# «Gadelrab, Hesham F. Estructura Factorial y Validez Predictora del cuestionario Approaches and Study Skills Inventory for Students en Egipto: ...»

It is hypothesized that the students' data would reproduce the three-factor structure of the ASSIST. It is also hypothesized to find a significant positive correlation between both the deep and strategic approach and the total assessment marks, and a significant negative correlation between the surface approach and the total assessment marks. In addition, negative correlation is expected between the deep and surface approaches, whereas positive correlation is expected between the deep and strategic approaches.

Method Participants The sample of the study consisted of (n=516) Egyptian students who studying engineering (116, 22.5%), computer science (120, 23.3%), business administration (202, 39.1%), and political sciences (78, 15.1%) in the British University in Egypt (BUE). Slightly more than half the sample was male (276, 53.5%). Age ranged from 17 to 23 years (mean= 17.2, SD= 1.2). Students mainly from high socioeconomic status (486, 94.2%). All participants were full-time undergraduate preparatory, first, or second year students. English is the language of instruction in the BUE. ASSIST was distributed to students during attending English classes.

**Estructura Factorial y Validez Predictora del cuestionario "Approaches and Study Skills Inventory for Students" en Egipto:**

Aproximación por Análisis Factorial Confirmatorio

** Procedures**

The purpose of the research was explained to students and the confidentiality of their responses was assured. Students were asked to respond to items with regard to the specific English module they were attending. The assessment marks for this module were extracted from the record system of the university. Three assessment components were used to assess students' learning: a group project, individual presentation, and final examination. Therefore, the assessment mark captured a broad range of learning outcomes including knowledge of the material, basic and advanced skills, as well as understanding and application of knowledge in new settings.

Missing values were minimal (1.5% at most, see Table 2) since the size of administration groups was small and students were instructed not to leave any item without response. Mean substitution was used as an imputation value if missing value is existed. Although it is not a preferred imputation technique due to its limitation in reducing the variance, mean substitution could be justifiable given that the missing values were very rare. The sample size was acceptable according to the rule of thumb recommendation of the minimum requirement of the case to variable ratio to be five (Bryant & Yarnold, 1995).

Gadelrab, H.F.

**Instrument- The ASSIST**

ASSIST consists of four sections. The first section is a six item measurement of the student’s own conception of the term ‘learning’. The second section consists of 52 items and students respond to items on a five-point Likert scale where 5=Agree, 4=Agree somewhat, 3=Unsure, 2=Disagree somewhat, 1=Disagree. These items are designed to measure the three main approaches to learning: deep, strategic and surface apathetic. Each approach to learning comprises of four or five subscales (see Table 2). Each subscale comprises four items.

Subscale scores are formed by adding together the responses on the items in that subscale.

Scores on the three main approaches are created by adding together the subscale scores which contribute to each approach (see Table 2). The third section of ASSIST is an eight item questionnaire measuring preferences for different types of learning and teaching. In the present study only the 52 items producing the three approaches to learning with their respective subscales were utilized.

**Data Analysis**

Data were analyzed using SPSS for windows, Rel. 15.0 (SPSS for Windows, 2006) and EQS for windows, Rel. 6.1 (Bentler, 2008). Confirmatory factor analysis (CFA) was performed to test the factorial structure of ASSIST. Maximum likelihood parameter estimation method was used. Assessment of overall goodness of fit of the model to the data was based on multiple criteria using both absolute and relative fit indices (Gadelrab, 2004;

Hu & Bentler, 1999; Marsh, Hau, & Wen, 2004). Root Mean Square Error of Approximation (RMSEA; Browne & Cudeck, 1993) was used with values less than 0.07 indicating acceptable fit and less than 0.05 indicating good fit. Relative and noncentrality-based goodness-of-fit indices were used in evaluating model fit as well; the Comparative Fit Index (CFI; Bentler, 1990), the Tucker-Lewis Index (TLI; Tucker & Lewis, 1973), and Incremental Fit index (IFI; Bollen, 1989a) with values of 0.95 and greater were indicative of good fit. In addition, Standardized Root Mean-squared Residuals (SRMR) was used, with values of less than 0.08 indicating relatively good fit between the hypothesized model and the observed data (Hu & Bentler, 1999). Values greater than 0.08 might indicate an area of local misfit (Raykov & Marcoulides, 2000).

**Estructura Factorial y Validez Predictora del cuestionario "Approaches and Study Skills Inventory for Students" en Egipto:**

Aproximación por Análisis Factorial Confirmatorio To assess local misfit standardized covariance residuals are consulted to locate the discrepancy between the observed and model-implied covariances. In order to overcome capitalizing on chance problem, cross-validation of the fitted model is needed (Raykov & Marcoulides, 2000), therefore the present full sample was randomly assigned to two equal subsamples of (n=258) using SPSS random selection algorithm. Sample 1 was used to investigate the three-factor structure of ASSIST, and sample 2 was used to cross-validate the factorial structure from sample 1.

** Results**

Unidimensionality and reliability of ASSIST subscales The unidimensionality of each of the ASSIST’s subscales was separately tested by fitting a single factor model to the corresponding four items. The results of separately testing each of the subscales are shown in Table 3. Excellent to perfect fits of the single factor models for all of the 13 subscales were supported and hence it is concluded that the items are unidimensional for each of the 13 subscales.

Although Cronbach’s alpha is a well-known measure of internal consistency, it does not take into account measurement error, and in most cases yields biased estimates of reliability, unless the items are parallel or tau-equivalent (Reuterberg & Gustafsson, 1992). Therefore, the reliability coefficients reported here (see Table 2) were computed using the formula given by Reuterberg and Gustafsson (1992), which is based on CFA results of testing unidimensionality and address some of the above mentioned weaknesses of Cronbach’s alpha. All ASSIST’s 13 subscales reliability coefficient values reached acceptable levels, indicating that the subscales can be interpreted as internally consistent.

Descriptive statistics, univariate and multivariate normality distributions for the 13 subscales (see Table 2) were examined using the total sample. The univariate skewness and kurtosis values of the indicators approximately ranged within ±1.0. Plots of normal probability showed approximate linear patterns for all of the 13 subscales. Shapiro-Wilk test values were not significant for any of the 13 subscales using the α level of 0.05. Therefore, the data were considered to approximate a normal distribution. To assess multivariate normality; "for each combination of two observed variables, a graph is created that plots the Mahalanobis distance for each observation against its ordered chi-square percentile value" (Marcoulides & Hershberger, 1997, pp. 48-52). An examination of the graphs revealed that the plotted values were reasonably close to a diagonal straight line, indicating that the data did not deviate considerably from multivariate normality.

** Models' evaluation and hypotheses testing**

Model specification of ASSIST is shown in Figure 1 (Model 1). The fit of this model to data (sample 1) was poor with goodness fit indices far from cutoff scores of acceptable fit as shown in Table 3. This might indicate that the model needs respecification. Checking the standardized residual matrix to locate the areas that might reflect misfit and consulting Lagrange multiplier test in EQS, five more paths were relaxed to be freely estimated, which were so selected because they appeared to make sense based on theory and previous

**Estructura Factorial y Validez Predictora del cuestionario "Approaches and Study Skills Inventory for Students" en Egipto:**

Aproximación por Análisis Factorial Confirmatorio empirical research (Model2, Figure 2). As shown in Table 4, a dramatic improvement in the model-data fit could be noted. All fit indices were consistent and showing a very good fit of the model to data.

To cross-validate Model 2, it was fitted to sample 2 data. Resultant fit indices showed very good fit between Model 2 and sample 2 data as shown in Table 4, which might refute the hypothesis that Model 2 fitted the sample 1 data only due to chance fluctuations.

Checking the standardized residual matrix revealed that most of the standardized residual were very close to zero, and the largest standardized residual was -0.063 which indicated that Model 2 fitted the data very well. Positive and moderate correlation coefficients between deep and strategic approaches and negative correlation coefficients between both deep and strategic approaches and surface approach were found (see Table 5).

**Predictive validity**

Predictive validity of the use of ASSIST main scales’ scores was tested using regression analysis by using the estimated factor scores (from model 2 on sample 2) as independent variables to predict the total assessment score. The expected significant regression weight of the deep score was supported (r=.56, p.001). In addition significant regression weight of the strategic factor score (r=.35, p.001) was found. However the negative regression weight of the surface approach score that have been found in some studies (Booth, Luckett, & Mladenovic, 1999; Entwistle & Ramsden, 1982) was not supported in the current study (r=.03, p=.09).

Aproximación por Análisis Factorial Confirmatorio

The findings of this research lent support to the three-factor structure of approaches to learning in higher education students. Some debate has been made about the notion and need of strategic approach (Kember & Leung, 1998; Wong, Lin, & Watkins, 1996; Biggs, Kember, & Leung, 2001; Kember, Biggs, & Leung, 2004). However, the existence of the strategic approach to learning in Egyptian college students was not unexpected. Assessment procedures applied in higher education system in Egypt which often reward those who are concerned with both the academic content and the course grades, who keep in mind how to organize answers in a way that impresses the marker, and who have also memorized material are legitimate reasons to expect such approach.

Gadelrab, H.F.

Three subscales that originally belong to strategic approach; organized studying, time management and achieving were loaded on the deep approach in addition to their anticipated loadings on the strategic approach. One subscale that originally belongs to deep approach, interest in ideas, and one subscale that originally belongs to surface approach were loaded on the strategic approach in addition to their expected loadings on their respective approaches.

This might support findings found in some studies (Biggs, 1987c; Biggs & Kirby, 1984;

Biggs et al., 2001) that most students combine strategic approach with either surface or deep approach. On the other hand many research results found similar patterns of cross loadings (Byrne, Flood, & Wills, 1999; Byrne, Flood, & Wills, 2004; Diseth, 2001; Entwistle, Tait, & McCune, 2000). Moreover, the original author of ASSIST, argued that some interconnection between domains should not be seen as a weakness, rather it is an inevitability of the seamlessness of human behavior (Entwistle & McCune, 2004). Additionally, Entwistle, Tait, and McCune (2000) commented on that cross loadings among the subscales of the approaches to learning as entirely understandable in conceptual terms.

Among all of the main subscale loadings on their respective approaches to learning, two of the subscales of strategic approach; alertness to assessment demands and monitoring effectiveness, showed relatively low factor loadings. Alertness to assessment demands was the last subscale to be added to ASSIST (Entwistle, Tait, & McCune, 2000), to be used particularly with students in the final stage of their studies, whereas the students in this study are mainly undergraduate preparatory, with some first and second year students. Byrne, Flood, and Wills (2004) and Diseth (2001) in their validation of ASSIST reported similar difficulties with the behavior of the alertness to assessment demands subscale. Monitoring effectiveness is a related subscale, which encompasses metacognition and self-regulation remarks, therefore this particular subscale is applicable primarily to graduate students more than undergraduate ones (Entwistle, Tait, & McCune, 2000). Caution should be taken with respect to the interpretation of such subscales with undergraduate-especially freshmanstudents. In sum, the results seem to indicate that the deep factor is a general factor accounting for inter-relationship among deep and surface sub-scales, whereas the strategy factor is accounting for unexplained variance above and beyond deep factor.

The negative correlation between surface approaches and achievement that have been found in some studies (Booth, Luckett, & Mladenovic, 1999; Diseth, 2010; Entwistle & Ramsden, 1982) was not supported in the current study, instead a nonsignificant positive