4.4 Data and sample
4.4.2 Validity of Identification Strategy
Our identification strategy relies on exploiting quasi-random variation in foreign peer exposure across cohorts within the same university and major. If variation across cohorts within university and major is as good as random, we should not observe any foreign peer effects on outcomes that pre-date entry into higher education. Here we use two outcomes that are determined before native students actually meet their peers: distance between home and chosen institution, and ability. Distance between home and university is chosen at the moment of applying and accepting an university offer and as such pre-dates interaction with foreign peer students and should not be influenced by them. Ability is measured using the UCAS score and as such is a summary index of pre-higher education educational achievement, therefore, it is predetermined at the entry in higher education.
In table 4.4 we provide estimates from regressing distance and tariff scores on foreign shares using the fix effect specification in (4.19). Our estimates show no sig- nificant effects on distance between home and university location. However, there are effects on tariffs and number of natives enrolled. Increasing the share of immi- grants by one percentage point leads to a .177% increase in the tariff and a -1.16% decrease on the number of enrolled natives. These effects are more strongly driven by EU students. This matches what we presented in our framework in section 4.2. EU students are treated by universities as home students and, as such, they put pressure on the number and the ability of natives across all universities. On the other hand, non-EU students only compete with natives in certain institutions, those that are
4.4. Data and sample 107
more selective. This can be seen in table 4.4. On average a one percentage point increase on the share of EU students lowers the number of natives by 2%. The same increase on non-EU produces a .93% decrease. However, this changes when we con- sider specific university groups. In selective universities such as those in the Russell Group,34increasing the share of EU and non-EU students has a similar effect. But in less selection universities, those in the 1994 group and all others, the effect of non-EU students gets attenuated. In the last column of table 4.4, increasing the share of EU and non-EU students produces a 1.98% and 2.53% decrease on natives enrolled in Russell Group universities, the baseline. The effect of non-EU students gets attenu- ated by 1.49 percentage points in the next most selective university group, i.e. 1994 Group, and by 2.04 percentage points in all other universities.
As we introduced in section 4.2, the asymmetry on the treatment of EU and non-EU students induces an asymmetry on the effect that this have on the number and ability of enrolled natives. Furthermore, this asymmetry is heterogeneous across universities depending on how selective they can be. Therefore, we interpret the effects on tariffs and course sizes as evidence showing that universities modify the distribution of student characteristics as a function of the foreign composition of their prospective students. This means that exploiting variation across cohorts within university and major cannot extract peer effects.
Table 4.4: Placebo Pre-entry Outcomes
(log-)Distance (log-)Ability (log-)Native Students
Foreign Peers 0.002 0.177 -1.164* (0.106) (0.109) (0.490) EU Peers 0.040 0.311* 0.284+ -2.055*** -1.980*** (0.196) (0.152) (0.154) (0.547) (0.546) non-EU Peers -0.013 0.132 -0.086 -0.923+ -2.523*** (0.126) (0.129) (0.145) (0.533) (0.667) 1994 Grp. * EU -0.279 -0.882 (0.229) (0.669) Other HEI * EU 0.151 0.210 (0.305) (0.929) 1994 Grp. * non-EU -0.086 1.487+ (0.447) (0.853)
Other HEI * non-EU 0.339 2.043*
(0.227) (0.916)
Observations 447,980 458,160 458,160 4,710 4,710
Note: Sample of undergraduate students enrolled in 2007/10 in all English higher eduction insti- tutions (HEI). Fixed effects: HEI, year (of enrolment), major, HEI-year, HEI-major, major-year. Observations are weighted with analytic weights to account for the different contribution of course size in our estimates. The overall foreign peer estimated is produced from a separate regression. Standard errors are clustered at HEI level and reported in parenthesis. Observations rounded to last unit. + p < .1 * p < 0.05, ** p < 0.01, *** p < 0.001
Selection carried by universities at entry would imply that our estimates return the effect of foreign students given the university selection mechanism. However, we can improve upon this thanks to the quality of our data. This is because we can control for ability using the UCAS score. We show that, conditional on individual ability, cross- cohort variation on foreign peer shares is uncorrelated with a comprehensive set of predetermined characteristics that are likely predictors of educational performance.35 Particularly, we provide estimates from regressing socio-economic background (figure 4.2), sex, age and private school (figure 4.6); ethnicity (figure 4.7) and disability (figure 4.8) on the share of EU and non-EU students. Where each of this variables is used in its own regression with our set of fixed effects and conditioning on ability score quartile.
In figure 4.2, all socio-economic groups show non significant correlations, at the 95%, with both the share of EU and non-EU peers and point estimates are typically close to zero. The only exception is the correlation between routine occupations and non-EU share for which we observe a marginally significant correlation. In figures 4.6,
35
This sort of balancing test have a long tradition on the peer effects literature (Anelli and Peri 2017; Chin et al. 2013; Cools et al. 2019; Lavy and Schlosser 2011).