The influence of a positive message on self-efficacy and performance

(1)

The influence of a positive message on

self-efficacy and performance

What is the influence of a positive message on self-efficacy and performance?

Shirley Goossens (11417420)

Masterthesis Business Economics – track Managerial Economics & Strategy 15 ECTS

(2)

2

Statement of Originality

This document is written by student Goossens, Shirley who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3 Abstract

A high self-efficacy leads to a greater performance (Collins, 1982). Especially among students higher self-efficacy results often in a better academic achievement (Lent et al., 1984). One can increase self-efficacy by giving a compliment, also when this compliment is not based on facts (Schunk, 1982; McAuley et al, 1999). This paper investigates whether a positive message increases self-efficacy. If this is the case then this enhanced self-efficacy can indirectly have an influence on grades, which will be tested.

This study is unique in a sense that the self-efficacy of students is enhanced by a tutor. Students are randomly assigned to a treatment and control group. All students receive an email in which the tutor wishes them good luck on the exam. However, the students in the treatment group receive an email indicating that the tutor is completely confident that he/she will pass the exam, while this specific information is not included in the email sent to the control group. The results of this experiment indicate treatment improves self-efficacy with marginal significance. However, when including variables about GPA, self-esteem and attitude towards the course, the influence of treatment on reported confidence of passing the exam is not significant anymore. The treatment in which self-efficacy should be positively influenced, did not result in different grades compared to the control group.

(4)

4

1. Introduction . Companies want their employees to generate the greatest output possible. In order for the employees to perform to their fullest capacity, they have to be emotionally stable and have the right attitude. Self-efficacy is a great influencer of performance, according to Lane et al. (2004). This is also proven by Fetz et al. (1988) who argue that people with high self-efficacy in sports perform better than their peers who have a lower level of self-self-efficacy. Even though self-efficacy itself has been greatly researched, the impact of a positive message on self-efficacy and performance has not been researched extensively. There are some informative studies concerning this subject. Alias et al. (2009) researched the impact of giving a positive influencer on the performance. He found that this improved performance. In addition, Schunk (1982) found that a compliment to enhance efficacy influences performance positively. It is interesting to find out whether this result also applies in different cases, since there are only few studies who did research in this topic.

This study will investigate whether a positive message has influence on self-efficacy and performance of students. This is a contribution to science because the influence of a positive message has never been studied in such a specific way, where a tutor gives a positive message to his or her apprentice. Companies can use the outcome of this paper to enhance efficacy of their employees. Furthermore, teachers can use the result to enhance self-efficacy of their students. Enhanced self-self-efficacy can then indirectly improve performance (Fetz et al., 1988).

This research is done by use of an experiment. The experiment is conducted amongst students who attended an exam training organized by a company. The exam training is focused on increasing knowledge and practicing difficult questions on a course in their bachelor. The students will receive a personalized email from their tutor in which he/she wishes them the best of luck at their exam. However, there will be two different versions of the email. One of these versions is randomly assigned to the student. The treatment group will receive a positive message from their tutor that is supposed to enhance self-efficacy and indirectly performance, while the control group receives an email without this positive message. It can be the case that this message has no effect, a positive effect or even a negative effect. The research question is:

(7)

7

The results of this study are that treatment improves self-efficacy with marginal significance. However, when including variables about GPA, self-esteem and attitude towards the course, the influence of treatment on reported confidence of passing the exam is not significant anymore. For people who are overconfident treatment influences self-efficacy significantly. The treatment in which self-efficacy should be positively influenced, did not result in different grades compared to the control group. Furthermore, it was found that the attitude towards a course is positively related to performance.

The thesis is based on different sections. Section 2 presents the literature, section 3 elaborates on different factors that are of influence on performance. Then section 4 describes the methodology, including the survey and the set-up of the experiment. Section 5 shows the analysis and in section 6 the discussion is given. Lastly, section 7 is about the conclusion.

(8)

8 2. Literature review

In this review self-perception will be explained in order to find out how different concepts about the self are related. This is helpful in understanding the core of self-efficacy and how it can be influenced and enhanced. Furthermore, this will serve as the basis for finding out in what way this can influence performance and to gain knowledge about previous studies.

2.1 Self-concept .

The view about the self all starts with self-concept. Self-concept is how a person sees him- or herself (Shavelson, Hubner & Stanton, 1976). It is defined as the thoughts and feelings a person has, referring to oneself as an object (Rosenberg, 1979). It can be influenced by environmental experiences or by good friends and family (Kelley, 1973). The perception of the self influences the way in which one acts. As a result, these actions will then affect the way in which this person perceives him/herself.

Researchers have different views on concept. Some of them argue that self-concept is a hierarchical category structure that is based on traits, values and memories of specific behaviors (Carver & Scheier, 1981; Rogers 1981). Others claim that self-concept is a system of self-schemas about the self, derived from previous social experiences (Neisser, 1976; Markus & Sentis, 1982). What these researchers agree upon in self-concept, is that it is active and changeable.

2.2 Self-esteem .

Another important element of how we see ourselves is self-esteem. Self-esteem is defined as a person’s appraisal of his value. It concerns how we feel about ourselves (Scheff, Retzinger & Ryan, 1989). It is an internal experience, a cognitive and emotional evaluation of the self and of our own value. Our relation to ourselves, to others and to the world is what it is focused on (Burton, 2015; Florence, 2015). Furthermore, it refers to the extent to which we accept ourselves (McLeod, 2008), in an overall way (Roberts, 2012). Self-esteem can be either global or domain-specific. The global self-esteem is a value judgement about the self, while the domain-specific self-esteem is specified as a value in a particular area (Leary & Baumeister, 1998).

Previous studies from Maruyama et al. (1981) and Baumeister et al. (2003) have investigated the influence of self-esteem on school performance. It was found that the level of

(9)

9

self-esteem has an impact on school performance. A distinction was made between a high and a low self-esteem. A high self-esteem implies that the person has a positive evaluation of the self, while a low self-esteem implies a low evaluation of the self. The experiment was set-up with random sampling to collect data by means of a questionnaire amongst 50 male and 50 female students. The Coopersmith’s questionnaire was used to measure self-esteem and academic achievement was measured by the average grade. The conclusion of this study was that there is a significant positive influence of self-esteem on academic performance (Aryana, 2010). This conclusion seems to be in line with other literature, however the sample of this study is rather small, which decreases the validity.

Another study examined the relationship between how one perceives oneself and one’s ability. It consists of a meta-analysis which is based on 128 studies, with a total sample of 202,823 participants. A significantly positive relationship between self-esteem and academic performance was found with a correlation of in between .21 and .26 (Hansford & Hattie’s, 1982). This study is very strong in a sense that it is based on a large amount of data. Unfortunately only 4 to 7 percent of the variance can be explained, which is very low.

Self-esteem is often mixed up with self-concept. This is due to the fact that there is a lot of overlap in the definition of these two concepts. The main similarity is that they are reflective processes. A reflective process is the ability to reflect on your own actions, which allows for continuous learning. However, there is also a large difference between self-concept and self-esteem. Self-concept is based on facts you know about yourself, while self-esteem is more about your feelings, considering the things that you know (Burton, 2015).

2.3 Self-confidence .

Self-confidence is about having trust in oneself and more specifically about having trust in your own ability to engage successfully or decently with the world. Someone who is self-confident is up to new challenges, opportunities and takes more responsibility (Burton, 2015). Confidence can be considered as an external experience, it places ourselves in the external world, in our relationship with others and how we deal with certain situations (Florence, 2015). The level of confidence one has, has an influence on how well this person performs in general. It is argued that people who are confident have more successful outcomes, while the ones who are not confident experience more failures (Bandura, 1986). Therefore this may indicate that the level of self-confidence is positively related to performance.

(10)

10

2.3.1 Overconfidence and underconfidence . It is not always the case that people with a high level of self-confidence perform better. Especially when this high level of self-confidence turns into overconfidence, the reverse may occur. Overconfidence appears when the confidence that people have exceeds their accuracy (Klayman, Soll, González-Vallejo & Barlas, 1999). This happens when people have strong beliefs that they can accomplish a specific task, while in fact they cannot (Moore & Healy, 2008).

A field study was conducted by Moores & Chang (2009) among 108 students to find out the relationship between (over)confidence, self-efficacy and performance. Overconfidence was measured as the difference between one’s actual performance and one’s expected performance. The performance of overconfident and underconfident participants was compared, and it showed differences. People who were overconfident were found to have a negative relation between self-efficacy and performance. This implies that a higher level of self-efficacy results in a lower performance. For the underconfident group, this result did not hold. This group had a positive relation between self-efficacy and performance. Therefore people with overconfidence and a high self-efficacy are expected to have a lower performance. This can be explained by the fact that overconfident people are less precise in performing a task, which leads to more mistakes and a worse performance.

On the contrary, according to Lundberg (2008), underconfidence could result in a lower performance. It is claimed that underconfident students may defer improvement opportunities and they may restrict themselves from asking questions during lectures. In addition, overconfidence and underconfidence is dependent on the performance of students. It was researched that those who belong to the lowest performers of a group tend to overestimate the most, while students with the highest performance of the group tend to underestimate the most. This estimation of performance by students happened immediately after finishing a test (Clayson, 2005). An explanation for this conclusion may be that incompetent students attribute their performance to external factors (Moreland, Miller & Laucka, 1981). Furthermore, it could be explained by the level of expectations that students have. Students with higher capacities may have high expectations, which are not easily met. The students with lower capacities have low expectations that are met more easily, which results in an overestimation of performance (Millea & Grimes, 2002). Whether students will be overconfident or underconfident also depends on the material that has to be studied for the

(11)

11

exam. When the material is perceived as not so challenging, overconfidence may occur. However, when the material is perceived as difficult, underconfidence will appear more easily (Klayman, Soll, González-Vallejo & Barlas, 1999). All-in-all being overconfident or underconfident can have a positive as well as a negative impact on performance.

2.4 Self-efficacy .

Self-efficacy is someone’s estimation of its fundamental ability to cope, perform and be successful (Judge & Bono, 2001). It influences how one feels, thinks, motivates oneself and how one behaves. Besides that, it is situational-specific (Murphy, Coover & Owen, 1988) and changes over periods of time (Gist & Mitchell, 1992). The four sources on which it is based are vicarious experiences, persuasive statements, psychological states and performance accomplishments.

Perceived self-efficacy is the faith that a person has that the wanted outcome can be achieved. It has a significantly strong positive relation with performance (Carmona, Buunk, Dijkstra & Peiro, 2008) and it has an influence on task choice, effort, persistence, resilience, and achievement (Bandura, 1997; Schunk, 1995).

The strong influence of self-efficacy on performance has been investigated in many studies. In one paper it is found that children with high self-efficacy finished more exercises accurately and they corrected more exercises that they failed to finish properly, compared to children with a lower self-efficacy, regardless of the ability level (Collins, 1982). Another study claims that there is a higher persistence and accuracy when increasing self-efficacy of children, which leads to an improved skill of children. This was investigated among 40 children in the age of 7 to 10. Before attending the experiment, they did a test about their ability of solving problems and they filled out a survey with which their self-efficacy was measured. Afterwards, the children had to solve exercises that could develop competencies. There were four treatment groups, including past attribution, future attribution, monitoring and training control. Past attribution implied a treatment in which feedback was given based on facts. Future attribution focused on the value of future effort. Monitoring meant that the students did not get any comments. The last treatment is training control, where the group served as a control group. It was found that attributional feedback improves self-efficacy. This group had the highest improved achievements. Therefore, it can be concluded that enhancing self-efficacy with attributional feedback results in the highest improvement of performance (Schunk, 1982).

(12)

12

Furthermore, self-efficacy is considered to be active, which makes it subject to change. Sending persuasive information with which people are told that they have the skills to perform a specific task can change the level of self-efficacy. When apprising someone “You can do this”, his or her self-efficacy increases, but this is only for a short period of time. It ends when the subsequent efforts turn out to be disappointing (Schunk, 1991). Furthermore, this was tested by McAuley et al. (1999). They conducted an experiment about self-efficacy for women about their physical ability. Women were asked to rate their confidence of being able to complete the task. Then participants were given high-efficacy or low-efficacy feedback, by stating that they performed at the top of their group (high-efficacy) or at the bottom of their group (low-efficacy). This feedback was purely random and not based on performance. It was found that performance was better for the group receiving high-efficacy feedback, which indicates that self-efficacy can be influenced.

It can be concluded that the way in which one views oneself has a major impact on performance. Of the different concepts in which we view ourselves, self-efficacy is one of the greatest drivers of performance.

2.4.1 Self-efficacy and self-confidence

Self-confidence is closely related to self-efficacy. However, the main difference between these two concepts is that self-confidence concerns how you feel about yourself in an overall way (Roberts, 2012), while self-efficacy is about how faithful you are that you are able to achieve a specific outcome (Gist & Mitchell, 1992). Hence, self-efficacy is more focused on one task and on how someone estimates its own ability to perform this task well. In this paper self-efficacy is similar to how confident people are, since confidence is asked for in one specific task. Therefore, the term confidence is used. Furthermore, the definition of confidence is expected to be easier to understand, because participants are more familiar with the term confidence than with the term self-efficacy.

(13)

13

3. Other factors of influence on performance . Besides self-efficacy, there are several factors of influence that can either positively or negatively affect performance, or in this case grades. These elements are discussed in this section and serve as the foundation for the survey.

3.1 Ability . First of all the ability of students has an influence on performance. Previous research by Gustafsson et al. (1996) shows that general mental ability has a great influence on school performance. It was also concluded that students’ ability and previous academic performance are positive predictors of success in college. Students’ ability was measured in terms of SAT or ACT scores, while previous performance was based on using high-school grades (Harackiewicz, Tauer, Barron & Elliot, 2002).

When it comes to students judging themselves on their own abilities, this is far less reliable. Students judge themselves to be incapable when they study hard for an exam, but fail anyways (Covington, 1992). When this situation occurs, they tend to use the discounting principle, in which they discount low ability as a cause of failure, even if they exerted great effort, when other factors could be in play (Kelley’s, 1973). That is why their opinion on their capabilities is considered unreliable. Besides, Spinath et al. (2006) show that own ability self-perceptions seem to have no influence on later achievement. These self-self-perceptions are claimed to be a result of prior achievements, but they seem to have no influence on later achievement.

3.2 Ability in combination with motivation .

Secondly, there is theoretical and empirical evidence that performance is influenced by ability and by motivation combined (Chan, Schmitt, Sacco, & DeShon, 1998; Chatman, 1989; Pinder, 1984). They concluded that performance, ability and motivation are interdependent. It was found that when there is a lack of ability, motivation itself will not cause high grades. On the other hand, a student with high ability and a lack of motivation will not have high grades either. Students with a high ability and a lot of motivation will excel (Nonis & Hudson, 2006).

3.3 Anxiety .

Anxiety is a factor that has a negative effect on performance. When students are not confident about their abilities to pass a course, they will be more anxious for the exam. This

(14)

14

will generally result in a lower performance (Pajares, 1996). However, the influence on performance depends on the level of anxiety. When students are very anxious, this will influence performance significantly in a negative way, while for students who are slightly anxious, the influence will be negligible (Seipp, 1991).

3.4 Time . Time is considered as a factor of influence on performance that impacts the learning of students. Students are having busy lives, which may decrease the time available that can be used for studying (Nonis & Hudson, 2006). This seems very plausible, however the evidence for the influence of study time on grades is mixed. It was found that study time has a positive influence on the performance of students (Douglas & Sulock, 1995). Another study found that in general study time had no effect on performance, but it positively affected grades for above average students and it negatively affected grades for below average students (Borg, Mason & Shapiro, 1989). Furthermore, there are studies who claim that study time does not have any influence on student performance (Lumsden & Scott, 1987; Gleason & Walstad, 1988; Park & Kerr, 1990). In addition, Didia & Hasnat (1998) found that study time is negatively related to performance. A possible explanation for this surprising conclusion is that the quality of time spent on studying is more important than the quantity of time spent. Unfortunately, the quality was not measured in this study. Therefore it is difficult to draw firm conclusions about the influence time has on grades.

3.5 Study methods and strategies . According to Okpala et al. (2000) students that have good study strategies, obtain higher grades. The strategies that students have depend on the goal they pursue. On one hand, they can have the intention to study courses in a deep way, which involves developing competencies during the course. On the other hand, they can have the intention to study the course in a superficial way, which involves only meeting course expectations. Unfortunately, it is unclear how this affects grades, since this is very specific and task-related. For example, when writing an essay, the strategy will be more deeply, while when studying for a multiple-choice exam, the approach will be more superficial (Elias, 2005).

3.6 Concentration .. Concentration is an important element in the effectiveness of studying, since it is the main study habit that positively impacts performance. Students who spend fewer time studying, but have a greater ability to concentrate, will in general have a better performance

(15)

15

compared to students who spend more time studying, but have a worse ability to concentrate Okpala et al. (2000).

3.7 Scheduling .. Another study habit that is claimed to influence grades, is the habit of scheduling. This was investigated by a study with 201 students who filled out a survey that was focused on scheduling, the ability to concentrate and access to notes (Nonis & Hudson, 2010). Furthermore they asked for the academic performance of the participants. Scheduling was found to have a small negative influence on performance, which may indicate that students who procrastinate perform better than students who divide the workload over a longer period of time. This is explained by the fact that crammers tend to study longer than people who use scheduling as a study habit. However, another study claims that scheduling has a positive influence on performance (Macan & Shahani, 1990). Scheduling reduces feelings of stress and stress often has a negative influence on performance. Therefore, this reduction of stress due to scheduling results in better performance.

3.8 Demographics .

Multiple demographic variables can have an impact on performance. One of them is gender, since according to Holmlund & Sund (2008) girls tend to outperform at school. This was also found by Maccoby & Jacklin (1974), who claim that girls tend to have better results than boys in subjects that require verbal competence. With mixed evidence, it is claimed that boys are slightly better than girls in analytic subjects (Burke, 1989). All-in-all, it can be concluded that there are differences based on gender for performance in school.

Furthermore, there is mixed evidence about the influence of age on school performance. Some studies claim that age negatively affects performance (Peiperl & Trevelyan, 1997), while other studies state that young students are more likely to have higher grades (Haist, Wilson, Elam, Blue & Fosson, 2000). These mixed conclusions can be explained by the fact that younger students may be more intelligent, for example when they skipped a class. On the other hand, older students may earn higher grades since they have more life-experience and therefore generated more knowledge in the past. Besides that, age has influence on the self-confidence people have. The older people get, the greater the bias in their confidence ratings (Stankov & Crawford, 1997).

(16)

16

In addition, it is found that socio-economic background has an impact on the performance of students. Students with a high socio-economic background tend to outperform students with a low socio-economic background (Beaumont, Walters & Soyibo, 2001).

Furthermore, there is a gap in educational performance across racial groups. In general, non-white students have lower grades than white students. This effect can be attributed to the fact that they lack material conditions that foster the development of skills, habitats and styles, which are appreciated and graded well by teachers (Ainsworth-Darnell & Downey, 1998).

3.9 Hypotheses . The hypotheses, based on literature, that will be tested by means of this experiment are:

1) Receiving a message in which the tutor tells the student that he/she believes in him/her, will positively influence the level of self-efficacy.

The expectation is that the level of self-efficacy of students who had received a positive message is higher than the students who did not receive this message. This expectation is based on the papers of Schunk (1982) and McAuley et al. (1999), in which it was concluded that a compliment enhances self-efficacy, even though this compliment was not based on facts. This enhanced self-efficacy will be measured by the grade that students give themselves for passing the exam. This grade is expected to be higher for the treatment group than for the control group.

2) High self-efficacy leads to better performance.

It is expected that higher self-efficacy leads to better performance, which in this case will result in a higher grade for the exam. This is based on previous research by Schunk (1983) and McAuley et al. (1999) who concluded that high self-efficacy improves performance.

(17)

17 4. Methodology

In this chapter the research design, the research population sample, the survey and the empirical relations are discussed. A distinction will be made between control variables and dependent variables.

4.1 Research design and population . The aim of this paper is to find out whether a positive message enhances self-efficacy of students and whether self-efficacy results in a higher performance. This idea was tested by means of an experiment. The experiment was conducted among customers of a tutoring organization. This organization provides crash courses in groups of approximately 16 students in which they get a rehearsal and practice additional exercises. The courses are given by an excellent student (tutor), who passed the course previously with an 8 or higher.

In the experiment the research population existed of 195 students who attended a crash course. They were randomly selected and assigned to either the treatment or control group. All students received an email from their tutor several hours before the exam, in which he or she wished them good luck for the exam. The treatment group received an email in which their self-efficacy was supposed to be enhanced, by including the message that the tutor is “certain that he or she will pass the exam”, while the other group received the same email without this sentence. A high self-efficacy was then supposed to positively influence grades (Collins, 1982; Schunk 1983). Besides that, in the email sent to students by the tutor they were asked to respond with their level of confidence that he or she would pass the exam on a scale of 1 to 10. Several weeks after the crash course, the students filled out an online survey using Qualtrics. This is an online tool for creating and distributing surveys. The survey could only be sent several weeks after the exam, since it takes some time till the exam grades are available.

4.2 Survey . Based on the literature review, a survey was developed. First of all, demographics like gender, age, and birthplace of parents were asked for. Secondly, questions concerning ability were included, since the influence of ability on performance is strong (Gustafsson & Undheim, 1996). Ability was measured by asking for the GPA from high-school and the current GPA at university. It is expected that a higher ability has a positive influence on performance. In this study this will most likely mean that a higher GPA resulted in a higher

(18)

18

grade for the exam. In addition students were asked which level of schooling they had at secondary school, to control for differences between students having a HAVO-diploma and students having a VWO-diploma.

Furthermore, questions to determine the dependent variables were included. These questions concerned how confident the students were right before the exam that they would pass. This question was incorporated to obtain a measure of self-efficacy. It is expected that the treatment group has a higher self-efficacy than the control group (Schunk, 1982). Furthermore, the result of a high self-efficacy is thought to be a higher grade. In addition, the participants were asked for the grade of the exam, to find out whether there was a difference in grades between treatment and control.

Moreover, questions regarding the level of self-esteem were added. With this information the general level of self-esteem can be tested to make sure that treatment and control group are balanced in terms of self-esteem. The well-known Rosenberg Self-esteem test was used as a measure of self-esteem, since this scale is internally consistent and very reliable (Albo, Núñez, Navarro & Grijalvo, 2007). Previous literature found the reliability of this scale to be 0.88 (Robins, Hendin & Trzesniewski, 1999). In the present study, the Cronbach’s alpha is 0.89, which indicates that the questions concerning self-esteem have a high reliability and consistency. Since there is a high alpha and only one single factor, the variable self-esteem is calculated by taking the average value of each question for every participant, after adjusting for reversed items.

There were also questions included concerning the attitude towards the course, which is closely related to self-efficacy (Hackett & Betz, 1989). The Attitude Towards Mathematics Inventory served as a base for these questions, since in the questions the course mathematics was changed to the course you attended an exam training for. Answers to these questions provide a measure of self-efficacy that is more stable and less influenced by treatment. This is due to the fact that it includes the attitude of the overall course, instead of only asking for the dependent variable about the confidence of passing the exam in the survey. Therefore it serves as a control variable for self-efficacy. The Cronbach’s alpha for attitude towards the course scale is 0.80. This value is very high, which indicates internal reliability and consistency. The variable was calculated by taking the average value of each participant on all items in the scale, after adjusting for reversed items.

(19)

19

Hence, there are also some influences on performance that were discussed in the literature, which are not taken into account in the survey. For example time is a variable that was discussed briefly, on which there was a lot of contradictory information. The mixed evidence makes it hard to draw firm conclusions on the influence study time has on the grades of students. This, along with the fact that students have difficulties in precisely estimating the time they spent studying for their exam, were reasons to exclude this variable from the questionnaire. Furthermore, the influence of anxiety on performance. Anxiety itself will not be taken into account as a variable in the experiment, since this is one of the four sources of self-efficacy (Bandura, 1986). If anxiety was taken into account as a separate variable, this would have caused multicollinearity. In addition, self-perceptions on ability are not taken into account in the questionnaire, since these perceptions are unreliable and do not have an impact on performance, as explained in the literature review. Furthermore the variables study method, concentration and scheduling were not included, because they were difficult to measure.

4.3 Empirical relation

In this section the empirical relation in hypothesis 1 and hypothesis 2 will be discussed.

Hypothesis 1 is about “Receiving a message in which the tutor tells the student that he/she believes in him/her, will positively influence the level of self-efficacy”. The empirical relation that is tested here is about the influence this specific email had on the variable confidence of passing the exam in the survey. The confidence of passing the exam was given a value on a scale of 1-10, filled out in the survey three weeks after doing the exam. It is expected that the positive message enhances self-efficacy (Schunk, 1982; McAuley et al., 1992)

Hypothesis 2 investigates whether “High self-efficacy leads to better performance”. The relation that is tested here is whether confidence of passing the exam in the survey, which indicates self-efficacy, is related to grades. To investigate this relationship, other variables must be taken into account as well. The GPA obtained in the bachelor is included, because it offers additional information about ability. This GPA is expected to have a positive effect on performance. Furthermore, the variable self-esteem has an influence on the relation of self-efficacy on performance. Participants with a high level of self-esteem will be likely to

(20)

20

be more positive about passing the exam, so they will score themselves higher on this variable (Korman, 1966). Therefore self-esteem and GPA have to be controlled for.

(21)

21 5. Analysis

In this section the variables of the experiment will be presented and explained. Furthermore, the results of the analysis will be explained.

5.1 Descriptive statistics

The experiment was carried out among 195 students, who all received a treatment or a control email. Unfortunately, not all of them filled out the survey that was sent to them afterwards. Table 1 shows the descriptive variables. The survey was completed by 78 students, while 18 students filled it out partly.

Table 1. Descriptive variables

Variable Observations Mean Standard dev. Minimum Maximum

Man 96 0.46 0.50 0.0 1.0 Age 78 20.04 1.83 18.0 27.0 GPAUni 96 6.85 0.47 5.5 8.3 GPASchool 96 6.99 0.67 5.5 9.0 VWO1 95 0.94 0.24 0.0 1.0 Self-esteem 78 3.66 0.59 2.3 4.6 Attitude Course 78 3.10 0.65 2.5 4.7

The participants have an age between 18 and 27, and the mean of the age is 20. The 96 participants are divided in 44 men and 52 women. The GPA that students have in their bachelor is between a 5.5 and an 8.3, with an average of 7.0. Furthermore the variable concerning the GPA in secondary school was included. The variable VWO is about the diploma participants had in their secondary school. A HAVO-diploma was given value 0 and a VWO-diploma was given the value 1. The variable self-esteem indicates how certain people are about themselves. It is based on a scale of 1 to 5, and for the participants the average is 3.7. This shows that the level of self-esteem is moderately high. The last variable is about the attitude towards the course, which informs us about the attitude participants have concerning the course. This value is based on a scale of 1 to 5, with an average item of 3.1. This tells us that participants are having a neutral attitude towards the course.

1

Participants who first obtained a HAVO-diploma and then continued to study for a VWO-diploma were also given value 1, since there were only six people in this situation. Furthermore, they are considered to be similar as they have the same diploma.

(22)

22

As can be seen in the table, the number of participants differed per variable. For the variable diploma one observation is missing, because the questionnaire was filled out in a trial version, in which the subject was allowed to skip this question. Furthermore, 18 people dropped out in the middle of the survey, so for them there was no information available concerning age, self-esteem and attitude towards the course. Preferably data of these 18 people would be included in the analysis, to have a higher number of observations. The data gathered from these 18 people could be included in the total sample if these people were similar to the people who did complete the questionnaire. If this was not the case, their answers would have to be dropped. Therefore, Wilcoxon Rank sum tests were conducted for significant differences. An overview is shown in table 2 where a division is made between people who finished the questionnaire and who did not.

Table 2. Respondents divided and tested on equality based on finishing the questionnaire Finished questionnaire Wilcoxon

Rank sum p-value on finished = did not

finish

Did not finish questionnaire

Variable Observ. Mean Std dev. Observ. Mean Std dev.

Man 78 0.51 0.50 0.60 18 0.44 0.51 GPAUni 78 6.88 0.47 0.20 18 6.75 0.46 GPASchool 78 7.01 0.71 0.59 18 6.91 0.48 VWO 77 0.95 0.22 0.36 18 0.88 0.32 Treatment 78 0.53 0.50 0.06* 18 0.27 0.46 * p < 0.10

Table 2 shows a mildly significant difference on treatment between the people who finished the questionnaire and the people who did not finish the questionnaire. This implies that subjects who did not get treatment dropped out more easily before finalizing the survey. This could be explained by the fact that reciprocity may have occurred in the treatment group, which caused these students to do something in return by filling out the complete questionnaire. Because of the mildly significant difference, tests were conducted for the complete sample, for the people who completed the questionnaire and for subjects who partly filled out the questionnaire.

Before doing the analysis it was essential that all participants are balanced on observable characteristics between treatment and control, to make sure that these groups

(23)

23

could be compared. In table 3 the group of participants is specified in treatment and control and a Wilcoxon Rank sum test was conducted to compare the treatment and control group on various characteristics.

Table 3. All participants divided into treatment and control and p-value on equality

Treatment Wilcoxon

Rank sum p-value on treatment =

control

Control

Variable Observ .

Mean Std dev. Observ. Mean Std dev.

Man 46 0.48 0.51 0.68 50 0.52 0.50 Age 41 19.98 1.82 0.67 37 20.11 1.85 GPAUni 46 6.95 0.43 0.12 50 6.77 0.49 GPASchool 46 7.06 0.64 0.21 50 6.93 0.70 VWO 46 0.93 0.25 0.94 49 0.94 0.24 Self-esteem 41 3.77 0.55 0.57 37 3.54 0.60 Attitude Course 41 3.10 0.53 0.60 37 3.18 0.57

In table 3 there seem to be no significant differences between the treatment and control group. Also the two-sample Wilcoxon rank-sum test conducted per variable did not show any significant difference.

Furthermore, it is necessary to test whether there were any differences in the treatment and control group for the subjects who completed the questionnaire. An overview of the variables and p-values of the Wilcoxon Rank sum test is shown in table 4.

Table 4. Participants who completed the questionnaire divided into treatment and control and p-value on equality

Treatment Wilcoxon

Rank sum p-value on treatment = control

Control

Man 41 0.49 0.51 0.64 37 0.54 0.51

Age 41 19.98 1.82 0.67 37 20.11 1.85

GPAUni 41 6.98 0.44 0.17 37 6.77 0.49

(24)

24

VWO 41 0.93 0.26 0.37 36 0.97 0.17

Self-esteem 41 3.77 0.55 0.36 37 3.54 0.60

Attitude Course 41 3.10 0.53 0.58 37 3.11 0.77

In table 4 no significant differences were found, which indicates that groups for treatment and control are also comparable for the participants who completed the survey.

Lastly, the group who did not complete the questionnaire was tested for differences in the treatment and control group, which is shown in table 5.

Table 5. Participants who partly finished the questionnaire divided into treatment and control and p-value on equality

Treatment Wilcoxon

Control

Man 5 0.40 0.55 0.82 13 0.46 0.52

GPAUni 5 6.74 0.36 0.92 13 6.75 0.51

GPASchool 5 6.98 0.53 0.66 13 6.88 0.48

VWO 5 1 0 0.37 13 0.85 0.38

Table 5 shows no significant difference between treatment and control for any of the observable characteristics. However, there is a misbalance in number of observations when comparing treatment and control. Therefore the conclusions that are drawn when comparing these groups can be considered less reliable.

5.2 Hypothesis testing

As mentioned in the previous section, treatment and control group are comparable. Therefore in this section the data was used to test whether there were differences between treatment and control group for the dependent variables confidence before exam reported in the survey and the grades obtained for the exam. This was tested with a two-sample Wilcoxon rank-sum test instead of a t-test, to take into account the low number of observations. The first analysis was on hypothesis 1. Hypothesis 1 is: “Receiving a message in which the tutor tells the student that he/she believes in him/her, will positively influence the level of self-efficacy.” This level of self-efficacy was included in the questionnaire by measuring confidence of passing the exam, since this was easier to understand for subjects.

(25)

25

The distribution of answers of this question is shown in histogram 1 for the treatment group and in histogram 2 for the control group.

Histogram 1. Distribution of confidence before exam reported in survey for treatment and control

The distribution of confidence before exam for the treatment and control group seems to be slightly different. The treatment group has a range of 3.5 to 8.5, while the control group has a range of 1 to 10. Furthermore, the density in the treatment group is mainly from 6 to 8, while there is a large peak in the control in the value 6 to 7. A two-sample Kolmogorov-Smirnov test was conducted to test for significant differences in distribution and a mildly significant difference was found with p-value 0.073.

Hypothesis 1 was tested by investigating whether the confidence of passing the exam reported in the survey is different between treatment and control. This test is shown in table 6, where a Two-sample Wilcoxon rank-sum between treatment and control was conducted.

Table 6. Confidence before exam tested between treatment and control

Treatment Wilcoxon

Rank sum P-value on treatment = control

Control

Confidence of passing the exam in survey for all

46 6.50 1.53 0.054* 50 6.01 1.01 0 5 10 15 20 25 30 35 40 45 50 1 2 3 4 5 6 7 8 9 10 Treatment Control

(26)

26 Confidence of passing the

exam in survey for partly filled out

5 7.66 0.53 0.003** 13 6.10 1.09

Confidence of passing the exam in survey for fully filled out

41 6.36 0.97 0.23 37 5.98 1.67

* p < 0.10 .

** p < 0.05

The tests show a significant p-value for the participants who filled out the questionnaire partly and for the entire group. In this case it can be concluded that there is a difference between treatment and control on confidence of passing the exam in the survey, which indicates that the treatment group was more confident about passing the exam than the control group. This conclusion holds for the overall group and for the group who dropped out during the survey. However, it is likely that the participants who did not complete the survey partly caused the significance, since the mean is very high in the treatment group. As the sample size is quite small, this has a big influence on significance. Hence, it can be concluded that the treatment had the desired result of increasing confidence of passing the exam for most participants.

Therefore hypothesis 1: “Receiving a message in which the tutor tells the student that he/she believes in him/her, will positively influence the level of self-efficacy” can be accepted for the whole group of participants and for the group who partly filled out the survey.

The next analysis that was done is concerning hypothesis 2, which is “A high self-efficacy leads to better performance”. The distribution of grades for treatment will be shown in histogram 2.

(27)

27

Histogram 2. Distribution of grade for treatment and control

Histogram 2 shows a quite similar distribution for the treatment and control groups. In the control group, the grades appeared to be higher than in the treatment group. The greatest density is between the grades 5 and 7. The difference is that the distribution of grades for the treatment group is from 4 to 9, while in the control group this distribution is from 1 to 9. A two-sample Kolmogorov-Smirnov test was conducted to test for significant differences in distribution, but with a p-value of 0.81 no significant difference was found.

A Wilcoxon rank sum test was conducted to test whether there are significant differences between treatment and control on the grades subjects scored for the exam. Table 7 shows that all p-values are insignificant, therefore it can be concluded that the grades for the exam in the treatment and control group do not differ.

Table 7. Grades tested between treatment and control

Treatment Wilcoxon

control

Control

Grade 46 5.83 1.39 0.68 50 5.85 1.58

Grade for partly filled out

5 5.86 0.91 0.92 13 5.82 1.48 0 5 10 15 20 25 30 1 2 3 4 5 6 7 8 9 10 Treatment Control

(28)

28 Grade for fully

filled out

41 5.83 1.45 0.61 37 5.85 1.4

The conclusion of this Wilcoxon rank-sum test is that hypothesis 2: “A high self-efficacy leads to a better performance”, is rejected.

5.3 Further analysis

In order to give a profound explanation to the conclusions that are drawn in the previous section, further analysis will be conducted. This includes the possible bias in answering the confidence of passing the exam item in the survey, due to the three weeks that were between answering the survey and entering the exam and a bias that may occur since subjects knew whether or not they passed the exam. Furthermore analysis on overconfidence will be done.

5.3.1 Bias in level of confidence of passing the exam reported in survey

In the survey participants were asked how confident they felt about passing before the exam took place. There is a possibility that participants are biased in their response, since when they filled out the survey, they already knew whether or not they have passed the exam. Besides that, it was approximately three weeks ago since they made their exam, which could also create a bias, especially when they could not remember their specific feelings of that time correctly anymore. It is tested whether there are differences in how the participants actually felt and how they think they felt. Right before the exam took place, a tutor sent an email to the participants with the question “How confident are you that you will pass the exam on a scale of 1 – 10?”. Only 30 participants replied to this email. Even though the number of responses is limited, it can still be used for a small-scale analysis that provides some extra information. First of all, their descriptive statistics are shown in table 8 for some background information.

Table 8. Descriptive statistics for participants who replied to the email

Variable Observations Mean Standard dev. Minimum Maximum

Man 30 0.57 0.50 0 1

Age 24 20.0 1.67 18 25

GPAUni 30 6.89 0.48 6.0 8.2

GPASchool 30 7.12 0.68 6.0 9.0

(29)

29

Self-esteem 24 3.54 0.54 2.5 4.4

Attitude Course 24 3.15 0.46 2.42 4

Treatment 30 0.3 0.47 0 1

In table 9 the participants who replied to the email are divided in treatment and control. The number of observations varies per characteristic, since the people who did not complete the survey are included in this sample as well. It is important that the group of participants is balanced between treatment and control on all variables. Therefore a Wilcoxon rank sum test was conducted for each variable to find out whether there were any significant differences.

Table 9. Participants who replied to the email specified in treatment and control

Treatment Wilcoxon

control

Control

Man 9 0.67 0.50 0.48 21 0.52 0.51 Age 8 19.5 1.51 0.34 16 20.25 1.73 GPAUni 9 7.07 0.47 0.36 21 6.81 0.47 GPASchool 9 7.29 0.68 0.34 21 7.04 0.69 VWO 9 1 0 0.50 20 0.95 0.22 Self-esteem 8 3.7 0.57 0.36 16 3.46 0.52 AtittudeCourse 9 0.67 0.50 0.48 21 0.52 0.51 .

Table 10. Participants who replied to the email specified in treatment and control on dependent variables.

Treatment Wilcoxon

Control

Confidence exam reported in survey 9 6.61 1.17 0.31 21 6.22 1.18 Confidence exam in reply to email 9 6.17 1.41 0.49 21 6.60 1.21

(30)

30

As can be seen in table 9 and 10 no significant differences were found. It is remarkable that more people from the control group replied to the email from the tutor than from the treatment group. The reason for this may be the positive message included in the treatment email. It could be the case that students did not respond if they were not so certain about passing the exam, and they did not want to inform the tutor personally.

This analysis was conducted to compare whether the variables about confidence of passing the exam in the survey and the level of confidence of passing the exam in reply to the tutor were different. This was tested with a Wilcoxon matched-pairs signed rank test. No significant difference was found, which indicates that participants are not significantly different in their reply to the tutor compared to their reply in the survey. Also passing or failing the exam did not significantly influence the level of confidence of passing the exam reported in the survey compared to the level of confidence of passing the exam in reply to the tutor.

5.3.2 Overconfidence

Overconfidence appears when the confidence people have exceeds their accuracy (Klayman, Soll, González-Vallejo & Barlas, 1999). In this study this would result in a difference for participants who scored themselves a 5.5 or higher on the question in the survey “How confident were you, right before the exam, that you would pass?”, compared to whether this student in fact passed or not. This was measured with dummy variables for overconfidence and underconfidence. Overconfidence occurred when the confidence of passing reported was a 5.5 or higher, while the grade of the exam was a 5.4 or lower. Underconfidence happened when the confidence of passing reported in the survey was below a 5.5, while the grade scored for the exam was higher than or equal to a 5.5. Overconfidence occurred 16 times in the treatment group and 14 times in the control group, so receiving treatment does not significantly increase the number of people who are overconfident.

A Wilocoxon rank-sum test was conducted for the influence of treatment for people who are overconfident, right estimation and underconfident.

Table 11. Participants divided in overconfidence, underconfidence and right estimation Overconfident Right estimation Underconfident p-value

(31)

31

58 0.383

7 0.478

** p < 0.05

Table 11 shows that treatment increased the confidence of passing the exam significantly for overconfident people, while this is not the case for underconfident people or people who estimate their abilities right.

5.3.3 Awareness of different mails

Participants were asked in the survey whether they were aware of the fact that two different mails were sent by the tutor. No significant difference was found in confidence of passing the exam in the survey for people who were aware of the varying information in the emails.

5.4 Regression

In this section a regression will be conducted. This regression is done since there are multiple related variables of influence for the grades for the exam and for the confidence of passing the exam in the survey. This regression will develop a fitting model for the available data.

Before conducting the regression, it is important to decide which variables to include. In order to make a firm decision, a correlation matrix is created for control variables to avoid including highly correlated variables that could cause multicollinearity.

Table 12. Correlation matrix

Confidence of passing exam in survey

Treatment Grade GPAUni VWO GPASchool Self-esteem Attitude Course Confidence of passing exam in survey 1.000 Treatment 0.132 1.000 Grade 0.162 0.004 1.000 GPAUni 0.141 0.227 0.311 1.000 VWO -0.105 -0.102 -0.032 0.075 1.000 GPASchool 0.221 0.103 0.202 0.417 -0.048 1.000

(32)

32

As can be seen in table 12, the correlations are quite low, except for attitude towards the course and self-esteem and GPA for university and secondary school. This is due to the fact that they capture partly the same characteristic. Only the GPA at university will be included in the regression. The GPA of university is chosen over the GPA of secondary school since the information is more recent and relevant. Concerning self-esteem and attitude towards the course both variables are taken into account, because the foundation of the two scales is different since one focused on the subject itself and one focused on the course that had to be passed.

A regression was conducted for the influence of variables on confidence of passing the exam in the survey. Three separate regressions were executed and for each regression an additional variable was added, to make the influence on the dependent variable more visible. The number of respondents in the third column is lower since participants who partly filled out the questionnaire are included.

Table 13. Regression on Confidence before exam in survey for the total number of participants Influence on Confidence

of passing exam in survey

(1) (2) (3) Treatment 0.494* (0.267) 0.438 (0.273) 0.362 (0.321) GPA 0.306 (0.291) 0.271 (0.3335) Self-esteem 0.043 (0.310) Attitude Course 0.491 (0.323) Constant 6.010 (0.185) 3.937 (1.977) 2.422 (2.392) R-squared 0.035 0.047 0.075 Adjusted R-squared 0.025 0.026 0.024 Self-esteem 0.157 0.200 0.160 0.171 0.075 0.103 1.000 Attitude Course 0.199 -0.085 0.242 -0.039 0.018 0.077 0.480 1.000

(33)

33

N 96 96 78

* p < 0.10

Table 13 shows that R-squared is very low (0.075). This implies that the variables included in the model only explain a small fraction of how confident people are before the exam. Treatment is found to have a moderate significant influence on confidence of passing the exam in the survey. This indicates that treatment influences efficacy positively. However, once the variables GPA at university, attitude towards the course and self-esteem are added, the influence is found to be insignificant. This indicates that in column 1 the influence of treatment was considered to be stronger than it actually is. The differences between treatment and control seem to be driven partly by GPA at university and self-esteem.

In addition a regression was conducted for the influence of different variables on grade obtained for the exam. Three regressions were executed. The first column only measured the influence of confidence of passing the exam in the survey on grade. In the second column the GPA at university is added and in the third column self-esteem is added. Since self-esteem is only known for the participants who completed the survey the number of participants is lower in this column.

Table 14. Regression on Grade for the complete group

Grade (1) (2) (3)

Confidence of passing exam in survey 0.174 (0.114) 0.126 (0.111) 0.076 (0.126) GPA at university 0.944** (0.312) 0.966** (0.357) Self-esteem -0.021 (0.326) Attitude Course 0.603 (0.347)* Constant 4.753 (3.144) -1.421 (2.154) -3.096 (2.578) R-squared 0.024 0.112 0.154 Adjusted R-squared 0.0137 0.093 0.107 N 0.174 (0.114) 0.126 (0.111) 0.076 (0.126)

(34)

34

* p < 0.10 ..

** p < 0.05 . Table 14 shows that R-squared is very low in all cases. This implies that the variables included in the model only explain a small fraction of the grade. The variable GPA at university is significant in table 14. This indicates that when the GPA of participants increases by 1, the grade of the exam increases by 0.94 or 0.97. This makes sense, since the GPA during the bachelor is based on the individual grades of each participant. When a student has a high GPA, this means that it has a high ability and understands the subjects well. Therefore it is more likely that this student will obtain a higher grade for the exam. Furthermore, the variable attitude towards the course is marginally significant. This indicates that participant have on average a 0.603 higher grade for each point they score higher on the variable attitude towards the course. This suggests that a higher level of attitude results in a higher grade. This conclusion was expected since a higher value on the attitude towards the course implies being less afraid of the course.

(35)

35 6. Discussion

In this section the previous literature will be compared to the findings in this study. Furthermore, the limitations will be discussed and explained.

6.1 Discussion

Previous literature has investigated self-efficacy and self-esteem intensively. It was argued that self-efficacy can be enhanced by giving a compliment, even though this compliment is not necessarily based on the truth (Schunk, 1982; McAuly et al., 1999). Besides that, it was found that both self-esteem and self-efficacy have a positive influence on performance (Aryana, 2010; Collins 1982). In this study, some different conclusions were drawn compared to previous literature.

The finding that self-efficacy can be enhanced by giving a compliment was found in this study as well with marginal significance. When a tutor sent a message to his or her student that “he was certain that he/she would pass the exam”, the level of confidence of passing the exam increased significantly for the total group and for the group who filled out the survey partly. It is remarkable that this increase in self-efficacy was not found for the group of 78 students who completed the survey. A possible explanation for this is that most of the people who dropped out did not receive treatment, reducing the sample size of the control group from 50 to 37. Because of this lower number of participants, the standard deviation has increased from 1.09 to 1.67, which made it harder to find significant results for this smaller group.

The influence of attitude towards the course on performance was found to be marginally significant. This scale measured attitude towards the course, which was closely related to self-efficacy. This indicates that having a better attitude and self-efficacy towards the course improved the grade, as expected by literature (Colling, 1982; Bandura, 1997).

It is argued that overconfidence occurs when people overestimate their capabilities and then perform worse (Klayman, Soll, González-Vallejo & Barlas, 1999). In this study there is a significant difference in the confidence students had for passing their exam, compared to whether they actually passed. Treatment of giving a compliment has shown a significant difference for people who are overconfident. This overconfidence could have resulted in people scoring worse, because they were less specific while doing a task, which lead to more mistakes and a worse performance (Moores & Chang, 2009).

(36)

36

The impact of self-efficacy, measured by confidence of passing the exam reported in the survey, on performance was expected to be positive, according to previous literature (Schunk, 1982). However, in this study the compliment did not influence grades at all. This can be explained by the fact that the given compliment was not based on facts. When this is the case, the effect of the message is expected to be less strong (Schunk, 1982).

Another reason for having a different outcome than expected is the sample size. The sample is quite small, which made it more difficult to find significant results. Especially the influence of the treatment on the confidence of passing the exam reported in the survey is very likely to be significant if the sample size would have been bigger.

Furthermore, the type of students that filled out the survey may have an influence. These students generally exert more effort in passing the exam compared to all students. Besides that, they have a difficulty with this specific course. This increased the chance that they were skeptical concerning the message sent by the tutor. If this were the case, it is very likely that the message did not have the desired effect.

6.2 Limitations

In this study there are several limitations. First of all, the number of respondents is limited and quite small. This makes the conclusions that were drawn less reliable and reduced the chance of finding significant results. With a larger number of respondents, the marginal significant results would probably turn out to be significant. Unfortunately, due to a time restriction, it was not possible to increase the number of respondents.

Another limitation is that the survey was conducted ex-post. The best way to measure the influence of the confidence would have been to have the level of confidence before receiving treatment and after receiving treatment. Unfortunately this was not realizable, due to restrictions from the company. To account for this, the level of confidence of the participants was asked before making the exam, by letting the tutor send an email with a question concerning confidence of passing the course. This value was considered to be more accurate, because it was not influenced by the outcome of the exam. Unfortunately, only 30 participants replied to this email, so the question in the survey concerning confidence of passing the exam had to be used. The reply on this question could be biased. The students already knew their grade, so this may have unconsciously altered the level of confidence of passing the exam reported in the survey. Besides, they took the exam approximately three weeks before, which increases the likelihood of them remembering their feelings incorrectly.

(37)

37

However, when comparing the grade of confidence of passing the exam reported in the survey to the grade participants answered to the tutor, no significant differences were found. In addition, there is a possibility that the subjects lied. They could have lied about their own grade for the exam, the confidence they had of passing the exam in the survey or in their reply to the tutor. Unfortunately, there was no other way of getting this information and it was assumed that students were honest in the survey and in the email.

Another limitation is that regular customers of the company may know that in general no email is sent by the tutor to the students. This may have decreased the credibility of the email. Especially the students who were aware of the fact that two different emails were sent to students, could have been skeptical about the sincerity of this email, even though no significant differences were found between the students who were aware of the different emails and the students who were not aware of the different emails.

Furthermore, endogeneity might have played a role in this study. The confidence students had before the exam was enhanced exogenously by sending an enhancing efficacy email to the treatment group. However, the variables self-esteem and attitude towards the course were included as control variables in the analysis, even though the exogenous treatment may have influenced these two variables. The decision to still have included self-esteem and attitude towards the course, despite potential endogeneity problems, was because of the fact these measures are considered to be stable (Roberts, 2012). Therefore the influence of the treatment on these scales was expected to be relatively small.

In addition, confidence of passing the exam and grades obtained for an exam were influenced by multiple factors that were difficult to measure. By including multiple variables in the survey and regression, it was tried to capture many causes that could explain the dependent variables. However, the regressions that were conducted showed that only a small part of r-squared could be explained. This indicates that the validity of this study is relatively low.

Lastly, it was found that treatment positively enhances self-efficacy for people who were overconfident. This could have been caused by reversed causality, since treatment could have caused subjects to become overconfident.

The influence of a positive message on self-efficacy and performance

The influence of a positive message on

self-efficacy and performance

Table of Contents