TPC-Journal-V4-Issue5

The Professional Counselor \Volume 4, Issue 5 547 successfully used in numerous studies to measure student change after educational interventions targeting these constructs. Self-Efficacy for Self-Regulated Learning scale. The SESRL was designed to measure students’ confidence in their abilities to perform self-regulatory strategies. It is a seven-item self-report instrument based on the Children’s Self-Efficacy Scale (Bandura, 2006; Usher & Pajares, 2008). Items (e.g., “How well can you motivate yourself to do schoolwork?”) reflect students’ judgments about their abilities to perform self-regulation strategies identified by teachers as being frequently used by students (Pajares & Valiante, 1999). The scale has been used successfully with older elementary students in a self-read format and with fourth graders in a read aloud administration format (Usher & Pajares, 2006). Cronbach’s alpha estimates of reliability have ranged between .78 and .84 (Britner & Pajares, 2006; Pajares & Valiante, 2002; Usher & Pajares, 2008). Factor analysis has suggested that the scale is unidimensional. Concurrent validity studies have indicated that the scale is related to measures of self-efficacy, task orientation and achievement (Usher & Pajares, 2006). Data Analysis In the initial analysis of the three SESSS subscales, the present authors used mean imputation to replace missing survey responses, by replacing a missing response with the overall mean for that survey item. For each of the 33 SESSS items, only 8.3%–9.1% of the responses were missing. Mean imputation is appropriate when the percentage of missing data is less than 10% and can be considered to be missing at random (Longford, 2005). In the current study, the students with missing survey data had an average SESSS score equal to that of the students with a complete response set, thus supporting the notion that the data were missing at random. Coefficient alpha, used as a measure of reliability, was calculated for each of the subscales before missing values were replaced. Both convergent and discriminant evidence is needed in the validation process (Campbell & Fiske, 1959; Messick, 1993). Messick (1993) argued that while convergent evidence is important, it can mask certain problems. For example, if all tests of a construct do not measure a particular facet of that construct, the tests could all correlate highly. Likewise, if all tests of a construct include some particular form of construct- irrelevant variance, then the tests may correlate even more strongly because of that fact. Due to these possible shortcomings of convergent evidence, discriminant evidence is needed to ensure that the test is not correlated with another construct that could account for the misleading convergent evidence. To determine the validity of the three SESSS subscales, the authors examined the correlations between each of the subscales with five other measures: four subscales of the MSLQ (Self-Efficacy, Cognitive Strategy Use, Self-Regulation and Test Anxiety), and the SESRL. Specifically, the authors considered the strength and direction of the SESSS subscales’ correlations with these other measures. Results Descriptive statistics and reliability estimates for the instruments used in this study are contained in Table 2. Coefficient alphas for the three SESSS subscales (Self-Direction of Learning, Support of Classmates’ Learning and Self-Regulation of Arousal), were 0.89, 0.79 and 0.68, respectively, and 0.90 for the SESSS as a whole. These results indicate good internal consistency (i.e., that the items within each instrument measure the same construct). All correlations between pairs of subscales appear in Table 3. Because of the large sample size in this study, statistical significance by itself could be misleading, so the authors used the magnitude and direction of the correlations for their interpretations. Correlation is an effect size reflecting the degree of association of