In 1991 some 30 countries participated in an international comparison organized by IEA of achievement in reading literacy in one population of 9-10 year olds and another population of 14-15 year olds. In 2001 a subset comprising 9 of these countries participated in a repeat study in which the same instruments were administered to samples of 9-10 year old students (Martin et al., 2003). The results from this study showed that reading performance had increased in several countries, and most strongly so in Iceland, Greece, Slovenia and Italy. For Sweden, however, results were poorer in 2001 than in 1991, and for the United States performance was unchanged. While Sweden was among the most well-performing countries in 1991 that was not the case in 2001, when Italy, Iceland and the United States were the three most highly performing countries. Rosén and Gustafsson (2003) made a somewhat closer analysis of the Swedish data. The reanalysis relied on a three-dimensional factor model for the scores on the items associated with the 15 text blocks included in the reading assessment (Gustafsson & Rosén, 2003). One factor was a general reading comprehension factor (ReadC) with the highest relations to performance on narrative and expository text blocks. Another factor (EoT, for End of Test) was related to performance on the last few blocks of each of the two separately timed booklets of reading tasks, and was interpreted to reflect reading speed. The third factor was a documents factor (Doc), with relations to the text blocks using non-continous forms of presentation in graphs, tables, lists and so on. The reanalysis showed that for Sweden there was a slight decrease in the level of the ReadC factor. For the EoT factor a larger decrease was observed, while for the Doc factor there was an increase. An analysis of which factors may explain these changes showed that changes in leisure time reading habits between 1991 and 2001 can account for the decrease in the ReadC factor. The purpose of the present paper is to extend the kinds of analyses done for Sweden to encompass all 9 countries that participated in the 10-Year Trend Study.

The study relies on the published data from the IEA 10-Year Trend Study concerning reading performance and information from the student questionnaire. The analysis of data proceeded in three steps. First, the properties of the three-factor confirmatory factor-analytic model were investigated, with particular attention to the invariance of factor loadings over countries. Second, differences in latent variable means were investigated for the countries in two different ways: through multiple-group modelling and through computing means on estimated factor scores. In the third step attempts were made to explain the pattern of changes over time for each country by fitting models in which possible mediators of the changes are introduced, using variables from the student questionnaire. The modelling was carried out with Mplus 5 (Muthén & Muthén, 2007) under the STREAMS modelling environment (Gustafsson & Stahl, 2005).

Preliminary results from the analyses indicated that the three-factor model (ReadC, Doc, EoT) obtained an excellent fit in all 9 countries in both 1991 and 2001. However, there were differences in some parameters of the latent model which indicated that parts of the country differences may have be caused by text-specificities in the reading tests rather than by the targeted reading ability dimension. This result calls for further research. The changes across time in these latent dimensions revealed a somewhat different pattern of change as compared with the change in IRT score in the international report. The improvements in reading comprehension were larger in the latent dimension compared to the IRT-score. Further, almost all countries showed a decrease in reading speed. Since reading speed is confounded with reading comprehension in the IRT score, this may also explain why the changes in the IRT-score of reading literacy where relatively smaller across time than in the ReadC in multivariate latent variable model. The opposite directions of mean change in reading speed and in reading comprehension cause the IRT-score change to be lower. One conclusion is thus that the reading speed factor has contributed further insights of the trend-lines between 1991 and 2001 in reading literacy.