The purpose of this study is to analyze the error facets in speaking assessment and explore optimal conditions to use the assessment in classroom.
This study explores reliability and error facet of the performance assessment for speaking. Also, it...
The purpose of this study is to analyze the error facets in speaking assessment and explore optimal conditions to use the assessment in classroom.
This study explores reliability and error facet of the performance assessment for speaking. Also, it suggest practical optimal conditions which are needed to conduct performance assessment. Practical optimal conditions are closely related with practicality of assessment. If reliability and practicality were added to speaking assessment which already equip with validity, it would be an ideal assessment for classroom.
Generalizability theory is a method to estimate reliability which can reveal error facet in assessment. In classical test theories, we can not find the error facet in assessment. Generalizability theory provides us information on what kinds of error facets exist through analysis of variance on real situations and conditions of assessment. Also, D study in generalizability theory provides optimal conditions on how to measure to get reliable results(genralizability coefficient is .7~.8 or dependability(Φ) coefficient is .6~.8).
After assessing speaking performance of Korean elementary school students' 5th graders directly and half-directly assessment method, this research found optimal conditions based on tasks of speaking assessment and the facet of raters with using GENOVA analysis of generalizability theory.
According to the result of G study in generalizability theory, the error facets which affects performance-based speaking assessment of fifth graders are tasks and raters. Among them, the weight of error in tasks(6.36%) and interaction of tasks and persons(8.69%) were high, but the weight of error in raters(0.88%) were low. It was not explained what the interaction among subjects, tasks, and raters(73.67%).
Next, accords exist result of D study in generalizability theory, to raise generalizability coefficient over .7, we should increase the number of tasks to 14 when there are 2 raters. If we want to fix the number of tasks into 7, at least 5 raters are needed. Also, to raise dependability coefficient over .6, the number of tasks should be at least 11 when there are 2 raters. In this case of 7 tasks, we need more than 4 raters. In the classroom where only 1 class teacher scores the performance assessment for speaking, more than more than 25 tasks are needed for generalizability coefficient over .7, and 19 tasks for over .6. The optimal assessment conditions can be changed according to the results of D study.
To get appropriate genralizability coefficient(0.7), the number of tasks should be 25 when 1 rater scores them, the number of tasks should be 14 when 2 raters scores them, the number of tasks should be 10 when 3 raters scores them, or the number of tasks should be 8 when 4 raters scores them. The numbers of tasks stated above are optimal conditions for practicing speaking assessments. The result of this research shows that speaking assessment should be conducted in the constant processes assessment or portfolio assessment. However, the result of this research which requires large number raters has relatively low realistic possibility. This is why it is needed to reconsider the design of GENOVA analysis for this research.
Also, to get appropriate dependability coefficients(0.6) the number of tasks should be 19 when 1 rater scores them, the number of tasks should be 10 when 2 raters scores them, the number of tasks should be 8 when 3 raters scores them, or the number of tasks should be 6 when 4 raters scores them. The numbers of tasks stated above are optimal conditions for practicing speaking assessments. This result shows that it is necessary for speaking assessment to be performed continually.
Other studies should be conducted on the appropriate methods for speaking assessment, solving the problems of reliability and practicality, and being helpful for practicing authentic speaking assessments.