Stata se v 14

We used Stata SE v. 14.0 (StataCorp, 2015, College Station, TX) to perform all statistical analyses, calculating descriptive statistics on targeted admissions strategies. Chi-square tests for independence were used to examine differences in school characteristics by each targeted admissions strategy. The significance level was set at P<.05. "Don't know," "Not applicable," and no responses were coded as missing values and excluded from the denominator when calculating percentages for each target group.

Descriptive statistics (Table 1) and estimated proportions (Figure 1) were weighted using calibrated cross-sectional weights provided by SHARE (Malter & B€ orsch-Supan, 2017) to make them representative of the target population. Separate multivariate logistic regressions were performed for each EOL attribute (0¼not important/not so important; 1¼important/very important) to determine their association with respondents' personal characteristics and the linguistic region in which they live. Analyses were conducted using Stata SE v.14.0 software (StataCorp, College Station, TX, USA). Estimated standard errors were adjusted for potential unobserved dependencies between partners' responses. Results of the full logistic regression models are presented in Annex 2 in the Supplementary Material. The figures of the result section below present the predicted margins (i.e. standardized proportions adjusting for all of the other model covariates) for a specific characteristic (e.g., gender) across the 23 multivariate logistic regression models. Finally, the value 1 on the outcomes corresponds to both answer categories, "important" and "very important", but we will refer to them collectively as "important" in the following paragraphs.