TSUNG-HAU JEN, CHE-DI LEE, CHIN-LUNG CHIEN, YING-SHAO HSU and KUAN-MING CHEN PERCEIVED SOCIAL RELATIONSHIPS AND SCIENCE LEARNING OUTCOMES FOR TAIWANESE EIGHTH GRADERS: STRUCTURAL EQUATION MODELING WITH A COMPLEX SAMPLING CONSIDERATION Received: 14 February 2012; Accepted: 28 June 2012 ABSTRACT. Based on the Trends in International Mathematics and Science Study 2007 study and a follow-up national survey, data for 3,901 Taiwanese grade 8 students were analyzed using structural equation modeling to confirm a social-relation-based affection- driven model (SRAM). SRAM hypothesized relationships among students’ perceived social relationships in science class and affective and cognitive learning outcomes to be examined. Furthermore, the path coefficients of SRAM for high- and low-achieving subgroups were compared. Given the 2-stage stratified clustering design for sampling, jackknife replications were conducted to estimate the sampling errors for all coefficients in SRAM. Results suggested that both perceived teacher–student relationships (PTSR) and perceived peer relationships (PPR) exert significant positive effects on students’ self- confidence in learning science (SCS) and on their positive attitude toward science (PATS). These affective learning outcomes (SCS and PATS) were found to play a significant role in mediating the perceived social relationships (PTSR and PPR) and science achievement. Further results regarding the differences in SRAM model fit between high- and low- achieving students are discussed, as are the educational and methodological implications of this study. KEY WORDS: complex sampling, large-scale survey, learning motivation, science achievement, self-determination theory, social relationships, structural equation modeling, TIMSS INTRODUCTION Contemporary science education reforms emphasize interactive and con- structive learning in which learners learn from the interaction among prior knowledge, concurrent experiences, and human interactions within socio- cultural contexts (United States National Research Council, 2007). However, how learners’ perceived social environment and attributes influence learning outcomes is complex and not well documented. Large international assessment data sets provide the basis for important secondary analyses that have implications regarding student, school, and cultural attributes far beyond the league tables about who’s in first place flowing from these International Journal of Science and Mathematics Education (2013) 11: 575Y600 # National Science Council, Taiwan 2012