Unveiling the effects of other-gender friendships: evidence from middle-school students in China

¹ A possible concern is that if a child performs well (poorly) academically, then a parent may become less (more) restrictive on a child’s activity. We take advantage of the rich information in our data to demonstrate that parents’ strictness with making friends has no direct effect on a child’s academic performance except indirectly through the child’s friendship choice. A more detailed discussion of parents’ strictness with making friends can be found in the appendix.

² It’s important to note that China predominantly follows a co-educational model. China has made significant efforts to promote gender equality in various aspects of society, including education. Co-educational schools align with the principle of providing equal opportunities and experiences for both genders. Furthermore, given the large population in China, building separate schools for boys and girls would require significant resources and infrastructure. Co-educational schools allow for more efficient use of educational facilities. There may still be single-sex schools or programs, but most are at higher education levels or in specific contexts. Single-sex schools do not exist in CEPS.

³ Detailed definitions of confidence, negative mood states, self-perception, and curiosity can be found in the appendix.

⁴ In contrast, raw scores of these exams are based on school-specific tests and are not comparable across schools.

⁵ As in many surveys, there is a restriction on the maximum number of friend nominations. This restriction, although standard and facilitates subjects’ responses, may cause some bias in our estimation due to the censoring of the data. However, we perform some robustness checks by including additional types of friends, i.e. non-school friends, and find that our results are robust. More details can be found in Section V. Therefore, the potential bias induced by this restriction should not be a serious concern in our sample.

⁶ Based on the CEPS data, we explore the important role of friendships for teen students and results are provided in the appendix.

⁷ To alleviate the concern that this strictness of making friends reflects students’ perceptions of parental strictness obtained only by students, we use parents’ responses about the strictness with making friends as an IV for robustness check. As is shown in Table A5 in the appendix, having at least one other-gender friend is estimated to have a statistically significant negative effect on Mathematics scores, which is consistent with findings from our main estimation.

⁸ A detailed discussion of the validity of IVs is provided in the appendix.

⁹ Using the standard deviations reported in Table 1, we calculate $(-10.5901\times 0.43/9.92)\times 100=45.9\mathrm{\%}$.

¹⁰ The standard deviations of having at least one other-gender friend, Chinese, Mathematics, English, and overall scores for girls is 0.4067, 8.3456, 9.4321, 8.4962, and 22.7947, respectively.

¹¹ As detailed in the appendix, we construct the negative mood states based on five categorical variables: blue, depressed, unhappy, bored, and sad. Table A7 in the appendix shows the estimates of having at least one other-gender friend on these five variables. As is evident in columns (1), (2), (4), and (5) from Table A7, having at least one other-gender friend is estimated to yield a statistically significant effect on increasing feelings of being blue, depressed, bored and sad.

¹² Table A8 in the appendix shows estimates of other-gender friends on five variables of students’ negative mood states by girls and boys. As is shown in columns (1), (3), (5), (7), and (9) from Table A8, having at least one other-gender friend is estimated to have a statistically significant effect on increasing girls’ feels of blue, depressed, unhappy, bored and sad. Notably, magnitudes of these estimates are higher than corresponding estimates in Table A7.

¹³ The standard deviation of negative mood states for girls is 3.9081. Using the standard deviation in Table 1 and estimates reported in Table 4, we calculate $(16.979\times 0.4067/3.9081)\times 100=176.69\mathrm{\%}$.

¹⁴ We explore a few different channels through which that other-gender friends influence students’ cognitive and noncognitive performance. Results are provided in the appendix.

¹⁵ The survey anonymously asked school administrators and teachers questions about rules that how they replaced students in classrooms. First, school officials were asked which one of the following methods they used to assign students to classrooms at the beginning of grade 7. We keep schools with random assignments. Second, the principal confirmed that the school will not rearrange the students in grades 8 or 9 into different classes.

¹⁶ The distinction between in-school and out-of-school friendships could significantly influence the extent to which other-gender friendships are linked to maladjustment. Out-of-school other-gender friendships may exert a particularly detrimental impact on in-school cognitive performance, especially for girls, as previously discussed. Table A6 in the appendix presents estimation results for girls with or without out-of-school friends. As is shown in columns (3) and (4), having at least one other-gender friend is estimated to have a statistically significant negative effect on Mathematics scores, with this effect being more pronounced among girls who have out-of-school friends.

¹⁷ We use a question in the survey that ‘How many of your best friends mentioned above (up to 5) doing well in academic performance?’. The students could circle one of the following three: ‘None of them’, ‘One or two of them’, or ‘Most of them’. The choice of ‘None of them’ is assigned a value of 1; ‘One or two of them’ are assigned a value of 2; ‘Most of them’ is assigned a value of 3.

¹⁸ All strictness measures have three levels. ‘Not care’ is assigned a value of 1, ‘Care but not strict’ is assigned a value of 2, and ‘Strict’ is assigned a value of 3.

¹⁹ We conduct a regression by using the number of school same-gender friends as a dependent variable, a dummy variable for having at least one other-gender friend, and other controls as independent variables. We find that the estimated coefficient for having at least one other-gender friend is negative and statistically significant. Estimation results are available upon request.

²⁰ The correlation coefficient between parents’ strictness on homework and parents’ strictness on friends’ choices is 0.23.

²¹ Detailed definitions of activities and behaviours can be found in the appendix. Table A9 shows the summary statistics for activities and behaviours.