Sie sind auf Seite 1von 5

The Muldimensionality of Test Anxiety: A Test of Different Models

Volker Hodapp and Jeri Benson

Purpose of the research


This research was conducted in order to better define the borders of text anxiety, to see which factors explain anxiety the best and which less. The motive behind the research was to see if "the dimensions of text anxiety which have been investigated previously adequately define the domain, or is it possible that additional dimensions which have been previously regarded as correlates of test anxiety be considered as dimensions within the content domain of test anxiety?" In order to answer this bigger question the researchers tested several multidimensional factor models of test anxiety starting with the traditional two factor model to a five factor model.

The models
4 different models were checked by the researchers from a two factor model to a five factor model. For each model both first and second order structures were checked. Model 1: Two factors of Worry and Emotionality Model 1a: Second-Order Structure of Model 1 Model 2: Three factors of Worry, Emotionality and Distraction Model 2a: Second-Order Structure of Model 2 Model 3: Four factors of Worry, Emotionality, Distraction and Lack of Confidence Model 3a: Second-Order Structure of Model 3 Model 4: Five factors of Worry, Emotionality, Distraction, Lack of confidence and Self-Efficacy. Model 4a: Second-Order Structure of Model 4 Model 4b: Two Second-Order Factors of Model 4. Following the analysis, two post-hoc models (5 and 5a) were proposed by the researchers at the phase of Model Modification. The post hoc models were based on the 3rd Model which appeared to fit the data best but excluded distraction from the model whose standard path coefficients were relatively lower and standardized residuals higher. While all of the models proposed by the researchers had a theoretical basis (except for the post-hoc models that were developed following the findings of this research) not all of them were tested previously by factorial models. The first model which explains test anxiety by two factors, `Worry` and `Emotionality` was based on the model of Liebert and Morris (1967) who identified worry as any cognitive expression of concern about one's performance and emotionality referred to autonomic reactions which occur under examinations stress. Spielberger, Gonzales, Taylor,Algaze and

Anton developed the Test Anxiety Inventory (TAI) to refine the dimensional concept of test anxiety. Several factor analyses of the TAI in different populations including American, Egyptian, German and Brazilian provided strong support for the two factor model. The second model adds to the first model the dimension of Distraction which appears differently in different researches prior to this one. The idea of distraction came up in Sarason's four factor model which he developed due to the lack of distinctiveness between worry and emotionality. Sarason divided the cognitive component into worry and test irrelevant thinking (distraction in the current research) and emotionality into Tension and Bodily symptoms. To test his model he developed the Reactions to Tests (RTT) scale. The RTT was used by several researchers afterwards; Benson and Bandalos, Zimmer, Hocevar Bachelor and Meinke and it was seen that there were problems with the construct validity of the scale thus the scales were further improved and combined by TAI (Benson et al. 1992) in order to explain the data better. Even after improvements it was seen that Test-Irrelevant Thinking was half the magnitude of the other three coefficients (worry, tension and bodily symptoms). In the third model the distinction was made with the addition of the "Lack of Confidence" dimension. The addition of this dimension arose from the need to reconceptualize and expand the content domain of the TAI-G. Items regarding lack of confidence had been previously included in many tools even Liebert and Morris's two factor model where it was included as a worry item. In 1964 in Sassenrath's work it was indeed added as a separate factor. There were other researchers who included it as a separate variable (Heckhausen, Salame, Stephan, Fischer and Stein). In the fourth model , self efficacy was also added to the previous factors due its obvious links to worry and confidence. "Self efficacy relates to the belief of being able to control environmental demands by means of instrumental activity" Thus individuals who feel confident also believe in themselves that they will achieve better. The relationship between selfefficacy and test anxiety is two way in that people who are more anxious lack self efficacy and people whose self efficacy is lower are more test anxious. Although self-efficacy has a theoretical contribution to the construct of test anxiety, in the present research there is no information on if it has yet been tested in a factorial model.

Tools
In the research three different tools were used to measure the different dimensions of test anxiety. 1. The Revised Test Anxiety Scale (RTA) (Benson et al., 1992) which constituted of 23 items tapping on the different dimensions: Worry (6), Tension (5), Test Irrelevant Thinking (6) and Bodily Symptoms (6).

2. German Test Anxiety Inventory (TAI-G) (Hodapp, 1991) which constituted of 28 items with focus on slightly different dimensions: Worry (10), Emotionality (6), Cognitive Interference (6) and Lack of Interference (6). 3. To measure self efficacy a generalized self efficacy was administered which contained 9 items. All three measures were based on a four point Likert rating scale.

Fit of the Models


After the scales were put together they were analyzed separately with exploratory and confirmatory analysis in order to determine which items failed to fit the four-factor structure of each scale. The reason they were analyzed separately, as explained above, is that although both included 4 dimensions they were slightly different in the two tests. After the EFAs and CFAs were run until acceptable model-data fit was obtained, remained 17 items of the 28 TAI-G items and 12 of the 23 RTA items. The self-efficacy scale was likewise analyzed and 5 items were reduced following EFA since they did not form a unidimensional scale. The remaining 29 items overall were used to create item parcels. The items parcels were created based on the content wording of the items and the item inter-correlations. Over all 14 item parcels were created. The competing models were tested by a covariance matrix of the 14 item parcels. For all of the models, chi-square test of fit, associated p-value and the x2/df ratio was found. Additionally, for each model Goodness of Fit Index (GFI), Root Mean Square Residual (RMSR), Comparative Fit Index (CFI) and Root Mean Square Error of approximations (RMSEA) were reported. The first order model of the first set containing the two factors Worry and emotionality was rejected since its overall fit in the confirmatory sample was not adequate x2/df = 3.00. Model 1 a which was a second order model was initially under-identified and when it could be identified with the fixation of the path from test anxiety to Worry, it turned out to be identical to Model 1 in the exploratory and confirmatory samples. In the second model although degrees of freedom increased thus providing more flexibility, in the first order model the x2/df ratio was still high (2.38) in the exploratory sample resulting in an inadequate fit. Model 2a which was the second order model of Model 2 was equivalent to it since the same number of parameters where being estimated in both models. Of all the models tested, Model 3 containing the factors, Worry, Emotionality, Distraction and Lack of Confidence was found to fit the data best according to the overall model data fit in that the p-values were non significant and the confidence intervals included zero. They were also the models with the best x2/df ratios 1.81 and 1.88 Model 3 and 3a

respectively in the confirmatory analysis. Amongst the two options in model 3, the first order and the second order it was seen that the second order model fit the data better in that path coefficients for all factors except Distraction were high and the standardized residuals for each of them except Distraction were low. Seeing the effect of Distraction on the four factor Model 3, the researchers developed 2 post-hoc models; Model 5 and 5a which constituted of the factors Worry, Emotionality and Lack of Confidence. It was seen that Model 5 which was a first order three factor model with the x2/df ratio equal to 1.83 was one of the best fitting first order models in the models that were tested. Regarding model 5a which was the second order version of Model 5, since the number of parameters which were estimated was the same, the fit was equivalent to that of model 5. Of interest was to see if the standardized path coefficients and standardized residuals had improved in the model and it was seen that it actually had. The standardized path coefficients and the standardized residuals(in parentheses were as follows in the confirmatory samples Worry .77(.41), Emotionality .92 (.16) and Lack of Confidence .63(.60). When this new three factor model was compared to the previous three factor model, Model 2, this new model is preferred over Model 2 for a couple reasons. First of all while for Model 2 x 2/df ratio was inadequate (2.38) in the fifth model the ratio decreased to 1.83 which provides for a much better fit. Second, the p-value statistic of the fifth model was .308 which means that the models 5 and 5a cannot be rejected. Additionally the 90% confidence interval in RMSEA included zero which also implies that the models cannot be rejected. Finally, while in model 2a the worry dimension was over-estimated it was not so in model 5a. Based on the above findings it can be said that Lack of Confidence was found to be a more relevant factor than Distraction in the construct of test anxiety. Regarding the 4th model which included 5 factors Worry, Emotionality, Distraction, Lack of confidence and Self-Efficacy, it was seen that both in exploratory and confirmatory samples the x2/df ratio was reasonable 1.43 and 1.76 respectively. But in model 4a which is a second order factor model of Model 4 the x2/df ratio deteriorated to 2.26 and the RMSEA could not be calculated indicating that the second-order factor model did not capture the correlations among the five factors very well. In the 4b model which was a two correlated second order factors model where Worry, Emotionality and Distraction form a test anxiety dimension and self efficacy and Lack of confidence were put together under the self esteem dimension, the fit improved both in exploratory and confirmatory samples resulting in x2/df ratio=1.76. As explained later on, even though the x2/df ratios in both Model 4 and 4b were adequate Model 3 was preferred over Model 4. It was seen that when self efficacy and lack of confidence were combined to form the self-esteem factor, the construct self esteem highly

correlated with the construct of test anxiety. This it was concluded that self efficacy is a correlate of test anxiety and not a part of the construct. Overall, it was found that in every model worry and emotionality contributed the most to the content domain of test anxiety as was shown in previous researches. This research broadened the construct in that it showed that the inclusion of lack of confidence was necessary as well as worry and emotionality.

Limitations of the Research


Besides the important findings of this research, there are some limitations of the research most of which are mentioned by the researchers. 1. There seems to be a confusion between the theoretical constructs and their operational definitions. In future research it is suggested that the operational definitions be carefully laid out and evaluated from the beginning.

2. Another limitation concerns the sample that was used in the study, it was noted that the sample may not be fully representative of all types of student populations. Regarding that especially in the American sample where most of the students were graduate students, it is apparent that even though they might have test anxiety it's impact on their performance is little since they reached graduate studies. 3. A limitation or more correct a criticism that was not mentioned by the researchers is the development of the post-Hoc models. As Geoffrey Maruyama mentions in his book Basics of Structural Equation Modeling "from a perspective more like Cliff's (1983) model modification is a substantial shift from the confirmatory intent of latent variable SEM approaches". He further goes on saying that from a conservative point of view models should purely be confirmatory and not be modified unless the intention is to plan the next study. In this research it was not clear what the intention of the researchers' was in suggesting a post-hoc model. The explanation that they provided was that they had seen that Distraction did not seem to fit in the construct so they checked the model without it. The researchers did not explain an intention of future research in doing so, so it can be assumed that it was done in order to ameliorate the model at hand which is less acceptable in the conservative circles.

Das könnte Ihnen auch gefallen