Gender and overconfidenceIntroductionThe empirical pictureExam resultsExam behaviorSome explanationsReferencesGender and overconfidenceClaes Bengtssona, Mats Perssonb,*, Peter WillenhagaaDepartment of Economics, Stockholm University, SE-106 91 Stockholm, SwedenbInstitute for International Economic Studies, Stockholm University, SE-106 91 Stockholm, SwedenReceived 7 February 2004; received in revised form 29 June 2004; accepted 20 July 2004Available online 10 November 2004AbstractDo males differ from females in terms of self-confidence? Using a large set of exam data from StockholmUniversity, we find that exam behavior is gender-specific: male students are more inclined than female students toaim for a higher grade.D 2004 Elsevier B.V. All rights reserved.Keywords: Overconfidence; Gender differences; Exam behavior; Exam resultsJEL classification: A2; J2; J161. IntroductionThere is a small but growing literature indicating that men are more overconfident than women.1While highly relevant to our understanding of social phenomena, studies of the link between genderand overconfidence are hampered by a lack of good data. The purpose of the present paper is to exploit anew data source to shed light on the issue.Stockholm University has used a particular design of the written exam for the first-year courses ineconomics. There are three grades: Very Good (VG), Pass (P), and Fail (F). The exam consists of fourquestions, and in order to get the grade P on the entire exam, you need a P on each one of these four0165-1765/$ - see front matter D 2004 Elsevier B.V. All rights reserved.doi:10.1016/j.econlet.2004.07.012* Corresponding author. Tel.: +46 8 162224; fax: +46 8 161443.E-mail address: [email protected] (M. Persson).1Cf. Barber and Odean (2001) and Correll (2001).Economics Letters 86 (2005) 199 –203www.elsevier.com/locate/econbasequestions. For the student aiming for a VG on the exam, however, there is a fifth question. In order to geta VG, you need firstly a VG on each of the first four questions and secondly a satisfactory answer toQuestion 5. If one has a mere P on one or more of the first four questions, one can never get a VG on theexam, regardless of whether or not one gives a good answer to the fifth question. At the time when onedecides whether or not to answer Question 5, one does not yet know how good the answers to Questions1–4 are. Thus self-assessment enters the picture. Furthermore, if one is satisfied with a mere P on theexam, one has no incentive to answer the fifth question—that would only be a waste of time.This exam structure provides an interesting opportunity for studying gender differences in self-assessment among the students. The issue we analyze is whether female students are less prone toanswer the fifth question than are male students. For this purpose, we use historical exam data consistingof enough observations so that reliable significance tests can be obtained.We find, in fact, that there is such a difference in exam behavior between male and female students;men tend to be more inclined to answer Question 5 than women. But this is not the entire story. There areinteresting patterns within each gender group. Dividing the students into bgoodQ and bmediocreQ ones(depending on their results on Questions 1–4), we find that the gender difference in self-assessment wassignificant in both groups, but more so in the mediocre one. Furthermore, we find an age effect: thegender difference in self-assessment is limited to the younger students.2. The empirical picture2.1. Exam resultsWe have used data from five Microeconomics I exams, from the Fall Term 2001 through the SpringTerm 2004. In total, there were 2217 students who took those exams, of which 1102 (49.7%) werefemale and 1115 (50.3%) were male.Among the female students, 78.8% (869 out of 1102) passed the exam, i.e. received the grade P orVG. Among the male students, 76.5% (853 out of 1115) passed the exam. Is 78.8% significantlydifferent from 76.5%?To answer this question, we use the test statistict ¼ˆllFˆllMffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiˆrr2FnF1þˆrr2MnM1q;where lˆjunji/nj, j=F, M, are the sample fractions among female and male students, respectively, whopassed the exam, and where rˆj2ulˆj(1lˆj), j=F, M, are the sample variances. This statistic is t distributedwith nF+nM2df.2The null hypothesis is now that the percentage of female students who pass the exam is the same asthe percentage of male students who pass the exam. Substituting nF=1102, nFi=869, nM=1115 andnMi=853, where i indicates bpassed the examQ, into the equation above, we obtain t=1.3312, which issignificant at the 90% confidence level. Thus the observed frequencies 78.8% and 76.5% are sufficiently2See Hogg and Tanis (1997, pp. 305–309).C. Bengtsson et al. / Economics Letters 86 (2005) 199–203200different for us to be able to reject the null hypothesis that male students pass the exam as often as femalestudents do. This result is in line with earlier evidence that females seem to perform slightly better thanmales at school.3What about the VG grade? There were 130 (11.8%) female students with a VG, and 180 (16.1%) malestudents; the difference in proportions achieving a VG are statistically significant at the 99% level(t=2.9574).There are thus striking differences between male and female students in terms of outcome: femalestudents are slightly better at passing the exam, but male students are much better at getting the highestgrade. Whether this pattern is due to differences in innate intellectual capacity, study habits, or someother factor, is an open question that deserves further investigation. In this paper, we confine the analysisto investigating whether there are any gender differences in exam behavior.2.2. Exam behaviorAmong the female students, there were 506 who were qualified to answer Question 5, i.e. who scoreda VG on each of the Questions 1–4. Of these, 424 (83.8%) answered Question 5. Among the malestudents, 480 were qualified, of whom 418 (87.1%) answered Question 5. Applying the above teststatistic, we find that qualified females are less prone to answer Question 5 than qualified males. Thedifference is significant at the 90% level (t=1.4653).There were 596 unqualified females (i.e. those who had not received a VG on each one of theQuestions 1–4). Of these, 248 (41.6%) still tried to answer Question 5. Among the 635 unqualifiedmales, 309 (48.6%) were