A framework for the use and interpretation of statistics in reading instruction
MetadataShow full item record
There are few instructional tasks more important than teaching children to read. The consequences of low achievement in reading are costly both to individuals and society. Low achievement in literacy correlates with high rates of school drop-out, poverty, and underemployment. The far-reaching effects of literacy achievement have heightened the interest of educators and non-educators alike in the teaching of reading. Successful efforts to improve reading achievement emphasise identification and implementation of evidence-based practices that promote high rates of achievement when used in classrooms by teachers with diverse instructional styles with children who have diverse instructional needs and interests. Being able to recognise what characterises rigorous evidence-based reading instruction is essential to choosing the right reading curriculum (i.e., what method or approach). It will be necessary to ensure that general classroom reading instruction is of universally high quality and that practitioners are prepared to effectively implement validated reading interventions. When educators are not familiar with research methodologies and findings, national and provincial departments of education may find themselves implementing fads or incomplete findings. The choice of method of instruction is very often based on empirical research studies. The selection of statistical procedures is an integral part of the research process. Statistical significance testing is a prominent feature of data analysis in language learning studies and also specifically, reading instruction studies. For many years, methodologists have debated what statistical significance testing means and how it should be used in the interpretation of substantive results. Researchers have long placed a premium on the use of statistical significance testing. However, criticisms of the statistical significance testing procedure are prevalent and occur across many scientific disciplines. Critics of statistical significance tests have made several suggestions, with the underlying theme being for researchers to examine and interpret their data carefully and thoroughly, rather than relying solely upon p values in determining which results are important enough to examine further and report in journals. Specific suggestions include the use of effect sizes, confidence intervals, and power. The purpose of this study was to: determine what the state of affairs is with regard to statistical significance testing in reading instruction research, with specific reference to post-1999 literature (post-I999 literature was selected because of the specific request, made by Wilkinson and the Task Force on Statistical Inference in 1999, to include the reporting of effect sizes in empirical research studies); determine what the criticisms as well as the defences are that have been offered for statistical significance testing; determine what the alternatives or supplements are to statistical significance testing in reading instruction research; To provide a framework for the most effective and appropriate selection, use and representation of statistical significance testing in the reading instruction research field. A comprehensive survey on the use of statistical significance testing, as manifested in randomly selected journals, was undertaken. Six journals (i.e., System, Language Learning and Technology, The Reading Matrix, Scientific Studies of Reading, Teaching English as a Second or Foreign Language (TESL-EJ); South African Journal for Language Teaching) regularly including articles related to reading instruction research and publishing articles reporting statistical analyses, were reviewed and analysed. All articles in these journals from 2000-2005, employing statistical analyses were reviewed and analysed. The data was analysed by means of descriptive statistics (i.e., frequency counts and percentages). Qualitative reporting was also utilized. A review of six readily accessible (online) journals publishing research on reading instruction indicated that researchers/authors rely very heavily on statistical significance testing and very seldom, if ever, report effect size/effect magnitude or confidence interval measures when documenting their results. A review of the literature indicates that null hypothesis significance testing has been and is a controversial method of extracting information from experimental data and of guiding the formation of scientific conclusions. Several alternatives or complements to null hypothesis significance testing, namely effect sizes, confidence intervals and power analysis have been suggested. The following central theoretical statement was formulated for this study: Statistical significance tests should be supplemented with accurate reports of effect size, power analyses and confidence intervals in reading research studies. In addition, quantitative studies, utilising statistics as stated in the previous sentence, should be supplemented with qualitative studies in order to obtain a more comprehensive picture of reading instruction research. Research indicates that no single study ever establishes a programme or practice as effective; moreover it is the convergence of evidence from a variety of study designs that is ultimately scientifically convincing. When evaluating studies and claims of evidence, educators must not determine whether the study is quantitative or qualitative in nature, but rather if the study meets the standards of scientific research. The proposed framework presented in this study consists of three main parts, namely, part one focuses on the study's description of the intervention and the random assignment process, part two focuses on the study's collection of data and part three focuses on the study's reporting of results, specifically the statistical reporting of the results.
- ETD@PUK