主题：从自然子刊的一篇文章看教育的不同，对问题的解决和看法影响 -- 实话实说

Graduate School of Education, Stanford University, Stanford, CA, USA. 2Freeman Spogli Institute for International Studies, Stanford University, Stanford, CA, USA. 3Educational Testing Service, Princeton, NJ, USA. 4International Center for Action Research on Education, School of Education, Henan University, Henan, China. 5Institute of Education, National Research University Higher School of Economics, Moscow, Russia. 6Center for Studies in Higher Education, Goldman School of Public Policy, University of California Berkeley, Berkeley, CA, USA. 7Faculty of Education, Sichuan Normal University, Sichuan, China. 8Institute of Higher Education Research, University of Jinan, Jinan, China. 9Graduate School of Education, Peking University, Beijing, China. 10Institute of Education, Tsinghua University, Beijing, China. 11World Bank, Washington, DC, USA

(1) Sampling in China, India, and Russia: We sampled computer science (CS) and electrical engineering (EE) major students that, taken together, comprise a large proportion of STEM undergraduates in China (34%), India (24%), and Russia (24%). We first identified all undergraduate (bachelor’s degree) CS and EE programs from China, India, and Russia that had comparable course requirements and content with undergraduate CS and EE programs in the United States. Using the population frame of all higher education institutions with these undergraduate CS and EE programs, we then randomly sampled institutions from these countries. Briefly, from China, we took a simple random sample of six institutions from each of six representative provinces. In India and Russia, we took stratified national random samples of 50 and 34 universities, respectively. Altogether, we sampled 7 elite and 29 non-elite institutions in China, 8 elite and 42 non-elite institutions in India, and 6 elite and 28 non-elite institutions in Russia. For more information about the sampling of institutions, see the SOM. We next randomly sampled administrative units within the sample institutions. In each randomly selected administrative unit, we sampled all first year (freshmen) and third year (junior) students. We randomly assigned half of the students in each year to take grade-specific math and physics exams, one quarter of the students to take a critical thinking exam, and one quarter of the students to take a quantitative literacy exam. Response rates in the baseline were high with 95% of enrolled students taking the exams in China, 95% in India, and 87% in Russia. Altogether, 5,102 freshmen and 4,145 juniors from China, 8,232 freshmen and 9,223 juniors from India, and 2,607 freshmen and 2,096 juniors from Russia participated. We conducted follow-up testing after almost two years with the different subsets of freshmen and junior students from the baseline (when they were at the end of their sophomore and senior years). Freshmen that had taken math and physics tests in the baseline took end-of-year 2-appropriate math and physics test in the follow-up, while freshmen and juniors that took critical thinking in the baseline took the critical thinking test in the follow-up. Response rates in the follow-up were again relatively high with 80% of enrolled students taking the exams in China, 95% in India, and 90% in Russia.

To ensure national representativeness, we adjusted our analytical estimates and standard errors for survey design features including multi-stage sampling and probability sampling weights (see the SOM). We also estimated both unadjusted (using listwise deletion) and adjusted (using multiple imputation—see the SOM) estimates of skill gains. Because skill gains estimates are substantively the same in either case, we only report unadjusted estimates in the main text (for adjusted estimates, see the SOM). (2) Sampling in the United States.—Data on the critical thinking skills of students in colleges in the United States were collected from 2016 to 2018 by Educational Testing Service (ETS). We use a subsample of STEM bachelor’s degree program students from a range of institutions in the United States to create comparative benchmarks of critical thinking skill levels. In terms of Carnegie classifications, the sample includes 11 doctoral research institutions (672 students or 69% of the sample), 17 masters institutions (245 students or 25% of the sample), and 8 baccalaureate institutions (56 students or 6% of the sample). Approximately 53% of the sampled students were in fact from the highest ranking R1 institutions: Doctoral Universities – Highest Research Activity. Since the distribution of STEM bachelor’s degree program students in the United States is 67%, 24%, and 9% across doctoral research, masters, and baccalaureate institutions (with 44% in R1 institutions), the across-institution distribution of students in the sample is similar to that of STEM students in bachelor’s degree programs in the United State.