The effect of queue style on server performance:
An individual perspective
Master’s Thesis, Technology and Operations Management
Name: Nick, M., van Gool Student number: s2221241 Supervisor: Msc. N. Ziengs Co-assessor: Dr. W.M.C. van Wezel
Date: 29-06-2015
Abstract
Several papers have examined the impact of different queue styles on the performance of the workers, but no prior work has looked at the influence of individual factors on worker performance. Using a controlled experiment, we investigated the influence of the strategic orientation – an individual’s way of approaching and performing tasks – on worker
performance in two different queue styles. A total of 39 students participated in an experiment with questionnaires and two server related tasks. We manipulated queue style (pooled and parallel) and measured the subjects’ strategic orientation (promotion and prevention focus). The results showed moderately significant evidence of people with a promotion focus having a higher performance than those with a prevention focus, when working in a pooled queue. Working speed – a component of performance – of these promotion focused individuals was slightly significantly higher. Performance in the parallel queue was unaffected by the
individuals’ strategic orientation. Furthermore, the highest performance was achieved by promotion focused individuals in the pooled queue. Our findings provide some important insights. For theory, our results imply that an individual’s strategic orientation moderates the causal effect of queue style on server performance in a pooled queue. For management, our findings suggest queue layout could be changed to a pooled queue, plus rewards could be tied to performance in order to induce a promotion focus, which results in the highest performance of the workers.
Contents
Preface ... 4
Introduction ... 5
Theoretical background ... 8
Regulatory focus in queuing systems ... 12
Conceptual model ... 15 Methodology ... 16 Task description ... 17 Experimental procedure ... 18 Parameter settings ... 21 Control questions ... 22 Results ... 24 Preliminary analyses ... 24 Descriptive statistics ... 27
Performance of the two foci ... 31
Post-hoc analyses ... 32
Comparison with working individually ... 38
Control variables ... 44
Discussion ... 46
Reflection on theory and method ... 48
Theoretical implications ... 48
Managerial implications ... 50
Limitations ... 50
Future research... 52
Preface
This thesis is the original intellectual product of my labor of the past six months. Those six months were months that have been predominantly exciting and interesting. I really enjoyed working on the project and increasing my understanding of the discipline. Besides the enjoyment, writing a thesis is also a process of coping with adjustments, reiterations and refinements. This has also been the case for me. At some moments I found it tough to have to rewrite pieces of text one, or sometimes multiple times. However, if you – in the end – are satisfied with the end product as it is, those moments of stress and hard work have been worth it. And I think this also counts for me.
Rarely, a thesis can be made without help or effort of others. Therefore, I would like to thank all the people that have participated in the experiment belonging to this thesis. Without all of your voluntary participation, I would not have had any data to analyze, making my thesis impracticable. But most of all, I want to thank my supervisor Nick Ziengs, without whom I would never have been able to come where I am. Thank you for providing the base for the experiments and for all the feedback and support during the process.
I hope you enjoy reading my thesis.
Introduction
In today’s competitive world, managers are constantly looking for ways to reduce service times, while simultaneously increasing the quality of the service in queuing systems. Yet, practice shows no optimal queue design with regard to these performance criteria. Large retailers such as Whole Foods, Trader Joe’s and Wal-Mart have experimented with the queue layout, testing both single waiting lines (pooled queue) and parallel waiting lines (parallel queue), with mixed results (Barbaro, 2007). Literature also does not directly provide outcome, as both pooled and parallel queues find support in literature. Analytical models from queuing theory found that a single queue with s servers leads to a lower average waiting time than s servers working in parallel (Smith and Whitt, 1981; Hopp and Spearman, 2008). However, later behavioral studies have shown that when humans are involved, parallel queues result in a higher server performance than a pooled queue (Shunko, Niederhoff and Rosokha, 2014; Song, Tucker and Murrell, 2013).
Another complicating thing is that these studies assume workers are stationary, meaning that they provide full quality levels and equal service times regardless of the state of the system (Boudreau, Hopp, McClain and Thomas, 2003). However, several empirical studies have shown that individuals are not static, but instead react to workload (Kc and Terwiesch, 2009; Batt and Terwiesch, 2012) by adjusting their speed and service quality (Tan and Netessine, 2014). Tan and Netessine (2014) showed that when workload gets high, people increase their working speed at the cost of the service quality.
However, no attention is being paid to how each individual makes this trade-off between speed and quality. These studies assume people to behave identically and to rationally weigh all factors at play (Gino and Pisano, 2008). But, since people are not
(Boudreau et al., 2003, p.185). This research aims to look for the individual factors that influence server performance in queuing systems and, thereby, provide more guidance in the practice of optimal queuing design.
Individual differences have been found to influence task motivation as well as the direction, intensity and persistence of the effort resulting from that motivation (Baddeley and Weiskrantz, 1993). Thus, individual differences can influence whether effort is directed towards quantity or quality, when there is a need to choose between both (Kanfer, 1990). One such individual factor that has been related to differences in motivation and performance is one’s regulatory focus (Higgins, 1998). Förster, Higgins and Bianco (2003) state that the different regulatory foci lead to differences in performances, since the way in which each focus approaches tasks is distinct. Regulatory focus theory discriminates between a promotion focus and prevention focus, which differ in way they strategically work to achieve goals and objectives (Higgins, 1998). The present study examines the influence of an individual’s regulatory focus on server performance among different queue styles. More specifically, we formulate the following research question: How does an individual’s strategic orientation influence server performance in queuing systems?
We answered our question by using a controlled experiment, which allowed us to tightly control the variables under investigation. The subjects worked with computerized, faster working, co-workers to handle a stream of customers at a checkout of a bicycle repair shop. In this experiment, we used the two different queue styles: a parallel queue and pooled queue. These are queues frequently used in real-life settings (Barbaro, 2007; Hiray, 2008) and provide us the opportunity to compare the most opposite queue designs.
decreases (Kostami and Rajagopalan, 2014), we use discretionary tasks to get a full picture of server performance. In order to isolate the effects of regulatory focus and queue style, the average workload is smoothed, which could otherwise affect server behavior (Tan and
Netessine, 2014). In this way, we are able to study regulatory focus and queue style so that we may consider their role in the performance differences of servers in queuing systems. This work contributes to, among others, the experimental results of Shunko et al. (2014) 1, the conceptual paper of Boudreau et al. (2003) and the empirical findings of Song et al. (2013).
Our findings showed that regulatory focus moderates the effect of queue style on server performance in the pooled queue. We found moderately significant evidence of people with a promotion focus having a higher performance than those with a prevention focus. This performance difference is caused by promotion focused individuals working faster than prevention focused ones, while quality level remained equal. In a parallel queue, performance (and its components speed and quality) did not differ across the two foci. Overall, the highest performance was achieved by people with a promotion focus in the pooled queue. For
management our results mean that switching the queue layout to a pooled queue and
equipping it with promotion oriented workers will provide the highest performance. To do so, a promotion focus can be induced by tying rewards to performance.
The rest of this article is structured as follows. First, literature on queuing and
regulatory focus is reviewed. Then, the methodology is outlined, in which the experiment will also be explained. Subsequently, the results of the experiment are provided, after which the discussion of these results and the main conclusions are treated.
1
Theoretical background
Early work in analytical modeling has shown that a single queue (or pooled queue) with a number of s servers leads to shorter average waiting times than s servers working in parallel (Smith and Whitt, 1981). The reason for this is that variability is reduced by pooling it in a single queue (Hopp and Spearman, 2008, ch.8). If a server in a pooled queue gets delayed by a difficult customer or by hitches in his working tools, the rest of the queue keeps flowing to the other servers. In a parallel queue, such a delay would cause all others in that queue to wait.
However, these analytical models rely on assumptions regarding human behavior that do not hold in real-life (Boudreau et al., 2003; Gino and Pisano, 2008; Bendoly, Donohue and Schultz, 2006). More recent work, and closely related to ours, has examined server
performance in different queue styles with real people (Shunko et al., 2014; Song et al., 2013). In their studies, Song et al. (2013) empirically investigated the influence of two different queue styles (dedicated and pooled) on the throughput time of a hospital’s
Emergency Department (ED). They found that patients’ length of stay was 10 percent shorter in dedicated queues compared to pooled queues. They state that, due to social loafing,
physicians worked slower under a pooled queue style than under a dedicated queue style. In the work of Shunko et al. (2014), the impact of social loafing and salient feedback on the productivity of workers in several queueing designs is investigated. In a controlled experiment, they also found that people worked faster in a parallel queue than in a pooled queue. Both studies state the interdependence between servers in the pooled queue, which is absent in the parallel queue, as influencing the workers’ performance. However, both studies examined the average performance in each queue and only considered speed as a performance measure. Our study will seek to highlight individual differences that cause variation in
The two queue styles (pooled and parallel) investigated in this paper have different levels of interdependence. The pooled queue, which combines all the incoming jobs into a single queue, falls within the category of the additive tasks of Steiner’s taxonomy (1972). Additive tasks are tasks which require the summing together of the inputs of all individuals to maximize the group’s product (Steiner, 1972). Thus, the contribution of each member adds to the total of the group. An example of such a task are ticket booths where the queues are shared and each employee contributes his amounts of sales to the total of the group (Edie, 1954).
In case of the parallel queue (without jockeying), in which each server has his own queue and carries the full responsibility for it, there is no interdependence between the workers. This is also called a coactive task (Smith, Kerr, Markus and Stasson, 2001, p.150), one in which individuals work in the presence of others but outcomes only depend on one’s own efforts. A GP practice is an example of such a coactive task; the general practitioners work in proximity of their colleagues but they each are responsible for their own patients.
Previous work on additive tasks mainly revolves around the possibility for social loafing (Smith et al., 2001; Karau and Williams, 1993; Weber and Hertel, 2007; Hertel, Deter and Konradt, 2003). It is found that under many conditions, people exert less effort when working on interdependent tasks (such as an additive task) than when working on a relatively independent task (such as a coactive task) (Karau and Williams, 1993). The works of Shunko et al. (2014) and Song et al. (2003) confirmed this relationship experimentally and
as the outcome of the group is determined by the sum of all group members’ contributions (Weber and Hertel, 2007, p.974).
As opposed to additive tasks, social loafing is not possible in coactive tasks. Since each person carries the full responsibility of performing the task at hand, reduced effort cannot be compensated for by others. Each individual experiences the full impact of the required task and will therefore work hard (Karau and Williams, 1993, p.682). However, in coactive tasks, people work in the proximity of others and thereby have the opportunity to compare their performance with that of others. Social comparison theory states that individuals usually adjust their performance towards standards created by their social
environment (Festinger, 1954). Namely, people have the constant need to evaluate themselves against a benchmark to maintain an accurate self-view (Corcoran, Crusius and Mussweiler, 2011, p.121). If objective benchmarks are absent or hard to compare with, people fall back on comparisons with others.
Thus, working with faster co-workers would on average lead to social loafing in the pooled queue and social comparison in the parallel queue. Based on the abovementioned, we form our first hypothesis:
H1: On average, performance in a parallel queue is higher than in a pooled queue.
Although social loafing and social comparison2 explain the performance differences between pooled and parallel queues, this only provides insight in average performance scores per queue. However, several papers have already shown that individuals differ in the degree in which they engage in social loafing (Weber and Hertel, 2007; Huguet, Charbonnier and Monteil, 1999; Smith et al., 2001; Hart, Karau, Stasson and Kerr, 2004) and social
comparison (Jones and Buckingham, 2005; Thwaites and Dagnan, 2004), leading to differences in performance. These performance differences might be explained by an
individual’s regulatory focus. Förster et al. (2003, p.149) state that both regulatory foci differ in their strategic orientation, which leads to differences in performance, since the way in which each focus approaches tasks is distinct. Hence, this seems as a plausible cause for differences in performance in both queue styles.
Regulatory focus theory states that there are two different self-regulation mechanisms, each having a distinct type of desired means and end-state; a promotion focus and a
prevention focus (Higgins, 1997, 1998). The promotion focus relates to aspirations,
accomplishments and hopes (Higgins, 1998, p.3). Cropanzano, Paddock, Rupp, Bagger and Baldwin (2008, p.37) describe people with a promotion focus as individuals that “strategically work to achieve a goal and pay special attention to advancement and gain”. Promotion
focused people evaluate their performance as having either the presence or absence of a positive outcome (gain vs. non-gain) (Higgins, 1998, p.4). Prevention focused people are concerned with safety, obligations and responsibility (Higgins, 1998, p.3). These “individuals strategically work to avoid mismatching goals and emphasize protection of the status quo” (Cropanzano et al., 2008, p.37). They value outcomes as having either the presence or absence of a negative outcome (loss vs. non-loss) (Higgins, 1998, p.5).
As an example of the distinction between the two regulatory foci, consider two students that have the goal to pass a certain course but have a different regulatory focus. The goal is the same for both, but the means through which this goal is to be obtained differs. The promotion focused student conceives the attainment of the goal as an accomplishment or aspiration, and uses eager means to achieve the goal such as reading related additional
information. Whether a feeling of cheerfulness or dejection is elicited depends on the outcome (i.e. success or failure). The student with a prevention focus, on the other hand, conceives the attainment of the goal as a responsibility or obligation, and uses vigilant means to achieve that goal such as making sure he or she has done everything that was required to be done during the course. The outcome (success or failure) will elicit a feeling of quiescence or agitation. Thus, although the goal to pass the course was the same for both students, the means to achieve this goal and the emotions experienced when attaining the goal differed depending on their regulatory focus.
Regulatory focus in queuing systems
In our setting, people only receive visible performance feedback about their speed. Just as in most real-life queue settings, the server can take the queue length of his queue or the shared queue as a measure of his own or the group’s speed. However, feedback about the servers’ quality is not readily available. Therefore, although the servers are supposed to work fast and accurately, the availability of feedback on only speed likely alters people’s goals toward speed. Furthermore, the individuals work with computerized co-workers who work faster than them to create an environment for social loafing in the pooled queue situation and for upward social comparison in the parallel queue.
In the pooled queue the group outcome consists of the summing together of all
the opportunity for social loafing. Furthermore, as stated in the previous paragraph, feedback is usually only available on speed, which provides the servers with just one benchmark of their performance. This leads us to predict the following. Since promotion focused individuals are eager to attain advancements, accomplishments and gains, they are expected to be exerting high efforts in order to maximize the output, regardless of working with faster co-workers. Since the queue length is the only form of feedback and is related to speed, these promotion focused individuals are expected to attain a high speed. Since value is only attained if subparts of the task are performed correctly, they are also expected to maintain a sufficient quality level. Prevention focused individuals, on the other hand, are vigilant to safety, duties and responsibilities. These people are concerned with avoiding losses and mismatches. When working in a pooled queue with faster co-workers, prevention focused individuals are expected to exert less effort than working individually, since the higher speed of the co-workers is assumed to be sufficient to keep up with the queue (and thereby the duty is
fulfilled). The prevention focused individuals will, therefore, lower their speed. This frees up time to perform the task correctly, which should lead to a sufficient quality level. We
therefore expect regulatory focus to moderate the extent to which an individual engages in social loafing when working with faster co-workers. More specifically, we hypothesize:
H2a: In a pooled queue, promotion focused individuals work faster than prevention focused individuals, when working with faster co-workers.
In the parallel queue, individual performance is directly dependent on one’s own inputs, ruling out the possibility for social loafing. The parallel queue falls within the category of the coactive tasks. As in the pooled queue setting, feedback is only available on speed (in terms of the queue lengths). Furthermore, the performance difference with the faster co-workers is now more readily visible than in the pooled queue, because jobs are not combined into a single queue. This means that the individuals can compare their speed with that of the co-workers. For the individuals with a promotion focus – who are aiming to maximize performance – an increase in speed is expected, because of social comparison. Since the promotion focused individuals are concerned with achievement and maximization, they are expected to increase their speed in an attempt to match the performance of their co-workers. Because of the inverse relationship between speed and quality (Förster et al., 2003), this would mean that quality level will decrease somewhat. Next, since prevention focused individuals are vigilant not to be the worst and obey to duties and responsibilities, it is expected that they will conform to the implicit norm set by the faster co-workers to avoid losses. Therefore, the prevention focused individuals will speed up with respect to working in a pooled queue to meet the standard. Because of the inverse relationship between speed and quality, quality level is expected to decrease. Although the motives to speed up and narrow the performance discrepancy between them and the co-workers is different among the two foci, the result is expected to be roughly equal. Namely, that individuals will speed up to close the performance gap between them and their co-workers. We therefore expect regulatory focus not to lead to performance differences in a parallel queue when working with faster co-workers. More specifically, we hypothesize:
H3b: In a parallel queue, quality level is unchanged by regulatory focus, when working with faster co-workers.
Conceptual model
To better visualize the aforementioned predicted relationships between the constructs, we brought them together in a conceptual model (Figure 1).
Methodology
Participants totaled 39 students, of which 33 were male and 6 female. The average age of the subjects was 25 years (SD = 2.6). To test our hypotheses, the subjects played the role of a cashier at a checkout in a bicycle repair shop. The experimental design concerned a 2x2 factorial design between subjects. This means that there are two independent variables with two levels each. These independent variables are the queue style (pooled vs. parallel) and regulatory focus (promotion focus vs. prevention focus) (see Table 1). The former is explicitly controlled, whereas the latter is measured by means of a questionnaire. Each individual worked on two consecutive trials, an individual and a group trial (for a picture of the settings see Appendix A). In the first trial, the subjects worked individually in order to determine a baseline measure of their performance. The second trial differed according to the between subjects design, which resulted in the 2x2 design mentioned above.
The influence of each combination of the levels of the independent variables on the performance of the servers (dependent variable) is investigated. Therefore, we measured the speed at which the customer requests were handled and the quality at which this was done. Speed was measured as the average processing time per customer (i.e. the time from starting with the first customer to finishing the last, divided by the number of customers). We
measured the quality level as the average number of sliders set correctly per customer (i.e. the total number of sliders set correctly divided by the number of customers). Furthermore, our total measure of performance combines speed and quality into a single measure. We took the average time per correct slider as the measure of performance (i.e. the speed measure
Table 1
Experimental conditions
Promotion focus Prevention focus Pooled queue Individual trial / Group
trial
Individual trial / Group trial
Parallel queue Individual trial / Group trial
Individual trial / Group trial
Task description
The subjects are put in the role of a cashier at the checkout of a bicycle repair shop, in which they perform a slider task. Slider tasks are used regularly in other papers and are considered to be real effort tasks (Gill and Prowse, 2012). Customers arrive at the cash desks with 5 products in their shopping cart. The items are all different from each other and priced between €0 and €12, in round figures. The subjects have to set the slider at the right price for each item by dragging the slider to the value that matches the price of the product (for a similar task see Shunko et al., 2014). The slider jumps in steps of €0,10 each. By design, the products are not priced at the extreme values €0 and €12 to ensure all sliders are equally difficult to set.
In the first trial, the subjects work individually to handle a predetermined stream of customers. A customer arrives at the cash desk with 5 items that have to be dealt with. The subjects have to set the slider of all items at the right corresponding value before they can proceed. Then they click on the submit button and the next customer arrives. No queue is displayed in this round, to avoid workload affecting the speed or accuracy of the subjects. This round serves as a measure of the subjects baseline speed when attaining a perfect quality level.
Furthermore, the participants’ objective was to maximize company revenues. In the pooled queue, the subjects work together with the co-workers to deal with the incoming stream of customers. The customers all join a single queue which has to be dealt with by the subjects and their co-workers. Here, subjects are not obliged to set the sliders of all items. They can proceed with the next customer at any time. In the parallel queue setting, the subjects work in proximity of the co-workers, but each faces its own stream of customers. Again, an incoming stream of customers has to be dealt with by the subjects, by dragging the slider to the value that corresponds with the price of the product. The subjects are not obliged to set the sliders of all items, they can choose to set the sliders they like with the objective of maximizing
company revenues in mind. Moreover, the subjects could not see the order fulfilment of their co-workers (i.e. they did not see the co-workers setting the sliders).
Experimental procedure
Participants were recruited by an announcement on the university’s blackboard website, telling that two vouchers would be allotted of €25,- each. Upon arrival, the subjects were guided to a computer, making sure there was a three meters distance with their
neighbors. When everyone had arrived, the subjects were welcomed by the experimenter and thanked for being present, after which the session started. The participants were asked to work diligently, since it concerned a task which required focus. Subsequently, subjects were
instructed to open Google Chrome and go to the URL-link provided on the blackboard in front of the room. They entered their e-mail address and started working on the questionnaires until they reached the instructions of the first round.
as their achievement motivation, need for cognition, self-efficacy and social comparison. To determine the subjects’ innate regulatory focus, the regulatory focus questionnaire of
Lockwood, Jordan and Kunda (2002) was used (see Appendix B). The participants answered 18 questions on a 7-point Likert scale. Half of the questions were related to a promotion focus, whereas the other half were related to a prevention focus. An example of such a promotion focus related question is: “I often think about the person I would ideally like to be
in the future”. An example of a prevention focus question is: “In general, I am focused on preventing negative events in my life”. Whereas the measure of the subjects’ regulatory focus
was part of the experimental design, the other constructs were included as control questions (see also paragraph ‘Control questions’).
After everyone has filled in the questionnaires and reached the instruction page on their screen, the experimenter classically elaborated on the instructions. The task was
explained, and a picture of the setting was shown as an example. The subjects were told to set all sliders at the right position before continuing with the next customer. Again, it was
mentioned to stay focused and work diligently before the subjects started working on the first round. The round was said to finish after a predetermined time had elapsed and that, therefore, some could finish earlier than others. In fact, participants worked on 30 customers and
completion time depended on their working speed3. Subjects were told to start working on the task and work through the questionnaires presented after the first round until they reached the instructions of the second round. The questionnaires contained some control questions about their perceived task ability and instrumentality among others (see paragraph ‘Control
questions’).
3
During the pre-questionnaire and the two rounds, the experimenter walked through the room evenly and steadily in order to check progress of the subjects and answer potential questions, but to avoid subjects feeling observed individually. When the participants finished the first round and the subsequent questionnaires, they reached the instructions of the second round. After everyone reached the instructions, the experimenter classically elaborated on them.
The subjects were instructed that they were going to work together with 3 other persons4 in the room, in either a pooled or parallel queue setting depending on the condition they were assigned to. The experimenter mentioned that the task the subjects would work on was the same as in the first round, in that they again would set sliders on the value
corresponding with the price of the product. In addition, the subjects were told that they could now choose the amount of sliders to set before continuing with the next customer, keeping in mind the objective of maximizing company revenues. Furthermore, the experimenter showed a picture of both the pooled and parallel queue setting as an example of what the next round would look like. Again, the subjects were asked to work as diligently as they had up to this point. The round was said to end after a predetermined time had elapsed, which could differ among the subjects. In fact, participants worked until they finished 30 customers (which arrived at 2 times their baseline speed). The subjects were indicated to start working on the second round, and to subsequently fill in a questionnaire. At the end, the subjects wrote down what they thought the purpose of the experiment was and if they had any thoughts regarding their co-workers, after which the program notified them that the experiment was finished.
4
After everyone had finished, the experimenter classically thanked the subjects for their participation and their diligent behavior during the experiment. The session was ended and shortly after, a lottery was conducted in accordance with the announcement and the winners were awarded the vouchers.
Parameter settings
In the first round of the experiment, subjects worked individually and could not see the queue. If the participants took the task seriously, a new customer was directly present when the previous was finished. This round served as the baseline measure of the subjects’ speed (when quality level is perfect).
In the second round, in which the pooled and parallel queues were introduced, the level of a few parameters had to be set. Shopping carts arrive at the subjects’ and their co-workers’ queues according to a Poisson distributed process (rate λ). The arrival rate was set at 200 percent of the subjects’ baseline speed. This means that the customers arrived on average 2 times as fast as the subjects’ normal working speed, which introduced the trade-off that could be made between speed and quality. The 200 percent level was chosen based on the effects of workload experience investigated in another study (see Waschull, 2015). Setting the percentage higher would lead to queue explosions and an unrealistic real-life working
condition, whereas a lower percentage diminished the need to trade-off (since the subjects could more easily keep up with the queue without speeding up or forgoing quality).
The average arrival rate (at 200 percent of the subjects baseline speed) was similar for all the queues in one system, in case of the parallel queues. Furthermore, the arriving
customers were randomly assigned to the cashiers. Since each worker had the full
pooled queue setting, which only has a single queue, the arrival rate was multiplied by the amount of servers in the system (including the subject).
Finally, the speed of the computerized co-workers was Poisson distributed, with a mean () slightly greater than the arrival rate. This means that the co-workers were faster than the baseline speed of the subjects and that they were on average able to keep up with the arrivals (Hopp and Spearman, 2008). All parameter settings are summarized in Table 2.
Table 2
Parameter settings
Individual trial
Group trial (pooled queue)
Group trial (parallel queue) Arrival rate (λ) - 2*Speedbaseline*#Servers 2*Speedbaseline
Co-worker speed () - 2*Speedbaseline 2*Speedbaseline
Control questions
The questionnaires filled in, at the beginning of the experiment and after both round 1 and 2, contained some control questions (see Appendix B for questionnaires). Questions about subjects’ need for cognition (Cacioppo, Petty and Kao, 1984), self-efficacy (Chen, Gully and Eden, 2001) and social comparison (Gibbons and Buunk, 1999) were included in the
questionnaire at the beginning of the experiment. This prevented these measures from being influenced by the task, since they were to be obtained in neutral state. As an example of a need for cognition and a social comparison question respectively, consider the following: “I
would prefer complex to simple problems” and “I always like to know what others in a similar situation would do”. All questions were answered on a 7-point Likert scale (Strongly
In the questionnaire after the first round, control questions were included concerning the subjects’ motivation (Bendoly and Prietula, 2008), their perceived task ability (McAuley, Duncan and Tammen, 1989), task meaningfulness (Hoigard and Ingvaldsen, 2006), fatigue and task difficulty (Chung and Monroe, 2000). An example of such a motivation related question is: “I was inspired to do well in this activity”. Again, questions were answered on a 7-point Likert scale ranging from ‘Strongly disagree’ to ‘Strongly agree’.
The questionnaire after the second round included, in addition to the questions presented after the first round, measures of the subjects’ perceived workload (Bendoly and Prietula, 2008), pressure (McAuley et al., 1989), group cohesiveness (Leana, 1985),
impressions about the co-workers, social loafing, performance discrepancy, instrumentality (Harmon, 2008), identifiability of effort (Hoigard and Ingvaldsen, 2006), task satisfaction and their performance orientation. An example of such a group cohesiveness and performance orientation question is, respectively: “I would not like to work with this group again on a
similar exercise” and “During the task, I focused on delivering high quality, and
compromised on speed (quantity)”. Similar to the other questionnaires, subjects answered
these questions on a 7-point Likert scale (Strongly disagree – Strongly agree).
Results
After the data was compiled, we had 20 subjects in the parallel queue and 19 subjects in the pooled queue. Since the customers in the first round could be clicked through without setting a single slider, we checked the accuracy of the subjects in the first round (see Figure 2). It turned out that 4 participants had an accuracy level of 20 percent or less, whereas all others had an accuracy of 100 percent or close to 100 percent. The 4 subjects with the low accuracy level did not perform the experiment as instructed and are, therefore, excluded from further analyses. This leaves us with 35 usable subjects (see Table 3).
Figure 2. Accuracy individual trial.
Preliminary analyses
First, we analyzed the individual trial to check whether our baseline measure of speed was valid enough to account for learning effects and if fatigue played a role. The average Processing Time (PT) was calculated (see Table 4) for the first 10 customers, for customers 11-20 and 21-30 for each participant and they were checked for differences. To start, 3, 1 and
Table 3
2 outliers (see Appendix C) were replaced5 by the first non-outlier value for the three interval groups respectively (this is called Winsorising, see Mulry, Oliver and Kaputa (2014)).
Comparing the PT of the first 10 customers (M = 25.21, SD = 6.27) with that of the following 10 customers (M = 20.76, SD = 6.04) reveals that there was a significant training effect (t(34) = 7.329, p < 0.01). On the other hand, comparing the group of the 11th to the 20th customer (M = 20.76, SD = 6.04) with that of the last 10 customers (M = 20.21, SD = 5.16) shows no statistical difference, t(34) = 1.151, p = 0.258. Therefore, we can assume that deleting the first 10 rounds – to account for learning effects – when calculating the participant’s baseline speed, has been appropriate and that fatigue did not play a role in this time duration.
Next, we checked for randomization of capabilities across the two queue styles, since we have manipulated queue style. First, data was checked for outliers and normality (see Appendix C). As can be seen in Table 5, there were no significant differences in the PT’s of the subjects between the two queue styles in the individual trial (trial 1) as determined by an independent samples t-test (t(25.499) = 1.026, p = 0.315). This means that the randomization of capabilities across the two queue styles has been successful. The averages of all four experimental groups are shown in Table 4.
To check whether our manipulation of queue style and faster co-workers has
succeeded, we looked at the subjective ratings of the participants regarding instrumentality of effort, identifiability of effort (queue style characteristics) and co-worker performance. In the questions on instrumentality and identifiability of effort, 1 and 2 outliers were replaced, and normality was checked6. The ratings on instrumentality are relatively high in both queue styles, parallel (M = 4.83, SD = 0.99) and pooled (M = 4.88, SD = 1.02) (rated on a 1-7 scale).
5
If we talk about outliers from this point onwards, we mean data points that fall outside the box plots. These outliers are replaced in all non-performance-related data. In case of outliers in performance related data of the group trial, they are left in to keep the average of all measures consistent (e.g. in this way the average PT in the pooled queue stays consistent with the average PT of its subgroups (i.e. the two foci)).
6
Difference between the queue styles, on the other hand, is small and not significant (t(33) = -.144, p = 0.886). The fact that instrumentality was also rated high in the pooled queue, might be due to the fact that there were 3 co-workers In this relatively small group, the subjects could have experienced their effort to be instrumental for the group outcome.
The scores on identifiability of effort were also above the mean of the scale (M = 4.72,
SD = 0.96, in the parallel queue and M = 4.35, SD = 1.50, in the pooled queue). The
identifiability of efforts were rated higher in the parallel queue, which is according to the characteristics of the queues. However, the differences failed to reach significance, U = 128.500, p = 0.406. A reason why the differences might be smaller than expected, is that people probably found their effort and performance were identifiable anyway because the computer was registering their actions.
The co-worker performance was rated by the subjects at the mean of the scale for both queue styles (M = 4.0 for both the parallel (SD = 0.71) and pooled queue (SD = 0.61)). This means that co-workers were experienced to be equally skilled in terms of speed and
capabilities in both queue styles. Running an independent samples t-test (after checking for outliers and normality, see Appendix C) confirms that there were no significant differences in perceived co-workers performance across the queue styles, t(33) < .000, p = 1.000. The fact that co-worker performance is not rated higher could be due to the fact that the average
participant set less than 5 sliders in the group trial. If they viewed queue length as the measure of performance, the subjects might think they were equally skilled as their co-workers.
Table 4
Descriptive statistics of learning effects and average PT trial 1 per experimental condition
Round 1-10 Round 11-20 Round 21-30
M SD M SD M SD
PT Trial 1 25.21 6.27 20.76 6.04 20.21 5.16
Parallel queue Pooled queue
Promotion focus (n = 9) Prevention focus (n = 9) Promotion focus (n = 8) Prevention focus (n = 9) M SD M SD M SD M SD PT Trial 1 18.42 4.79 23.80 7.56 18.43 3.98 20.88 3.70
Note. M = Mean. SD = Standard Deviation. PT = Processing Time.
Table 5
Randomization and manipulation check
Mean (SD) Test
Check type Comparative measure Parallel Pooled stat. p-value
Randomization Average PT individual trial 21.83 (7.74) 19.72 (3.92) 1.026 0.315 Manipulation Average rating
instrumentality
4.83 (0.99) 4.88 (1.02) -.144 0.886 Manipulation Average rating
identifiability
4.72 (0.96) 4.35 (1.50) -.832 0.406 Manipulation Average rating co-worker
performance
4.00 (0.71) 4.00 (0.61) .000 1.000
Note. SD = Standard Deviation. PT = Processing Time.
Descriptive statistics
The average Processing Time (PT), average Number of Correct Sliders (NCS) and Average Time per Correct Slider (ATCS) of the two queue styles (pooled and parallel), and of both foci within each queue style are shown in Table 6 and Figure 3.
the parallel queue (M = 3.84, SD = 0.9, as opposed to M = 4.02, SD = 0.9, in the pooled queue). Combining the speed and quality measure into a single measure of performance (ATCS), reveals that performance was higher in the parallel queue7 (M = 3.68, SD = 1.12) than in the pooled queue (M = 3.86, SD = 1.25). The averages of the performance in both queue styles support hypothesis 1.
Next, we look at the averages of both foci within the pooled queue (lower part of Table 6 and lower row of Figure 3). People with a promotion focus (M = 13.40, SD = 2.6) have a lower PT than those with a prevention focus (M = 16.74, SD = 5.7). The quality level, on the other hand, was higher for promotion focused individuals (M = 4.12, SD = 0.9, as opposed to M = 3.93, SD = 0.9 for prevention focused individuals). If we then look at the ATCS per foci, we see that performance is higher for people with a promotion focus (M = 3.36, SD = 0.8) than for those with a prevention focus (M = 4.31, SD = 1.5). The averages concerning PT support hypothesis 2a. The numbers about the quality level do not clearly support or tone down hypothesis 2b.
Finally, if we look at the averages within the parallel queue (lower part of Table 6 and lower row of Figure 3), we see that promotion focused individuals have a lower PT (M = 13.46, SD = 3.9) than prevention focused individuals (M = 14.06, SD = 4.1). The quality level was slightly lower for those with a promotion focus. The NCS was on average 3.80 (SD = 0.9) for promotion focused and 3.88 (SD = 0.9) for prevention focused people. If we look at the ATCS as the total measure of performance, we see that those with a promotion focus (M = 3.71, SD = 1.5) have a slightly lower performance than prevention focused individuals (M = 3.63, SD = 0.7). The averages of the PT’s do not clearly support or tone down hypothesis 3a. On the other hand, the averages of the quality level seem to support hypothesis 3b.
7
As can be seen in Table 6 and Figure 3, there is spread present in all measures (PT, NCS and ATCS), indicating that there is variation in the behavior of the individuals.
Furthermore, the spread is roughly equal for most of the groups, which is important for a valid comparison.
Table 6
Descriptive statistics per queue style and per experimental condition
Parallel queue Pooled queue
M SD M SD PT Trial 2 13.76 3.9 15.178 4.7 NCS 3.84 0.9 4.02 0.9 ATCS 3.68 1.12 3.86 1.25 Promotion focus (n = 9) Prevention focus (n = 9) Promotion focus (n = 8) Prevention focus (n = 9) M SD M SD M SD M SD PT Trial 2 13.46 3.9 14.06 4.1 13.40 2.6 16.74 5.7 NCS 3.80 0.9 3.88 0.9 4.12 0.9 3.93 0.9 ATCS 3.71 1.5 3.63 0.7 3.36 0.8 4.31 1.5
Note. M = Mean. SD = Standard Deviation. PT = Processing Time. NCS = No. of Correct Sliders. ATCS = Average Time per Correct Slider.
8
Figure 3. Plots of average processing time (seconds), average number of sliders set correctly and average time per correct slider (seconds) with
Performance of the two foci
In order to determine whether an individual’s regulatory focus influences server performance in a parallel and a pooled queue, we compared the average PT (measure of speed) and the average NCS (measure of quality) between both foci within each queue style. In the post-hoc section we also compare the Average Time per Correct Slider (ATCS) between the foci as a total measure of performance.
First of all, we compared the average performance in the parallel queue with that of the pooled queue. The ATCS data of both queues was not normally distributed and contained one outlier. Therefore, we used a Mann-Whitney U test to test our first hypothesis.
Furthermore, we checked if the PT and NCS data of both foci within each queue style were free of outliers and if the data were normally distributed for all four experimental conditions. This was the case for all separate measures (see Appendix D).
To test our second hypothesis, we performed two independent samples t-tests (since observations were independent). For hypothesis 2a, we compared the PT’s of the two foci in the pooled queue to determine differences in working speed. The other independent samples t-test was used to compare the NCS of both foci to determine the differences in service quality. In the same way, we used two independent samples t-tests to test our third hypothesis.
Thereby, we compared speed and quality of both foci within the parallel queue. For the results of the statistical tests, see Table 7.
Table 7
Summary of statistical tests of hypotheses
Test
stat. p-value
H1 On average, performance in a parallel queue is higher than in a pooled queue
U = 137 0.597
H2a In a pooled queue, promotion focused individuals work faster than prevention focused individuals.
H2b In a pooled queue, service quality is unchanged by regulatory focus.
t(15) = .426 0.676 H3a In a parallel queue, server speed is unchanged by
regulatory focus.
t(16) = -.321 0.752 H3b In a parallel queue, service quality is unchanged by
regulatory focus.
t(16) = -.193 0.849
Regarding the results of the tests, hypothesis 1 cannot be statistically confirmed. Thus, performance in the parallel queue was not statistically significantly higher than in the pooled queue.
Next, hypothesis 2a cannot be confirmed. However, the difference in server speed between the two foci is slightly significant (p = 0.147). This is probably due to the small sample size: a slightly larger sample size will likely show a significant difference at the 0.05 alpha level. This suggests that the difference in server speed between the two foci probably exists. Next, hypothesis 2b is confirmed (p = 0.676), meaning that there were no significant differences in the quality levels of both foci in the pooled queue.
Furthermore, the results confirm the third hypothesis (both 3a and 3b). There are no significant differences in both server speed and quality level between the two foci in the parallel queue. Although the sample size is also small in this case, the results are amply insignificant, meaning that a larger sample size would likely not make the results significant.
Post-hoc analyses
In order to fully explore the differences between the two queue styles and two foci, we performed several post-hoc analyses. First of all, we took ATCS as a total measure of
foci within both queues and three of the four groups contain one outlier9, we use the two Mann-Whitney U tests to test for significant differences (see row 1 and 2 in Table 8). The results show that the ATCS of the two foci within the pooled queue (M = 3.36 for promotion focused and M = 4.31 for prevention focused individuals) are statistically significantly
different from each other at the alpha 0.10 level and nearly significant at an alpha of 0.05 (p = 0.054). This hints at the existence of a difference in performance between the promotion and prevention focused individuals within the pooled queue. This outcome also hints at either social loafing of prevention focused individuals or social competition of those with a promotion focus, or both.
Within the parallel queue, there is no significant difference in ATCS between the two foci (M = 3.71 for those with a promotion focus and M = 3.63 for prevention focus, p = 0.566). This is, of course, in line with the fact that working speed and quality (the components of ATCS) also did not differ significantly between the two foci in the parallel queue (see our third hypothesis).
Next, we compared the average PT and average NCS between the two queue styles. Both are checked with a Mann-Whitney U test (because of one outlier in the PT-data in the pooled queue and non-normality of the NCS-data in the parallel queue). Although the average PT (M = 15.17, in the pooled and M = 13.76 in the parallel queue) and average NCS (M = 4.02, and M = 3.84 for pooled and parallel queue respectively) are both higher in the pooled queue than in the parallel queue, the differences fail to reach statistical significance (p = 0.448 and p = 0.419 respectively). The higher PT’s in the pooled queue is in line with the findings of Shunko et al. (2014), however, the average NCS is also higher in the pooled queue. In the study of Shunko et al. (2014), speed was the measure of performance, since every slider had
9
to be set correctly. In our study, individuals could trade speed for quality and vice versa. Therefore, a better comparison of our findings with those of Shunko et al. (2014) is to take our total measure of performance. We tested performance in both queues with our first hypothesis. Although not statistically significant, we did find that performance was lower in the pooled queue, which is in line with the findings of Shunko et al. (2014). For a summary of the post-hoc statistical tests see Table 8. For box plots, normality tables and test output of the post-hoc analyses see Appendix E.
Table 8
Summary of post-hoc statistical tests
Test
Comparative measure stat. p-value
Performance (ATCS) of both foci in pooled queue U = 16 0.054*
Performance (ATCS) of both foci in parallel queue U = 34 0.566
PT in pooled queue vs. parallel queue U = 130 0.448
NCS in pooled queue vs. parallel queue U = 128.500 0.419
Note. *p < 0.1; **p < 0.05. PT = Processing Time. NCS = No. of Correct Sliders. ATCS =
Average Time per Correct Slider.
In addition, we looked at the average PT and average NCS per round (i.e. per
customer) for each experimental condition (see Figure 4). As can be seen from the graphs, the PT in the first one or two rounds is higher than the average PT, after which the PT declines and swings around the mean evenly. This could have two reasons: the subjects have to start-up again or the system has not reached steady state yet (since the system starts with only one customer in queue). Looking at the average NCS per round, we see no strange things. The average per round swings around the mean evenly. Furthermore, we see that the average PT per customer and the average NCS per customer follow the same patterns in each
the figures display the averages mentioned earlier, showing that speed and quality level of both foci is on average roughly equal in the parallel queue. In the pooled queue, it can be seen that the average speed and quality level are both higher10 for the promotion focused
individuals than for those with a prevention focus.
We also visualized the PT and NCS per round (i.e. per customer) for the most extreme cases in each experimental condition (see Figure 5). This means that we looked at the
behavior of those with the highest promotion and those with the highest prevention scores in both queue styles. Again, the PT and NCS follow the same patterns most of the time, which is logical since setting more sliders correctly requires more time.
Looking at the two foci in the parallel queue (upper row of Figure 5), we see that the average PT (M = 15.20 and M = 15.27) and average NCS (M = 3.40 and M = 3.33) of the extreme promotion and prevention focused persons are roughly equal. This is in line with our third hypotheses (that working speed and quality are approximately equal in a parallel queue). If we look at the pooled queue, we see that the average PT of the extreme promotion focused individual (M = 14.60 sec) is lower than that of the extreme prevention focused one (M = 27.53 sec). This is also in line with hypothesis 2a (that in a pooled queue, promotion focused individuals work faster than those with a prevention focus). Furthermore, the average NCS is also higher for the promotion focused individual (M = 4.43 against M = 3.43 for the
prevention focused person). The higher quality level (i.e. NCS) for the promotion focus is not what we expected, but it does confirm that the promotion oriented extreme at least loafed less than the prevention oriented one. Altogether, the extremes of each experimental group have performed quite in line with what we would expect.
Figure 4. Average PT (= Processing Time) and average NCS (= No. of Correct Sliders) per customer per experimental condition. Note. Pa =
Figure 5. PT (= Processing Time) and NCS (= No. of Correct Sliders) per customer per experimental condition of the extremes. Note. Pa =
Comparison with working individually
In order to assign the performance discrepancy between the pooled and parallel queue, and between promotion and prevention focused individuals to either social loafing or social comparison, we compare the results of our two queues with that of subjects who have worked individually. Data of subjects that worked individually comes from a closely related study (see Waschull, 2015), in which the experiment was conducted under the same conditions and with the same measurements. The averages and spreads of the PT, NCS and ATCS for each division (i.e. queue style and regulatory focus within each queue style) are shown in Figure 6.
As can be seen from Table 9 and Figure 6 the mean PT of the subjects that worked individually (M = 11.20, SD = 2.4) is lower than those in the parallel (M = 13.76, SD = 3.9) and pooled queue (M = 15.17, SD = 4.7). After the data was checked for outliers11, normality and homogeneity of variances (see Appendix III), we performed a Kruskal-Wallis H test to check whether there are significant differences between these groups. The results show that there are significant differences between at least two of the groups, X2(2) = 8.353, p = 0.015. To check which groups’ means are significantly different from each other, we performed three post-hoc Mann-Whitney U tests. These test show that the PT in the pooled queue is statistically significantly higher than working individually, p = 0.005. The differences in PT between the parallel queue and working individually (p = 0.046) were also significant. Comparing the parallel and the pooled queue showed no significance (p = 0.448). The results of all the statistical tests in this section are summarized in Table 10.
The average NCS is the lowest in the individual scenario (M = 3.15, SD = 0.8), then comes the parallel queue (M = 3.84, SD = 0.9) and the highest average is in the pooled queue (M = 4.02, SD = 0.9). To check for differences, we first checked for outliers and normality.
11
The data in the individual and parallel scenario is not normally distributed. Therefore, we used a Kruskal-Wallis H test to check for differences in the average NCS between the three groups. The results show that there is a significant difference between the groups, X2(2) = 8.485, p = 0.014. In order to determine which groups differ from each other, three Mann-Whitney U tests are performed as post-hoc tests. From these results it can be concluded that the average NCS in the parallel queue was not statistically significantly different from that in the pooled queue (U = 128.500, p = 0.419). On the other hand, the difference between the parallel queue and the individual scenario (U = 94, p = 0.031), and that between the pooled queue and individual scenario (U = 71, p = 0.007), are significant.
In the same way the ATCS for the parallel queue (M = 3.68, SD = 1.1), pooled queue (M = 3.86, SD = 1.3) and the individual scenario (M = 3.69, SD = 1.0) are compared for differences. This data contains two outliers and is not normally distributed for two of the three groups. Therefore, we again used a Kruskal-Wallis H test to check for differences. The results indicate no significant differences between any of the groups, X2(2) = .289, p = 0.866.
Table 9
Copy of table 5 including descriptive statistics of working individually
Parallel queue Pooled queue Individual
M SD M SD M SD PT Trial 2 13.76 3.9 15.1712 4.7 11.20 2.4 NCS 3.84 0.9 4.02 0.9 3.15 0.8 ATCS 3.68 1.1 3.86 1.3 3.69 1.0 PRO focus (n = 9) PRE focus (n = 9) PRO focus (n = 8) PRE focus (n = 9) PRO focus (n = 9) PRE focus (n = 9) M SD M SD M SD M SD M SD M SD PT Trial 2 13.46 3.9 14.06 4.1 13.40 2.6 16.74 5.7 11.69 2.8 10.71 1.9 NCS 3.80 0.9 3.88 0.9 4.12 0.9 3.93 0.9 3.31 1.0 2.99 0.5 ATCS 3.71 1.5 3.63 0.7 3.36 0.8 4.31 1.5 3.64 0.8 3.73 1.2
Note. M = Mean. SD = Standard Deviation. PRO focus = Promotion focus. PRE focus = Prevention focus. PT = Processing Time. NCS = No. of Correct Sliders. ATCS = Average Time per Correct Slider.
12
Now that we have compared the average PT, average NCS and ATCS between the three queue styles, we see that the total performance (ATCS) in the parallel queue and working individually is roughly equal, whereas it is lower in the pooled queue. Although the results are statistically insignificant, which could be due to our small sample sizes, we do find performance is lower in a pooled queue. This all would mean that, on average, working in a parallel queue results in the same performance as working individually (i.e. no overall effect of social comparison). Furthermore, on average, people loaf in a pooled queue resulting in a lower performance than working individually.
In order to discriminate social loafing and social comparison between the two foci in each queue style, we compared the PT, NCS and the ATCS of both foci with that of those who worked individually.
First, we compared the average PT, average NCS and the ATCS of those with a promotion focus across the three queue styles. Concerning the average PT, all three groups are free of outliers, data is normally distributed and there is homogeneity of variances, so we used a one-way ANOVA to check for differences. The results of the test show no statistically significant differences in PT of promotion focused individuals across the three queue styles,
F(2,23) = .899, p = 0.421. Next, we compared the average NCS by promotion focused
We then compared the average PT, average NCS and ATCS between the three queue styles for prevention focused individuals. Data was normally distributed and contained no outliers for the PT and NCS data. Thus, we used a one-way ANOVA, which was statistically significant for both PT and NCS, F(2,24) = 4.682, p = 0.019 and F(2,24) = 4.027, p = 0.031, respectively. From a post-hoc Tukey test, it appears that the both the PT (p = 0.015) and NCS (p = 0.047) was different between the pooled queue and working individually. For the ATCS, there were outliers and data was non-normal, so we used a Kruskal-Wallis H test. Results show no significant differences for prevention focused individuals across the three queue styles, X2(2) = 1.337, p = 0.513.
Although not statistically significant, performance (ATCS) of promotion focused individuals is roughly equal in the parallel queue and working individually. This would indicate they worked equally hard. Performance in the pooled queue was somewhat higher than that of those who worked individually. Thus, these promotion oriented people seem to have increased their effort in the pooled queue. Next, the prevention focused individuals’ performance was also roughly equal in the parallel and individual setting, meaning that they worked equally hard. In the pooled queue, however, performance was lower than that of those who worked individually. These prevention oriented people, thus, decreased efforts in the pooled queue.
Table 10
Summary of statistical tests: comparison with working individually
Test
stat. p-value
Comparing PT (parallel vs. pooled vs. individual) X2(2) = 8.353 0.015**
PT parallel vs. pooled U = 130 0.448
PT parallel vs. individual U = 99 0.046**
PT pooled vs. individual U = 68 0.005**
Comparing NCS (parallel vs. pooled vs. individual) X2(2) = 8.485 0.014**
NCS parallel vs. individual U = 94 0.031**
NCS pooled vs. individual U = 71 0.007**
Comparing ATCS (parallel vs. pooled vs. individual) X2(2) = .289 0.866 Comparing PT of promotion focus (parallel vs. pooled vs.
individual)
F(2,23) = .899 0.421
Comparing NCS of promotion focus (parallel vs. pooled vs. individual)
X2(2) = 2.561 0.278 Comparing ATCS of promotion focus (parallel vs. pooled
vs. individual)
X2(2) = .540 0.760 Comparing PT of prevention focus (parallel vs. pooled vs.
individual)
F(2,24) = 4.682 0.019**
PT prevention focus parallel vs. pooled -2.681 0.378
PT prevention focus parallel vs. individual 3.348 0.227
PT prevention focus pooled vs. individual 6.030 0.015**
Comparing NCS of prevention focus (parallel vs. pooled vs. individual)
F(2,24) = 4.027 0.031**
NCS prevention focus parallel vs. pooled -.052 0.989
NCS prevention focus parallel vs. individual .889 0.063*
NCS prevention focus pooled vs. individual .941 0.047**
Comparing ATCS of prevention focus (parallel vs. pooled vs. individual)
X2(2) = 1.337 0.513
Note. *p < 0.1; **p < 0.05. PT = Processing Time. NCS = No. of Correct Sliders. ATCS =
Figure 6. Plots of average processing time (seconds), average amount of sliders set correctly and average time per correct slider (seconds) with
Control variables
Finally, we checked whether the control variables identified earlier have an effect on the speed, quality and performance of the subjects. We separated the promotion and
prevention scale (instead of a single continuum) and we included four more variables (age, self-efficacy, need for cognition and social comparison).
Figures of the scatter plots of the average PT, average NCS and ATCS plotted against a list of control variables are shown in Appendix F. We first look at the scatter plots of the average PT. The plots of the PT against the prevention scale and the PT against age show a weak linear relationship. This means that the higher the prevention score or age, the higher the PT. The first seems logical, as those with a prevention score were expected to be slower in the pooled queue than promotion focused individuals, being more prevention focused leading to higher PT’s matches this expectation. Furthermore, a higher age being related to a higher PT is somewhat more difficult. It could be due to younger people being more experienced with digital devices. However, in this age range (20-30 years), this would be remarkable. The other four variables do not show a relationship with the PT (plots seem random).
From the scatter plots of the average NCS and ATCS against the control variables it can be seen that none of the six control variables show a clear relationship with the average NCS or ATCS (i.e. the plots appear random).
As outlined above, the interaction between the control variables and the three measures (PT, NCS and ATCS) was not noteworthy, but, in addition, we check if the
distribution of the control variables across the two queue styles was randomized. We checked for randomization by means of three Mann-Whitney U tests and three independent samples t tests (depending on normality and outliers). The results (see Table 11) show that the
distribution of the six variables is random across the two queue styles. This increases the believe that our findings are attributable to the dependent and independent variables under investigation in this paper.
Table 11
Summary of randomization check of control variables
Test
Randomization across queue styles stat. p-value
Promotion focus scale t(33) = -1.670 0.104
Prevention focus scale U = 144 0.766
Age t(33) = -.817 0.421
Self-efficacy U = 148 0.869
Need for cognition t(33) = .243 0.810
Social comparison U = 152.500 0.987
Discussion
Our results showed that, overall, performance in the parallel queue was higher than in the pooled queue. Although this performance difference failed to reach statistical significance (future research could find out whether this is due to our small sample size or not), the
numerical difference is consistent with the experimental and empirical findings of Shunko et al. (2014) and Song et al. (2013), respectively. Moreover, this numerical difference is also in accordance with the theory on which our hypothesis was based.
Comparing the performance of the two queue styles with working individually showed no statistically significant differences. Performance in the parallel queue was roughly equal to working individually, whereas the performance in the pooled queue was somewhat lower (see paragraph above). However, speed and quality in both the parallel and pooled queue were significantly lower and higher than working individually, respectively. Thus, the trade-off between speed and quality is made differently when co-workers are present than when they are not. When working alone, a clear benchmark is missing, so people might just work fast in order to keep the queue as short as possible. When the co-workers are present, the joined responsibility in the pooled queue and the visible queue lengths of the co-workers in the parallel queue, could cause the speed of the workers to be lowered. This on its turn frees up time to increase the quality level.
that the subjects worked collectively in our experiment as opposed to individually in the work of Förster et al. (2003). The collective task structure brings to the fore additional mechanisms, such as interdependence between the workers (Karau and Williams, 1993).
Looking at the single measure of performance, we found performance of individuals with a promotion focus to be moderately significantly higher than that of prevention focused people. An explanation of the higher performance might be that individuals with a promotion focus set higher goals and perceived their effort more instrumental to a positive outcome than prevention focused people did (Hart et al., 2004, p.996). Comparing the total performance scores in the pooled queue with working individually shows that promotion focused individuals increased performance in the pooled queue. A reason for this might be that co-worker effort was rated at the mean of the scale, leading the promotion oriented people to feel a good group result depended on their effort. People with a prevention focus, on the other hand, have a lower performance in a pooled queue compared to working individually. This indicates that individuals with a prevention focus exert less effort and, thus, loaf when working in a pooled queue. This is in line with what we supposed in the lead-up to our hypothesis and with the fact that individual differences can moderate social loafing (e.g. Smith et al., 2001; Hart et al., 2004).
