Statistical Analyses

In order to test the validity and reliability of the questionnaire used in this study and to answer the research questions, a number of statistical treatments were applied to the quantitative data. These data were analysed by both descriptive and inferential analyses including factor analysis, descriptive statistics, Pearson's

correlation coefficient, and multiple regression. Discussions of these statistical procedures are presented below.

Factor Analysis

Two types of factor analysis, namely Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA), were used in the present study in order to establish the construct validity of the scales of TAIS and TEBS. Once the factor structure was derived by the EFA procedure, the second step involved using CFA to provide confirmation of the instruments’ factorial structures.

For a long time, factor analysis has been largely used in researchers' and psychometricians' attempts to establish construct validity (Goodwin, 1999), which concerns whether a number of items on a questionnaire actually represent the

theoretical latent construct those items were developed to measure (Hair et al., 2010). Factor analysis is a statistical technique that is most commonly used to analyse the structure of the intercorrelations among a large number of variables (or items) by identifying groups of variables that are strongly intercorrelated, known as factors. The main objective of factor analysis is to determine the underlying factor structure among the variables in the data analysis (Hair et al., 2010). Establishing the factor structure of an instrument is an important consideration for theory development (Byrne, 1998; Peter, 1981). There are two main types of factor analysis, namely EFA and CFA, that can be used to achieve the primary purposes of factor analysis (Hair et al., 2010; Tabachnick & Fidell, 2007).

EFA is basically used for the situation where the correlations between the items (known as observed variables) and the underlying dimensions (known as latent variables or factors) are uncertain or unknown, and the analysis proceeds in an exploratory manner to determine how and to what degree the observed variables are

related to their associated latent factors (Byrne, 1998). With EFA, all observed variables are correlated with every factor, and these correlations are represented by factor loading estimates (Hair et al., 2010). EFA enables the researcher to reduce a set of variables to a smaller set of representative factors, which can then be used for further analysis (Hair et al., 2010; Ho, 2006). Although an EFA analysis does not give the verification needed to determine the construct validity (Miller et al., 2011), conducting EFA is recommended in order to provide a preliminary investigation of the number of factors prior to performing the CFA to test the measurement model (Hair et al., 2010).

Unlike EFA, CFA is performed with a strong prior knowledge about the structure of the factor model. In other words, before the analysis of CFA starts, it is already known which variables load on which factors (Lattin, Carroll, & Green, 2003). Thus, CFA is a technique that enables a researcher not to “explore” but to “confirm” or “reject” the prior knowledge regarding the factor structure. Specifically, it is a method for examining how well observed variables represent a smaller set of factors (Hair et al., 2010). The use of CFA serves multiple objectives, including but not limited to, assessing the psychometric properties of instruments, designing new instruments, and examining method effects (Harrington, 2009). CFA is also

considered to be one of the most rigorous methodological techniques for assessing construct validity (Byrne, 2001; Harrington, 2009; Miller et al., 2011).

Descriptive Analysis

A number of Descriptive statistics were analysed to answer a number of the research questions in the current study. Specifically, the mean, standard deviation, and percentage were used to address the research sub-question #1.1, while frequency and percentage were used to address the research question #2. That is in order to

describe teachers’ knowledge of AD/HD and their attitude towards the inclusion of students with AD/HD-related behaviours. SPSS 20.0 was used to generate tables and graphs describing the data. Descriptive statistics are techniques that assist researchers in organising, summarizing, and simplifying the findings derived from the data set (Gravetter & Forzano, 2012). There are various types of Descriptive statistics, such as mean, standard deviation, frequency, percentage and mode. Mean and standard deviation, particularly, are the most frequently used measures for central tendency and variability (Gay et al., 2009).

Pearson's Correlation Coefficient

Pearson's correlation coefficient was conducted to answer the third research question of the present study. Specifically, this technique was used to examine the relationship between teachers’ knowledge of AD/HD and their efficacy beliefs for teaching students with behavioural problems. SPSS 20.0 was used to perform a Pearson's correlation analysis.

Pearson's correlation coefficient, known as Pearson's r, is a parametric statistical technique developed to measure the linear relationship between two variables (Norusis, 2008). “Essentially, it works out a measure of how much the scores of the two variables vary together (their ‘product’) and then contrasts this with how much they vary on their own” (Hinton et al., 2004, p. 297). Pearson's r was the initial formal association measure (Lee Rodgers & Nicewander, 1988), and it remains one of the most frequently used correlation coefficients (Croux & Dehon, 2010; Lee Rodgers & Nicewander, 1988). Similar to other parametric methods, there are important assumptions underlying the Pearson's correlation coefficient. These assumptions include: (a) the variables are at an interval or ratio level; (b) the data are

normally distributed; (c) the variables are linearly related: and (d) homoscedasticity (Field, 2009; Hinton et al., 2004).

In Pearson's analyses, values of relationships can range from -1 to +1, and the direction of a relationship (positive or negative association) is indicated by the coefficient's sign. The strength of the association is represented by the size of the absolute value. A relationship of +1 or -1 indicates that there is a perfect relationship between two variables whereas a relationship of 0 means that there is no relationship between the variables (Pallant, 2011). For interpreting correlation coefficients, this study used the guidelines proposed by Cohen (1988) and shown in Table 3.3.

Table ‎3.4

Interpretation of Pearson's Correlation Coefficient

Small r =.10 to .29 OR r = -.10 to .29

Medium r =.30 to .49 OR r = -.30 to .49

Large r =.50 to 1.0 OR r = -.50 to 1.0

Multiple Regression

In order to answer the third research question, multiple regression was used. It was performed to determine whether the variances in teachers’ attitude towards the inclusion of students with AD/HD-related behaviours can be explained by the

independent variables, which included efficacy beliefs, teacher age, class size, training about students with AD/HD-related behaviours, years of teaching

experience, prior experience with a child with attentional problems, prior experience with a child with behavioural problems, and the subtype of AD/HD (see Figure 3.1). All experience variables besides the subtype of AD/HD were dummy coded for the regression analyses. All independent variables were entered into the analysis

simultaneously because these variables were chosen on a theoretical basis and the order of importance among these variables was not theoretically supported (Field, 2009). The multiple regression was performed using SPSS 20.0.

Multiple regression analysis is a statistical method that can be used to examine the relationship between one dependent variable and two or more independent variables (Tabachnick & Fidell, 2007). That is, “multiple-regression analysis can provide a scatterplot and equation for a situation in which two or more independent variables work collectively to predict scores on the dependent variable” (Wetcher-Hendricks, 2011, p. 245). Prior to performing a multiple regression

analysis, there are several assumptions that should be met. These assumptions include the following: (a) the independent variables are measured at the quantitative or categorical (with two categories) level whereas the dependent variable is measured at the continuous level; (b) normality of the residuals distribution; (c) the absence of multicollinearity (a high relationship is present between the independent variables); (d) the relationship between the dependent variable and each of the independent variable is a linear one; (e) the errors are dependent; and (f) the variance of residual terms is equal (homoscedasticity) (Field, 2009).

Figure ‎3.2.The Multiple regression model.