Understanding impact of supervisory support on work outcomes using agent based simulation Suman Kumar, Mayuri Duggirala, Harshal G. Hayatnagarkar, Vivek Balaraman {suman.kumar4,mayuri.duggirala,h.hayatnagarkar2,vivek.balaraman}@tcs.com TCS Research Pune, MH 411013, INDIA ABSTRACT our fine-grained approach to composing behavior models and our Support service environments are stressful with stringent demands use of these to study how individual behavioral dimensions such on individual and workgroup performance that have to be met day as affect, conscientiousness and stress impact work outcomes. after day. In earlier work we have modeled the impact of stress In this work, we extend these models to examine how organiza- within such environments on the performance of the individual tional social dimensions impact workplace outcomes. In particular and correspondingly that of the team. Since teams are social envi- we study how the organizational social dimension called supervi- ronments, we can intuitively realise that social dimensions such as sory support may impact workplace outcomes in case of a support supervisory support would impact a team member’s performance services organization. Past studies including our own show that for the better or the worse. But what is the precise impact of su- supervisory support impacts team member characteristics such as pervisory support on a team’s macro outcome parameters such engagement, job satisfaction, absenteeism and productivity. In this as productivity and performance? Using the results of a ground work we use an agent based system to study the dynamic implica- study of a support services organization, we use an agent based tions of supervisory support on macro parameters of a prototypical simulation approach to understand the dynamics and the implica- support services team. tions of supervisory support on individuals and consequently the macro parameters of the team. We show that supervisory support 2 CONTEXT AND PAST WORK plays a critical role in ensuring that the team as a whole meets its Past research on the role of supervisory support has highlighted its performance parameters particularly in the presence of disruptive beneficial impact on a range of individual, team and organizational factors such as work spikes. outcomes. Supervisory support is described as the employees’ per- ception of the extent to which supervisors value their contributions CCS CONCEPTS and care about their wellbeing [10]. The role of supervisory support •Computing methodologies → Agent / discrete models; as a buffer for job stress in individuals has been well documented [2]. Supervisory support has also been found to raise levels of KEYWORDS employees’ trust in the organization with supervisors embodying Agent-based modeling, Agent-based simulation, human behavior the organization’s goals, values and priorities which in turn was model found to positively influence the employee-organization relation- ship over and above impersonal formal organizational structures [20]. With respect to innovation, studies have indicated how super- 1 INTRODUCTION visory support behaviors of encouraging innovation, skill building, Employees in support services organizations work as a part of large open communication, rewards and recognition and effective man- teams. These teams are expected to reach very competitive targets agement of responsibilities led employees to willingly participate in from their business clients in industries such as finance, retail, promoting initiatives aimed at promoting innovative environmental banking, health-care etc. The targets are specified in service level policies [11]. Other individual level outcomes being influenced by agreements (SLAs) which indicate aspects such as the Mean Time to supervisory support include career satisfaction [19], low emotional Resolution (MTR), Turn Around Time (TAT) for different categories exhaustion and depersonalization [15] and low turnover intent [9]. of tasks as well as the escalation hierarchy in case of emergencies. Thus past research establishes supervisory support as an impor- The organizational environment in which these associates work is tant construct in organizational behavior research and justifies its stressful and requires individuals within the team to rely on each inclusion in the present study. other as well as their supervisors and leadership in order that the Before we go on to discuss the context, we introduce a few terms tasks are done as specified in the SLA. Studies in such environments that will be used in rest of the paper. Below we define some of [4, 17, 18] including our own [14] indicate that psychological, social, the study variables that have been referred to in the following dis- cognitive and environmental factors play a considerable role in cussion: Emotional state refers to an individualfis experience of impacting organizational metrics of interest such as productivity positive and negative emotion with respect to their work, at a spe- and job satisfaction. We have already discussed in [1, 3, 6, 13, 14] cific point of time during the work day, namely at the start of their work day and at the end of their work day. Momentary stress Copyright 2017 for the individual papers by the papers’ authors. Copying permitted refers to the perception of stress related to work at the start and end for private and academic purposes. This volume is published and copyrighted by its editors. of the individualfis work day. Workload refers to the number of tasks arriving on a day, to be completed by an individual before end Conference’17, July 2017, Washington, DC, USA S. Kumar et. al. (a) Without supervisory support (b) With supervisory support Figure 1: Affective stress dynamics model Table 1: Behaviour relations Relation Model Description Source Affect ← Work- Affect= 0.106*(workload) + 0.14 Perception of workload has a positive impact [8] load on negative affect Stress ← Affect Stress= 0.093*(Affect) + 0.547 Without interaction of moderating supervisory Field Study support Stress ← Affect Stress= 0.023*(Affect) + 0.547 With interaction of moderating supervisory Field Study *Supervisory Sup- support port Productivity ← Productivity = M * BaseProductivity Stress has an impact on decision making and [12, 16] Stress If(Stress ≤ 0.1) then M = 0.5 hence influences productivity. This follows the If (Stress > 0.1 and ≥ 0.25) then M = 1.0 inverted-U model which suggests that an opti- If (Stress > 0.25 and ≥ 0.75) then M = 1.25 mal amount of stress is required for best perfor- If (Stress > 0.75 and ≥ 0.9) then M = 1.0 mance, very low and very high stress degrades If(Stress > 0.9) then M = 0.5 performance. P(Absenteeism) If(stress > 0.9) then NORMAL DIST(0.1, 0.1) High stress (> 0.9) was correlated with high Field Study ← Stress absenteeism of the day. Affect is the extent to which the associate experiences decreases in team productivity. The organizational structure had positive or negative mood during the course of the work day. In one supervisor leading a team of several hundred associates. The this study, we are focusing only on the negative affect. Workload supervisor was responsible not only for ensuring that SLAs were spike refers to a 1.75 times increase in workload on a particular met on daily basis, but also were required to frequently monitor day (exceptional day). Backlog refers to the number of pending individual learning and performance particularly for newcomers to tasks for an individual at an instance of time. Bench strength the team. It was also the supervisorfis role to maintain team morale refers to individuals in the workforce that are used only during on days when there was a heavy spike or accumulated workload crisis situations like: heavy workload arrival or large number of due to absentees among the team, seasonality or other factors. unplanned absentees on a day, etc. Supervisory support refers to We had carried out an exploratory study in the account teams perception of employees regarding the degree to which the supervi- identified by the support services organization to examine the im- sors value their contributions and care about their well-being [10]. pact of static (trait) as well as dynamic (state) behavioral factors This is expressed as a percentage of the total available workforce. on the outcomes of interest, i.e. absenteeism and productivity. Ele- Turn-around time (TAT) is the time taken by the simulated team ments of our study findings pertaining to individual traits and states to complete a newly arrived task. Absenteeism refers to the num- such as conscientiousness, affect and stress, have been reported ber of unplanned leaves taken by an individual participating in the in [1, 3, 6, 13, 14] where we have also discussed the dynamics or study. Productivity was measured via self-reports, i.e. using a implications of those findings. survey where the individual rated themselves in terms of whether In the field study, we also observed that, associates who per- they had achieved their daily goals and targets, and whether they ceived lower supervisory and coworker support reported lower had achieved all that they had planned to do. Objective productivity engagement and job satisfaction (p < 0.05). Similarly, higher per- metrics were also collected for the participating individuals, from ceived supervisor support was linked to higher objective ratings, the support services organization in terms of their performance productivity and quality as well as perceived engagement and satis- ratings, quality and productivity. faction (p < 0.05). In parallel, higher coworker support was linked A large support services organization had been facing issues to objective ratings and perceived engagement and job satisfaction. with its employees of unscheduled leave or absenteeism as well as Understanding impact of supervisory support… Conference’17, July 2017, Washington, DC, USA Figure 2: Process Model In addition to the above analysis, multiple regression also showed implications of these findings. We have been using a grounded fine the significant impact of supervisory support on job satisfaction grained agent based simulation approach to explore these implica- (b = 0.27, p < 0.05) and stress (b=-0.27, p < 0.01). The combined tions and which have been reported in [1, 3, 6, 13, 14]. We compose effect of stress and supervisory support on productivity was also a simulation as a directed graph of relations that tie together be- significant (b = −0.21, p < 0.01) indicating an indirect effect of havior variables with outcome variables of interest and where each supervisory support on productivity in the presence of stress. In relation comes either from past literature or from our own study. other words, the buffering effects of supervisory support described We have used this approach to both explore different models for in past research were also supported by the empirical findings the same situation but different variables of interest or explore the in our field study. This finding also lends support to our model use of the same model in different situations. In the current work, presented in section 3.1 where we include supervisory support as a we extend the basic stress model reported in [13, 14] to factor in moderator in the stress→productivity relationship. In our review of the impact of supervisory support. the research on supervisory support we have yet to find a study that models the dynamic effects of supervisory support on productivity. 3.1 Simulation Model This therefore is a key contribution of the present study. Fig. 1a depicts the basic stress dynamics model used for the sim- Thus, support from the larger organization, particularly the su- ulation and which has also been discussed in [13, 14]. This model pervisor emerged as one of the important insights from this study ignores the role of Supervisory Support. In Fig. 1b we factor in as we found that supervisory support was linked to both objective supervisory support which been added in the role of a moderator performance outcomes measured by the HR team as well as per- variable. ceived outcomes measured in our survey as discussed above. This Details of the model are described in table 1. Fig. 2 describes the result from the study was further supported by in depth interviews overall simulation process. In this work, we do not use demand with the associates, supervisors and senior leadership in both the management strategy. teams that participated in the study. These demonstrated the close As with every agent model we need to make some assumptions: ties that the supervisor had with the rest of the team despite the We assume that agents have uniform skills and competency level large spans of control. to complete the given tasks and does not have a fixed deadline Given the importance of supervisor support in terms of providing to adhere. Tasks also have equal difficulty levels and workload consistent role modeling, mentorship, counseling and guidance to only corresponds to the number of extra tasks getting assigned to their reportees, the present study examines the following research an agent. The overall performance of the team is monitored via questions: ”How does supervisory support at an individual level average TAT and backlog accumulated over the period of time. affect dynamics at the team level, in its presence and absence?” The next section presents the model, experiments, and results obtained. 3.2 Experiment For conducting the simulated experiments for our process model, 3 MODEL, EXPERIMENT, RESULTS AND we have chosen the GIS and Agent-based Modelling Architecture DISCUSSION (GAMA) [5]. The model uses the specification language GAML These insights on dimensions of behavior and potential for impact to describe the environment, process and behavior of agents. We on outcomes, led naturally to the need to understand the dynamic simulated the experiment using a team of 50 agents. Conference’17, July 2017, Washington, DC, USA S. Kumar et. al. Figure 3: Average Turn-around Time v/s Bench Strength. Figure 4: Average Backlog v/s Bench Strength. The tasks are assigned on daily basis with a mean of 1000 tasks was 16 days, which jumps to 65 days in presence of spike. With per day and std. deviation of 10%. We monitor the running simula- bench strength of 6%, the average TAT falls from 16 days to 8 days tion for 1200 cycles which are equivalent of 120 simulation-days. without spike, and from 65 days to 55 days in handling a spike, Average tasks received per day are 1000, with variation of 10%. A which is a small 10% reduction. spike in workload implies 2000 tasks. With supervisory support, the team with 0% bench strength can Each simulation of is executed 10 times for every combination turn a task around in 5 days in absence of spike, which is approxi- and the mean is reported as the final parameter value. During these mately a third of earlier 16 days, and same for bench strength of runs, we collected data for variables such as average turn-around 6%, which is 33% reduction. However, interestingly, the team also time, average backlog, and average stress. These data are visualized mitigates spike in the workload in the same envelope of 5 days, a in following charts. reduction from 65 days and from 55 days respectively. Thus, we see that teams with high supervisory support can deal even with work spikes without significant impact on TAT, while 3.3 Results and Discussion a team that lacks supervisory support shows both higher average In this section, we discuss impact of supervisory support on average TAT without a spike as well as a significant jump when there is a TAT, backlog, and stress in presence and absence of workload spike. spike. This is macro-level effect, and it can be attributed to a slower In addition, these scenarios are simulated against bench strengths rise in the stress at an individual level, when supervisory support of 0% and 6%. is available. Fig. 3 informs us of the importance of supervisory support. The Similarly, we see in Fig. 4 that the average team backlog increases chart shows us effect of spike on average TAT for different bench without supervisory support in absence of workload spike, and in- strengths, in presence and absence of supervisory support. creases substantially further in presence of such a spike. Please First, we will discuss case without supervisory support. If the team has the 0% bench strength, then without spike average TAT Understanding impact of supervisory support… Conference’17, July 2017, Washington, DC, USA Figure 5: Average Cumulative Stress v/s Bench Strength. note that this chart uses the log axis for representing average back- to perceived organizational support and employee retention. Journal of applied log, to accommodate large differences. The primary reason why psychology 87, 3 (2002), 565. [5] Arnaud Grignard, Patrick Taillandier, Benoit Gaudou, Duc An Vo, Nghi Quang supervisor support has such a dramatic effect on TAT and backlog is Huynh, and Alexis Drogoul. 2013. GAMA 1.6: Advancing the art of complex because supervisory support reduces stress levels of team members. agent-based modeling and simulation. In International Conference on Principles and Practice of Multi-Agent Systems. Springer, 117–131. Past research also supports this result wherein supervisory support [6] Harshal Hayatnagarkar, Meghendra Singh, Suman Kumar, Mayuri Duggirala, is linked to lower levels of stress among employees by acting as a and Vivek Balaraman. 2016. Can a buffering strategy reduce workload related buffer against work related stress [7]. stress? (2016). [7] David P Himle, Srinika Jayaratne, and Paul A Thyness. 1989. The buffering effects Fig. 5 shows average cumulative stress levels of individual team of four types of supervisory support on work stress. Administration in Social members (accumulated per member over simulation duration)with Work 13, 1 (1989), 19–34. and without supervisory support and with different bench strength. [8] Remus Ilies, Megan Huth, Ann Marie Ryan, and Nikolaos Dimotakis. 2015. Ex- plaining the links between workload, distress, and work–family conflict among With supervisory support, the stress goes down visibly from approx. school employees: Physical, cognitive, and emotional fatigue. Journal of Educa- 850 to 700. In the presence of moderating interaction of supervisory tional Psychology 107, 4 (2015), 1136. [9] Ipek Kalemci Tuzun and R Arzu Kalemci. 2012. Organizational and supervisory support, the stress grows much slowly than in absence of super- support in relation to employee turnover intentions. Journal of Managerial visory support. This slower growth has a substantive impact on Psychology 27, 5 (2012), 518–534. various outcome parameters. We also see that increase in the bench [10] Janet L Kottke and Clare E Sharafinski. 1988. Measuring perceived supervisory and organizational support. Educational and psychological Measurement 48, 4 strength from 0% to 6% to counter poor supervisory support does (1988), 1075–1079. not reduce average stress in the team members. [11] Catherine A Ramus and Ulrich Steger. 2000. The Roles of Supervisory Support Behaviors and Environmental Policy in Employee fiEcoinitiativesfi at Leading- Edge European Companies. Academy of Management journal 43, 4 (2000), 605– 4 CONCLUSIONS AND FUTURE WORK 626. [12] Barry G Silverman. 2001. More realistic human behavior models for agents in In this paper, we have explored the implications of supervisory virtual worlds: emotion, stress, and value ontologies. (2001). support on work outcomes using an agent based model. We show [13] Meghendra Singh, Mayuri Duggirala, Harshal Hayatnagarkar, and Vivek Balara- that supervisory support has significant impact on TAT and work man. 2016. A Multi-Agent Model of Workgroup Behaviour in an Enterprise using a Compositional Approach. backlog. We conjecture using our model that this impact is because [14] Meghendra Singh, Mayuri Duggirala, Harshal Hayatnagarkar, Sachin Patel, and supervisory support helps mitigate stress caused by negative affect. Vivek Balaraman. 2016. TOWARDS FINE GRAINED HUMAN BEHAVIOUR This work extended our earlier work on how individual traits SIMULATION MODELS. In Winter Simulation Conference 2016. [15] Louise Tourigny, Vishwanath V Baba, and Terri R Lituchy. 2005. Job Burnout and states impact work outcomes by considering the organizational among Airline Employees in Japan A Study of the Buffering Effects of Absence social dimension of supervisory support. We plan to further extend and Supervisory Support. International Journal of Cross Cultural Management 5, 1 (2005), 67–85. this by studying peer to peer impacts as well as group level effects. [16] Michael Van Lent, Ryan McAlinden, Paul Probst, Barry G Silverman, Kevin O’Brien, and Jason Cornwell. 2004. Enhancing the behaviorial fidelity of synthetic REFERENCES entities with human behavior models. Departmental Papers (ESE) (2004), 300. [17] David Watson and Lee Anna Clark. 1999. The PANAS-X: Manual for the positive [1] Vivek Balaraman, Harshal Hayatnagarkar, Meghendra Singh, and Mayuri Duggi- and negative affect schedule-expanded form. (1999). rala. 2016. Towards better crisis management in support services organizations [18] Howard M Weiss and Russell Cropanzano. 1996. Affective events theory: A using fine grained agent based simulation. In International Conference on Prin- theoretical discussion of the structure, causes and consequences of affective ciples and Practice of Multi-Agent Systems. Springer International Publishing, experiences at work. (1996). 366–375. [19] Vathsala Wickramasinghe and Mayura Jayaweera. 2010. Impact of career plateau [2] Robert C Cummins. 1990. Job stress and the buffering effect of supervisory and supervisory support on career satisfaction: A study in offshore outsourced support. Group & Organization Management 15, 1 (1990), 92–104. IT firms in Sri Lanka. Career Development International 15, 6 (2010), 544–561. [3] Mayuri Duggirala, Meghendra Singh, Harshal Hayatnagarkar, Sachin Patel, and [20] Ann Yan Zhang, Anne S Tsui, Lynda Jiwen Song, Chaoping Li, and Liangding Vivek Balaraman. 2016. Understanding impact of stress on workplace outcomes Jia. 2008. How do I trust thee? The employee-organization relationship, supervi- using an agent based simulation. In Proceedings of the Summer Computer Simula- sory support, and middle manager trust in the organization. Human Resource tion Conference. Society for Computer Simulation International, 35. Management 47, 1 (2008), 111–132. [4] Robert Eisenberger, Florence Stinglhamber, Christian Vandenberghe, Ivan L Sucharski, and Linda Rhoades. 2002. Perceived supervisor support: contributions