One of the roles of education in people’s lives is to make changes and to empower them to make changes in their own lives. Although education aims to change many aspects of people’s lives such as values, attitudes, and academic achievements, the research usually focuses on how the attitudes, values, and other variables affect the academic achievement of individuals. Educational experiences of individuals have implications for regions and countries, therefore international assessment studies have been done such as Trends in Mathematics and Science Study (TIMSS, 1995, 1999,2003, 2007, 2011, 2015), Program for International Student Assessment (PISA, 2000, 2003, 2006, 2009, 2012, 2015) to compare the achievement of countries and to investigate the factors affecting the achievement of students and countries. (PISA OECD, 2015; Martin, Mullis, & Foy, 2015)
There are many studies on TIMSS and PISA data from different perspectives. Some of them focuses on measurement issues namely measurement invariance (e.g. Lee, 2009) and differential item functioning (e.g. Klieme & Baumert, 2001), and some aims to determine the factors affecting student and country level achievement (e.g. Fertig, 2003; Dincer & Uysal, 2010). In this study, the factors related to the Turkish students’ achievement level are investigated by quantile regression and results are compared with ordinary linear regression.
In standard regression analysis, based on the Ordinary Least Squares (OLS), the effect of independent or predictor variables on the dependent variables are estimated. OLS predictors show the effect of predictor variables at the conditional mean of the distribution of the dependent variable, therefore OLS regression provide only limited information to this specific point of the distribution. In PISA and TIMSS studies, dependent variable is usually student achievement and dependent variables are students’ socioeconomic status, gender, teaching method or such. When the data is analyzed with OLS, analysis results can yield to incomplete findings when the effects of predictors vary along the distribution. In recent years with the development of the computational techniques with computers, new alternatives to ordinary linear regression methods were introduced. Koenker and Basset introduced quantile regression (1978) to deal with the research questions in the field of Econometrics. Quantile regression analysis method allows determining the effect of predictor variables on dependent variable’s at different points of its distribution. In other words, quantile regression (QR) approach provide information to describe the effect of predictor variables along the entire distribution of the dependent variable. In this way the researchers can have information at all quantiles of the distribution, including high, low and medium parts of it (Hao & Naiman, 2007.
The purpose of this study is to examine some of the factors influincing Turkish students' achievement by quantile regression, so that more detailed suggestions to improve the current achivement status can be made. In the next part this text, quantile regression and the study method are explained in more detail.