The moderating effect of coworkers’ training participation on the influence of peer support in the transfer process

Abstract

Purpose

The impact of the number of coworkers participating in training on transfer outcomes has largely been overlooked. This paper aims to examine whether the number of coworkers participating in training interacts with peer support (PS) to influence training motivation and transfer.

Design/methodology/approach

Data were collected using a cross-sectional survey from a sample of 688 employees working in 14 midsize and large companies. All participants were recent trainees in various open skill (e.g. leadership) training programs. Moderated mediation was used to test the hypotheses.

Findings

Motivation to transfer (MTT) mediated the relationship between PS and perceived training transfer. When more coworkers participated in the training, PS had a stronger influence on trainee MTT.

Practical implications

Organizations should consider training coworker cohorts at the same time to influence MTT and training transfer. Generally, whole-team training programs could be used to boost training transfer outcomes, although it could potentially have a negative impact on transfer if PS is low.

Originality/value

To the best of the authors’ knowledge, this was the first study to demonstrate that the number of coworkers participating in training can moderate the effect of PS on MTT and training transfer.

Citation

Salamon, J., Blume, B.D., Orosz, G. and Nagy, T. (2023), "The moderating effect of coworkers’ training participation on the influence of peer support in the transfer process", European Journal of Training and Development, Vol. 47 No. 10, pp. 15-36. https://doi.org/10.1108/EJTD-07-2021-0102

Introduction

To maintain competitive advantage, organizations are compelled to train and develop employees and managers (Noe and Tews, 2012; Reio, 2020). The necessary training is often being fulfilled by providing corporate training programs. Regardless of their targeted skills, organizations need to ensure that the significant amount of dedicated resources invested into workforce training leads to a return on investment (i.e. in the form of individual and organizational benefits). For ensuring these benefits, it is essential for participants to transfer the training by applying the newfound knowledge, skills and attitudes on the job (Baldwin abd Ford, 1988; Burke and Hutchins, 2007; Ford et al., 2018). The study of the training transfer process has identified factors that are associated with successful training transfer (Ford et al., 2018; Kraiger and Ford, 2021). Among these, well-established factors, peer support has received ample research interest and its positive effect on training transfer is supported by several previous studies (Ford et al., 2018; Massenberg et al., 2015). Despite the well-known importance of PS, the potential moderating effect of how many coworkers participate in the training has largely been overlooked. Yet, this aspect of the training programs may be important in that the number of peers participating in the training could enhance the influence of PS. If this is the case, organizations could more closely consider the impact of coordinating coworker training programs. As training coworker cohorts at the same time may be relatively easy for organizations to implement, this could be a way to increase the positive effect of PS on training transfer.

This study aims to address this research gap and investigate the impact of the number of coworkers participating in training on the training transfer process. In the proposed conceptual model (Figure 1), it is assumed that the number of coworkers participating in training influences the positive relationship between PS and both motivation to transfer and training transfer. Below, we outline the hypotheses and underlying theoretical background based on the proposed model.

Peer support, motivation to transfer and training transfer

The importance of the social environment in work-related factors has been recognized by researchers of health and social psychology for decades. In their meta-analytic study, Humphrey et al. (2007) provided evidence for the positive relationship of social support with beneficial aspects of work like job-satisfaction, work motivation and performance. Social support can be defined as “the extent to which a job provides opportunities for getting assistance and advice from either supervisors or coworkers” (Humphrey et al., 2007, p. 1336). Similarly, social support is found to be an important antecedent of both MTT (Gegenfurtner et al., 2009; Richter and Kauffeld, 2020; Seyler et al., 1998) and training transfer (Blume et al., 2010; Burke and Hutchins, 2007; Ford et al., 2018). The important role of social support (e.g. peer/workgroup support, supervisor support) was identified in the early empirical work of Ford et al. (1992), Facteau et al. (1995), Holton et al. (1997) and Bates et al. (2000). Moreover, Hughes et al. (2020) in their recent meta-analysis demonstrated that social support could account for 32% of the variance in transfer. Previous studies differentiated social support types by their sources (e.g. top management, organizational, supervisory, peer and subordinate support; Facteau et al., 1995; Tracey et al., 2001) and their supportive functions (e.g. instrumental, informational, emotional and appraisal support; House, 1981). According to Hughes et al. (2020), organizational support, supervisor support and PS each have a unique contribution to training transfer. Although several studies demonstrated that support provided by different sources has an important effect on the transfer process, the current study solely examines PS with a focus on its relation to the number of coworkers participating in training.

Previous studies have proposed that social support may not directly cause the occurrence of behavior but rather functions as an environmental trigger that exerts its effect through increased motivation (Baldwin and Ford, 1988; Gegenfurtner et al., 2009). The prominent role of motivation in performing a behavior is highlighted in several models and theories of behavior change (Michie et al., 2014). Similarly, MTT – defined as the “trainees’ desire to use the knowledge and skills mastered in the training program on the job” (Noe, 1986, p. 743) – has been found to be a key determinant of training transfer (Axtell et al., 1997; Baldwin and Ford, 1988; Gegenfurtner et al., 2009). Van den Bossche et al. (2010) found evidence for the partially mediating effect of MTT between the feedback provided by the work-related social network (i.e. PS) and training transfer. Similarly, Bhatti et al. (2013) and Massenberg et al. (2015) found that MTT fully mediated the effect of training interventions between PS and training transfer.

The first hypotheses serve as a foundation for this study and are directly based on these previous findings in the training transfer literature as follows:

H1a.

PS is positively related to perceived training transfer.

H1b.

PS is positively related to MTT.

H1c.

MTT is positively related to perceived training transfer.

H1d.

MTT mediates the effect of PS on perceived training transfer.

Coworkers’ training participation

The training transfer literature differentiates two main categories of training programs regarding the amount of team members participating in the program. Individual training (i.e. the training of individuals) refers to the programs where training participants attend the program independently from their teams, mainly with other members of the organization. In contrast, intact-team training or whole-team training refers to the interventions where all team members participate on the same program together (Kozlowski and Ilgen, 2006; Mathieu et al., 2008). Training transfer studies support the impact of PS and MTT on training transfer for training directed at individuals (Blume et al., 2010; Burke and Hutchins, 2007; Ford et al., 2018; Gegenfurtner et al., 2009) and there is also supporting evidence for similar patterns in whole-team training interventions (Massenberg et al., 2015). Cannon-Bowers et al. (2003) assumed that team members can increase transfer by providing the opportunity to model and reinforce the trained behavior. They also argue that the transfer of whole-team training programs could be even more successful, as team members can provide mutual support.

While previous research shows or assumes a similar mechanism and important antecedents in both individual and team transfer processes, empirical evidence is lacking on whether there is a difference on transfer outcomes when employees participate in training alone, with some coworkers or with all coworkers. The literature suggests that training programs targeting task-relevant skills are more effective when directed to individual team members while the programs targeting knowledge, skills, attitudes necessary for effective team functioning are best to be delivered to whole-teams (Mathieu et al., 2008). The reasoning behind these assumptions is that whole-team training programs provide the opportunities for participants to integrate and jointly practice their newly acquired skills (Mathieu et al., 2008). Although the underlying logic of this assumption is clear, empirical evidence is needed.

The advantage of whole-team training over training directed at individuals can be explained by different theoretical frameworks. According to the social information processing perspective (Salancik and Pfeffer, 1978), an individual’s attitude is formed by cues from their social environment beyond their own past behaviors and experiences. For example, an employee’s attitude about using a new software is influenced by peer opinions shared with him at lunch. These social cues can affect attitudes in four different ways. First, overt statements and observable behavior of other people directly cues attitudes (e.g. coworker says: “I don’t like this training.”). Second, social influence directs the attention of individuals on certain aspects of favorable or unfavorable information, shifting their attitude toward the direction that the social cue made more salient (e.g. coworker says: “This part of the training is useful.”). Third, the social environment also forms attitudes by providing interpretations of environmental cues, such as events, job characteristics and behavior (e.g. coworker says: “I can use this technique when I negotiate with my clients.”). Finally, it shapes the interpretation of individual needs, highlighting the presence or absence of specific aspects that should be important to the person (e.g. coworker says: “We could not manage these sort of client complaints, we badly needed this part of the training.”). These are four ways that social cues from peers in the training environment could impact trainee perceptions.

The multiple pathways of social influence (direct, attentional, interpretation and learning pathways; Grant et al., 2010; Salancik and Pfeffer, 1978) indicate that the knowledge, skills and attitudes that are targeted in a training program should be acceptable by the social environment and fit into its norms to promote application. When more coworkers participate in a training program, it is likely that the coworkers’ direct experience with the training content makes their reactions more observable, and these social cues and the impact of the social environment will have a stronger impact on trainees’ motivation and attitudes. For example, coworkers’ reactions and behaviors would be more directly observable and more likely to lead to shared interpretation and learning pathways for the trainee. For these reasons, we expect that trainees will have a higher motivation and willingness to transfer the training when a higher percentage of those in their social environment are also receiving training.

Furthermore, a broad social consensus about the targeted skills is essential for ensuring that more people are able to provide relevant and useful feedback, which were found to be crucial for successful behavior change and training transfer (Van den Bossche et al., 2010). The amount of feedback and PS are also found to increase MTT (Kirwan and Birchall, 2006; Van den Bossche et al., 2010). According to Russ-Eft (2002), the information provided by the feedback allows the learner to compare their current and the desired behavior, which increases their motivation to dedicate more effort to change their behavior. Additionally, Gilpin-Jackson and Bushe (2007) highlighted the positive effect of observing the on-the-job application of the learned skills that would likely occur more frequently if coworkers participated in the training program. Based on the Social Cognitive Theory (Bandura, 1986), these observations not only increase the observers’ knowledge acquisition via vicarious learning but also have a positive impact on their motivation to perform the behavior.

While the current study is the first that we are aware of to examine the effect of the different number of coworkers participating in training, based on the above reasoning and theoretical backgrounds, we propose the following regarding the moderating role of the number of coworkers participating in training:

H2a.

Coworkers’ training participation (CTP) moderates the relationship between PS and perceived training transfer; such that the effect of PS on perceived training transfer will be stronger when more coworkers participate in the training.

H2b.

CTP moderates the relationship between PS and MTT; such that the effect of PS on MTT will be stronger when more coworkers participate in the training.

H2c.

CTP moderates the MTT mediated effect of PS on perceived training transfer in such a way that the positive, mediated effect is stronger when more coworkers participate in the training program and weaker when fewer coworkers participate in the training.

Method

Procedure and participants

The study was conducted in accordance with the Declaration of Helsinki, approved by the Institutional Review Board of the Eötvös Loránd University Faculty of Education and Psychology, and is in line with the European Union’s General Data Protection Regulation (2016). Data collection was conducted in 14 mid- to large-size Hungarian companies. The invitation letter to participate in the study was sent to employees who had attended a training program in the prior six months. Voluntary participation was encouraged by a lottery drawing that awarded a total of 150 small prizes, each worth about US$15.

From a total of 864 survey respondents, the final sample included those who participated in a company-organized, open/soft-skill training program (e.g. leadership development, assertive communication, time management, sales, stress management) and who responded to the survey between 13 and 120 days after training. With the chosen timeframe, participants had at least two weeks after the classroom session to transfer the training to their job, and less than four months after training to ensure that the training was recent enough to accurately recall relevant aspects of the transfer process. In the online survey, respondents were instructed to consider the last training program in which they participated.

The final sample consisted of 688 working adults (48% female) who were between 22 and 67 years old (M_age = 39, SD_age = 8.88). Regarding their organizational levels, 403 (58.6%) worked at a nonmanagerial level, whereas 285 (41.1%) worked at a managerial level. The 14 participating companies (workforce ranged between 500 and 15,000 employees) operate in the accounting, automotive, chemical, energy, financial, insurance, pharmaceutical, retail and telecommunications sectors. Detailed company characteristics are shown in table S1 at the online supplementary materials on the project’s open science framework (OSF) page: https://osf.io/kf9yn/?view_only=a654da486f67403f8c35a88d1f3a432c. A study that used a subset of the current database has been published (Salamon et al., 2021). However, the present study contains significantly more data, focuses on distinct research questions and includes different predictor and moderator variables.

Measures

Data collection was conducted in Hungarian. To support the potential application of the shared materials in future research, the original materials were translated into English, following a standardized translation-back translation protocol proposed by Beaton et al. (2000). The full questionnaire and related materials are available on the project’s OSF page: https://osf.io/aw2kg/?view_only=a654da486f67403f8c35a88d1f3a432c. Items/responses for the first three measures listed below were provided on a seven-point Likert-scale (1 = Not true at all, 7 = Completely true) as follows:

• Perceived training transfer (outcome). To assess the application of learned techniques on the job, a four-item scale from Salamon et al. (2021) was used. The items reflect a topic-independent, general behavior transferred to the job (e.g. “At my workplace, I applied the methods acquired during training.”; α = 0.96).

• PS (predictor). A three-item scale was developed to measure the extent of perceived support from colleagues in the on-the-job application of the techniques learned during the training. The items were formulated based on Holton et al. (1997, p. 110) definition of PS (i.e. “the extent to which peers reinforce and support the use of learning on the job”). A sample item is “My coworkers encouraged me to use what I learned at the training.” This scale had good reliability (α = 0.87).

• MTT (mediator). A three-item scale from Salamon et al. (2021) was used to measure participants’ MTT the new techniques after the training (e.g. “By the end of the training, I was determined to use the new techniques I learned at the training.”). This measure has strong theoretical and empirical underpinnings (Nijman and Gelissen, 2011) and reliability was good (α = 0.92).

• CTP (moderator). Respondents were given the following four options and indicated whether their direct colleagues (with whom they work daily) participated in the same training program: “Yes, almost all of my direct colleagues participated in this training (in the same training session)”; “Yes, some of my direct colleagues attended this training (in the same training session).”; “Yes, my colleagues participated in this training before (but in a different training group).”; “No, I was the only one to participate in this training among my direct colleagues.” From these four response options a three-level scale was created by combining the two response options indicating that “some” coworkers participated in the training program. The final scale reflected three levels of CTP, including none, some and nearly all.

• Time lag (control). In line with the suggestion of Taylor et al. (2009), Salamon et al. (2021) found that participants reported less MTT when more time had passed after training. Consequently, we controlled for time lag by measuring the number of days between the end of the last training session and the response date on the survey.

• Organizational level (control). According to Chen et al. (2006), the job type or position in the organizational hierarchy can have an effect on the perceived transfer-related variables such as MTT. Furthermore, Salamon et al. (2021) found that trainees from higher organizational levels reported higher perceived training transfer. Respondents indicated their positions and we classified these as either employee-level or manager-level.

Statistical analysis

Statistical analyses were performed with R 4.0.5 (R Core Team, 2020) using the lavaan package (Rosseel, 2012) for structural equation modeling (SEM). First, a preliminary measurement model was estimated, using a confirmatory factor analytic approach, to confirm the factor structure and the psychometric adequacy of the measures used in this study. For the main analyses, this measurement model was converted into the proposed predictive model (Figure 1) in which PS predicted training transfer directly and indirectly through MTT. In addition, the direct path (between PS and training transfer) and the mediation path between PS and MTT were moderated by CTP. Furthermore, the control variables time lag and organizational level were included as predictors of both training transfer and MTT. In the analysis, 1,000 bootstrap replication samples were used for estimating the 95% bias-corrected confidence intervals (CIs). For estimating the interaction between the observed moderator variable (CTP) and the latent variables (PS and MTT) in the moderated mediation model, the product indicator approach (PI; Kenny and Judd, 1984) with the double-mean-centering strategy (Lin et al., 2010) was used with SEM. Following the recommendations of Yzerbyt et al. (2018) the component approach inspired joint-significance testing of multiple parameter estimates was applied to identify the presence of the indirect effect in moderated mediation.

The models were evaluated on the basis of common goodness of fit indices and interpreted along commonly-used cut-off values (Hu and Bentler, 1999; Marsh et al., 2005) the comparative fit index (CFI; ≥0.95 for excellent, ≥0.90 for good), the Tucker–Lewis Index (TLI; ≥0.95 for excellent, ≥0.90 for good) and the root mean square error of approximation (RMSEA; ≤0.06 for excellent, ≤0.08 for good) with its 90% CI. Furthermore, we calculated model-based composite reliability indices (ω; McDonald, 1970) which may better represent the construct, relative to Cronbach’s alpha, by estimating reliability from the factor loadings and their respective measurement errors. To establish convergent validity on the construct level we calculated average variance extracted (AVE) of the constructs. Moreover, to detect any potential problems regarding discriminant validity we used Heterotrait-monotrait (HTMT) ratio analysis (Henseler et al., 2015). Finally, we estimated the variance inflation factor (VIF) for the predictor, mediator and moderator to detect potential issues of multicollinearity.

Within R 4.0.5 (R Core Team, 2020), the tidyverse package (version 1.3.0.; Wickham et al., 2019) was used for data transformation and visualization, lavaan package (version 0.6–8; Rosseel, 2012) for SEM and calculating the omega composite reliability indices, semTools package (version 0.5–4; Jorgensen et al., 2021) for conducting moderated mediation analysis. The data, a more detailed analytic plan, and the analysis code can be found on the project’s OSF page: https://osf.io/aw2kg/?view_only=a654da486f67403f8c35a88d1f3a432c.

Results

Preliminary analyses

The goodness-of-fit indices showed excellent fit of the preliminary factor analytic model (χ² = 106.916, df = 46, comparative fit index (CFI) = 0.987, Tucker–Lewis Index (TLI) = 0.982, root mean square error of approximation (RMSEA) = 0.044 [90% CI 0.034, 0.054]). Parameter estimates (reported in Table 1) revealed well-defined and reliable factors for PS (λ = 0.757 to 0.871, ω = 0.873, AVE = 0.696), MTT (λ = 0.869 to 0.902, ω = 0.919, AVE = 0.791) and training transfer (λ = 0.893 to 0.952, ω = 0.958, AVE = 0.850). Furthermore, the results of the preliminary analyses show that our constructs were distinct (highest HTMT ratio = 0.799; values reported in table S2 at the online supplementary materials) and as VIF values did not exceed the threshold of 5 (highest VIF = 1.198), multicollinearity was unlikely a problem (Hair et al., 2018; Henseler et al., 2015). Bivariate correlations from the preliminary measurement model are reported in Table 2.

Main analyses

The fit indices of the latent moderated mediation model indicated an excellent fit (χ² = 229.809, df = 91, CFI = 0.982, TLI = 0.976, RMSEA = 0.047 [90% CI 0.040, 0.055]). The results of the model are shown in Figure 2 and the parameter estimates are presented in Table 3. Regarding control variables, the time lag between training and the outcome measures showed a significant negative effect on MTT (β = −0.09, p = 0.010), and no effect on training transfer (β = 0.01, p = 0.854). In contrast, organizational level showed significant association with training transfer (β = 0.06, p = 0.016) and nonsignificant association with MTT (β = −0.01, p = 0.777).

In line with our expectations, H1a and H1b were supported. PS had a significant, direct effect both on training transfer (β = 0.14, p < 0.001) and MTT (β = 0.47, p < 0.001). Furthermore, in support of H1c, MTT had a positive, significant relationship with training transfer (β = 0.72, p < 0.001). The joint-significance of individual parameter estimates of the indirect effects show the presence of MTT’s mediating effect on the relationship between PS and training transfer (β = 0.34, p < 0.001). This result supported our hypothesis H1d.

In contrast to our expectations, H2a was not supported. The interaction of PS and CTP (β = −0.01, p = 0.776) on the relationship between PS and training transfer was not significant (i.e. the moderating effect of CTP was not supported). H2b was supported as results indicate that the interaction between PS and CTP had a positive, significant effect on the MTT (β = 0.11, p = 0.004). The joint-significance of individual parameter estimates for the indirect effects also provided support for the first-stage moderated mediation (β = 0.08, p = 0.006) in H2c.

The significant interaction between PS and CTP on MTT (Figure 3) represents the importance of training cohort composition. As more coworkers participated in a training program, the peers’ supportive or non-supportive behavior had more of an impact on trainee MTT.

Discussion

In the present study, we successfully replicated previous findings regarding the association between PS, MTT and training transfer. More importantly, we also found that CTP moderated the effect of PS on the transfer process. Although the number of coworkers participating in training did not moderate the direct relationship between PS and perceived training transfer, its moderation effect was present in the MTT mediated path.

The moderated mediation effect highlights the importance of the number of coworkers participating in training. This is likely related to the shared understanding and attitude about the knowledge/skill/attitude targeted in the training program. The results suggest that the positive effect of PS and MTT on training transfer can be increased by more coworkers participating in training. A higher percentage of coworker participation can increase shared understanding, the positive attitude toward the targeted behavior, the opportunity to observe the target behavior on-the-job, and can strengthen the on-the-job feedback loop. These results are in line with the assumptions of prior studies that emphasized the advantages of whole-team training programs compared to training programs directed at individuals (Mathieu et al., 2008). However, the results also show that training interventions which target nearly all coworkers could also have an undesirable effect. For example, when there is a high percentage of coworkers participating in the training, but participants do not experience support from their peers, it can negatively affect employees’ MTT.

Theoretical implications

The present study aimed to investigate how the number of coworkers participating in training impacts the relationship between PS, MTT and training transfer. The findings address the previous research gap and are in line with the Social Information Processing Theory (Salancik and Pfeffer, 1978), which emphasizes the important role of the social context in attitude formation and the occurrence of a behavior. Based on this theory, a probable explanation for the moderating effect of CTP is that the higher the number of coworkers participating in a training program, the more information is available about the direct environments’ attitude (acceptance or refusal) regarding the learned skills. As high coworker participation in training makes these attitudes more perceptible, it makes it easier for participants to adjust their own attitudes and behaviors to the social consensus or norm. Through direct, attentional, interpretation and learning pathways (Grant et al., 2010; Salancik and Pfeffer, 1978) these social cues influence participants’ attitudes and MTT the learned skills.

Furthermore, previous studies have shown that employees who participate in a training together with more of their coworkers have a better chance to observe the learned skills that are applied on the job by their coworkers (Gilpin-Jackson and Bushe, 2007). This positive effect of observation on learning and motivation to perform the observed behavior is consistent with the Social Cognitive Theory (Bandura, 1986). Accordingly, the observation of a behavior not only increases the observers’ knowledge acquisition via vicarious learning but also has a positive impact on their motivation to perform the behavior. The results of the current study show that the effect of PS (e.g. acting/leading by example) on MTT is stronger when a higher number of coworkers participate in a training program. On the one hand, this result implies that when participants receive support and observe the trained skills performed by their coworkers, they will also be motivated to apply those skills. On the other hand, participants’ motivation can be diminished if they do not receive support from their coworkers and do not observe their coworkers implementing these learned skills to the job.

In the present study, 40% of the respondents participated in the training independent from the majority of their direct workgroup. This means that their coworkers likely had less knowledge or information about the topic, which could reduce their ability to provide support or motivation for training transfer (e.g. encouraging; modeling behavior; providing positive and developmental feedback). Previous research has shown that providing relevant and useful feedback is critical for successful training transfer (Van den Bossche et al., 2010). However, to provide useful and relevant feedback, a broad social consensus, common knowledge and positive attitude related to the targeted training is important. Whole-team training interventions can contribute to the learning of these shared models in the workgroup (Mathieu et al., 2005).

Practical implications

On average, respondents in this study did not experience high levels of PS. Looking more closely at the extent of PS, the results show that it is the lowest when no coworkers participate in a training, moderate when some coworkers participate and the highest level of PS is shown when nearly all coworkers participate in a training program. The fact that whole-team training programs show higher levels of PS highlights an important area for managers to focus on when they are looking for ways to enhance the transfer of training. It is likely that organizing and delivering training programs for employees who work together daily will increase the understanding, positive attitude, knowledge and skills of these teams/coworkers. This can lead to increased PS, MTT and training transfer.

It is important to note that training programs which target nearly all coworkers without a supportive performance environment could result in worse effects on MTT than training programs directed at individuals in a similarly non-supportive environment. However, participating alone or with only a couple others from a workgroup means that the knowledge about the targeted skills and the useful feedback loop is less likely to exist when the transfer attempts occur. According to Kozlowski (2018, p. 210) whole-team training programs “are key interventions for enhancing team processes and effectiveness, but the extent to which they are used routinely by organizations (outside of the military) is limited. This is an extraordinary untapped opportunity to enhance organizational effectiveness across a broad swath of the economy.” Consequently, when implementing whole-team training rather than training programs directed at individuals, it is crucial for learning and development professionals and managers to identify the barriers of PS in these programs. The lack of support may occur if peers have negative attitudes toward the training or there could be a general lack of cooperation in the team that could be improved. Alternatively, it could be that the direct social environment faces too many job demands (e.g. lack of time, extreme workload/pressure), which discourage supportive behaviors and derail trainees from transferring learned skills.

Given the above points, it seems especially important in these programs to ensure that the necessary PS will occur and to be aware of potential barriers to social support. One way to accomplish this could be to examine peer/trainee reactions toward the training via a survey to identify any negative attitudes toward training. In addition, practitioners can follow findings and suggestions in the human resource development literature to deliver effective training programs that are useful to trainees (Ford, 2020; Kraiger and Ford, 2021). For example, organizations should consider developing a “feedback-friendly” culture and investing in training supervisors to foster a climate in which psychological safety is high, peers support training transfer and employees proactively ask for high-quality feedback (Facteau et al., 1995; van der Rijt et al., 2012). Moreover, managers who dedicate resources to train their team should provide feedback on transfer attempts and consider whether trainees have enough dedicated time to practice and apply learned skills.

Limitations and future research

While the current study has several strengths (e.g. data collection conducted in multiple companies including multiple soft-skill training programs), there are also several limitations that should be considered in interpreting these findings. First, the current study included soft-skill training programs where workgroup support is especially important (Blume et al., 2010). Consequently, the generalizability of the findings to hard-skill training programs requires further research investigation.

Second, this study was cross-sectional with self-report measures, limiting causal interpretation of the investigated relationships. However, the implemented procedural remedies of common method variance (Podsakoff et al., 2012; Reio, 2010) incorporated into our study could alleviate the concerns of these problems. The issue of common method variance is probably the most relevant in connection with the first hypotheses (H1a–H1d), where these relationships have been supported by several previous studies (Blume et al., 2010; Ford et al., 2018). In addition, self-report data collection is more acceptable regarding variables that measure respondents’ internal states, such as the perceived support and MTT variables in this study (Spector, 2019). Furthermore, the other hypotheses tested interaction effects, and research suggests that although interaction effects can be deflated by method bias, they are unlikely to be artifacts of it (Podsakoff et al., 2012; Siemsen et al., 2010). Nevertheless, replication of the study in conditions where other-reports of transfer are analyzed and a time lag between surveys is applied would provide further evidence.

Third, the CTP moderator variable had unequal subgroup sample sizes. In total, 40% of the respondents were the only one in their direct workgroup to participate in the training program, 52% of the respondents had some coworkers who also participated in the training program, and only 8% of the respondents indicated that nearly all their coworkers participated in the training program. As the unequal sample sizes can significantly decrease the power to detect the effect of a moderator variable (Aguinis, 1995), it could be a potential reason for the nonsignificant effect of the number of coworkers participating in training on the relationship between PS and training transfer. Consequently, to handle this issue, future research that investigates similar relationships should aim for more equal subgroup sample sizes (Aguinis, 1995).

A promising direction for future research would be to investigate the individual characteristics and contextual factors that may moderate the effect of the number of CTP on transfer outcomes. Self-monitoring is an example of a trainee characteristic that could moderate the influence of CTP in the transfer process. The extent of self-monitoring was found to be an important moderator of social adjustment. High self-monitoring individuals are more sensitive to and tend to follow social norms, whereas low self-monitors are more likely to follow their internal attitudes and beliefs to guide their behaviors (Snyder, 1974; Burkhardt, 1994). Another important variable in connection with the effect of social cues is related to self-identity variables like organizational or workgroup identification (Chen et al., 2013).

An additional direction for future research could be to explore the percentage of coworkers as training participants required to have high PS. For example, is it necessary for 100% of coworkers to participate or could 50% or 75% of a team training participation still provide a positive environment of PS to facilitate trainee motivation and transfer? A related question is whether the percentage of coworkers (i.e. the ratio of the social environment affected by the training initiative) or a specified “high” number of coworkers has more impact on the effect of PS in the transfer process. Furthermore, while the current study assumed that each coworker would have the same effect on a trainee, future research could consider investigating how certain peers may have a larger influence on a group (Chen et al., 2013). For example, these could consider the impact of a coworker’s network position (social network approach) or the different quality of relationships, such as their strength and influence (relational approach). It can be assumed that the following statement is applicable in the context of training transfer: “those who are socially more important will exert a greater influence on the focal employee than others” (Chen et al., 2013, p. 1621). Furthermore, future research would be necessary to provide further information regarding the reasons for the presence or absence of PS in a work environment.

Conclusion

The present study unveiled an important but relatively neglected aspect of training programs. In line with theory that emphasizes the impact of the social context on individuals’ attitudes and learning, the results show that when more coworkers from a team participate in a training, PS has a stronger influence on MTT. Therefore, organizations should consider how the decision of when to train coworkers may influence transfer outcomes; and they should consider training cohorts at the same time.

Figures

Figure 1.

Latent moderated mediation concept model, representing the effect of PS on training transfer mediated by MTT and moderated by CTP

Figure 2.

Latent moderated mediation statistical model, representing the effect of PS on training transfer mediated by MTT and moderated by CTP

Figure 3.

Interaction effect of CTP and PS on MTT

Table 1.

Standardized parameter estimates from the preliminary model

	PS (λ)	MTT (λ)	TT (λ)	δ
Peer Support (PS)
Item 1. My coworkers encouraged me to use what I learned at the training	0.871**			0.241
Item 2. My colleagues helped me when I had difficulties applying the new methods I had learned during training	0.871**			0.241
Item 3. I regularly talked with my colleagues about how to best use the methods I had learned	0.757**			0.427
Motivation to Transfer (MTT)
Item 1. After completing the training, I was excited to use the techniques I learned there		0.902**		0.187
Item 2. By the end of the training, I felt that I would love to use what I learned immediately in my job		0.896**		0.197
Item 3. By the end of the training, I was determined to use the new techniques I learned at the training		0.869**		0.245
Perceived Training Transfer
Item 1. In my workplace, I used what I learned during the training			0.920**	0.153
Item 2. I tried the techniques at work I had learned at the training			0.922**	0.151
Item 3. At my workplace, I applied the methods acquired during training			0.952**	0.094
Item 4. In my day-to-day work, I implement the knowledge that I had acquired at the training			0.893**	0.202
ω	0.873	0.919	0.958
AVE	0.696	0.791	0.850

Notes:

**

p < 0.01; λ = Factor loading; δ = item uniqueness; ω = model-based omega composite reliability; AVE = average variance extracted

Table 2.

Descriptive statistics and bivariate correlations between latent variables

Variables	M	SD	1	2	3	4	5
1. Organizational level	0.41	0.49	–
2. Time lag	51.51	29.13	0.03	–
3. Coworkers’ participation	0.67	0.61	0.04	0.06	–
4. Peer Support	3.30	1.55	0.14**	0.07	0.32**	–
5. Motivation to Transfer	5.09	1.38	0.05	–0.06	0.06	0.44**	–
6. Perceived training transfer	5.00	1.38	0.12**	–0.03	0.10*	0.47**	0.79**

Notes: N = 688, Time lag: Days elapsed between training and data collection.

*

p < 0.05;

**

p < 0.01

Table 3.

Summary table of the parameter estimates from the latent moderated mediation model

Hypothesis	Involved variables	b	SE	[95% CI]	β	p
H1a	PS → TT	0.237	0.049	[ 0.134, 0.324]	0.14	<0.001
H1b	PS → MTT	0.535	0.052	[ 0.432, 0.635]	0.47	<0.001
H1c	MTT → TT	1.061	0.090	[ 0.902, 1.247]	0.72	<0.001
H1d	PS → MTT → TT	0.567	0.069	[ 0.438, 0.696]	0.34	<0.001
H2a	PS × CTP → TT	–0.012	0.041	[–0.093, 0.069]	–0.01	0.776
H2b	PS × CTP → MTT	0.128	0.044	[ 0.043, 0.217]	0.11	0.004
H2c	PS × CTP → MTT → TT	0.136	0.049	[ 0.046, 0.244]	0.08	0.006
Control	Time lag → MTT	–0.104	0.040	[–0.186, –0.029]	–0.09	0.010
Control	Time lag → TT	0.008	0.044	[–0.076, 0.100]	0.01	0.854
Control	Organizational level → MTT	–0.012	0.042	[–0.106, 0.061]	–0.01	0.777
Control	Organizational level → TT	0.105	0.043	[ 0.022, 0.188]	0.06	0.016

Notes: The table represents unstandardized (b) and standardized (β) parameter estimates with standard errors (SE) and 95% bias-corrected bootstrapped confidence intervals ([95% CI]). N = 688; MTT = motivation to transfer, TT = perceived training transfer and CTP = coworkers’ training participation; Time lag: Days elapsed between training and data collection; PS = Peer Support

Table A1.

Descriptive statistics by participating companies

Variables/company codes	01	02	03	04	05	06	07	08	09	10	11	12	13	14	Total
N	149	14	75	101	49	7	59	3	29	57	21	34	54	36	688
Gender (Female)	23%	43%	52%	65%	73%	57%	58%	100%	48%	63%	67%	32%	57%	72%	52%
Mean age (SD)	37 (8)	34 (8)	40 (9)	39 (9)	41 (7)	41 (7)	42 (10)	36 (13)	41 (7)	38 (8)	32 (7)	41 (9)	40 (10)	43 (11)	39 (9)
Role (manager)	41%	21%	33%	43%	10%	86%	42%	33%	93%	39%	48%	71%	37%	36%	41%
Mean time lag (SD)	57 (29)	42 (24)	46 (23)	72 (31)	46 (20)	72 (23)	44 (25)	25 (12)	60 (20)	59 (40)	50 (18)	31 (15)	32 (21)	35 (15)	52 (29)
Coworkers’ participation
None	45%	29%	52%	43%	37%	14%	29%	0%	41%	19%	29%	38%	56%	47%	40%
Some	51%	50%	41%	50%	61%	86%	63%	100%	21%	68%	57%	62%	43%	47%	52%
Nearly all	4%	21%	7%	7.9%	2%	0%	9%	0%	38%	12%	14%	0%	2%	6%	8%
Mean peer support (SD)	3.1 (1.4)	3.7 (1.8)	3.0 (1.6)	3.4 (1.5)	2.8 (1.2)	3.7 (1.5)	4.6 (1.8)	4.1 (1.4)	4.1 (1.5)	3.3 (1.4)	3.8 (1.3)	3.1 (1.7)	2.9 (1.4)	3.0 (1.5)	3.3 (1.6)
Mean mot. to tran. (SD)	5.1 (1.3)	5.3 (1.7)	5.0 (1.5)	4.7 (1.6)	5.0 (0.9)	6.3 (0.6)	5.6 (1.6)	5.1 (1.0)	5.5 (1.2)	5.2 (1.2)	4.8 (1.4)	4.7 (1.3)	5.1 (1.1)	5.3 (1.3)	5.1 (1.4)
Mean train. transfer (SD)	4.9 (1.4)	5.1 (1.4)	4.8 (1.5)	4.9 (1.5)	5.1 (1.2)	6.1 (1.0)	5.4 (1.6)	5.9 (0.6)	5.3 (1.4)	4.9 (1.4)	5.1 (0.9)	4.8 (1.5)	5.1 (1.1)	5.0 (0.9)	5.0 (1.4)

Notes: The table shows the descriptive statistics (means of continuous variables and frequencies of categorical variables) by companies. Time lag = time lag between training and data collection (in days); mot. to tran. = motivation to transfer; and train. transfer = training transfer

Table A2.

Heterotrait–monotrait ratio of correlations

Variables	PS	MTT	TT
Peer support (PS)	–
Motivation to transfer (MTT)	0.447	–
Perceived training transfer (TT)	0.487	0.799	–

Notes: Entries larger than 1 suggest discriminant validity problems; entries lower than 0.90 indicate sufficient discriminant validity between associated constructs

Appendix. Detailed analytic plan

Statistical analysis

Statistical analyses were performed with R 4.0.5 (R Core Team, 2020) using the robust maximum likelihood (MLR) estimator, which provides tests of model fit and standard errors that are robust to non-normality. First, a preliminary measurement model was estimated using a confirmatory factor analytic approach to confirm the factor structure and the psychometric adequacy of the measures used in this study. In this preliminary model, scale items loaded on their corresponding latent factors, and the factors were freely allowed to correlate with one another. Relying on fully latent variables also allowed us to reduce the biasing effect of item-level measurement error, in turn obtaining more accurate parameter estimates (Cheung et al., 2017; Cortina et al., 2021; Finkel, 1995).

For the main analyses, this measurement model was converted into the proposed predictive model (Figure 1) in which peer support predicted training transfer directly and indirectly through motivation to transfer. In addition to that, in the proposed predictive model, the direct path (between peer support and training transfer), and the mediation path between peer support and motivation to transfer were moderated by coworkers’ training participation. Furthermore, both training transfer and motivation to transfer were also predicted by time lag and organizational level. In the analysis, 1,000 bootstrap replication samples were used for estimating the 95% bias-corrected confidence intervals (CIs). These were computed with the maximum likelihood estimator, as bootstrapping is not available with the MLR estimator. For estimating the interaction between the observed moderator variable (coworkers’ training participation) and the latent variables (peer support and motivation to transfer) in the moderated mediation model, the product indicator approach (PI; Kenny and Judd, 1984) with the double-mean-centering strategy (Lin et al., 2010) was used with structural equation modeling (SEM). This approach is a well-performing alternative to latent moderated structural equations (LMS, which are only available in Mplus; Muthén and Muthén, 2012). PI is also robust to heteroskedasticity (Kolbe et al., 2020), which obtains similar power, but lower Type 1 error rates than LMS (Kolbe and Jorgensen, 2019), can be implemented into any software program, and provides more traditional SEM fit indices that are not available when using LMS in Mplus (Kolbe and Jorgensen, 2018). Following the recommendations of Yzerbyt et al. (2018), the component approach inspired joint-significance testing of multiple parameter estimates was applied to identify the presence of the indirect effect in moderated mediation.

The models were evaluated on the basis of common goodness-of-fit indices and interpreted along commonly-used cut-off values (Hu and Bentler, 1999; Marsh et al., 2005) the comparative fit index (CFI; ≥0.95 for excellent, ≥0.90 for good), the Tucker–Lewis Index (TLI; ≥0.95 for excellent, ≥0.90 for good) and the root mean square error of approximation (RMSEA; ≤0.06 for excellent, ≤0.08 for good) with its 90% CI. In the preliminary measurement model, the definition of the factors was interpreted based on the magnitude of their factor loadings. Following the recommendation of Morin et al. (2020) factor loadings greater than 500 were accepted as satisfactory. This simple rule is generally in line with the more detailed guidelines of Comrey and Lee (1992) who suggested to interpret factor loadings as excellent above 0.71, very good between 0.63 and 0.70, good between 0.55 and 0.62, fair between 0.44 and 0.33 and poor below 0.32.

Furthermore, we calculated model-based composite reliability indices (ω; McDonald, 1970; Morin et al., 2020) as ω = (Σ|λ_i|)²/([Σ|λ_i|]² +Σδ_ii) where λ_i represents the factor loadings, and δ_ii indicates the error variances. Its advantage is that it may better represent the construct, relative to Cronbach’s alpha, by estimating reliability from the factor loadings (λ_i) and their respective measurement errors (δ_ii). Its values above 0.60 are considered acceptable and good above 0.70 (Bagozzi and Yi, 1988; Perreira et al., 2018). To establish convergent validity on the construct level, we calculated average variance extracted (AVE) of the constructs, which is considered acceptable above the threshold of 0.50 (Hair et al., 2014). It was calculated as follows: AVE = (Σ|λ_i²|)/N), where λ_i² indicates the squared factor loadings, and N indicates the number of indicators. Moreover, to detect any potential problems regarding discriminant validity, we used Heterotrait–monotrait (HTMT) ratio analysis, which indicates discriminant validity issues when HTMT ratios are higher than 0.90 (Henseler et al., 2015). Finally, we estimated the variance inflation factor (VIF) for the predictor, mediator and moderator to detect potential issues of multicollinearity. The values of VIF indicate no multicollinearity below the threshold of 5 (Hair et al., 2018).

Within R 4.0.5 (R Core Team, 2020), the tidyverse package (Version 1.3.0.; Wickham et al., 2019) was used for data transformation, the cfa function of the lavaan package (Version 0.6–8; Rosseel, 2012) for calculating the omega composite reliability indices, the sem function of the lavaan package (Version 0.6–8; Rosseel, 2012) and indProd function of the semTools package (Version 0.5–4; Jorgensen et al., 2021) were used for conducting moderated mediation analysis. The data and analysis code can be found on the project’s OSF page: https://osf.io/aw2kg/?view_only=a654da486f67403f8c35a88d1f3a432c (blinded for review).