The aim of the present study was to conduct a systematic review and meta-analysis to synthesize the evidence base of interventions focused on promoting work engagement by developing workplace resources from bottom-up. It is our hope that the findings will guide not only future work engagement research and practice, but also that of the broader organizational psychology field. Specifically, we addressed the following research questions:

What is the evidence base for the effectiveness of bottom-up, resource-developing interventions targeting employees in the promotion of work engagement?

We conducted the current study in accordance with the guidelines presented in the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (Moher et al., 2009) to the extent that they apply to non-medical research. These guidelines include following a checklist for reporting (see Supplementary Data Sheet S1). Our study approach (e.g., search strategies and data extraction) was also consistent with that of ample review exercises on work engagement published in the past (e.g., Knight et al., 2019).

Our comprehensive search strategy included searches in seven international, scientific online databases, chosen with regard to the interdisciplinary nature of the research topic. Four of these were specialized EBSCO databases: Academic Search Complete, Business Source Ultimate, PsycInfo, and PsycArticles. The three additional online databases that we conducted searches in were Web of Science, SCOPUS, and Google Scholar. We included research published between January 2000 and December 2020. The main searches in databases were conducted between September 25 and October 14, 2020, and the same searches were repeated on February 22–23, 2021 in order to include records from the end of year 2020. The selected databases along with database-specific search strategies are described in detail in the supplemental material (see Supplementary Data Sheet S2).

In accordance with the standard PICOS approach (Participants, Interventions, Comparisons, Outcomes, and Study design; Moher et al., 2009), we defined the following eligibility criteria for the systematic review:

Additionally, we adopted eligibility criteria relevant to our systematic review but not specified in PICOS. Specifically, these criteria were that the included studies should be published in peer-reviewed established journals (i.e., journals with an impact factor, not conference papers, dissertations, or books); written in English; focused on the promotion of work engagement (i.e., not focused on how to prevent decreased work engagement); and the presented study findings should be based on completed intervention studies (i.e., not study protocols). We included intervention studies in which individual bottom-up approaches and individual-level outcomes were in focus (i.e., participatory action interventions and/or aggregated outcome measures were not considered), although the interventions included could be delivered in various ways (e.g., target groups of employees, individual employees, and online or face-to-face). Due to the psychological nature of the review primary outcome, we excluded studies that emphasized physiological resources related to lifestyle and bodily health (e.g., low blood pressure, yoga, and diet), rather than psychosocial resources related to the interaction between the individual and the workplace (which can be inherent in the individual, reside in the social context, or in the way work is organized). Since the target population was working individuals, we excluded studies focusing on the work engagement of other groups of individuals (e.g., students). No limitations were applied regarding the duration of the intervention program.

The meta-analysis was conducted on a sub-set of the studies included in the systematic review. To be eligible for the meta-analysis, the studies had to include a control group (i.e., waiting list, inactive, or other intervention) and provide eligible information to compute pooled effect sizes (alternatively information retrievable from other sources than the actual report).

The first author managed the abstract screening process independently. The number of retrieved records from the selected databases and the process of screening and selecting studies can be viewed in the PRISMA Flow Diagram (Moher et al., 2009, see Figure 1). Next, the first and third author screened the full-text of records that had been assessed as eligible based on their abstracts. The quantified agreement between the raters was high (97% agreement, Cohen’s k=0.91; Landis and Koch, 1977). In case of disagreement, the second author assessed the study and discussions were held until agreement was reached. When the final dataset of included studies and their reports was decided upon, the first author independently extracted and coded data available according to the Data Extraction Form (see Supplementary Data Sheet S3). Discussions regarding the data extraction, including the study categorization, were held between the three authors to ensure consistency. Data extracted from each included study were, e.g., author(s), year of publication, method, study setting (country of origin; industry), and key findings.

Quality assessment of the included intervention studies was conducted utilizing the recognized NICE checklist for intervention studies (National Institute for Health and Care Excellence, 2012, based on Spencer et al., 2003; Jackson et al., 2006). The study quality was primarily assessed by the first author followed by discussions among the authors, revealing no discrepancy between the authors’ ratings. A summative quality score was coded for each study as ++, +, or − based on the assessed quality of study population, allocation of participants, outcomes, analyses, and internal and external validity. The highest quality rating (++) indicated low risk of bias, and this rating was given to studies that fulfilled all or most checklist criteria (and it was unlikely that the study conclusions would have been different if the few unfulfilled criteria had been fulfilled). Similarly, a moderate-quality rating (+) indicated moderate risk of bias and this rating was given to studies in which some of the checklist criteria had been fulfilled. The conclusions would likely have remained the same if unfulfilled criteria had been fulfilled, or if poor descriptions of criteria had been adequate. Finally, the lowest quality rating (−) indicated high risk of bias. Studies that received this rating fulfilled few or no criteria and the study conclusions would likely have been different if the missing criteria had been fulfilled.

The effect sizes of the interventions were calculated by Review Manager 5.4.1 software (The Cochrane Collaboration, 2020) for the primary outcome under study (i.e., work engagement). Data from all the publications that provided eligible post-test or follow-up data on overall work engagement measured by the UWES-scale (i.e., no sub-scale data considered) were extracted from the study reports by the first author and then double checked and entered into the Review Manager by the third author. Both the weighted mean difference (WMD) and the standardized mean difference (SMD) were calculated as appropriate for the continuously distributed outcome using a random effects model. The random effects model was chosen based on guidelines and recommendations provided by, e.g., APA Publication Manual (Cohen, 1988) for increased interpretability and generalizability. Endpoint continuous data for intervention completers were used in these calculations. With regard to eligible studies with more than two arms, only the intervention-arm and the control-arm that received no intervention were considered in the meta-analysis. If measures of variance of outcomes could not be found in the study publications or through calculations, the corresponding authors of the identified publications were contacted with data requests. If the missing data could not be retrieved, the study was excluded from the meta-analysis. Substantially skewed data (where the standard deviation was greater than double the mean value) were not entered in the meta-analysis. The impact of statistical heterogeneity on the meta-analysis was assessed by quantifying inconsistency among the studies with the I² Index test (Deeks et al., 2008). This test describes the percentage of the variability in effect estimates that is due to heterogeneity rather than sampling error (chance). All calculated I²-values were deemed acceptable, however, all over 50% indicating the proportion of the variation in point estimates due to among-study differences being moderate to large. A sensitivity analysis was conducted to test the robustness of the performed analysis and related findings. Only the interventions that retrieved the highest quality rating (++) in the methodological quality assessment exercise were included in this sensitivity analysis. The extracted data also allowed for three post-hoc sub-group analyses; two of them according to two of the explored potential underlying mechanisms and one of them only including studies that applied the short version of the UWES (Schaufeli et al., 2006). The extracted data also allowed for a meta-analysis of pooled effect sizes for role performance (secondary outcome).

The total number of records originally identified in the systematic database searches was 1,988. After duplicates were removed, the abstracts of 1,468 unique records were screened according to the eligibility criteria. During this abstract screening process, an additional 1,341 records were excluded, leaving us with 127 records. Main reasons for exclusion of records at this stage were that they were not intervention studies, did not have work engagement as the primary outcome of the study, and/or were not targeted at working individuals. Following a careful assessment of full-text articles, the final number of articles included in the systematic review was 30, of which one contained two included studies (Gordon et al., 2018), resulting in 31 independent studies (see Figure 1 and Table 1). The main reasons for exclusion of articles that were assessed for eligibility in full-text were that they were judged to have a top-down rather than a bottom-up approach, emphasized physiological resources related to lifestyle and bodily health rather than psychosocial resources related to the interaction between the individual and the workplace, and/or did not use the UWES-scale for measuring work engagement. Also, a few of the excluded intervention studies were organizational level studies that did not target workers at the individual level. The number of studies that contributed with data to the meta-analysis was 24, and the sample size at baseline for these studies can be viewed in Table 2.

The quality assessment exercise was challenging due to scant reporting in several studies. Poor descriptions of population, allocation of participants (if applicable), and statistical analyses performed were especially common shortcomings of the study reports assessed. Based on the reported information, three studies received a low-quality score, indicating high risk of bias. In comparison, 16 studies received a moderate-quality score, indicating moderate risk of bias, and 12 received a high-quality score, indicating low risk of bias. The overall quality score for each study can be viewed in Table 1.

The included studies reported several limitations. Commonly reported shortcomings of the interventions were reliance on self-reports (risk of common method variance), small sample size or high dropout rates (risk of low statistical power), limited generalizability of the study findings (focus on certain industrial and geographical contexts and groups of workers) and that the results were short-term ones (no information on long-term effects). Intervention studies with no comparator and studies with non-randomized intervention and comparator groups often reported these study characteristics as important limitations.

The systematic review and meta-analysis have several limitations. First, the data were collected from studies with varying design and characteristics, also revealing moderate to high inconsistency based on high heterogeneity. Further, our study highlighted risks of reporting bias. Some of the studies included in the systematic review lacked the required information to be included in the meta-analysis (e.g., two arms, means and standard deviation values, and measurement points) but many more provided insufficient descriptions of the study design, sample, and procedure – all of which complicated the assessment of study quality and publication bias. Further, several of the studies included in the meta-analysis were based on a low sample size and thus reduced the statistical power. The lack of statistical significance in some of the findings is probably the result of a combination of small effect sizes and lack of statistical power due to the low number of studies, many of which included small samples. All these factors limited the extent to which conclusions can be drawn from this study regarding the evidenced effectiveness of interventions. However, in order to nuance the information on the evidence identified, as well as to test the robustness of the findings from the meta-analysis exercise, we performed several sub-group analyses. For example, the sensitivity analysis that included only high-quality studies showed a lower but still statistically significant pooled effect on overall work engagement. The reason behind a lowered pooled effect size estimate among the high-quality studies only compared to all included studies could be explained by an on average smaller difference between the intervention and control group in relation to the measured outcome, which in turn implies a slightly weakened relevant effect in practice among these studies. Not only does this call for more intervention studies applying high-quality research design and methods, but it also points out the need for a more nuanced examination of the mechanisms underlying the effectiveness of the studies aiming to promote work engagement.

The second limitation pertains to the categorizations of the included studies. While we did our best to classify the studies in a meaningful way that would further the understanding of how work engagement can be promoted, there is always a risk of mis-categorization due to inconsistency in how information is reported.

A third limitation is that we only included studies that measured work engagement using the UWES-scale (Schaufeli et al., 2006). Although this scale is widely used in the work engagement literature (Bailey et al., 2017; Shuck et al., 2017; Kelders et al., 2020), a recurring criticism concerns its robustness, which is argued to be weakened due to the three-factor structure (Wefald et al., 2012). At the same time, applying use of the UWES-scale as one of the eligibility criteria for this study could be viewed as a strength. One reason for this is that the validity and reliability of the UWES-scale are supported in several studies and in several settings (Schaufeli, 2014). It is also likely that an inclusion of the studies that we excluded on this basis would have aggravated the work with this systematic review and meta-analysis to the extent that the meaningfulness and robustness of the study results had been diminished.

We provide researchers with a checklist that could be used when conducting future studies on bottom-up work engagement interventions (see Table A1). Future intervention research and practice can build upon the aggregated results of our systematic review and meta-analysis in at least three different ways. First, robustness of study findings should be ensured in future bottom-up intervention studies investigating the effect on work engagement. Here, ensuring robustness especially entails ensuring that the study sample is representative of the investigated population, the statistical power is sufficient, and a comparison group is included. Further, the participants should be allocated randomly, or baseline differences between the intervention and the comparison group should at least be controlled for. In the current study, only 12 of the included studies were rated with the highest quality score. For example, a statistically significant increase in work engagement was reported in a clear majority of the systematically reviewed job crafting interventions, while the aggregated results in our meta-analysis showed that this dominant category of intervention focus had no statistically significant effect on work engagement. Moreover, one third of the intervention studies that focused on the promotion of work engagement through job crafting was conducted by van Wingerden et al. (2016, 2017a,b,c) and the resemblance between these studies is high. It is our interpretation that these ground-breaking studies set the tone for most of the subsequent studies that shared this intervention focus, which illustrates the danger in relying on the results of single intervention studies, especially if they can be associated with methodological flaws and risk of bias. From a practical point of view, this learning is also relevant for practitioners, since it suggests that popular practice does not necessarily constitute best practice.

Second, more studies investigating the effects of bottom-up interventions on sub-components of work engagement are warranted. Such studies could deepen our understanding of how bottom-up interventions aimed at promoting sub-components of work engagement stand in comparison with those aimed at promoting overall work engagement. However, based on the synthesized evidence, practitioners are guided to educate, facilitate, and encourage individual bottom-up approaches that promote the overall work engagement of employees.

Third, the evidence retrieved from the meta-analysis suggests that future intervention research should apply universal approaches rather than tailored ones. In practice, these results can be interpreted to imply that similar training, methods, and techniques should be applied to all kinds of employees when organizations want to facilitate the process in which employees learn, practice, and eventually use bottom-up approaches for the development of workplace resources.