Discussion and Conclusions

Abstract

The main purpose of this research was to develop a psychometric framework to assess the relationship between parental involvement and reading literacy. The framework incorporates country specific differences, both at the item level and the scale level, to gain insight into cultural differences in the parental involvement component and its relation to student achievement in reading literacy. We conducted secondary analyses on the PIRLS-2011 data of 41 countries. A review of the research literature distinguished four dimensions of parental involvement:(1) home-based involvement from a parent perspective; (2) school-based involvement and home-school communication from a parent perspective; (3) home-based involvement from a student perspective; and (4) school-based involvement and home-school communication from a school perspective. Based on items available in the PIRLS data, the first dimension was split in two components: early literacy activities and helping with homework. IRT analyses provided item-by-country interactions indicating CDIF. The five components were first modeled using the unidimensional GPCM. Using these analyses, potential items with CDIF were identified and subsequently modeled using country-specific parameters for the 10 and 20 % most extreme interactions. These methods for identifying and modeling CDIF were compared with two other models. The first was the GPCM with random item parameters, where the variance of the parameters across countries provided an indication of possible CDIF. The second was a bi-factor GPCM where a country-specific covariance matrix gave an indication of the extent to which the scale loaded on the intended latent variable and the extent of loading on a country-specific dimension. Finally, multilevel analyses were conducted to explore the association between parental involvement and student achievement for all countries that participated in PIRLS-2011. A three-level (student, school and country) random intercept model was explored, as well as a random three-level model.This study addressed three central research questions.