University of Amsterdam
The Importance of Social Ties
on Productivity
By Eva Verhoef (10003622)
Specialization: Finance and Organization, Organization Economics Supervisor: Eszter Czibor
Date: 16/07/2014
Faculty of Economics and Business
Abstract
In this paper we discuss the effect of social ties. Do strong social ties increase productivity
through positive behavioural motives? Several behavioural motives are discussed, hence the
motives that are important in this paper are altruism and image concern. An experiment is
designed to test three hypotheses and identify the effects – (i) whether an employee performs
better when social ties between a manager and employee are strong; (ii) whether an employee
performs better when the payoff of the manager depends on his/her productivity; and, (iii)
whether the employee performs better when his/her performance can be seen by the manager.
The experiment is conducted via Facebook. No significant results are found, however the
subjects in a real experiment are more controllable. It is interesting to conduct the experiment
with a more controllable group.
Contents
1. Introduction and Motivation ... 4
2. Literature Review ... 5 3. Hypotheses ... 6 4. Experimental Design ... 7 4.1 Treatments ... 7 4.2 Participants ... 8 4.3 Design ... 8 4.4 Game ... 8 4.4.1 Payoffs ... 9 5. Descriptive statistics ... 9 5.1 Correlations ... 10 5.2 Mean estimations ... 11 6. Results ... 12 6.1 T-tests ... 12 6.2 OLS-regression ... 14
7. Conclusion and Discussion... 17
8. Appendices ... 19
Appendix 1; Experimental form ... 19
Appendix 2; Desciptives ... 23
9. References ... 25
1.
Introduction and Motivation
Companies are constantly trying to increase productivity. To increase productivity there is need for incentives. The principal-agent theory motivates the need for incentive provision. The stockholders (principal) select a CEO (agent) to run the organization. Ownership and control is separated. However, the stockholders want the CEO to act in their best interest. Incentives are introduced to align the behaviour of the agent with the interest of the principal (Garen, 1994). Prendergast (1999) summed some evidence that agents respond to incentives.
There are non-monetary and monetary incentive provisions in the workplace. A monetary incentive is based on a reward that includes money. If the employer accomplishes a task or performs better than expected a reward is given. So employees can earn more compensation depending on their
performance. An example is the introduction of a piece rate, where workers are paid based on the amount of output (Lazear, 1986). Non-monetary incentives associates rewards excluding money. These kinds of rewards can simply be recognition or some privileges in working scheme. Bandiera et al. (2009) shows that good social connections, between managers and their workers, can provide good non-monetary incentives.
Human behaviour is influenced by two broad motivations: extrinsic and intrinsic motivation. A monetary incentive is an extrinsic motivation due to the material reward provided. The incentives following from a non-monetary treatment are intrinsic concerned. This motivation comes from the inside and encompasses pro social behaviour (Ariely et al., 2009). There is already a motivation on its own, independent of any reward (Gneezy and Rustichini, 2000). Intrinsic motivation is stimulated by social norms and habits. However, financial incentives can crowd out this motivation. In other words, the total contribution of the agent diminishes through monetary rewards (Frey and Oberholzer-Gee, 1997).
Increased pro social behaviour through social connections can be beneficial for productivity. Although it is recognized that social norms and habits are playing a significant role in economic context,
economists are reluctant to assign these factors to their theories. Normally the assumption is that the subjects maximize their money. Since social norms play a role in economic theory there are constant deviations from money maximization. As a consequence the competitive equilibrium cannot
unambiguously be confirmed (Fehr et al. 1998).
In addition, contriving alignment between principals and agents is not simply to pay for performance, because the performance of an agent is not always measurable. And the theory is not always confirmed by the observations. Therefore it is interesting to examine what incentivizes the agent in the first place, regarding social norms and habits. Intrinsic motivation may be stimulated by strong social ties. So when the social ties between managers and employees are strong it might stimulate the performance among people in the company. If we know more about social norms and habits we can focus on that, besides measures that cost money. The aim of this paper is to think of the influence of social ties on
productivity. Therefore I formulated this research question: What is the effect of social ties between a manager and an employee on productivity.
2.
Literature Review
There are several channels via which social ties influence workers’ productivity. In this section I will discuss trust, favouritism, image concern, altruism, reciprocity and fairness and equity motives. One factor that influences the behaviour in the workplace is trust. There are numerous studies
concerning the effect of trust in a workplace. One thing causing trust is having a good connection with one another. And a good social connection between the manager and his employees can be beneficial in reducing asymmetric information and can provide intrinsic motivation (Bandiera et al., 2009). Another study showed the hidden costs of control. Here it seemed that it is not always better to control than to trust. (Falk and Kosfeld, 2006). Therefore it seems plausible to aim social control in a
workplace. Meaning that social control contributes to pro social behaviour. Although it has important benefits to organizations, it is not straight forward for every organization and many outcomes are not statistically significant (Dirks et al., 2001).
Another factor that influences the behaviour in the workplace is favouritism. Favouritism involves distinctiveness in treatment between friends and non-friends. Bandiera et al. (2009) showed that when favouritism is present, social connections can be harmful for a company. Although favouritism can increase individual productivity when the manager favours the worker, it can be detrimental for firm’s average productivity. This results from assigning tasks or positions to the managers good connected workers instead of less connected workers who can execute the task better.
Image concern is a channel that can increase productivity and may be influenced by social ties. Image concern relates to the desire to give a good picture of yourself towards others. Containing not being selfish, but also the longing of showing abilities (Bénabou and Tirole, 2009). Hoffman et al. (1994) showed that lower social isolation in the dictator game causes more offers towards the counterparts. But still in total isolation some dictators gave away some money. However, Hoffmann et al. (1994) dot not motivate this by ‘other-regarding’ preferences. According to this paper image concern still can play a role if someone do not believe the total anonymity. Hoffman et al. (1994) named it the
experimenter effect, offering money because the experimenter can observe the decision of a subject. Furthermore altruism is a channel we discuss. Pure altruism is hard to distinguish from image concerns and ‘other-regarding’ preferences. Altruism is a private preference to care about each other’s sake (Ariely et al., 2009). Grossman and Eckel (1994) interpret the offers in the total anonymous dictator game as altruistic behaviour. Instead of offering money for a counterpart, the money offered is sent to charity. As a result the offers increase. The experimenter still does not know the offer, whereby image concern cannot be present. This experiment provides evidence for the presence of altruistic motives in human behaviour.
Suppose the anonymity weakens regarding the dictator game, reciprocity may become a motive to offer. Fehr et al. (1998) conclude that if someone treats you well you will treat the other with the same respect. This is called reciprocal behaviour, which is a basic element in human behaviour. Reciprocity is not only present in a repeated game since, subjects take into account that they can meet again in another game. Reciprocal behaviour is efficiency enhancing. When social ties are strong reciprocal behaviour become more naturally and there may be less deviations.
Additionally, the provision of incentives becomes more ambiguous through the existence of another channel, fairness and equity motives. Fehr and Schmidt (2000) discuss these motives of subjects. Fairness concerns impact the contractual choices of principals and the behavioural responses of the agents significantly. Their experiment provided an inequity aversion model. In this model the utility of subjects depend on another subjects payoff. Subjects do not accept unfair outcomes. Utility decreases when the payoff of the other subject is higher, but also when the payoff of the other subject is lower. In this extent the utility of a subject decreases with inequitable or unfair outcomes. There is non-negligible evidence that reciprocally fair behaviour is of impact on the interaction between subjects. In addition, it is interesting to think of the influence of social relations on inequity and the effect on subjects utility.
There is plenty of evidence that all these channels are relevant for motivating human behaviour. I am interested in this motivations linked to social ties. From the channels above, I hypothesize altruism is of high impact on productivity when social ties become stronger. If the connection between manager and employee is strong the employee may want to work harder, also for the sake of the manager. But altruism is hard to distinguish from image concern, because working hard as a private motivation or working hard for your image is different. Accordingly, the aim of this paper is to focus on the effect of social ties on productivity via the channels altruism and image concern. The existence of a remaining effect will be speculated via the other channels. Some studies named above explore this field and found evidence, however not every study is related to social ties. To my knowledge, this study is the first to analyse how altruism and image concerns influence the effect of social ties on productivity.
3.
Hypotheses
Based on the above evidence, I will try to find support for three hypotheses. I expect strong social ties may increase behavioural motives. When a managers payoff depends on performance altruistic behaviour may become stronger by an employee. And if answers are absolutely anonymous, so when the answers are not seen by a manager, image concern decreases. Further, when neither the payoff depends on performance nor the answers are seen, this may lead to diminished performance, since there are less motives to perform well.
Hypothesis 1: Employee performs better when the social tie between a manager and employee is strong.
Hypothesis 2: Employee performs better if the payoff of the manager depends on his/her productivity. 6
Hypothesis 3: Employee performs better when his/her performance can be seen by the manager.
4.
Experimental Design
In order to test the hypotheses of this study, an experiment is conducted to find the effect of social ties between a manager and an employee on productivity. The experiment includes three different
treatments to test the three hypotheses. In the next subsection the treatments are described.
Subsequently, the second subsection describes how the participants are gathered. The third subsection presents the design and the last subsection explains the game.
4.1 Treatments
The experiment is designed in three treatments. Each treatment has a different group, this is called between subject design. So the subjects participate in one single treatment. Within this design the treatments cannot affect each other. First, it will be of interest to see if there is a difference in
performance between close friends and acquaintance. After that we will question what this difference is related to. The aim of this paper is not to identify intrinsic motivation. The channels altruism and image concern are tested. If I find an effect of social ties even in the absence of altruism and image concern, I will tentatively attribute this to different intrinsic motivations discussed in section 2. First I try to find a variation in performance between weak and strong social ties. Then the different characteristics of the three treatments explore the underlying causes of that variation. On the next page, table 1 specifies the different treatments. Before specifying the differences, I will discuss the general framework of the experiment.
In total, six subjects are assigned as manager and the other subjects participate voluntarily. These are the employees. The employee is asked to do a task. During the implementation of this task the manager is supervising. This is the only task of the manager and cannot otherwise influence the performance of the employee. The employee is not paid for implementing the task. However, the payoff of the manager differ per treatment, as well the evaluation of the employee’s performance by the manager.
The difference between treatment one and two is whether the manager’s payment depends on the performance of the employee (in the form of a bonus). In both treatments, the employee’s performance is known by the manager. Comparing these treatments a deviation can be ascribed both to doing a favour for the manager as well as employees being concerned about their image. A favour is done by working hard so the manager earns a bonus. And image concerns are evoked when the manager sees the answers when performance is not that good. As a consequence, the treatments one and two split these two channels. The characteristics of treatment three diverge in the performance being known by the manager. The manager does not get a bonus and the performance is not public. If the employee still works hard under these circumstances, there is no image concern or altruism present. Remember, the worker is not paid for performance, so if he still works harder in the presence of strong social ties,
we attribute this to the brought concept of intrinsic motivation, which embodies lots of channels, some of which are discussed in the first paragraph.
Performance known
Yes No
Manager bonus Yes Treatment one:
Altruism + image concern No Treatment two: Image concern Treatment three: Intrinsic motivations Table 1: Differences between treatments
4.2 Participants
Ideally data of social ties and productivity would be gathered from a firm. However, this data is generally not available. In addition, running the experiment with real managers and employees at a firm is too costly. Therefore my plan was to run the experiment on a high school. Children of this age are not totally comparable, but in this way the experiment is better to control. Nevertheless the schools are at the end of the school year and have no time. Instead, I run the experiment via Facebook. I choose three managers in advance and they invited their friends. The participants participated voluntarily.
4.3 Design
The experiment is executed via Facebook. The managers do not have an active task, hence they are picked in advance. Three managers are chosen and they send all their Facebook friends an invitation to participate in the experiment. Via their Facebook wall a message is posted with the link to the survey. This is done two times at the same time, two days a row. The message is the same for all treatments. The third day three more managers are included in the experiment in order to collect enough responses. The survey includes a questionnaire. This will define the relationship of the participants beforehand.
4.4 Game
To operationalize the research question I propose the following game. The participants are divided by being a manager or an employee. The employee is asked to solve a task that is delegated by the manager. Women have different attitudes towards competition then boys. Hence, this must be taken into account by choosing a task. When a task is “masculine” women are more reluctant to compete. Therefore I chose a verbal task, which is more “gender-neutral” (Grosse and Riener, 2010). The employee is asked by the manager to solve thirteen anagrams in six minutes. I repeat this in three different treatments, as discussed in section 4.1 (detailed instructions are included in Appendix 1). Before the employee start the task, they are required to fill out a questionnaire. This questionnaire is
mentioned to be supportive to the game and it is designed to measure the relationship between the manager and employee. I decided to clarify the relationship beforehand. Then, if the participant decides to stop during the experiment the relationship is already clarified. The social tie is measured by questions concerning the relationship and questions about the employees Facebook activities connected to the managers page. Thus, the social tie is measured from the employee’s point of view. After the task is completed, a few questions are asked about the manager, difficulty of the task and explanation in the survey. An example of the questions before and after the implementation of the task is: How is your relationship towards this person? How often do you like a photo posted by this person? How do you feel about this person being your manager?
4.4.1 Payoffs
In the first two treatments the payoff of the manager is known by the employee and the manager. In treatment one the payoff for the manager is 10 points if the task is well implemented and 0 points when the task is not well implemented. So, in case of good implementation the manager is rewarded with a bonus. Well implemented means eight or more out of thirteen good answers. In treatment two the task is the same. The only difference is that the payoff of the manager is 5 points. This is a flat payment, meaning that this payment does not depend on the performance of the employee. In treatment three the payoff is equal to the payoff in treatment two. However the number of the employee’s correct answers of the task is not public to the manager.
By keeping the payoff of the employee zero during the treatments I control for the managers influence on payoff. The manager cannot influence the employees payoff. If there is a difference between the treatments it can be ascribed to the different motivations during the performance of the task when being friends or not. The payoff will always be zero points, independent from the performance. Gneezy and Rustichini (2000) have evidence that offering money is not always increasing the performance. They showed that not paying is better than paying a little amount. It would cost too much money to pay out all three managers for every participant. Therefore the employees do not receive any payment. Concerning the manager, five participants are picked randomly per treatment at the end of the experiment. According to the points the selected participants earn for them, the
managers get paid.
5.
Descriptive statistics
In this section the data collected with the experiment is examined. The data of the experiment is gathered via Facebook. In total 98 subjects participated with 6 different managers.
Treatment Manager # participants
1 1 11
2 10
3 14
2 4 32
3 5 20
6 11
Table 2: Participant overview
5.1 Correlations
For the significance of the correlations I use a p-value of 0.10. In the experiment the first five
questions were about the social tie with the manager. Two of them concerning Facebook relationship. The social tie questions are positively related with each other. All the correlations are significant, but the correlations between the first three questions are higher (0.8397; 0.7276; 0.7823) than the
correlations of the two Facebook questions (0.5328 and lower; Appendix 2, table 1A). Thus, the first three questions may be more relevant for defining social ties, so these three questions are combined. The Facebook questions are also combined and included as a robustness check. The first three questions are combined as variable social tie and the other two questions are combined as variable Facebook connection. These two variables are scaled from 0 to 1.
The two questions concerning feedback by the manager and the feeling about this person is being your manager are positively correlated with each other (0.2773). And these two questions are also
positively correlated with the number of good answers (0.2849; 0.2100). An explanation for this correlation may be that if a participant performed well, the likeliness for feedback agreement and the acceptance of the manager increases. On the other hand, the feedback and the manager acceptance is not significantly correlated with social tie. So when the feedback and manager acceptance is high, this cannot be related by the good relationship between the manager and employee and the employee. These two variables, Facebook connection and manager acceptance are also scaled from 0 to 1. There is a strong variance in the time participants spend on the survey. There are three participants that spend more time on the survey then most participants (1h45; 12h48; 2h21). The rest of the participants spend under the 15min to finish the survey, so these are outliers. It could be that subjects spending more time on the task influences performance positively. More time can be interpreted as more effort. With the outliers included the correlation is negative (-0.1482), which means that spending more time on the task has a negative influence on performance. If the outliers are excluded the correlation is positive but very small (0.0056; Appendix 2, table 2A). However, both correlations between the number of good answers (NGA) and the time spent on the task are not significant. Besides this
correlation, time spent on the task could be influenced by the strength of the social tie. This correlation is 0.1126 and not significant. Both correlations are not significant and therefore not included in testing the hypotheses.
Furthermore, difficulty and number of good answers are negatively correlated (-0.2055; Appendix2, table 3A). When the subjects indicate the task as more difficult the number of good answers
diminishes. The correlation is not as strong as expected. Difficulty is scaled from 1 till 7.
5.2 Mean estimations
In table 3 the mean and standard deviation is listed of all the variables that are relevant.
Table 3: Mean estimations
In the survey the participants were asked to solve thirteen anagrams. Furthermore, some questions were asked before about the social tie which I combined to get the variables Social tie and Facebook connection, as explained in section 5.1. After the participants solved the anagrams some questions were asked about the manager acceptance, feedback and difficulty of the anagrams (see Appendix 1). In table 3 the means of the answers are listed. The mean of the number of good answers in the three treatments is diminishing from treatment one to three. In treatment one the manager is paid when the employee had 8 or more good answers. The mean is 8.114, so on average the manager had a payoff. In treatment three, when the answers are totally anonymous and the payment of the employee does not depend on the performance, the mean is lower and the standard deviation is bigger.
In appendix 2, table 4A, a histogram shows the distribution of the number of good answers (NGA). The minimum value of NGA is one and the maximum value thirteen good answers. The distribution is close to normal, hence more right tailed. This explains the mean in table 3, which is closer to the maximum value. Table 5A, appendix 2, shows the number of good answers per treatment in a scatterplot. This scatterplot shows that the variance in treatment two and three is higher than in treatment one. The number of good answers in treatment one are more concentrated around the mean.
Variable Treatment Manager Mean Standard deviation
NGA 1 8.114 0.354 2 7.969 0.463 3 7.839 0.506 Social tie 1,2,3 1,2,3,4,5,6 0.586 0.260 1 1,2,3 0.526 0.282 1 0.470 0.192 2 0.722 0.291 3 0.416 0.297 2 4 0.605 0.224 3 5,6 0.631 0.265 5 0.544 0.234 6 0.790 0.254 Facebook connection 0.292 0.144 Manager acceptance 0.701 0.219 Feedback 1,2 0.963 0.198 Difficulty 4.980 0.962 Time (MM:SS) 1,2,3 07:40 02:34 1 07:30 02:20 2 07:17 02:25 3 08:16 02:58 11
This is seen in table 3, the standard deviation increases from treatment one to three.
The social tie mean is 0.586, and the standard deviation is 0.260. Indeed, the social tie between the managers and employees in this experiment is higher than the median. The social tie per treatment is also listed in table 3 and the social tie under the subjects in treatment three is the highest (0.631). The lowest mean of 0.526 is in treatment one, so in treatment three the subjects have stronger social ties with the managers than in treatment one. The standard deviations of social tie per treatment are close to each other (0.282;0.224;0.265). Per manager, the differences between social ties are much bigger. The mean of 0.416 of manager 3 in treatment one is the lowest, and the highest mean is social tie is in treatment 3 with manager 6 (0.790). So treatment three ends up with much stronger social ties than treatment one.
The mean of the Facebook connection (0.292) is lower than the social tie mean. This difference in means, as well as the relatively low correlation between the two measures, explains that it is not straightforward that Facebook activities between good friends are high.
Furthermore, the mean of the difficulty (4.980) shows that on average, the participants experienced the difficulty of the anagrams as a bit difficult. With a standard deviation of 0.962, the deviation from a bit difficult is not extreme. The anagrams were of a good level to test performance.
Lastly, the mean of time spent on the survey (07:40) is higher than the maximum time subjects could spent on the task. The participants in treatment three spent the most time on the survey on average (08:16). Treatment two has the lowest mean (07:17). On average the participants spent enough time to solve the anagrams.
6.
Results
After analyzing the data in de latter section, the hypotheses will be test in this section. To test the hypotheses a two sample t-test, a pairwise comparison of mean test and OLS-regression are used. All the tests are based on a 90% confidence level. I use two outcome measures, number of good answers and time spent on the survey. Number of good answers as a measure of productivity and time spent on the survey as a proxy for effort. This is a noisy proxy, but assume that if the subjects spent more time on the complete survey, they spent more time on the task as well.
6.1 T-tests
In order to test the first hypothesis, all treatments are combined. A two sample t-test is used to test whether employees productivity is higher when the social tie with the manager is stronger. Two groups are made to do the two sample t-test. I split social ties in weak social tie and strong social tie. Social tie is scaled from 0 till 1, and social tie is “weak” when the value is below 0.5 and “strong” when the value is above or equal to 0.5. This results in a mean of 8.049 when social tie is weak and 7.930 when social tie is strong. The difference of 0.119 is small, but the performance is worse when the social tie is strong. In addition, the result of the t-test is a t-value of 0.2320 with a p-value of
0.5915. The null-hypothesis, no difference in productivity between weak or strong social ties cannot be rejected. There is not enough statistical evidence to accept hypothesis 1 (Employee performs better when the social tie between a manager and employee is strong).
Besides all treatments combined, a second test is conducted to see the effect of social ties within the treatments. In treatment one as well as in treatment three the performance is worse when the social tie is strong. In the second treatment the performance is higher with strong social ties. However, in this treatment there are a lot more subjects having a strong social tie compared to subjects that have a weak social tie with the manager. The difference in performance within the treatments is bigger than
combined (0.211; 0.575; 0.543). A sample t-test is used to test if the number of good answers are significantly different by the employees who have a strong social tie with the manager, within the treatments. These p-values, treatment one (0.6139), treatment two (0.2927) and treatment three (0.6974), are not high enough to reject the null-hypothesis. There is no significant difference of productivity of employees through social ties within treatments.
To test the second and third hypotheses, the number of good answers across treatments are compared with a pairwise comparison of mean test. The mean of number of good answers per treatment is compared to the mean of the other treatments. Comparing number of good answers between treatment one and two, hypothesis 2 is tested (Employee performs better if the payoff of the manager depends on his/her productivity). The mean of number of good answers in treatment one is little higher compared to treatment two. The comparison between treatment one and two result in a t-value of -0.24 with a p-value of 0.813. This p-p-value is too high to reject the null-hypothesis. There is no significant difference in productivity by the employee when the payoff of the manager depends on number of good answers. Number of good answers compared between treatment two and three give clarity to hypotheses 3 (Employee performs better when his/her performance can be seen by the manager). The mean of the number of good answers in treatment two is negligibly higher. Between these two treatments the t-test shows a t-value of -0.21 with a p-value of 0.838. This p-value is higher than 0.05, so there is no statistical support to reject the null-hypothesis, no difference between number of good answers between treatment two and three. Hence, there is no support that productivity is higher when the answers are seen by the manager.
In order to test the hypotheses with the outcome measure time spent on the survey, I use the same tests. This outcome measure is more noisy, because the time spent on the survey is not the same as time spent on the anagrams. But we assume that subjects spending more time on the complete survey have spent more time on solving the anagrams as well. First, for all treatments combined a t-test is used to test whether effort is higher when the social tie with the manager is strong. This results in a difference of 37 seconds, where the mean of strong social ties is higher (07:56). However, the difference is not significant with a t-value of -1.1733 and a p-value of 0.2437. The null-hypothesis, there is no difference in effort between weak and strong social ties, cannot be rejected.
To show the effect of effort within the treatments I use a t-test. The mean of time spent on the survey
in treatment one and three are higher when social ties are strong (1:00; 1:36). In treatment two the time spent on the survey is 40 seconds lower when social ties are strong. These differences are higher than the treatments combined but still negligible. The p-values, treatment one (0.2088), treatment two (0.4834) and treatment three (0.1508) are not high enough to reject the null-hypothesis. There is no significant difference in effort of the employees through social ties within treatments.
Furthermore, a pairwise comparison of mean test is used to test whether effort is higher when comparing the mean of the treatments. The comparison between treatment one and two shows a t-value of -0.35 with a p-t-value of 0.726. The null-hypothesis cannot be rejected, so there is no significant difference in effort level of the employee when the managers payoff is dependent on performance.
Comparing the effort level between treatment two and three a t-test shows a t-value of 1.48 with a p-value of 0.143. There is no significant difference in effort between treatment two and three. Thus, there is no support that effort is diminishing when the answers are not seen by the manager.
6.2 OLS-regression
Although the t-tests were not supportive for my hypotheses, I want to show the effect of the variables on productivity (number of good answers). I include all the questions that were correlated with the number of good answers. First, I use an univariate analysis on number of good answers and then I use a regression model. Besides showing the effect on performance, time spend on the survey can be used as a proxy of effort. I will use a univariate analysis and a regression model to show the effect of the variables on effort.
The results of the first univariate analysis are shown in table 4.
NGA NGA NGA NGA NGA NGA
Constant 8.011174 10.56585 6.305711 5.902142 7.83871 8.642857 (0.000)* (0.000)* (0.000)* (0.000)* (0.000)* (0.000)* social tie -0.03682 (0.971) difficulty -0.51113 (0.045)* manager acceptance 2.387281 (0.038)* feedback 3.225657 (0.020)* treatment1 0.275576 (0.658) treatment2 0.13004 (0.838) manager1 -1.09740 (0.286) manager2 -0.64286 (0.542) manager3 -0.67411 (0.410) manager4 -0.79286 (0.373) manager5 -0.82468 14
(0.422)
R-squared (0.0000) (0.0422) (0.0441) (0.0811) (0.0021) (0.0147)
* denote statistical significant difference at 10% level.
Table 4: Univariate regression on number of good answers
In the regressions in table 4 only two variables are significantly different from zero. Manager
acceptance has a positive coefficient, which means that if manager acceptance increases the number of good answers also increase. Further, the acceptance of giving feedback by the manager is positively related to number of good answers. So, when the employee has a better feeling about the manager and about getting feedback of the manager, number of good answers increase significantly. And the other way around, when subjects performed better they filled in a higher manager acceptance. Better performance increases manager acceptance. When people are proud of their performance they may feel better about getting feedback.
In table 5, I estimate a model including all possible control variables. The dependent variable is number of good answers and the independent variables are social tie, difficulty, manager acceptance and the different managers. Feedback is excluded because of multicollinearity with manager 4 and 5. Between the dummy variables treatments and managers is also multicollinearity. This means that managers and treatments are highly correlated and essentially they convey the same information. Indeed, I only include the treatments in the regression. For the robustness check I include the variable Facebook connection instead of social tie. This lead to the following OLS-regression:
NGA = β0+ β1*socialtie + β2* difficulty + β3*manageracceptance + β4*treatment1 +
β5*treatment2 +µ NGA NGA Constant 9.361243 9.409028 (0.000)* (0.000)* social tie -0.2804523 (0.774) Facebook connection -0.8095577 (0.677) difficulty -0.4917937 -0.4988438 (0.062)* (0.059)* manager acceptance 1.834087 1.909138 (0.127) (0.117) treatment1 0.2447853 0.2613045 (0.694) (0.671) treatment2 -0.3269476 -0.3272914 (0.603) (0.602) R-squared (0.0746) (0.0760)
* denote statistical significant difference at 10% level.
Table 5: Regression on number of good answers
The first regression (Table 5) shows that the coefficient of social tie is not significantly different from zero. The coefficients of the different treatments are all not significantly different from zero. So, social tie and the different treatments have no significant effect on productivity. Table 5 shows that the coefficient of the variable difficulty is significantly different from zero. So this variable explains part of the number of good answers. Table 4 already showed the negative effect of difficulty on number of good answers.
In the second regression the variable social tie is exchanged for Facebook connection as a Robustness check. This coefficient is as well as social tie not significantly different from zero.
These results are supportive for the t-tests, not rejecting the null-hypothesis. This is not in accordance to the three hypotheses in this paper.
To show the effect of the variables on effort I repeat the analyses I used before. In table 6 a univariate analysis is shown. The effect of the variables individually are not significantly different from zero, except the effect of manager 4. The coefficient of manager 4 is positive, which means that when this subject is being the manager effort significantly increases.
Time Time Time Time Time Time Time Constant 7.029317 6.995595 8.122824 7.938031 7.625987 8.275862 7.357143 (0.000)* (0.000)* (0.000)* (0.000)* (0.000)* (0.000)* (0.000)* social tie 1.140562 (0.282) difficulty 0.1409692 (0.624) manager acceptance -0.6382003 (0.602) feedback -0.6681659 (0.651) NGA 0.0059779 (0.957) treatment1 -0.7615764 (0.243) treatment2 -0.9855395 (0.143) manager1 0.0064935 (0.995) manager2 0.5428571 (0.612) manager3 -0.0668203 (0.936) manager4 1.537594 (0.093)* manager5 -0.2571492 (0.810) R-squared 0.0127 0.0027 0.0029 0.0033 0.0000 0.0254 0.0622
* denote statistical significant difference at 10% level.
Table 6: Univariate regression on time
Now I will include all possible control variables to estimate a model. The dependent variable is the time spend on the survey and the independent variables are social tie, difficulty, manager acceptance,
number of good answers and the treatments. Equally to the above regression model (Table 5), feedback and the managers are excluded because of multicollinearity. And the variable Facebook connection is included as a robustness check. This lead to the following OLS-regression:
Time = β0+ β1*socialtie + β2* difficulty + β3*manageracceptance + β4*NGA + β5treatment1
+ β6*treatment2 +µ Time Time Constant 6.674953 6.80385 (0.005)* (0.003)* social tie 1.080409 (0.332) Facebook connection 2.233299 (0.264) difficulty 0.1695235 0.1701916 (0.585) (0.583) manager acceptance -0.827833 -0.9390552 (0.547) (0.497) NGA 0.0793997 0.0716341 (0.535) (0.574) treatment1 -0.621253 -0.6754792 (0.337) (0.332) treatment2 -0.797482 -0.778784 (0.264) (0.275) R-squared 0.0404 0.0439
* denote statistical significant difference at 10% level.
Table 7: Regression on time
None of the coefficients in table 7 are significantly different from zero. Neither the variables in the first regression nor the variables in the second regression have a significant effect on effort. These results are supportive for the t-tests, not rejecting the null-hypotheses.
7.
Conclusion and Discussion
This study attempted to answer the following question: ‘What is the effect of social ties between a manager and employee on productivity?’. To answer this question three hypotheses were formulated, an employee performs better when the social tie between a manager and employee is strong, employee performs better if the payoff of the manager depends on his/her productivity, and employee performs better when his/her performance can be seen by the manager. An experiment with three different treatments was conducted to test these hypotheses. For all three hypotheses I do not find enough statistical support. Besides productivity this study tried to find an effect of social ties on effort. Moreover, for this effect is not found enough statistical support. Even though several studies showed effects of social ties and behavioural motives, this study is not supportive for both effects. Nor does it show the effect of social ties on productivity via the channels altruism and image concern.
An explanation for the results can be the difference in real experiments and internet experiments. In this paper, the experiment is conducted via Facebook which is not comparable to real experiments. Anderhuba et al. (2001) shows that exerting control via Facebook is not optimal. There is no
experimenter that can provide help if needed, and there is no control that ensures serious participation. Furthermore, the task of the manager, supervision, may have a stronger effect in real interaction. In a real experiment the participants may take the time to complete the task as good as possible in six minutes. In this survey the participants participated on the computer when doing other things. It may have not felt real to imagine a manager.
The first intention of this paper was to conduct the experiment on a high school were the participants are more controllable and from the same age and education level. The participants via Facebook were not of the same age and education level. This may have an effect on the productivity per participants and on the average performance. For further research I recommend to do the same experiment with a more controllable group and participants with the same characteristics concerning age and education. Fehr et al. (1998) showed that subjects deviate from money maximization, through the existence of pro social behaviour. The fact that all the participants participated voluntary and deviated from money maximization supports the paper of Fehr at al. (1998). Although it is not supportive for the hypotheses of this paper, it gives evidence for pro social behaviour.
8.
Appendices
Appendix 1; Experimental form
Treatment 1
Welkom en alvast dank voor het meewerken aan mijn scriptie! Deze informatie word alleen gebruikt voor scriptie doeleinden.
Voor het doel van deze survey zijn er een manager en een werknemer. Beeld in dat Josien de Heul uw manager is en dat u haar werknemer bent. Josien deelt een taak aan u uit en u wordt gevraagd deze uit te voeren. Josien houdt toezicht gedurende de uitvoering. Voor het uitvoeren van de taak wordt u gevraagd eerst een paar vragen te beantwoorden. De vragen en antwoorden voor en na het uitvoeren
van de taak zijn anoniem. Deze informatie word dus niet gedeeld met Josien.
Na het beantwoorden van de vragen zal een uitleg volgen omtrent de taak. Hoe is jou relatie tot deze Josien?
o Geen bekende o Verre bekende o Bekende o Goede bekende o Vriend/familie o Goede vriend/familie
Hoe vaak heb je contact met Josien?+ o Nooit
o Soms o Vaak
Hoe vaak sms/bel je met Josien? o Dagelijks
o een paar keer per week o 3-5 keer per maand o 1-2 keer per maand
o minder dan 1 keer per maand
Hoe vaak like of reageer je op een foto of post van Josien?
o Dagelijks
o een paar keer per week o 3-5 keer per maand o 1-2 keer per maand
o minder dan 1 keer per maand
Hoe vaak stuur je een bericht op Facebook naar Josien? o Dagelijks
o een paar keer per week o 3-5 keer per maand o 1-2 keer per maand
o minder dan 1 keer per maand
U wordt gevraagd 13 anagrammen op te lossen. Hiervoor heeft u 6 minuten de tijd. Als u naar de volgende pagina klikt gaat de tijd automatisch lopen.
Een anagram is een woord, waarvan de letters gehusseld staan. De begin letter blijft hetzelfde en u verschuift de letters tot u op het bestaande woord komt. Een voorbeeld:
B G I J P L I K E R E J Antwoord:
BEGRIJPELIJK
Josien kan punten verdienen. Deze punten zijn afhankelijk van uw prestatie. Als u 8 of meer goede antwoorden geeft ontvangt Josien 10 punten (1punt = €0.50).
De manager kijkt mee met het uitvoeren van de taak. Na het uitvoeren van de taak zal de manager uw antwoorden nakijken en een bericht sturen met de goede antwoorden en het aantal punten dat hij heeft behaald. Na het experiment worden 5 deelnemers random gekozen. De punten die Josien bij
deze deelnemers heeft ontvangen worden opgeteld en omgerekend naar euro's. Dit bedrag krijgt Josien uitbetaald.
Anagrammen oplossen 1. S E L P E N
2. L S U I E R T
3. A N B O E E N 4. O L E M T E 5. D I N R K N E 6. R Z A N D E 7. B Z O D I N E R J 8. Y D I L H C L S I 9. V G E L O 10. K N N E K O E A P 11. B K E D N E 12. S H E I N E L D 13. C P U C I N C A O P 21
Hoe moeilijk/makkelijk vond je deze taak? o Extreem makkelijk
o Makkelijk
o Een beetje makkelijk o Neutraal
o Een beetje moeilijk o Moeilijk
o Extreem moeilijk
Hoe vind je het dat Josien jou manager is? o Helemaal niet leuk
o Niet leuk o Matig o Neutraal o Een beetje leuk o Leuk
o Heel leuk
Wat vind je ervan dat Josien jouw antwoorden te zien krijgt en feedback geeft? o Helemaal niet leuk
o Niet leuk o Matig o Neutraal o Een beetje leuk o Leuk
o Heel leuk
Vul hieronder uw naam in, zodat de manager u een bericht kan sturen met het aantal juiste antwoorden. ………..
Hoe vond u de uitleg in deze survey? o Extreem onduidelijk o o o o o 22
o o o
o Extreem duidelijk
Appendix 2; Descriptives
Table 1A. Correlation between social tie questions
Social tie questions Q1 Q2 Q3 Q4 Q5 Q1 1.0000 Q2 0.8397 1.0000 Q3 0.7276 0.7823 1.0000 Q4 0.5127 0.5110 0.4843 1.0000 Q5 0.3251 0.3315 0.3436 0.5328 1.0000
Table 2A. Correlation between time and number of good answers
Time and number of good answers Time NGA Time 1.0000 NGA Outliers Incl. -0.1482 1.0000 Outliers Excl. 0.0056
Table 3A. Correlation between difficulty and number of good answers
Number of good answers and
NGA Difficulty
difficulty
NGA 1.0000
Difficulty -0.2055 1.0000
Table 4A. Histogram: Frequency of number of good answers
Table 5A. Scatterplot: Frequency of number of good answers per treatment
0 5 10 15 F req ue nc y 0 5 10 15 NGA 0 5 10 15 NG A 1 1.5 2 2.5 3 treatment 24
9.
References
Anderhuba, V., R. Müllerb, C. Schmidtc (2001). “Design and evaluation of an economic experiment via the Internet”, Journal of Economic Behavior & Organization, Vol. 46, 227–247. Ariely, D., A. Bracha and S. Meier (2009). “Doing Good or Doing Well? Image Motivation and Monetary Incentives in Behaving Prosocially”, The American Economic Review, Vol. 99, No. 1, pp. 544-555.
Bandiera, O., I. Barankay and I. Rasul (2009). “Social connections and incentives in the workplace: Evidence from personal data”, Econometrica, Vol. 77, 1047–1094.
Bénabou, R., and J. Tirole (2006). “Incentives and Prosocial Behavior”, American Economic Review, 96(5): 1652-78.
Cassidy, C.M., R. Kreitner (2011). Principles of management, 12th edition. South-Western Cengage Learning.
Dirks, Kurt T., Donald L. Ferrin, (2001). “The Role of Trust in Organizational Settings”, Organization Science, 12(4):450-467.
Falk, A., and M. Kosfeld (2006). “The hidden costs of control”, The American Economic Review, Vol. 96, pp. 1611-1630.
Fehr, A., E. Kirchler, A. Weichbold, and S. Gächter (1998). “When Social Norms Overpower Competition: Gift Exchange in Experimental Labor Markets”, Journal of labor economics. Vol. 16, No. 2, pp. 324-351.
Fehr, A., K.M. Schimdt (2000). “Fairness, incentives and contractual choises”, European Economic Review, Volume 44, Issues 4–6, May 2000, Pages 1057–1068
Frey, B. S., and F. Oberholzer-Gee, (1997). “The Cost of Price Incentives: An Empirical Analysis of Motivation Crowding-out”, American Economic Review, Vol. 87, No. 4, pp. 746-755.
Garen, J.E., (1994). “Executive Compensation and Principal-Agent Theory”, Journal of Political Economy, Vol. 102, No. 6, pp. 1175-1199.
Gneezy, U., and A. Rustichini (2000). “Pay Enough, or Don’t Pay at All”, Quarterly Journal of Economics, 115(3), pp. 791–810.
Grossman, P., and C.C. Eckel (1994). “Altruism in Anonymous Dictator Games”. Games and Economic Behavior. Volume 16, Issue 2, October 1996, Pages 181–191.
Grosse, N.D., and G. Riener, (2010). Explaining gender differences in competitiveness: gender-task stereotypes. Work. Pap., Friedrich-Schiller-University Jena, Germany.
Hoffman, E., K. McCabe, K. Shachat and V. Smith. (1996). “Social Distance and Other-Regarding Behavior in Dictator Games,” American Economic Review 86, 653–660.
Lazear, E.P., (1986). “Salaries and Piece Rates”, The Journal of Business, Vol. 59, No. 3, pp. 405-431. Prendergast, C. (1999). “The Provision of Incentives in Firms”, Journal of Economic Literature, Vol. 37, No. 1, pp. 7-63.
