Working Paper: CEPR ID: DP5660
Authors: Andreas Ammermüller; Jörn-Steffen Pischke
Abstract: We estimate peer effects for fourth graders in six European countries. The identification relies on variation across classes within schools. We argue that classes within primary schools are formed roughly randomly with respect to family background. Similar to previous studies, we find sizeable estimates of peer effects in standard OLS specifications. The size of the estimate is much reduced within schools. This could be explained either by selection into schools or by measurement error in the peer background variable. When we correct for measurement error we find within school estimates close to the original OLS estimates. Our results suggest that the peer effect is modestly large, measurement error is important in our survey data, and selection plays little role in biasing peer effects estimates. We find no significant evidence of non-linear peer effects.
Keywords: measurement error; peer effects
JEL Codes: I21; J24
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| measurement error (C20) | biased estimates of peer effects (C92) |
| correcting for measurement error (C20) | reliable estimates of peer effects (C92) |
| one standard deviation increase in peer composition (C92) | 0.11 standard deviation increase in reading test scores (I24) |