC (Sclove, 1987), as well as the entropy and the model-comparative LMR test (Nylund et al., 2007).

Among the possible profile solutions, AIC, BIC, and SABIC were first considered for the model fit. Although all information criteria decrease with increasing profiles, there is only minimal improvement above the 5-profile solution, favouring the 5-profile model. Additionally, the LMR test was conducted, comparing the respective k-profile model with the (k − 1)-profile model, with significant results indicating a better fit for the k-profile model (Nylund et al., 2007). In the present case, the values also favoured the 5-profile model because the 6-profile solution was not significantly better than the 5-profile solution, which, on the other hand, was better than the 4-profile solution.

We also report entropy, indicating that individuals are classified with more confidence (Van De Schoot et al., 2017). Entropy can take values from 0 to 1 and receives a maximum value of 1 when each participant perfectly fits to only one of the profiles, which would indicate that the latent profiles are completely discrete partitions (Van De Schoot et al., 2017). Following Greenbaum et al. (2005), the entropy should not be used to choose the number of profiles. Models with higher entropy should only be decisive for the decision if also supported by similar relative fit indexes (e.g. BIC; Ram and Grimm, 2009; Van De Schoot et al., 2017). In addition to the statistical fit of profile solutions, according to Uebersax (1994), model selection should be based on a model’s interpretability, parsimony, and meaningfulness. Even though the five-factor solution had a slightly lower entropy, as compared to solutions with less profiles, a further differentiation emerged (concerning average crafters, showing either more or less social crafting) compared to the four profiles. Therefore, we used the five profiles for further analyses.

Figure 1 shows the mean centered characteristic response behaviour of the five profiles in the final solution. The above average crafting profile was assigned to 4.8% of university students in our sample and is characterized by above average crafting in all three facets, with especially high values for social crafting. Two further profiles can be labeled as average crafters. These two profiles differ mainly in the usage of social crafting strategy. Hence we labeled these profiles as ‘average crafting more social crafting’ with a share of 22.2% and ‘average crafting, less social crafting’ with a share of 17.3%. Two further profiles show below average crafting in all three facets. These two profiles mainly differ in the usage of increasing structural resources, and were thus labeled ‘below average, more structural crafting’ represented by 45.7% of respondents, and ‘below average, less structural crafting,’ represented by 10.0% of our sample.

Table 4 shows the means, standard deviations, correlations, and Cronbach’s alphas of all used variables. The correlations show negative relationships between study crafting strategies and emotional exhaustion and positive relationships with study engagement and well-being.

In order to test for the discriminant validity of the five profiles, we investigated differences in the concurrent assessment of emotional exhaustion, engagement, and well-being. Table 5 summarises the results from the automatic BCH approach (Asparouhov and Muthén, 2021) and reports chi-square estimates and p values of each Wald test to assess the equality of the profile-specific mean levels for each outcome (Bakk and Vermunt, 2015). Profile-specific mean levels for each outcome are illustrated in Figure 2.

As expected, inclusion in the profiles with high study crafting strategies (above average crafters) resulted in the lowest level of emotional exhaustion and the highest level of study engagement, as well as well-being. The above average crafters scored significantly higher in study engagement than any other group. Emotional exhaustion was only significantly lower in the above average group when compared to the two below average crafter groups. Regarding well-being significant differences were observed to three of the four other groups, with average, less social crafter being the exception. The average crafter groups showed significantly lower levels of emotional exhaustion and higher levels in study engagement, when compared to the two below average crafter groups. Regarding well-being only the average, less social crafter group, but not the more social crafter group scored higher than the below average, more structural crafting group. The two below average crafting groups showed significant differences in all three outcome variables, with the group below average, more structural crafting reporting lower emotional exhaustion, higher study engagement, and higher well-being.

Overall, this confirms the expected negative relationship between study crafting strategies and emotional exhaustion as proposed in Hypothesis 1a. The results are mainly consistent with the presumed positive influence of study crafting on study engagement and well-being, lending support to H1b and H1c.

This study provides further evidence (e.g., Dormann and Guthier, 2019; Ferreira, 2020; Körner et al., 2021) that the job crafting concept can be adapted successfully to the higher education context. Study crafting, in general, showed comparably positive associations with engagement and general well-being, as well as a negative association with emotional exhaustion, as has been reported for employees (Rudolph et al., 2017; Lichtenthaler and Fischbach, 2019; Frederick and VanderWeele, 2020; Hu et al., 2020). In addition, the presented results align with previous studies on study crafting (Ferreira, 2020; Körner et al., 2021).

Our study suggests that the concept of approach or active crafting can further be differentiated. However, the five clusters in our study differ mainly with regard to the overall level of the three selected crafting dimensions. Increasing social resources turned out to be the driver for a more fine-grained differentiation among students using crafting strategies on an average level. Surprisingly, profile 2 (average, less social crafter), with a lower increase in social resources, combined with average other crafting strategies, showed higher study engagement compared to profile 3 (average, more social crafter), which is characterized by higher values in increasing social resources. Combining an average increase in structural resources and an average increase in challenging demands seems to be better for engagement when combined with a lower increase in social resources. This might indicate that students who already have sufficient social resources and thus have a lower need in further increase of social resources might be better off. This finding provides a more fine-grained picture compared to previous research on job crafting in variable-centered studies, where higher crafting behaviour in all dimensions has been consistently related to higher engagement (Rudolph et al., 2017; Lichtenthaler and Fischbach, 2019). For the interpretation of these results, the cross-sectional study design needs to be considered as a limitation. In addition, we do not know the individual reasons why some students engage in more or less certain crafting strategies. One potential reason to engage less in gaining a certain resource can be that this resource is already sufficiently available. Future studies might thus consider the motivation to engage in crafting behaviour, as well as an evaluation of the current status regarding structural and social resources as well as challenging demands.

A similar pattern emerged for well-being. Again, profile 2 (average, less social crafter) was associated with significantly higher well-being than profile 3 (average, more social crafter). Profiles 2 (average, less social crafter) and 3 (average, more social crafter) did not differ concerning emotional exhaustion, and no significant differences between profiles 1 (average, less social crafter) and profile 3 (average, more social crafter) were observed, despite the fact that profile 1 (above average crafter) is characterized by overall higher values in all study crafting dimensions. From these results, it can be concluded that, at least in the study context, it is not only the overall extent of crafting behaviour that is essential for health and well-being. Even though increasing social resources in isolation showed a significant negative correlation with emotional exhaustion and significant positive correlations with engagement and well-being, these effects were not observable when combined with increasing structural resources and increasing challenge demands.

This research is a cross-sectional investigation of study crafting and emotional exhaustion, work engagement, and well-being. Consequently, we cannot draw any causal conclusions. Whether study crafting leads to more health and well-being or health and well-being lead to more or less study crafting remains an open question that should be addressed in studies using multiple waves of data and cross-lagged panel analyses.

This study offers some insight into person-centered study crafting dimensions, despite its exploratory nature. Even though we used data from a considerably large sample, we cannot be sure that other samples will not show different combinations of study crafting behaviour. Even though the entropy should not be the reason to choose the number of profiles (Greenbaum et al., 2005), the entropy value of the five profile solution was the lowest (0.66), which indicates that some other profile solutions (e.g., four profiles) might provide a slightly more reliable categorization of students. However, the five profile solution provided a further differentiation, which proved to show incremental validity. Further research should examine whether the same or similar profiles are found in other student samples. Particular attention should also be paid to examining students in other parts of the world, including other European samples at large universities, as study conditions vary considerably internationally.

It should be noted that the study crafting items were not yet validated and that the short form of the job crafting scale by Tims et al. (2012) was adapted. We want to highlight the new combination and adaption of existing items. The items about increasing social resources included the support and feedback of lecturers and not fellow students. Therefore, the contradictory results of increasing social resources should be considered cautiously.

When interpreting the data, it should be noted that the survey took place during the summer of 2020, with study conditions characterized by online teaching due to the pandemic situation. It seems plausible that these circumstances impacted opportunities to engage in study crafting.

The findings of this study have several important implications for practice. Since study crafting has been shown to be positively related to engagement and general well-being, as well as negatively related to emotional exhaustion, it seems warranted to search for ways to train students in study crafting. To the best of our knowledge, no study crafting interventions have been reported in research to date. Preventive interventions focussing on the reduction of stress in students mainly focused on stress regulation techniques such as mindfulness (Voss et al., 2020; Světlák et al., 2021). One meta-analysis reports that job crafting is trainable overall, but not all dimensions seem to be equally malleable. According to Oprea et al. (2019), overall structural and social resources increases are not reported.

However, according to our results in the study context, it is essential to increase structural resources, which should also be promoted in combination with increasing challenging demands. Moreover, we cannot make any recommendation regarding the reducing hindering demands dimension in studies because it was not considered. However, in interventions, students could reflect on their personal values, strengths, and passions in a person and task analysis. In addition, the content of a study crafting training could include sharing information about the study crafting dimensions and then setting SMART goals. SMART is an acronym that stands for Specific—Measurable—Achievable—Realistic—Timely. For instance, a goal in increasing structural resources could be to learn new methodological development outside of the regular curriculum; resources are gained, and study-related self-efficacy may grow. Students could keep a checklist to solidify their goals (Gordon et al., 2018).