Abstract Recent studies have shown the appropriateness of unfolding models in the analysis of binary disagree-agree responses as opposed to the use of traditional cumulative measurement models. An adaptation of Cohen’s (1941) Scale of Attitude towards Aesthetic Value was used for the purpose of attitude measurement toward the visual arts. The study aimed to: (i) examine reliability of the modified scale and consistency of the items; (ii) clarify and facilitate interpretation of results obtained from the Thurstone-type scale; and (iii) improve the quality of attitude measurement. Pupils (n = 131) from two South Australian secondary schools completed two equivalent forms that comprised the Visual Arts Attitude Scale on a single occasion. The study demonstrated the usefulness of the HCM in analysing binary responses and in explaining the disagree responses in terms of their constituent elements. The manifest disagree responses were resolved into two separate latent response curves, which characterised the directions of disagree responses. These were unfolded to correspond to the data, thus providing a description of the effect that is expected as a function of the distances of persons from statements. Key Words: Attitude Measurement, Binary Coded Data, Hyperbolic Cosine Method, Unfolding, Single Peaked Response Function

Abstract The Development of the Unfolding Model The Hyperbolic Cosine Model For Unfolding Single Stimulus Responses Method Results Discussion References

 

The Development of the Unfolding Model

Recent studies have shown the appropriateness of unfolding models in the analysis of binary coded data (Roberts, 1995; van Schuur & Kiers, 1994). The use of binary disagree-agree responses is traced back to Thurstone (1928, 1931). Andrich and Styles (1998, p. 454) argue that Thurstone’s rationale for measuring attitudes can be summarised by two features: (i) statements ranging from positive to ambivalent to negative affect, with respect to a construct being measured, are located numerically on an attitude continuum; and (ii) persons are located on the continuum according to the statements they endorse. Under Thurstone’s process referred to as the pair comparisons design, the locations of statements on a scale are quantified by requiring judges to compare the selected statements in pairs, and to indicate which statement of the pair presents greater intensity of the construct. In order to analyse data derived from the pair comparisons design, Thurstone (1928) formulated the law of comparative judgement (LCJ) which is a probabilistic expression wherein a person declares that one statement has a stronger affective value than another with respect to the attitude being measured, independent of a person’s location on the continuum. Thurstone used the normal distribution for the latent error and formulated a number of different cases for the LCJ depending on the constraints imposed on this error (Andrich & Styles, 1998). One of the easiest and most common cases used is that in which the error distribution for every statement is identical. This case, according to Andrich (1978a) is equivalent to the logistic expression: Pr{Xij=1} = exp(di-dj)[1 + exp(di-dj)] (1) where Pr{Xij=1} indicates the probability that statement i is considered to have a greater affective value than statement j; and di and dj are the locations of statements i and j, respectively. In order to estimate the location of statements from the LCJ, it was possible to examine the consistency of responses to the requirements of the model, ensuring they were located on a single continuum. Location of a person on the attitude continuum was calculated by Thurstone, based on an individual’s responses to the scaled statements. This method implied a single peaked response function characterising the probability that the person agreed with a statement, with the peak occurring where the person’s ideal point was at the location of the statement. Thurstone’s attitude measurement procedure is used to analyse the binary responses following the ideal point process (Coombs, 1964) wherein, a person is assumed to agree with an attitude item to an extent that the item content satisfactorily represents the person’s own opinion. With this perspective, it is appropriate to analyse binary responses with the single-peaked response function implemented by the unfolding model. Coombs (1964) was the first to formalise analyses of binary responses using the term unfolding, and from which the location of the items and respondents can be identified simultaneously. In order to avoid Thurstone’s time consuming and onerous approach, other formal stochastic models have been developed following the work of Likert (1932), that advanced a procedure to measure attitudes where a cumulative mechanism was presupposed. Under this procedure, the probability of an agree response (or strongly agree) increases as a function of the location of that person. Unlike the Thurstone scales, where the responses need to be unfolded to find their locations, person measures are obtained from summing the responses across statements, and by applying the correlational procedures of classical test theory. However, the method does not require that statements be located on an attitude continuum (Andrich & Styles, 1998). Traditionally, binary response data are analysed with the use of cumulative measurement models. Likert’s (1932) procedure increased the number of different response categories, but did not require scaling of the statements. He also included an undecided or neutral category between disagree and agree. Other researchers have shown that this category does not necessarily operate as a middle category (Bock & Jones, 1968; Dubois & Burns, 1975; Andrich, de Jong & Sheridan, 1997). By using traditional checks of statements, such as item-total correlations and factor analyses, Likert failed to check the internal consistency of the statements (Andrich & Styles, 1998). Nevertheless, the total score of persons across statements was taken as a measure of attitude, which implied a cumulative mechanism, such that the greater the location of a person, the greater the total score. In Thurstone's method, two people can have the same number of statements to which they agree, but their respective attitude values differ because these values depend on the type of statements with which individuals agree. In contrast, in Likert’s method (1932), the greater the number of statements agreed to, the greater the person’s attitude toward the construct being measured. Therefore, Likert needed to convert the single peaked unfolding Thurstone method into a cumulative method by reversing certain scores (Andrich & Styles, 1998). Consequently, the sum of the scores across statements could then provide an index of a person’s attitude. Statements that reflected an ambivalent attitude toward the relevant construct could not be included in the Likert scale, since scoring would not be clear (whether to score as a negative or a positive). For this reason, statements that reflect an ambivalent attitude have no place in a Likert questionnaire (Andrich & Styles, 1998). Hence, by using the simpler Likert procedure, the location of the items on the scale which is central to Thurstone scaling is lost. Furthermore, van Schuur and Kiers (1994) and van Schuur and Kruijtbosch (1995) argued that a factor analysis of responses to scales, which included positive and negative worded statements, gave the impression that positive and negative statements with respect to a construct are on separate dimensions, whereas when analysed according to an unfolding model, the items are located at different parts of a single continuum. Since traditional checks of statements, total correlations, and factor analyses cannot possibly test for the internal consistency of items scaled from an ideal point perspective, the necessity of analysing binary response scales with an explicit probabilistic model is mandatory. In this study binary graded responses to attitude items were collected for the purpose of attitude measurement toward the visual arts from two secondary public schools. The instrument piloted was an adaptation of Cohen’s (1941) Scale of Attitude toward Aesthetic Value. The purpose of the study was to examine the structure and reliability of the modified scale and the item consistency. In order to clarify and facilitate the interpretation of results obtained from the Thurstone-type scale, the dimensionality of the measuring instrument was explored, thus ascertaining the quality of attitude measurement to be included in subsequent multilevel and multitrait analyses of data. Because of the aforementioned limitations of cumulative deterministic methods of analysis the study considered the use of the probabilistic Hyperdolic Cosine Method of unfolding.

The Hyperbolic Cosine Model For Unfolding Single Stimulus Responses

The probabilistic Hyperbolic Cosine Model (HCM) for dichotomous responses of each person to each item, assumes a single peaked item response function for the items, in which the probability of a correct response decreases as the distance between a person’s trait level and the item’s location, increases in either direction above or below the items. Data obtained from direct responses to a dichotomous attitude item are analysed to provide estimates of item location parameters. The unfolding method is developed from two assumptions: (i) the respondent tends to agree with the item located closest to his/her position on a unidimensional latent attitude continuum; and (ii) the individual may choose a disagree response category from one of two possible locations on the scale of attitude. Thus a respondent might disagree with a statement when holding a very positive or a very negative attitude. If a statement is located far below the respondent’s position on the attitude continuum, then the statement is more negative than the respondent’s attitude and thus the person disagrees from above the item (DA), since the person is at an attitude level higher up the attitude scale than the statement. If however, the item is above the person’s location on the attitude continuum, then the item is more favourable than the individual’s attitude and thus the person disagrees from below the item (DB). In this case, the unfolding model postulates two subjective responses for each observable disagree response on a rating scale. Figure 1 gives a graphical representation of the three categories. Here, the single disagree response is composed of the sum of two latent responses, DA and DB. In addition to these two responses, there is also the single response (A-agree), which implies that the respondent’s location is close to that of the statement. Figure 1. Response functions of the HCM, including the resolved disagree response.   By Andrich and Luo (1993) the HCM model takes the form: Pr{xni =1} = exp(qi)/[exp(qi)+2cosh(bn-di)] Pr{xni =0} = 2cosh(bn-di)/[exp(qi)+2cosh(bn-di)] (2) where x = 1 is a positive response; x = 0 is a negative response; bn is the location of person n; di is the location of item i; qi is the unit of item i; and cosh(u) = [exp(-u)+exp(u)]/2 is the hyperbolic cosine function. The model used to construct the HCM in the case of three ordered categories (where DB = 0, DA = 2 and A = 1) as shown by Andrich and Styles (1998), takes the explicit form: Pr{Yni=0} = 1/hni Pr{Yni=1} = exp[qi+(bn-di)]/hni Pr{Yni=2} = exp[2(bn-di)]/hni (3) where bn and di are the locations of person n and statement i respectively; qi reflects the distance from the statement location di where the DB and DA response functions intersect with the A response function; and hni is a normalising factor that is the sum of the numerators and ensures that the probabilities of the responses sum to one.

Method

Sample, measures and procedure A total of 131 students (Year 8 to Year 11) from two South Australian secondary schools participated in the study, which was a trial for a more extensive investigation. Binary graded disagree-agree responses to attitude items were obtained for the purpose of piloting the Visual Arts Attitude Scale adapted from Cohen’s (1941) Scale of Attitude toward Aesthetic Value. The attitude scale consisted of two equivalent forms each comprised of 20 statements. Form A is presented in Table 6 and Form B in Table 10. They provided two sets of scores that were analysed for an estimate of reliability. Statements included in the original two forms had been scaled by the paired comparison process and were placed in decreasing intensity of affect (from extreme positive to extreme negative). The equivalent-forms reliability method is a traditional method that estimates reliability based on the administration of two forms that may be considered equivalent. In this method, the two forms provide two sets of scores that are correlated to obtain an estimate of reliability. An average of the intercorrelations among the forms may be taken as the estimate of reliability. Provided both forms actually measure the same attitude, the correlation coefficient will reflect consistency of measurement. The purpose of this pilot study was to examine the reliability of the modified scale and consistency of items within the scale. Both forms of the attitude scale were administered in the classroom on a single occasion. The scale was not timed, thus allowing ample time for reading and understanding statements. Respondents were requested to agree or disagree with each statement.  Statistical analyses The following statistical analyses were performed on SPSS version 8.0, QUEST (Adams & Khoo, 1993) and RUMMFOLDss (Andrich & Luo, 1998). The basis of RUMMFOLDss lies in the principles of the Rasch unidimensional measurement model (RUMM) for three ordered categories, which gives the Hyperbolic Cosine Model (HCM) for unfolding. In the Rasch model for a simple dichotomous item (agree-disagree), an individual’s (n) response to item (i) can be expressed as a function of the person’s ability (bn) on the trait being tested as well as item difficulty (di). The probabilistic HCM for the dichotomous responses (‘1’ = agree, ‘0’ = disagree) of each respondent to each item is single peaked. The HCM resolves the disagree responses into two components (Andrich & Luo, 1993). An individual may regard the location of an item as being below himself or herself, or an individual may perceive himself or herself as being lower than the item. An individual gives an agree response to an item when he or she perceives him or herself to be close to the location of the item. Thus with the aid of RUMMFOLDss, responses are examined using the Rasch model for three categories (Andrich, 1982). Figure 1 contains the probabilistic curves of the HCM (x = 0, 1) and the Rasch model for three ordered categories (y = 0, 1, 2) (Andrich & Luo, 1998). The HCM equation is expressed by Andrich and Luo (1993) as: P{xni=1} = coshri/[coshri+cosh(bn-di)] P{xni=0} = cosh(bn-di)/[coshri+cosh(bn-di)] (4) where the parameter ri characterises the latitude of acceptance of the item/statement. The relationship between ri and qi is expqi = 2coshri. The model deals here with dichotomised responses, and since the HCM is single-peaked and symmetric about bn-di = 0, for each of the possible values there are two corresponding positions on the latent trait continuum. Therefore, a sign analysis is performed where the program identifies a pair of items that have the minimum correlation coefficient, so that the signs of correlation coefficients of items with either one of these two items, is chosen as their initial sign. There are four major cycles performed using the Joint Maximum Likelihood (JML) principle, with a constraint that the mean of item locations are equal to zero (0). These four cycles are defined as follows:  (i) Cycle 1 - Parameters are estimated with the additional constraint that all item units have the same value q0; (ii) Cycle 2 - Parameters are estimated with item units allowed to vary, but with their mean value constrained to be equal to q0; (iii) Cycle 3 - Parameter estimates of Cycle 1 are corrected because of inconsistencies in the estimates of parameters generated from the JML procedure; and (iv) Cycle 4 - Parameter estimates of Cycle 2 are corrected for the same reasons stated in Cycle 3.  From the location estimates for each person and their associated standard errors, an index of (person) separation is constructed according to the expression: Separation Index = (V[b]-V[e])/V[b] (5) where V[e] is the average variance of the errors from estimates of b among persons. It can be interpreted as Cronbach’s alpha reliability coefficient, that is, as indicators of the items being responded to consistently by subjects or whether the set of items are measuring several different incoherent attributes. In particular, V[e] is a characteristic of a set of persons in the context of a particular set of items. Individuals are ordered according to the estimates of their locations on the continuum and grouped into three class intervals. The number of observed agree (A) responses to each statement by individuals in each class interval are then compared with the expected number according to HCM. This comparison is formalised as an approximate chi-square statistic for each item i, on G-1 degrees of freedom, according to the following equation:  ci2 = SGg=1[SnÎgxni-E(SnÎgxni)]2/V[SnÎgxni] (6)  where g = 1, 2.…G is the number of class intervals; Sxni, (nÎ g) is the total number of individuals in class interval g who agreed to statement i; E[Sxni], (nÎ g) is the expected number of A responses according to the model; and V[Sxni], (nÎ g) is the variance of expected number of A responses. Since the model is single-peaked, if responses are consistent with the model and if individuals are located around the location of the statement, then they too will be single-peaked (Andrich & Styles, 1998).

Results

Exploratory Factor Analysis of the Visual Arts Attitude Scale (Forms A and B) Initial analyses of data were conducted with principal component analysis followed by varimax rotation, in order to determine the factor structure of the Visual Arts Attitude Scale. A number of factors resulted from the principal components analysis that did not coincide with the hypothesised underlying unidimensionality of the instrument, but rather confirmed the argument stated by van Schuur and Kiers (1994) and van Schuur and Kruijtbosch (1995), that a factor analysis of responses to scales gave the impression that negative and positive statements with respect to a construct located on separate dimensions.  From the principal components analysis there were six components extracted with eigenvalues greater than unity. Since some items had substantial loadings on more than one factor, a varimax rotation was performed to elucidate the results. Tables 1 and 2 contain respectively the factor loadings (Form A) for the initial principal components analysis which resulted in the extraction of six factors, and from the varimax rotation of the three strongest factors.  Table 1. Principal Components Analysis of Visual Arts Attitude Scale Form A Item No. Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 13 11 12 16 14 01 15 17 18 19 02 20 03 04 05 06 07 08 09 10 -0.33 0.13 0.21 -0.47 -0.29 0.71* -0.52* -0.62* -0.61* -0.63* 0.75* -0.57* 0.67* 0.82* 0.75* 0.78* 0.61* 0.64* 0.68* 0.44* 0.46 0.43 0.31 0.66* 0.58* 0.44 0.43 0.45 0.46 0.38 0.21 0.48 0.14 0.15 0.37 0.38 0.39 0.11 0.07 0.43 0.23 0.17 -0.63* 0.27 -0.07 -0.24 0.29 0.06 -0.07 -0.07 -0.03 -0.13 0.22 -0.11 -0.34 0.12 -0.21 0.51 0.59 0.03 0.23 -0.69* -0.22 -0.10 -0.22 0.07 0.03 -0.27 0.35 -0.26 -0.01 0.26 0.30 0.13 0.04 0.18 -0.19 -0.22 -0.04 0.31 0.22 -0.23 0.32 -0.07 -0.48 0.11 0.51* -0.28 0.08 0.32 0.14 0.00 -0.16 0.01 0.18 -0.03 -0.13 0.35 0.00 -0.29 0.52* -0.12 0.02 -0.04 0.01 -0.30 -0.13 0.11 0.11 -0.22 0.35 0.14 0.22 -0.08 -0.11 0.04 -0.24 0.28 0.00 0.40 Total 7.02 3.16 1.57 1.23 1.18 1.02 %Variance 35.08 15.81 7.83 6.15 5.89 5.09 * Highest loadings of items contributing to each factor  The results from the two analyses presented in Tables 1 and 2 showed contrasting factor structures as several items that previously had recorded the highest factor loadings on a specific component, following the varimax rotation showed the highest factor loadings on an alternate component. For example, Item 12 initially had the highest factor loading (-0.63) on Factor 3, while following varimax rotation its highest factor loading was 0.58 on Factor 1. Similarly, Item 15 had its highest loading (-0.52) on Factor 1 and following varimax rotation its highest loading (0.70) was on Factor 2. Items 17, 18, 19 and 20 had high negative loadings (-0.62, -0.61, -0.63 and -0.57 respectively) on Factor 1, while following varimax rotation their highest positive loadings (0.72, 0.69, 0.64 and 0.67 respectively) were on Factor 2. Finally, Items 3, 8 and 9 had their highest factor loadings (0.67, 0.64 and 0.68 respectively) on Component 1 whereas following the varimax rotation their highest factor loadings (0.51, 0.75 and 0.83 respectively) were on Component 3. Table 2. Varimax rotation for Visual Arts Attitude Scale Form A with limited number of factors extracted Item No. Factor 1 Factor 2 Factor 3 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 0.85* 0.67* 0.47 0.71* 0.88* 0.72* 0.75* 0.30 0.28 0.55* 0.27 0.58* -0.05 0.16 -0.24 -0.05 -0.20 -0.14 -0.20 -0.07 -0.07 -0.24 -0.20 -0.35 -0.18 -0.08 -0.02 -0.15 -0.19 0.12 0.32 0.00 0.60* 0.62* 0.70* 0.84* 0.72* 0.69* 0.64* 0.67* 0.14 0.33 0.51* 0.28 0.06 0.49 0.19 0.75* 0.83* 0.26 0.23 -0.44 0.08 -0.15 0.05 0.06 -0.20 -0.31 -0.33 -0.34 Total 4.99 3.37 1.96 %Variance 24.97 16.88 9.82 * Highest loadings of items contributing to each factor  A principal components analysis was also undertaken on data collected from Form B of the Visual Arts Attitude Scale. From the analysis, five components were extracted with eigenvalues greater than unity. Examination of the factor loadings of each item, showed that further analysis using varimax rotation was necessary to clarify the results. Table 3 contains the factor loadings for the initial principal components.  Table 3. Factor analysis of responses to Visual Arts Attitude Scale Form B Item No. Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 08 17 16 03 13 09 10 01 15 14 18 19 02 20 11 12 04 05 06 07 -0.09 0.26 0.42 0.25 0.37 0.35 0.13 -0.07 0.66* 0.62* 0.74* 0.78* -0.55* 0.73* 0.59* 0.59* -0.67* -0.58* -0.74* -0.60* 0.34 0.11 0.30 0.45 0.61* 0.51* 0.47* 0.44* 0.53 0.16 0.31 0.04 0.54 0.01 0.30 0.07 0.50 0.50 0.48 0.36 0.19 0.45 0.11 -0.68* 0.28 0.07 -0.40 0.03 0.21 0.37 0.09 -0.39 -0.10 -0.35 -0.11 0.32 0.02 0.21 -0.08 0.08 0.43 -0.06 0.51* 0.30 0.07 -0.52 -0.30 -0.08 -0.01 0.00 0.33 0.25 0.34 0.16 -0.45 -0.28 0.30 -0.03 0.08 0.20 -0.53* 0.57* -0.40 0.04 0.12 -0.15 -0.11 0.38 0.13 -0.10 -0.08 -0.13 0.01 -0.12 -0.06 -0.19 -0.01 -0.15 0.04 -0.19 Total 5.51 3.53 1.95 1.65 1.13 %Variance 27.57 17.67 1.95 1.65 1.13 * Highest loadings of items contributing to each factor   Table 4 contains factor loadings resulting from the varimax rotation. The factor loadings demonstrated that Items 10, 11, 12, 14, 15, 16, 17, 18 and 8 contributed to a different factor from the one observed prior to varimax rotation. For example, Item 10 presented the highest loading (0.60) on Factor 3 as opposed to the highest loading (0.47) on Factor 2 presented prior to varimax rotation. Consequently, results extracted from the exploratory factor analysis did not facilitate interpretation of data obtained from the administration of the Visual Arts Attitude Scale (Forms A and B). In view of the results obtained by factor analyses conducted on responses from both Forms A and B, it appeared that there were more factors than desired, as the ideal instrument would need to conform to a unidimensional model for measuring a single attribute based on the analysis of individual responses. Therefore, in order to facilitate further understanding and interpretation of the data, a Rasch analysis was undertaken. Table 4. Factor loadings following a varimax rotation on Visual Arts Attitude Scale Form B Item No. Factor 1 Factor 2 Factor 3 01 03 10 09 12 13 14 15 16 17 18 04 05 06 07 08 11 02 19 20 0.18 -0.08 0.09 0.06 -0.32 0.16 -0.14 -0.14 -0.11 -0.02 -0.35 0.74* 0.79* 0.83* 0.70* 0.32* 0.48* 0.72* -0.66* -0.62* -0.05 0.11 0.18 0.57* 0.57* 0.74* 0.81* 0.81* 0.50* 0.45* 0.70* -0.18 0.02 -0.23 -0.13 0.24 0.46 -0.07 0.34 0.32 0.71* 0.84* 0.60* 0.24 -0.15 0.12 0.25 0.25 0.12 -0.30 0.18 0.03 -0.00 0.21 0.02 -0.00 0.04 0.29 0.47 0.41 Total 4.37 3.47 2.26 %Variance 21.85 17.34 11.29 * Highest loadings of items contributing to each factor     Rasch analysis of the Visual Arts Attitude Scale (Form A and Form B)  The Rasch measurement model is employed to produce a scale on which all items in an instrument and all respondents can be placed, and embodies the assumption that all responses are described by a response model. Particular interest is given to the item fit through the examination of the infit mean square (MNSQ) values which are calculated. Figure 2, displays the input file ‘att30.inp’ used in the Rasch analysis. Key codes were inserted so that the program identified items coded ‘1’ as having a positive attitude toward the visual arts and items coded ‘0’ as having a negative attitude.   Figure 2. Input file for Visual Arts Attitude Scales    However, in using this coding system it is difficult to determine the exact location of central items that involve an ambivalent response toward the visual arts. From this system, it is presumed that individuals who agree with the statements in items coded ‘1’ and disagree with statements in items coded ‘0’, are characterised as having demonstrated a positive attitude towards the visual arts. Moreover, students who disagree with the statements in items coded ‘1’ and agree with statements in items coded ‘0’, are subsequently characterised as having demonstrated a negative attitude toward the visual arts. Table 5 contains the item estimates for Forms A and B of the Attitude Scale.  Table 5(a). Item estimates* for Form A and Form B of the Visual Arts Attitude Scale (N = 131 Probability Level = 0.50) Form A ITEM No. SCORE MAXSCR THRSH SE INFT MNSQ OUTFT MNSQ INFT t OUTFT t 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 38 54 54 40 35 34 33 78 78 55 89 84 37 63 62 69 74 75 63 84 111 123 120 117 106 98 107 123 117 111 119 111 103 98 110 109 105 114 117 116 1.00 0.53 0.52 1.05 1.06 0.93 1.22 -0.47 -0.66 0.20 -1.13 -1.33 0.87 -0.60 -0.13 -0.53 -0.90 -0.61 -0.01 -1.02 0.23 0.21 0.22 0.23 0.24 0.25 0.25 0.22 0.23 0.22 0.25 0.25 0.24 0.25 0.23 0.23 0.25 0.23 0.22 0.25 1.02 0.76 0.85 0.80 0.92 1.13 0.94 0.99 0.85 1.47 1.45 1.53 1.11 1.13 0.93 0.87 0.83 0.82 0.79 0.87 0.97 0.70 0.75 0.69 0.77 1.01 0.97 0.91 0.66 2.06 1.61 3.70 1.13 1.22 0.79 0.80 0.66 0.68 0.66 0.70 0.20 -3.00 -1.80 -1.90 -0.60 1.10 -0.40 0.00 -1.50 5.00 3.10 3.40 1.00 1.20 -0.80 -1.40 -1.40 -1.80 -2.70 -1.10 -0.10 -1.80 -1.40 -1.60 -1.00 0.10 0.00 -0.30 -1.40 4.40 1.70 5.00 0.70 0.90 -1.00 -0.80 -1.20 -1.40 -2.00 -1.00 Mean     0.00   1.00 1.07 -0.20 -0.10 SD     0.84   0.24 0.71 2.10 1.90 Form B ITEM No. SCORE MAXSCR THRSH SE INFT MNSQ OUTFT MNSQ INFT t OUTFT t 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 31 56 25 42 73 66 64 42 51 83 60 51 71 64 79 64 76 88 88 84 124 121 111 106 102 106 107 90 109 117 107 120 116 106 116 100 113 121 120 119 1.61 0.46 1.75 0.82 -0.73 -0.28 -0.16 0.35 0.43 -0.73 0.00 0.64 -0.23 -0.19 -0.56 -0.38 -0.48 -0.80 -0.85 -0.69 0.23 0.20 0.25 0.22 0.24 0.22 0.22 0.23 0.21 0.22 0.22 0.21 0.21 0.22 0.22 0.23 0.22 0.22 0.23 0.22 0.98 0.90 1.11 0.75 0.87 0.75 0.88 1.40 1.64 1.35 1.31 0.83 1.17 0.85 0.79 0.95 1.02 0.81 0.77 0.82 1.11 0.86 1.24 0.67 0.71 0.65 0.85 1.43 2.16 1.59 1.31 0.86 1.30 0.79 0.68 1.07 1.16 0.68 0.61 0.66 -0.10 -1.30 0.80 -2.80 -1.20 -3.00 -1.50 4.00 6.60 3.10 3.40 -2.20 2.00 -1.70 -2.30 -0.50 0.20 -2.00 -2.20 -1.90 0.50 -1.00 0.90 -2.00 -1.30 -2.20 -0.80 2.30 5.60 2.40 1.80 -0.90 1.60 -1.20 -1.80 0.40 0.80 -1.60 -1.90 -1.80 Mean     0.00   1.00 1.02 -0.10 0.00 SD     0.76   0.25 0.40 2.60 2.00 * where SCORE = raw score; MAXSCR = maximum possible score; THRSH = item difficulty estimate; SE = standard error; and INFT (OUTFT) MNSQ = infit (oufit) mean squares. Bold indicates very high infit MNSQ values, a misfitting of the central items.   From the results it was observed that for Form A of the Visual Arts Attitude Scale Items 10, 11, and 12 presented very high infit MNSQ values of 1.47, 1.45, and 1.53 respectively, while for Form B, Items 8, 9, 10, and 11 presented infit MNSQ values of 1.40, 1.64, 1.35 and 1.31 respectively. A clearer view of these misfitting items is shown in Figures 3(a) and 3(b) where the infit MNSQ values are plotted for each item. These items fall outside the range of acceptance, or close to the margins of the established range (between 0.77 and 1.30). For both forms, misfitting items appear in the centre of the scale, indicating a difficulty in resolving negative responses related to the degree of affect expressed in these ambivalent central statements. This effect is a consequence of the structure of the scales, since the scales are designed to contain items indicating extremely positive or extremely negative attitudes on either end of the scale, and items with an ambivalent position in the centre. As a result, students with very positive attitudes could agree strongly with items on the top end of the scale and disagree strongly with items on the bottom end of the scale. However, students agreement or disagreement with central items of the scale, would be difficult to define as to whether students disagreed because of a stronger positive approach or because of a stronger negative approach toward the statement contained in the item (Andrich, 1996). For this reason, it was important to identify students whose negative responses to the central items were due to an overall positive attitude toward the visual arts, or students whose negative responses to central items were due to an overall negative attitude toward the visual arts. In other words, careful attention was necessary for interpreting negative responses given to items located in the centre of the Visual Arts Attitude Scale. A method of resolving the analysis and interpretation of such data was obtained from the Hyperbolic Cosine Model (Andrich & Luo, 1993) for unfolding single stimulus responses. Consequently, an analysis using the unfolding model could result in proving that the components generated from exploratory factor analyses and the inconsistency present in items demonstrated through Rasch analyses of equivalent forms of the Visual Arts Attitude Scale, were just another expression of describing items located at different parts of a single attitude continuum.    Figure 3(a). Plot of item fit for Form A of the Visual Arts Attitude Scale Figure 3(b). Plot of item fit for Form B of the Visual Arts Attitude Scale   HCM analysis of the Visual Arts Attitude Scale (Form A and B)  Since the HCM is single-peaked and symmetric about bn -di = 0, then for each of the possible values, there are two corresponding positions on the latent trait continuum. The assignment of a sign to each item is needed in order to initiate the estimation process. The sign analysis performed by RUMMFOLDss identifies pairs of items with the minimum correlation coefficient, so that the signs of correlation coefficients of items with either one of these items is selected as their initial sign. In this study, the sign analysis identified the lowest correlation between pairs of items being Item 2 and Item 19 (-0.53), with which the correlations between items were exported into the input card. Item 2 was chosen as the initial sign, representing the extreme positive attitude. Table 6 contains the correlation coefficients of the items with Item 2. The correlation matrix of the ranked items resulted in Item 2 and Item 19 being the two extreme opposite statements on the scale.  Table 6. Correlation coefficient exported with Item 2 (Form A) Item No. Statements Correlation coefficient 01 It is in the visual artistic experiences of life that I find my greatest satisfaction. 0.39 02  I have a great interest in matters of the visual arts. 1.00  03  I am interested in anything in which I can see a visual artistic quality. 0.42  04  Attendance at a visual arts exhibition gives me inspiration. 0.46  05  I believe that the pursuit of visual artistic interests increase one’s satisfaction in living. 0.50  06  I believe that visual arts promote desirable relationships between nations 0.17  07  I am attracted to individuals that pursue visual artistic interests. 0.49  08  I believe that everyone should have a little training in the visual arts. 0.37  09  I am in favour of visual arts exhibitions for they do no harm to anyone. 0.33  10  I believe that the teaching of the visual arts subject is O.K., but the type of person now teaching it fails to ‘get it across’. 0.08  11  Visual arts do not interest me now, but I expect that sometime I shall find time to pursue them actively. -0.07  12  I would be willing to give money to support visual arts enterprises if it were not for the ‘highbrow’ atmosphere surrounding them. 0.18  13  Practical considerations should come first, visual arts second. -0.26  14  I believe that individuals engaged in purely visual arts occupations are parasites on society. -0.14  15  I do not believe I would receive any benefit from lectures concerning visual arts subjects. -0.34  16  I do not care for visual artists because their interests seem to me to be more emotional than rational. -0.31  17  I see no reason for the government to spend money on aesthetic objects and activities. -0.28  18  I have no desire to join or have anything to do with any organisation developed to visual arts activities. -0.33  19  I see very little worth while in visual artistic interests. -0.53  20  Visual arts education is nonsense. -0.27 Bold indicates the smallest correlation coefficient found between Item 2 and Item 19.  After completion of the four cycles of the estimation, item locations, item units and associated standard errors per cycle were presented in the output. Table 7 contains the estimates of item units, with the four columns each corresponding to a different cycle of estimation. The first column, ‘theta1’, contains the estimates for item units, where all item units have the same value 3.06, or, all items have the same width of acceptance. The item unit value is 2.74 in the column ‘theta3’ which includes a correction of equal value to all item units for JML inconsistencies which stretch out the estimates of the parameters. In columns ‘theta2’ and ‘theta4’, the item units differ in value as they are allowed to float, as would be observed in the case of the Partial Credit Model (Masters, 1982), in which the item width varies. These item unit values are associated with corresponding standard errors for each cycle. Thetas can be regarded as proxy performance or attitude level parameters, since large thetas can be used to distinguish between items and individuals. Positive +q is the boundary set for the top end on the attitude continuum, while negative -q sets the boundary for the bottom end on the attitude continuum. From the results observed in column ‘theta4’, Items 9 and 12 appear to have the largest latitude of acceptance with theta values of 3.64 and 4.00 respectively, after allowing for correction. These items may be regarded as having good discrimination indices. Standard errors which in conjunction with a person’s location can also function as a discrimination index. These estimates are also provided by RUMMFOLDss. For example, if a person has a very negative attitude toward the visual arts, yet presents a high standard error, this shows an inconsistency in that person’s behaviour.  Table 7. Estimates of item units for each corresponding cycle of estimation (Form A) Item No. theta1 theta2 theta3 theta4 StdEr1 StdEr2 StdEr3 StdEr4 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 3.06 2.55 2.72 2.73 2.31 2.44 2.53 2.15 3.40 4.02 2.56 3.85 4.22 2.16 3.33 3.18 3.22 3.59 3.53 2.92 3.76 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.74 2.26 2.38 2.40 2.02 2.13 2.22 1.84 3.04 3.64 2.24 3.54 4.00 1.85 3.03 2.84 2.91 3.29 3.19 2.59 3.43 0.24 0.21 0.21 0.23 0.23 0.24 0.23 0.21 0.21 0.21 0.23 0.24 0.22 0.22 0.21 0.22 0.23 0.21 0.21 0.22 0.24 0.22 0.22 0.23 0.24 0.23 0.23 0.21 0.23 0.20 0.24 0.28 0.21 0.22 0.21 0.21 0.23 0.22 0.20 0.23 0.23 0.21 0.21 0.23 0.23 0.23 0.22 0.21 0.21 0.21 0.22 0.22 0.22 0.21 0.21 0.21 0.22 0.21 0.21 0.21 0.23 0.21 0.21 0.22 0.23 0.23 0.22 0.21 0.22 0.20 0.24 0.27 0.21 0.22 0.21 0.21 0.23 0.21 0.20 0.23 Bold indicates items with the largest latitude of acceptance  A generalised c2 test of fit is performed for the fourth cycle, with different item units with correction, including three person-groups or class intervals because of the single-peaked structure of the HCM function. The three class intervals correspond to three person-groups, ordered according to the estimates of their locations on the continuum and represent three graded responses. The number of observed responses to each statement by individuals in each class interval, are compared to the expected number of responses according to the HCM, resulting in the calculation of an approximate c2 statistic for each item. The reason for selecting different unit values for analysis of the fit statistic, lies in the nature of the instrument and context of the items. Attitude items express diverse emotive intensity for each person, as they elicit highly subjective responses. As a consequence, the latitude of acceptance is allowed to float and vary according to the sample, and is not restricted after implementation of a single unit value for all items. Table 8 shows the observed and expected agreements and disagreements for three class intervals of an extreme positive statement (Item 1), a fairly negative statement (Item 14) and an extreme negative statement (Item 20). Figures 3, 4 and 5 provide the location and the theta value (unit) for Items 1, 14 and 20.   Table 8. Expected and observed responses for Item 1, Item 14 and Item 20 Item/ statement Class Interval c2 (2, N = 131) p 1 2 3 PA1 Obs. Exp. 0 37.00 33.95 1 3.00 6.05 0 20.00 22.30 1 17.00 14.70 0 10.00 12.34 1 21.00 18.66 c2 1.946 0.610 1.042 3.60 0.14 NA14 Obs. Exp. 0 16.00 14.15 1 18.00 19.85 0 20.00 19.43 1 10.00 10.57 0 27.00 30.06 1 4.00 0.94 c2 0.442 0.048 10.923 11.41 0.00 NA20 Obs. Exp. 0 21.00 17.22 1 15.00 18.78 0 24.00 29.22 1 13.00 7.78 0 40.00 39.37 1 0.00 0.63 c2 1.78 4.49 0.66 6.94 0.03 Note. NA = negative worded attitude statement; PA = positive worded attitude statement; Obs. = observed responses; Exp. = expected responses  In Table 8, the expected and observed number of responses (score) and the chi-square value for each item are displayed. Item 1 has 37 observed negative responses and 3 observed positive responses from the first class interval, the group with an extremely negative attitude. This item also has 20 observed negative responses and 17 positive responses from the second class interval, the group with an ambivalent attitude toward the visual arts. Finally, the third interval, representing individuals with an extreme positive attitude toward the visual arts gave 10 negative responses and 21 positive responses to Item 1. The number of observed responses does not differ significantly from the expected responses, an indicator of fit of this item to the expected or ideal model. Figure 4 contains the plot of the average value for each group and the related observed proportion of positive responses for Item 1. Observed proportions are close to the curve indicating that the item fits the model. The rectangle presents the latitude of acceptance of the item, and it is evident that Item 1 is an extremely positive statement. From Figure 4, Item 1 location is 2.28 with a unit value of q = 2.28, which demonstrates it is a highly positive statement placed close to the top of the attitude continuum. It has a chi-squared value of 3.60 for two degrees of freedom and fits the model well.   Figure 4. Observed responses, latitude of acceptance and location for Item 1 (Form A) Figure 5 shows a plot of the average value for each group and the related observed proportion of positive responses for Item 14. Observed proportions for the third group are not close to the curve, indicating that this item does not fit the model. The rectangle presents the latitude of acceptance of the item, indicating that Item 14 is a negative statement.   Figure 5. Observed responses, latitude of acceptance and location for Item 14 (Form A) Figure 6 contains a plot of the average value for each group and the related observed proportion of positive responses for Item 20. Observed proportions for the first and second group are not close to the curve indicating that this item does not fit the model. The rectangle presents the latitude of acceptance of the item and shows that Item 20 is a negative statement.   Figure 6. Observed responses, latitude of acceptance and location for item 20 (Form A). For the three items presented above, the response patterns across the class intervals are monotonic. As the size of the difference between the locations of individuals and of items increases, the probability of an agree response decreases. Table 9 contains a summary of the goodness of fit test while listing the items in location order. It is also interesting to note that the original order of items has been changed to reflect their location on the attitude continuum. From an analysis using a three class interval model, two statements, Item 14 and Item 20, with large fit statistics showed probability values of less than 0.05 thus misfitting the model. Item 14 has a single peaked structure, but it overdiscriminated relative to other statements such that the third interval has a lower number of observed disagreements than expected as well as a higher number of observed agreements than expected. Thus the item has 16 observed negative responses as opposed to 16 expected, and 18 observed positive responses as opposed to 20 expected from the first class interval, the group with the extremely negative attitude. The same item has 20 observed negative responses as opposed to 19 expected, and 10 observed and expected positive responses from the second class interval, the group with an ambivalent attitude toward the visual arts. However, the third interval, representing individuals with an extreme positive attitude toward the visual arts, gave 27 negative responses as opposed to the expected 30, and 4 positive responses as opposed to the expected 1. The latter inconsistency caused an inflated chi-squared value of 10.92 with a significant probability value of 0.00, indicating a significant difference between expected and observed responses in the third interval, and verifying a problem with the fit. The item location was -2.66 with a unit value of 1.62, demonstrating that it is a fairly negative statement placed closer to the lower end of the attitude continuum.   Table 9. Summary of goodness of fit test for Form A with three and four class interval models Item Statements Location Unit Three class interval Four class interval No.       c2 Deg. Signific. c2 Deg. Signific. 17 I see no reason for the government to spend money on aesthetic objects and activities. -4.40 2.68 1.62 2 0.37 1.93 3 0.59 18 I have no desire to join or have anything to do with any organisation developed to visual arts activities -4.16 2.85 2.53 2 0.22 1.17 3 0.76 16 I do not care for visual artists because their interests seem to me to be more emotional than rational. -4.14 2.84 1.52  2  0.37  0.64  3  0.89  20  Visual arts education is nonsense.  -3.63  1.89  6.94  2  0.03  2.52  3  0.47  19  I see very little worth while in visual artistic interests.  -3.17  2.65  0.74  2  0.61  4.13  3  0.25  15  I do not believe I would receive any benefit from lectures concerning visual arts subjects.  -3.00  2.42  0.08  2  1.0  0.94  3  0.82  14  I believe that individuals engaged in purely visual arts occupations are parasites on society.  -2.66  1.62  11.41  2  0.00  4.11  3  0.25  13  Practical considerations should come first, visual arts second.  -2.55  3.20 2.79 2  0.22 4.58 3 0.21  11  Visual arts do not interest me now, but I expect that sometime I shall find time to pursue them actively.  0.14  0.36  2.15  2  0.37  4.20 3  0.24  10  I believe that the teaching of the visual arts subject is O.K., but the type of person now teaching it fails to ‘get it across’. 0.86  2.25  0.67  2  0.61  1.86 3  0.60  12  I would be willing to give money to support visual arts enterprises if it were not for the ‘highbrow’ atmosphere surrounding them. 1.22  0.81  1.43 2 0.61  1.19 3  0.76 06  I believe the visual arts promote desirable relationships between nations. 1.76  2.02  0.93  2  0.61  1.00 3  0.80  01  It is in the visual artistic experiences of life that I find my greatest satisfaction.  2.28  2.28  3.60  2  0.14  2.63  3  0.45  05  I believe that the pursuit of visual artistic interests increase one’s satisfaction in living.  2.30  2.24  3.08  2  0.22  4.62  3  0.20  07  I am attracted to individuals that pursue visual artistic interests.  2.65  2.30  2.11  2  0.37  3.85  3  0.28  04  Attendance at a visual arts exhibition gives me inspiration.  2.87  2.60  3.28  2  0.22  6.83  3  0.08  08  I believe that everyone should have a little training in the visual arts.  3.18  4.55  1.04  2  0.61  1.28  3  0.73  03  I am interested in anything in which I can see a visual artistic quality.  3.21  3.44  1.06  2  0.61  2.24  3  0.53  09  I am in favour of visual arts exhibitions for they do no harm to anyone.  3.47  5.07  1.91 2  0.37  2.28  3  0.52  02  I have a great interest in matters of the visual arts.  3.78  3.86  1.73  2  0.37  4.53  3  0.21  Total        50.62  38  0.07  56.52  57  0.49 Bold indicates misfitting items for Form A for three class intervals and the same items improvement in fit after adjustment was made by increasing the number of class intervals from three to four.    Item 20 was another item found to have similar problems. This item had 21 observed negative responses as opposed to 17 expected, and 15 observed positive responses as opposed to 19 expected from the first class interval. The same item had 24 observed negative responses as opposed to 29 expected, and 13 positive responses as opposed to 8 expected from the second class interval. Finally, the third interval gave 40 observed negative responses as opposed to 39 expected and 0 positive responses as expected. The second class interval presented a statistically significant difference between the number of observed and the number of expected responses. The item location was -3.63 and had a unit value of 1.89, indicating that it was a highly negative statement placed close to the lower end of the attitude continuum. An inflated chi-squared value of 6.94 and the significant probability value of 0.03, demonstrated that Item 20 was sensitive at this extreme. From the chi-squared and probability values it was evident that all other items apart from Item 14 and Item 20 fitted the HCM. In order to investigate further the misfitting items, an adjustment was made in the model by increasing the number of class intervals from three to four (Table 9). It was observed that the chi square values for Items 14 and 20 did not appear inflated after this adjustment, but rather, all items presented a good fit.  Figure 7 shows a plot of the statements’ locations and respondents’ frequency. An obvious gap was observed in the continuum, with items being polarised at either extreme corresponding to the negative and positive statements. Andrich (1996) has argued that when statements from Likert type questionnaires are located on a continuum, based on Thurstone’s principles, they tend to be located at extremes with a gap in the middle, whereas Thurstone-type statements are required to be located more or less uniformly on the continuum. This contrasts with the present findings, where Thurstone-type positively worded statements were located uniformly across the continuum while negatively worded statements formed a gap between 0.00 and &endash;3.00 logits. This indicated that there was an uneven distribution of items in the negative extreme. However, the locations of individuals on the same continuum were primarily between the locations of positive and negative statements. A slightly larger number of individuals was located among the negative statements than those among positive statements, suggesting that there were more individuals with a negative attitude than a positive attitude toward the visual arts.   Figure 7. Frequency plot of persons with item locations for Form A (different units with correction) The same procedure was repeated during the analysis of the data originating from Form B of the Visual Arts Attitude Scale. A sign analysis identified Item 4 and Item 12 as the two extreme opposite items on the attitude continuum with a correlation coefficient of -0.49. The correlations of items were exported with Item 4, being the extreme positive statement. Table 10 contains the correlation coefficients of the items with Item 4.  Table 10. Correlation coefficients exported with Item 4 (Form B)  Item No. Statements Correlation coefficient  01  I find more satisfaction in artistic pursuits than in anything else. 0.37  02  I like art works because they give me genuine pleasure. 0.55  03  I believe that the great leaders of the world come from the ranks of those individuals who are artistically inclined. 0.28  04  Appreciation of the visual arts aids in making my life happier. 1.00  05  I believe that artistically sensitive people are fine people. 0.40  06  I believe that artistic pursuits are satisfying. 0.48  07  Artistic interests are not essential but make for happy existence. 0.49  08  I believe in the value of artistic interests but I do not like the stilted way in which the ideas on this subject are presented to me. -0.05  09  I believe in the value of artistic interests but I seldom take time to pursue them. -0.32  10  I go to such things as art exhibitions, installations etc., occasionally, but I have no strong liking for them. 0.07  11  Sometimes I believe that artistic interests are necessary and sometimes I doubt it. -0.16  12  The visual arts do not play an especially large part in my life. -0.49  13  I can enjoy the beauty of such things as paintings, sculpture and prints only occasionally, for I feel they are impractical. -0.08  14  I have no interest in visual aesthetic objects (such as fine paintings and pottery) because I do not understand their technical aspects. -0.32  15  I find the life of people pursuing artistic interests too slow and uninteresting. -0.22  16  The ‘highbrow’ attitude of individuals having a great deal of artistic interest is quite distasteful. -0.10  17  I believe that artistic interests are rarely genuine and sincere. -0.08  18  It is hard for me to understand how anybody can be stupid enough to concentrate all his/her energies on artistic activities. -0.24  19  Education in artistic things is a waste of public funds. -0.39  20  The pursuit of visual arts activities is a sheer waste of time. -0.36 Bold indicates the smallest correlation coefficient found between Item 4 and Item 12.  Table 11 contains the theta estimates of parameters resulting from the four estimation cycles, after which the item locations and item units were calculated. From these results, it appears that Items 5 and 6 have the largest latitude of acceptance with theta values of 4.44 and 4.10 respectively. These items have good discrimination indices as they have a broad bandwidth.  Table 11. Theta estimate parameters for Form B  Item No.  theta1  theta2  theta3  theta4  StdEr1  StdEr2  StdEr3  StdEr4  01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20  2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56 2.56  1.97 3.12 0.78 3.12  4.24 4.13 3.78 1.53 2.14 0.59 1.36 3.71 1.20 3.43 2.27 2.38 1.70 2.85 3.95 2.92  2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28 2.28  0.60 2.88 0.39 2.63  4.44 4.10 3.85 1.69 2.04 0.75 1.50 3.33 1.30 2.45 1.94 2.03 1.60 2.48 3.00 2.54  0.23 0.21 0.24 0.21 0.22 0.22 0.22 0.20 0.20 0.22 0.20 0.20 0.20 0.21 0.22 0.21 0.21 0.23 0.26 0.23  0.23 0.22 0.22 0.22 0.24 0.23 0.23 0.20 0.21 0.21 0.20 0.22 0.20 0.22 0.22 0.22 0.22 0.24 0.25 0.23  0.23 0.21 0.24 0.21 0.22 0.21 0.21 0.20 0.20 0.22 0.20 0.20 0.20 0.21 0.22 0.21 0.21 0.24 0.25 0.23  0.22 0.23 0.22 0.23 0.25 0.24 0.23 0.20 0.21 0.21 0.20 0.23 0.20 0.22 0.23 0.22 0.22 0.24 0.25 0.24 Bold indicates items with the largest latitude of acceptance Table 12 shows the fit statistics for Form B. Overall, the items in this form have a good fit, although Item 6 presents a marginally high chi-squared value of 5.85 (p = 0.049).  Table 12. Summary of goodness of fit test for Form B with three and four class interval models  ItemNo.  Statements  Location  Unit  Three class interval  c2  Deg.  Significance  19  Education in artistic things is a waste of public funds.  -4.10  3.00  1.44  2  0.61  18  It is hard for me to understand how anybody can be stupid enough to concentrate all his/her energies on artistic activities.  -3.42  2.48  0.12  2  1.00  20  The pursuit of visual arts activities is a sheer waste of time.  -3.29  2.54  3.51  2  0.14  15  I find the life of people pursuing artistic interests too slow and uninteresting  -2.29  1.94  2.86  2  0.22  14  I have no interest in visual aesthetic objects (such as fine paintings and pottery) because I do not understand their technical aspects  -2.25  2.45  0.83  2  0.61  16  The ‘highbrow’ attitude of individuals having a great deal of artistic interest is quite distasteful.  -2.11  2.03  4.74  2  0.08  12  The visual arts do not play an especially large part in my life.  -1.96  3.33  0.05  2  1.00  17  I believe that artistic interests are rarely genuine and sincere.  -1.54  1.60  2.01  2  0.37  09  I believe in the value of artistic interests but I seldom take time to pursue them.  -1.01  2.04  3.77  2  0.14  13  I can enjoy the beauty of such things as paintings, sculpture and prints only occasionally, for I feel they are impractical  0.02  1.30  0.19  2  1.00  08  I believe in the value of artistic interests but I do not like the stilted way in which the ideas on this subject are presented to me.  0.27  1.69  2.04  2  0.37  11  Sometimes I believe that artistic interests are necessary and sometimes I doubt it.  0.45  1.50  0.53  2  0.61  10  I go to such things as art exhibitions, installations etc., occasionally, but I have no strong liking for them  0.87  0.75  0.96  2  0.61  03  I believe that the great leaders of the world come from the ranks of those individuals who are artistically inclined  1.05  0.39  0.26  2  1.00  01  I find more satisfaction in artistic pursuits than in anything else.  1.53  0.60  0.51  2  0.61  02  I like art works because they give me genuine pleasure.  3.41  2.88  4.56  2  0.08  05  I believe that artistically sensitive people are fine people.  3.42  4.44  1.58  2  0.37  07  Artistic interests are not essential but make for happy existence  3.57  3.85  2.93  2  0.22  04  Appreciation of the visual arts aids in making my life happier.  3.65  2.63  0.80  2  0.61  06  I believe that artistic pursuits are satisfying.  3.75  4.10  5.85  2  0.05  Total        39.54  38  .32 Figure 8 contains the frequency plot of persons with item locations for Form B. This frequency plot revealed that gaps were located amongst the statements between +1.00 and +3.00 logits, 0.00 and &endash;1.00 logit and &endash;2.00 and &endash;3.00 logits. This is a disturbing result considering the expected uniformity found in Thurstone-type item-distributions. In addition, contrary to observations made for data collected with Form A, the majority of persons (according to the statements endorsed from Form B) are located among the statements reflecting a positive attitude toward the visual arts and fewer persons are located among more negative statements. Finally, the equivalent forms reliability coefficients were estimated, which indicated that with this sample of students, the two forms were operating in an equivalent way to the extent of a correlation of 0.76 between Form A and Form B and very similar Cronbach alpha coefficients of 0.87 and 0.85 for Forms A and B respectively.   Figure 8. Frequency plot of persons with item locations for Form B (different units with correction)

Discussion

Binary graded disagree-agree responses to attitude items for the purpose of attitude measurement toward the visual arts, were collected and examined using three methods of analysis. The first method used was exploratory factor analysis in an attempt to determine the factor structure and provide a framework for understanding the network of relationships among measures included in the scale. The exploratory factor analysis did not provide adequate information to facilitate interpretation of data. Conversely, such traditional cumulative methods of analysis of responses to Thurstone scales have proved to be misleading, giving the impression of statements with respect to a construct are on separate dimensions (Van Schuur & Kiers, 1994; Van Schuur & Kruijtbosch, 1995). The second method used was Rasch analysis of the Attitude Scale, particularly employing the partial credit model. Results obtained from the analyses implied an additional difficulty in explaining non-fitting items located in the middle of the attitude continuum. Observed non-fitting items appeared in the centre of the scale, indicating a difficulty in discriminating between positive and negative attitudes towards the visual arts. Because of the necessity of identifying and interpreting students’ negative responses to central items on the continuum, a third analysis was conducted using the Hyperbolic Cosine Model (Andrich & Luo, 1993) for unfolding single stimulus responses. The HCM calculated responses to follow the Rasch model for three categories. Thus, the manifest dichotomous response (Agree/Disagree) was resolved into three latent responses &endash; Disagree Below (DB) because a person was below the statement, Disagree Above (DA) because the person was above the statement and Agree (A) because the person was close to the statement. In this HCM analysis, an unfolding model was applied to the latent responses, in which a direction was implied. Two response curves which characterised the directions of the Disagree response, were unfolded to correspond to the data. So that the dichotomous observed data is unfolded by the HCM, in order to explain the Disagree response in terms of its constituent elements. The HCM described an effect that was expected as a function of the distances of persons from the statements. Furthermore, from the location estimate of each person and the associated standard error, an index of (person) separation was constructed, which is an indicator of items being answered either inconsistently by respondents, or of items measuring different incoherent attributes. From the results obtained using the HCM, items that did not fit this model were detected, while the location of each item and each person on the attitude continuum was clearly interpreted. Item locations and unit values were of particular interest in demonstrating a lack of uniformity in certain areas of the distribution of statements along the continuum. However, the results did provide person location on the attitude continuum, hence an accurate determination of an individuals’ degree of affect toward the visual arts was derived, which may be used for further analyses in multilevel and multivariate studies. The information obtained supported the underlying unidimensionality of the measurement model employed and was particularly useful in the selection of items that in fact fitted the model and had good discrimination indices. Furthermore, the item location could be accurately estimated and thus statements could be put in location order on the scale. Single-peaked response functions have been rarely used in substantive research although modern computing algorithms have overcome the problems of time-consuming and laborious traditional analysis of single-peaked response data. This study has demonstrated how a single-peaked response function can be employed for the development of a scale of attitude measurement toward the visual arts. Nevertheless, single-peaked response functions have further implications for the investigation of stage developmental constructs (Inhelder & Piaget 1958), cognitive stages and the application of models in psychological, social and educational development and measurement (Leik & Matthews, 1968; Birnbaum, 1968; Coombs & Smith, 1973; Andrich & Styles, 1998).

   

References

Adams, R. J & Khoo S. T. (1993) QUEST: The Interactive Test Analysis System. Hawthorn, Victoria: Australian Council For Educational Research. Andrich, D. (1978) Relationships between the Thurstone and Rasch approaches to item scaling. Applied Psychological Measurement. 2, 449-460. Andrich, D. (1982). An extension of the Rasch model for ratings providing both location and dispersion parameters. Psychometrika. 47, 105-113. Andrich, D. (1995). Distinctive and Incompatible Properties of the two common classes of the IRT Models for graded responses. Applied Psychological Measurement, 19(1), 101-119. Andrich, D. (1996). A hyperbolic cosine latent trait model for unfolding polytomous responses: Reconciling Thurstone and Likert methodologies. British Journal of Mathematical and Statistical Psychology. 49, 347-365. Andrich, D., & Luo, G. (1993). A hyperbolic cosine latent trait model for unfolding dichotomous single-stimulus responses. Applied Psychological Measurement. 17, 253-276. Andrich, D., de Jong, J. H. A., & Sheridan, B.E. (1997). Diagnostic opportunities with the Rasch model for ordered response categories. In J. Rost & R.Langeheine (Eds.). Applications of latent trait and latent class models in the social sciences (pp. 58-68). Munster, Germany: Waxmann Verlag. Andrich, D., & Styles, I.M. (1998). The structural relationships between attitude and behaviour statements from the unfolding perspective. Psychological Methods. 3(4), 454-469. Bock, R.D. (1972). Estimating item parameters and latent ability when responses are scored in two or more nominal categories. Psychometrika. 37, 29-51. Bock, R.D., & Jones L.V. (1968). The measurement and prediction of judgement and choice. San Francisco: Holden Day. Birnbaum, A. (1968). Some latent trait models and their use in inferring an examinee’s ability. In F.M. Lord & M.R. Novick, Statistical Theories of Mental Test Scores (pp. 397-479). Reading MA: Addison-Wesley. Cohen, J.A. (1941). A scale for the measurement of attitude toward the aesthetic value. J.Psychol., 12, 75-79. Coombs, C. H. (1964). A Theory of Data. New York: Wiley. Coombs, C.H., & Smith, J.E. (1973). On the detection of structure in attitudes and developmental process. Psychological Review, 80, 337-351. Donoghue, J. R. (1994). An examination of the IRT information of polytomously scored reading items under the generalised partial credit model. Journal of Educational Measurement. 41, 295-311. Dubois, B., & Burns, J.A. (1975). An analysis of the meaning of the question mark response category in attitude scales. Educational and Psychological Measurement. 35, 869-884. Guttman, L. (1950). The basis for scalogram analysis. In S.A. Stouffer, L. Guttman, E. Suchman, P.E. Lazarsfeld, S.A. Star, & J.A. Gardner (Eds.). Measurement and prediction (pp. 60-69). Princeton NJ: Princeton University Press. Inhelder, B., & Piaget, J. (1958). The growth of logical thinking from childhood to adolescence. London: Routledge & Kegan Paul. Leik, R.K., & Matthews, M. (1968). A scale for the development processes. American Sociological Review, 33, 62-75. Likert, R. (1932). A technique for the measurement of attitudes. Archives of Psychology: No 140. Masters, G.N. (1982). A Rasch model for partial credit scoring, Psychometrika. 47, 149-174. Roberts, J.S. (1995). Item response theory approaches to attitude measurement. (Doctoral dissertation, University of South Carolina, Columbia, 1995). Dissertation Abstracts International. 56, 7089B. Roberts, J.S., Laughlin, J.E.M & Wedell, D.H. (1999). Validity issues in the Likert and Thurstone approaches to attitude measurement. Educational and Psychological Measurement, 59 (2), 211-233. Silance, E.B., & Remmers, H.H. (1934). An experimental generalised master scale: A scale to measure attitudes toward any school subject. Purdue Univ. Stud. High. Educ., XXVI, 35, 84-88. Thissen, D.M. (1976). Information in wrong responses to the Raven’s Progressive Matrices. Journal of Educational Measurement. 13, 201-214. Thurstone, L.L. (1928). Attitudes can be measured. The American Journal of Sociology. 33, 529-554. Thurstone, L.L. (1928). The measurement of social attitudes. Journal of Abnormal and Social Psychology. 26, 249-269. Van Schuur, W. H., Kiers, H.A.L. (1994). Why factor analysis is often the wrong model for analysing bipolar concepts and what model to use instead. Applied Psychological Measurement. 18, 97-110. Van Schuur, W.H., & Kruijtbosch, M. (1995). Measuring subjective well being: Unfolding the Bradburn Affect Balance Scale. Social Indicators Research. 36, 49-74.   International Education Journal, 1 (2) 2000 http://iej.cjb.net