Library or Bookstores with the Child. Parents reported how often they or someone else in their home went to the library or a bookstore with the child. Once again, the response options were: every day or almost every day; once or twice a week; once or twice a month; never or almost never. The responses were also recoded into the number of times per month (i.e., 30, 6, 1.5, or 0, respectively). The fifth variable was Parental Attitudes towards Reading. Parents indicated the extent to which they agreed or disagreed with the following reading-related statements along a four-point scale: (a) I read only when I have to (reverse coded); (b) I like talking about books with other people; (c) I like to spend my spare time reading; (d) I read only when I need information (reverse coded); and (e) Reading is an important activity in my home. The responses to the different statements were averaged for each respondent, and the Cronbach's Alpha was 0.76. The sixth variable was Parents Read for Information. Parents indicated the extent to which they read for work, to get news, or for their own education at home along a four-point scale. Once again, the responses were recoded into the number of times per month and summed for each respondent.
PIRLS Teacher Questionnaire. Unfortunately, only one variable was selected for
further analysis from the teacher questionnaire. This variable was Class Size, which was simply the number of fourth-grade students in the class.
PIRLS School Questionnaire. Six variables were selected for further analysis from
the school questionnaire. The first variable was the Number of Instruction Hours per School Year. For France, Singapore, and Hong Kong, the Ministries of Education provided information on the mean number of hours of reading instruction, which was therefore equal across the different schools in these countries. In all of the other countries, the total number of hours of instruction per year was calculated on the basis of the number of hours of instruction per day multiplied by the number of days that the school was open for instruction per year, both reported by the principals. The second variable was the Number of Books in the School Library. The response options for this variable were: more than 10,000 books; 5,001 to 10,000 books; 2,001 to 5,000 books; 501 to 2,000 books; 251 to 500 books; 250 or fewer books; or no school library. To be able to perform multilevel analyses, the ranges were recoded to create an interval variable representing the number of books in the school library (i.e., 15,000; 7500; 3500; 1250; 375; 125; and 0, respectively). The third variable was the Percentage of first through fourth grade Students in the School with Another Home Language than the test language. The fourth variable was the Percentage of Students in the School coming from Economically Disadvantaged Homes. For both variables, the response options were: 0 to 10%; 11 to 25%; 26 to 50%; or more than 50%. The responses were again recoded as 5%, 18%, 38%, or 75.5%, respectively. The fifth variable was the Point in the Curriculum at which major emphasis was placed on Decoding Skills. Principals reported the grade in which the following decoding skills first received major instructional emphasis: (a) knowing the letters of the alphabet; (b) knowing letter-sound relationships; (c) reading words; and (d)
reading isolated sentences. The response options were: first grade; second grade; third grade; fourth grade; or not in these grades. The responses were averaged across the four decoding skills; the Cronbach's Alpha was 0.83. The sixth variable was the Point in the Curriculum at which major emphasis was placed on Reading Literacy Skills. Principals reported the grade in which the following reading literacy skills and strategies first received major instructional emphasis: (a) identification of the main idea in a text; (b) explanation or support of understanding of text; (c) comparison of a text with personal experiences; (d) comparison of different texts; (e) prediction of what will happen next in the text; (f) generalization and inference based on text; and (g) description of the style and structure of the text. Once again, the response options were: first grade; second grade; third grade; fourth grade; or not in these grades. The responses were averaged across the difference reading literacy skills, and the Cronbach's Alpha was 0.91.
Replacement of Missing Values
Of the 49,889 students in the database, only 13,966 were found to have no missing values whatsoever. For this reason, a procedure was established to replace missing values wherever possible. Each missing value within a unit (i.e., the student, the class, or the school) was replaced by a random draw from a normal distribution with a variance equal to the variance for the group to which the unit belonged and a mean equal to the mean for that group. For example, a missing student score was replaced by a random draw from a normal distribution with variance equal to the variance for the classroom of the student and a mean equal to the mean for the classroom of the student. For class size and the percentage of second-language-speakers in the school, the missing values could be derived from alternative sources of information. In order to assess the influence of the replacement procedure on the results, the procedure was repeated four times with the same conditions but different random draws. When more than one third of the cases for a particular variable were missing, the replacement procedure was not applied because the variance estimates and means were judged to not be sufficiently reliable under such circumstances. For this reason, almost 25% of the cases still had a missing value for at least one of the variables of interest. These cases were removed from the database, leaving 38,954 valid cases for analysis, or 78% of the original sample. An overview of the final number of schools, number of classes, and number of students included in the analyses for the eleven countries in the present study can be found in Appendix B.
Adjustment of PIRLS weights
The PIRLS weights (see PIRLS Procedure above) had to be adjusted for two reasons. First, only 78% of the sample was available for analysis. The sum of the PIRLS student weights for a country represents the complete target student population for that country; the sum of the PIRLS class weights for a country
represents the complete target classroom population for that country (i.e. the number of classes with the highest percentage of nine- and ten-year olds); and the sum of the PIRLS school weights for a country represents the complete target school population (i.e. all those schools in the country including classes with the highest