Association for Public Policy Analysis & Management

Existing literature generally apply experimental or quasi-experimental methods to eliminate the self-selection problem (Angrist & Lavy,1999; Hoxby, 1998). Despite the lack of consensus on class size effect, many countries still invested additional resources in reducing the size of the class of K12 education in the last decades. However, only a few of them found it is beneficial to student achievement (OECD,2012). China, as the most prominent developing country, invested plethora to lower the class size of compulsory education from 2000 as well (State Statistical Bureau, 2001-2015).

Nevertheless, in most cases, previous studies analyses the class size effect within a specific region or country, which may produce the inconsistent findings of the impact of class size. However, only a few of studies dedicate to the international comparison to the class size effect, using the global database from standardized tests (e.g., TIMSS, PISA), and find only several countries with specific characteristics benefit from the reduction of class size (Wößmann & West, 2006). Nevertheless, these studies still hypothesize that the effect of class size is linear. In our former study in China, a non-linear correlation is found in China’s junior high schools. A question raised, is adequate size exist of class? Because of the diversity of class sizes across the world, teachers of different areas may response differently towards the change of class size.

Using PISA 2015 database, we categorize 15 countries of the top 20 performing in science into three types based on countries’ class size, defining the level of development. Most western countries are identified as the advanced group with most students (more than 80%) is being in a class less than 30 people. Most East Asian nations are in the intermediate group, in which most secondary students are in classes around 30-40 people. And we put China and Vietnam in the initial group, whose students are mainly placed in classes over 40 and even 50.

Our HLM result shows that the positive correlation of smaller class exists only in a few countries, condition on students’ personal, peer and school-level characteristics, and no consist association with our postulated developing stages. Surprising, when we categorize the original class size variable into five categories dummy variables, 30-40 class shows best or not worse no matter the stage of development. Some countries’ small classes even show worse like Germany and Netherlands. We divide the sample into junior high schools and senior high schools since grade10 is not inclusive in some countries compulsory education. The result is robust, and only some East Asian countries and regions with high-selective-entering test entice larger class is better because of sorting.

Furthermore, we investigate whether preference exists within different countries, and find that parents with the higher social-economic background will not prefer smaller size when country’s most classes are under 30.