Friend Effects and Racial Disparities in Academic Achievement

Jennifer Flashman

Sociological Science, July 7, 2014
DOI 10.15195/v1.a17

Racial disparities in achievement are a persistent fact of the US educational system. An often cited but rarely directly studied explanation for these disparities is that adolescents from different racial and ethnic backgrounds are exposed to different peers and have different friends. In this article I identify the impact of friends on racial and ethnic achievement disparities. Using data from Add Health and an instrumental variable approach, I show that the achievement characteristics of youths’ friends drive friend effects; adolescents with friends with higher grades are more likely to increase their grades compared to those with lower-achieving friends. Although these effects do not differ across race/ethnicity, given differences in friendship patterns, if black and Latino adolescents had friends with the achievement characteristics of white students, the GPA gap would be 17 to 19 percent smaller. Although modest, this effect represents an important and often overlooked source of difference among black and Latino youth.

Jennifer Flashman: Center for Research on Educational Opportunity, University of Notre Dame. E-mail:

  • Citation: Jennifer Flashman. 2014. “Friend Effects and Racial Disparities in Academic Achievement.” Sociological Science 1: 260-276.
  • Received: March 27, 2014
  • Accepted: April 29, 2014
  • Editors: Jesper Sørensen, Stephen L. Morgan
  • DOI: 10.15195/v1.a17

, , , , ,

2 Reactions to Friend Effects and Racial Disparities in Academic Achievement

  1. syed ali July 10, 2014 at 2:21 pm #

    you’re totally right that it’s all about peers (though why you would ignore judith harris’s “the nurture assumption” is baffling). i make a similar argument for economic and cultural assimilation that might be interesting. or not.

  2. Joeri van Hugten July 22, 2014 at 5:34 am #

    Three ideas to expand your work given the data you already have:

    Because you control for GPAwave1, you are predicting a change in GPA from wave 1 to wave 2. That is, you do not predict how high grades are, but you predict how greatly grades improve from wave1 to wave2 (as you mention on p.264).
    This leads to my first question: is it that friends improve grades OR friends prevent decreases in grades as the curriculum becomes more challenging as students move to the next grade between wave1 and wave2? Knowing this helps narrow down the mechanism.
    One possible answer is in Table A10: the friendship effect is stronger for mediumhigh performers (and females, but gender and performance are correlated). One would find this pattern if friends prevent decreases rather than cause increases. Low performers cannot really decrease in grades further (especially without dropping out of the sample), and therefore friends are not necessary to prevent decrease in grades. Therefore, the effect you find is weak for this group.

    Another question from the finding that high performers benefit while low performers do not: Is the effect driven by people that have GPAwave1 above their (average or bestperforming) friends’ level? If A has higher GPA than his friends, some mechanisms for the effect become less plausible (e.g. even high GPA friends cannot set a good example for A, because A would set the example for them). Conversely, a mechanism that becomes more convincing is that higher GPA friends cause more competitive pressure on A to stay ‘on top’. I think you can check this logic by looking at the GPA of respondents relative to their friends’ GPA at wave1. If this variable “the degree of respondent’s GPA above their friends’ GPA” has an effect, it would narrow down the mechanism.
    It is difficult to figure out what adding this variable to your model would mean for the causal interpretation of your main variable of interest “friends’ average GPA W1”. On the one hand, the new variable’s effect does not seem to me as ‘an aspect of’ the causal effect of “friends’ average GPA W1”. On the other hand, I think it would not make the instrumental variables invalid either, because this new variable’s effect would be picked up by the instruments as part of the ‘except through the characteristics of friends’ (as you phrase it on p.265). Or, maybe adding this variable doesn’t affect the main estimates at all.
    All in all, a ‘relative-to-respondent’-interpretation of friend GPA seems intuitively important and plausible enough to try.

    One more suggestion: as I understand it, you have taken the list of indirect friends and removed those that are also direct friends. You could do something more with this difference between indirect friends and direct friends. High overlap between the list of direct friends and the list of indirect friends could be interpreted to mean that the friendship network is highly integrated. Highly integrated friendship networks may have different effects than less integrated networks. Integrated networks are associated with strong ties, high trust, tight cliques, strong norms, gossip, whereas less integrated networks are associated with weak ties, boundary spanning, new information. If the effect of friend’s GPA is stronger for highly integrated networks, for example, a whole host of mechanisms (those associated with low integration networks) could be excluded.