An Empirical Study of Peer Effects on Academic Achievement in an Elite College

• In Event: Roundtable Session (Wednesday IV)
  In Refereed Round-Table Session: Student achievement in China and Brazil

Proposal

Keywords：peer effects; academic achievement

During the sixties of the last century peer effects began to draw researchers’ attention. More and more economists and sociologists have begun to study peer effects since it was found that peer’s interaction impact academic performance significantly in the famous Coleman report. In the primary and secondary schools peer effects happen in the classroom; by contrast peer effects happen mostly in the dormitories in colleges. There are very extensive empirical studies on the roommates’ peer effects on students’ achievement (Foster, 2006; Sacerdote, 2001; Zimmerman, 2003), on students behaviors, such as alcoholism, drug abuse, sexual intercourse etc. (Duncan et al., 2005), on community participation, professional choice (McEwan and Soderberg, 2006). During the four years’ college life the pattern, frequency and content of interaction among peers changes with the development of college students. However, there is little evidence focusing on the changing trend of peer effects.

This paper takes advantage of a natural experiment of the randomly assigned dormitory in a top research university in China to explore the characteristics of peer effects both on academic achievement and behaviors students in each year, New students were randomly assigned to dorms and roommates, thus eliminating the selective bias. Random distribution means that all roommates background variables and their background features are not related. The data are merged administrative data from the Department of Education, the International Exchange and Cooperation Office, the Ministry of Logistics and the Ministry of Employment and other administrative departments. Overall there are 87 variables and 5176 undergraduate samples, of which 2671 undergraduate students enrolled in 2009 and 2505 undergraduate students enrolled in 2010. There are 2859 boys, 2317 girls, they come from 31 provinces (municipalities, autonomous regions), distributed in 48 professional, 1294 quarters. Dormitory distribution follows the principle of random, dormitory is mainly 4 people.

With regard to research methods, the standard method of measuring peer effect is to use observational data to return to their behavior or results with observed peer behavior or results, but as Manski (1993) says, such research methods do not make an accurate estimate of the impact of peer effects because of the existence of endogenous problems, for now, more and more studies are mainly to find the school randomly assigned companion data, and in the analysis of the use of tool variables or adding a fixed effect to get a more accurate estimate of the companion effect.

The basic regression model is expressed as follows:
GPAi=α+β1SATi+β2SATiRM+β3Xi+εic （1）
Where GPA represents the students after entering the school's academic performance, SAT represents the college entrance examination results, SATRM represents the student group members of the college entrance examination results, X refers to the student's personal characteristics (such as gender, urban and rural household registration), ε is the error term. If the population is randomly assigned, then β2 will be unbiased estimate; if the error term ε is not related to other explanatory variables, then the equation is estimated to be unbiased.

While peer effects can vary with individual learning ability or their peer learning ability,As shown in equation (2): SATD igRM represents the different category of individual or individual group members' learning ability, n represents the number of such categories, βg represents the companion effect corresponding to different learning ability:
GPAig=α+β1SATi+∑_(g=1)^n▒βgSATigDRM+β3Xi+εic （2）
Because of the particularity of our college entrance examination, students from the same major may be affected by the same factors, and thus have some common characteristics. In order to solve these problems, this study transforms the scores of the provincial candidates into the standard percentile scores. On the basis of the control profession (m represents the specialty), the influence of the dormitory groups on the student's academic performance is investigated.
GPAim=α+β1SATim+β2SATimRM+β3Xim+ Vm （3）
GPAimg=α+β1SATim+∑_(g=1)^n▒βgSATimgDRM+β3Xim+ Vm （4）
While examining whether the influence of the peer effect in the first and second year, the third and fourth year is the same, and eventually becomes equation (5)，Where GPAimsg-1 represents the grade point of the previous school year：
GPAimsg=α+β1SATim+∑_(g=1)^n▒βgSATimsgDRM+β3GPAimsg-1+β4Xims+ Vm （5）

To examine peer effects and it’s trend this paper set up several linear regression models with dependent variables of major choice, GPA, financial aids and scholarships in each year, career planning, and employment choice before graduation respectively. To obtain a more accurate estimate of peer effects and solve the problem of endogenous，fixed effect parts are added in the regression models. The independent variable in the models include a serial of roommates’ characteristics, such as gender, ability, political status, parents’ education, SES, hometown provinces, the way of admission etc.

In the specific measurement of variables, we use the average grade of four academic years as the agent variable of academic achievement; learning ability is the college entrance examination results（(we use college students' entry as a tool variable for their learning ability to solve the problem of synchronicity. The college entrance examination results is not affected by the peer behavior because it is generated before the individual entering the same dormitory and class, and, thus avoiding the synchronization of individual behavior and companion behavior And mutual decisive, there is no two-way causal relationship problem)）; family background and individual characteristics include urban and rural household registration and gender, ethnic, family economic situation, parental education, Whether it is a party member. Family background is an important part of peer theory research model, such as parental education, family income, etc.

Based on the above results this paper provides administration departments with some operational suggestions in the allocation of dormitories to improve positive peer effects.

Authors

• Yifan Huang, Graduate School of Education, Peking University, Beijing , China

• Liping Ma, Peking University