Differential Psychological Processes Underlying the Skill-Development Model and Self-Enhancement Model Across Mathematics and Science in 28 Countries

Abstract

The skill-development model contends that achievements have an effect on academic self-confidences, while the self-enhancement model contends that selfconfidences have an effect on achievements. Differential psychological processes underlying the 2 models across the domains of mathematics and science were posited and examined with structural equation modeling using the data of grade 8 students from the Trends in International Mathematics and Science Study across 28 countries (N = 144,069), which generated 2 major findings. In statistical terms, (1) there were negative cross-domain paths leading from achievements to self-confidences, controlling for the positive matching domain paths, as predicted by the skill-development model across domains; (2) there were positive cross-domain paths leading from self-confidences to achievements, controlling for the positive matching domain paths, as predicted by the selfenhancement model across domains. There were, however, qualitative variations in the degrees of support for the 2 models across countries.

TIMSS data in an African comparative perspective: Investigating the factors influencing achievement in mathematics and their psychometric properties

“…This finding supports our hypothesis that achieving an acceptable model fit requires that steps should be taken to control for the method effects associated with the combined use of negatively and positively worded items in our data set. Moreover, it supports the literature (e.g., Chiu 2012;Marsh 1996;Marsh et al 2013). Model M1 depicted in Fig.…”

“…This helps eliminate any irrelevant empirical relationship with the constructs or variables in the study. Method effects related to negatively worded items using the CTCU method has been reported in both TIMSS 2003 (e.g., Chiu 2008Chiu , 2012 and TIMSS 2007 (e.g., Marsh et al 2013Marsh et al , 2014, and the National Educational Longitudinal Survey (NELS:88) (Marsh 1994) for the self-concept variate.…”

“…Studies have indicated that the method effect can be seen if at least about 10 % of respondents fail to acknowledge the presence of negatively worded items in a survey instrument and respond appropriately (Woods 2006;Schmitt and Stults 1985). Moreover, many studies on constructs validation (e.g., even for older students and adults) have shown that positively and negatively worded items in a survey often form two distinct factors structures in factor analysis (e.g., Marsh 1986Marsh , 1996Metsämuu-ronen 2012a;Roszkowski and Soven 2010;Chiu 2008Chiu , 2012. However, the split in factors is very much associated with uni-dimensional construct.…”

“…When method effects are not accounted for, they have been shown to affect the goodness of fit statistics, lead to biased parameter estimates (e.g., Chiu 2012Chiu , 2008Bagozzi 1993;Quilty et al 2009;Marsh 1994Marsh , 1996Distefano and Motl 2009;Marsh et al 2013;Lindwall et al 2012;Tomǭs et al 2013;Horan et al 2003;Magazine et al 1996) and influence the validity and reliability of the scale (e.g., Roszkowski and Soven 2010;Woods 2006;Raykov 2001;Brown 2015). Moreover, the method effect can lead to wrong inferences by suppressing or inflating the relationships between constructs by contributing to Type I or Type II errors if not incorporated into the measurement model (e.g., Bagozzi 1993; Magazine et al 1996).…”

“…The 6 negatively worded items (see Additional file 1: Table S1), 4 in the confidence in mathematics scale, 2 in the like mathematics scale were set correlated in order to control the method effect associated with including negatively worded items in the questionnaire. The rationale for this approach has been outlined and illustrated by Chiu (2008Chiu ( , 2012; Marsh (1994Marsh ( , 1996 Marsh et al (2013) and there are appeals for extensively in the section discussing the method effect associated with negatively worded items. As can be seen in Additional file 2: Table S2, the fit indices for section A (part without the correlated uniqueness of the negative items) tend to be outside the range of what is generally considered to be an adequate fit (i.e., CFI, >0.90) for each country.…”