test anxiety, perceived test difficulty, and test ... - Science Direct

12 downloads 0 Views 1MB Size Report
the exam both had effects on test anxiety perceived at each corresponding time, ... on the arousal of test anxiety than did test difficulty perceived before the exam.
TEST ANXIETY, PERCEIVED TEST DIFFICULTY, ANDTEST PERFORMANCE: TEMPORAL PATTERNS OF THEIR EFFECTS EUNSOOKHONG UNIVERSITY

OF NEVADA, LAS VEGAS

ABSTRACT: Two hypothesized models that represent relationships among test anxiety, perceived test difficulty, and test performance observed immediately before and after a final examination were tested. The structural equation modeling was used to examine the hypothesized

relationships

by imposing the structure

of

direct and indirect effects on the current data collected from 208 undergraduate students. Students’ perception of test difficulty had a significant effect on the arousal of both worry and emotionality. Temporal changes were observed in the degrees of these relationships; while test difficulty perceived before and during the exam both had effects on test anxiety perceived at each corresponding time, students’ perception of test difficulty during the exam had a greater direct effect on the arousal of test anxiety than did test difficulty perceived before the exam. Worry, but not emotionality,

had a strong inverse relationship

with performance.

That perceived test difficulty did not have a direct effect on test performance, but had an indirect effect mediated by worry, indicates that test performance was not directly influenced by how they perceive the test difficulty, but worry aroused by the difficulty perception influenced their performance. Importance of understanding students’ perceived test difficulty was discussed for its effect on arousing worry anxiety.

Direct all correspondence lo: Eunsook Hong, Department of Educatronal Psychology, University of Nevada, Las Vegas, Nevada 89154.3003, USA. E-mail. iEhongQnevada.edu> Learning and Individual Differences, Volume 11, Number 4.1999, pages 431-447. All rights of reproduction in any form reserved.

CopyrighlO

2001 by Elsevier Science Inc. ISSN: 1041-6080

432

LEARNlNGAPlD/ND/V/DUAL DFFERENCES

VOLUMEll,NUMBER4.1999

Studies on test anxiety have provided evidence that test anxiety has a negative effect on performance (e.g., Calvo, Eysenck, Ramos, & Jimenez, 1994; Hembree, 1988; Kivimaki, 1995; Seipp, 1991), is a multidimensional construct (Liebert & Morris, 1967; Sarason, 1984), and could be treated with appropriately developed cognitive or behavioral treatments (e.g., Anup, 1994; Nicaise, 1995; Sapp, 1993; Sud & Sharma, 1990). Continuous efforts have been made to understand better the dimensional structure of test anxiety, differential effects of its components on performance, and anxiety treatments (e.g., Gierl & Rogers, 1996; Hong, 1998; Sapp, 1996; Williams, 1996). In order to further the understanding of test anxiety phenomenon underlying an entire testing period, the current study examined anxiety of students immediately before and after an examination. Test difficulty perceived by the students was also included to determine its relationship with test anxiety and test performance. Two hypothesized models that represent two theoretical views of temporal relationships among test anxiety, perceived test difficulty, and test performance were examined.

DIMENSIONS OF TEST ANXIETY AND THEIR DIFFERENTIAL EFFECTS ON PERFORMANCE Among the various conceptualizations of test anxiety, Sarason’s (1984) four-factor model (i.e., worry, tension, test irrelevant thinking, and bodily symptoms) and Liebert & Morris’ (1967) two-factor model (i.e., worry and emotionality) have been widely discussed (e.g., Benson & Bandalos, 1992; Hong, 1998; O’Neil & Fukumura, 1992). In preparing the measure of test anxiety in the current study, Liebert & Morris’ two-factor conceptualization (1967) was adopted. The worry vs. emotiona2ity distinction has been evidenced by factorial validation studies (e.g., Benson & Tippets, 1990) and by their differential effects on performance (e.g., Morris, Davis, & Hutchings, 1981). Worry refers to the cognitive concern about test-taking and performance, such as negative expectations and selfevaluation and preoccupation with performance and potential consequences. Emotionality refers to perceived physiological reactions, that is, autonomic arousal and somatic reactions to testing situations such as nervousness, discomfort, and tension (Liebert & Morris, 1967; Morris et al., 1981; Schwarzer, 1984). However, some factor analytic studies have revealed conflicting findings on the dimensional structure of test anxiety (Richardson, O’Neil, Whitmore, & Judd, 1977; Zimmer, Hocevar, Bachelor, & Meinke, 1992). Accordingly, attempts have been made to examine further by testing different models of test anxiety (Hodapp & Benson, 1997; Nasser & Takahashi, 1996). For example, whereas a clear two-factor solution was demonstrated in the state measures of worry and emotionality, the same solution was not evident in the trait measures (O’Neil & Abedi, 1992). Factor analyses of the 37-item Test Anxiety Scale (Sarason, 1972) did not support the two-component view of test anxiety, with the first factor indicating both cognitive worry about self and performance and emotionality or intense autonomic

433

TEMPORAL PATTERNSOF TESTANXIETY

arousal, whereas the second factor reflected a reaction of emotional distress that does not involve the cognitive worry or excessive autonomic arousal (Richardson et al., 1977). The items used in this study were adapted from the scale that has shown to produce the two-factor structure. In addition, state test anxiety, as opposed to trait test anxiety, that has demonstrated a clear worry-emotionality distinction was examined in the current study. Although worry and emotional&y are strongly correlated due to the testing situations that contain elements related to the arousal of each, their patterns of correlations with performance scores are distinguishable. Worry is the test anxiety component that has shown a consistent and strong inverse relationship with performance, whereas emotionality has a weak or no relationship with performance (e.g., Kim & Rocklin, 1994; Morris et al., 1981; O’Neil & Abedi, 1992; Zeidner & Nevo, 1992). The cognitive-attentional view of test anxiety (Wine, 1971,198O) may explain the negative role of worry in performance. The view suggests that it is how individuals allocate their attention in testing situations; highly anxious individuals pay more attention to task-irrelevant cues than task-relevant cues, and thus their performance suffers due to their division of attention to both relevant and irrelevant tasks (Sarason, 1972). However, a few studies demonstrated that the relationship between test anxiety and performance may be more complex than the simple inverse relationship reported. When other factors were included in the anxiety study, for example, performance during the lessons versus tests (Kivimaki, 1995) or massed vs. distributed practice of test taking (Zimmer & Hocevar, 1994), the strength of the inverse relationship changed. Item difficulty, because different levels of difficulty demand different levels of cognitive processing and attention, is another variable that mediates differential effects of worry and emotionality on performance (e.g., Kim & Rocklin, 1994; Morris & Liebert, 1969).

TEMPORAL

PATTERNS OF TEST ANXIETY

Although cognitive and emotional anxiety that students experience in testing situations are highly related, worry and emotionality anxiety are aroused and maintained by different situational conditions. Compared to worry, the elements that arouse emotionality experience are typically of shorter duration and consist of initial, nonevaluative cues such as student conversations about the test, which occurs at the initiation of the testing period and soon dissipates as attention is turned to the test itself (Morris et al., 1981; Morris & Fulmer, 1976). Worry anxiety are aroused and maintained by evaluative situations in which the anxiety-provoking elements that influence individuals’ cognitive functioning continue to exist throughout the evaluation period. Thus, as the testing session progresses, emotionality typically decreases significantly from the beginning to the end of an examination (e.g., Morris & Engle, 1981; Smith & Morris, 1976), whereas worry does not (Morris et al., 1981).

434

LfARN/NGAND/ND/L/IDUAL DFFERENCES

VOLUMEll.NUMBER4.1999

Other studies, however, observed mixed findings on the temporal patterns. The temporal change in worry and emotionality depended on the student’s performance expectancy. Students who were able to form their performance expectancy through feedback received during academic testing, showed decrements in worry as well as emotional&y (Morris & Fulmer, 1976). In a few studies, only minimal differences in the arousal patterns for worry and emotionality were observed (Holroyd, 1978; Holroyed, Westbrook, Wolf, & Badhorn, 1978). Kim and Rocklin (1994), in studying temporal patterns of worry and emotionality, found significant interaction effects between the time period and other variables such as item difficulty. For instance, students’ worry gradually increased as the time progressed if they were attempting difficult items, but slightly increased first and then decreased if they were attempting easy items. Students’ emotionality generally decreased across time unless they were attempting difficult items, in which case it was stable or slightly increased. Depending on how students perceive the difficulty of the test in the course of a testing period, test anxiety in students may be aroused to different degrees at different times. The level of test difficulty perceived by students at the beginning of the test may be different from that perceived during the test, dependent upon various cues or indicators they receive during the test (e.g., estimation of their own test performance). Thus, it is conceivable that different levels of test difficulty perceived at two different occasions would have differential effects on test anxiety measured at each corresponding occasion. This temporal pattern of the relationship between students’ perception of test difficulty and test anxiety has not been tested. In an effort to determine the relationships among test anxiety, perceived test difficulty, and test performance discussed above, the current study examined (a) the relationships among the two components of test anxiety-worry and emotionality-and perceived test difficulty, and their impacts on test performance and (b) temporal patterns of the effects perceived test difficulty has on worry and emotionality. Test anxiety was measured during the final examination in statistics classes, where a fair amount of test anxiety can be expected. To address these questions, we compared two models that reflect different theoretical views of the relationships among the constructs. The constructs examined in this study were perceived test difficulty, worry, and emotionality measured at two different occasions in a testing situation, and test performance. Model 1 depicts the hypothesized relationships that perceived test difficulty has a direct effect on worry and emotionality, and in turn, worry and emotionality differentially influence performance on the exam. Perceived test difficulty also is hypothesized to have a direct effect on test performance. The relationship between students’ perception of test difficulty and test performance has not been studied previously. Thus, we explored the relationship by testing the path between the two constructs. That is, in Model 1 perceived test difficulty is hypothesized to have a direct effect on test performance, whereas in Model 2 this path is removed. During the exam, students might alter their perceptions about the difficulty of the exam, and accordingly the altered perception would differentially impact the level of worry and emotionality students experience during the exam. Figure 1 presents the model that reflects these relationships.

435

TEMPORAL PATTERNSOFTESTANXIETY

FIGURE 1 The hypothesized model 1. (“B” indicates the data collected before the exam, and “A” indicates those collected after the exam. wo = worry; em = emotionality; diff = perceived test difficulty; perform = performance on the final examination.)

Model 2 is similar to the first model except that (a) perceived test difficulty has no direct effect but has an indirect effect on performance through the mediation of the worry component of test anxiety and (b) emotionality is hypothesized to have no effect on test performance (see Figure 2). As indicated earlier, research on the latter relationship has produced conflicting findings.

METHOD PARTICIPANTS

AND PROCEDURE

Undergraduate students enrolled in a statistics course offered by the department of economics at a large western university volunteered to participate in the study. Two instructors taught two and three classes each. The statistics course was a required course to the students volunteered in this study. Data were collected during two semesters. Appropriate informed consent was obtained, and students were assured anonymity. No remuneration or extra credit was provided. Among 216 volunteered (not including those who dropped the course after the first class), 7 students completed only 1 of the 2 questionnaires, leaving 209 students (120 males, 82 females, 7 unspecified) with complete data. Of 209 students,

436

LfARNlNGAND /ND/V/DUALDlFFfRfNCfS

V0LlJME11.NUMBER4.1999

FIGURE 2 The hypothesized model 2. (“B” indicates the data collected before the exam, and “A” indicates those collected after the exam. wo = worry; em = emotionality; diff = perceived test difficulty; perform = performance on the final examination.)

1 case was found to be an outlier in a preliminary data analysis, resulting in 208 participants for the study. The questionnaires (see Instruments, below) were attached to the final examination, one on top and the other at the bottom of the test. Students were both told and read the instructions on the questionnaires. The questionnaire given immediately before the test asked participants to indicate how they think or feel now before they read the exam items, and the one given immediately after they completed their test asked them to indicate how they thought or felt during the test. Participants were instructed not to spend too much time on any one item.

INSTRUMENTS The modified version of the Test Anxiety Inventory (TAI; Spielberger, 1980) was employed to measure participants’ worry and emotionality at the time of their final examination. The questionnaires were designed to address state test anxiety for two different occasions, i.e., immediately before and immediately after the examination. Eight items were used with four items measuring worry and four items measuring emotionality. These items have been used in numerous studies, and their reliability and validity have been established in those studies (e.g., Hong, 1998; O/Neil & Abedi, 1992; O/Neil & Fukumura, 1992). The internal consistency estimates (coefficient alpha) for the worry and emotionality subscales

437

TEMPORAL PATTERNSOFTESTANXIETY

distributed at the beginning of the exam were 80 and .91, respectively, and those distributed at the end of the exam were 88 and .93, respectively. The number and content of items for the two questionnaires given at the beginning and at the end were the same except for the differences in the verb tense. For example, the worry item, “I am thinking about how important the test is for me,” given at the beginning, was changed in the verb tense at the end to read, “. . . thought. . . was. . . .” In addition, some clauses that were found in TAI were omitted in this measure. For example, the TAI item, “I feel very panicky when I take an important test,” was changed to “I feel very panicky,” thus deleting, “when I take an important test” from the original TAI item. The complete item set is presented in the appendix. Participants responded to each test anxiety item by rating themselves on the following four-point scale: (1) not at all; (2) somewhat; (3) moderately so; and (4) very much so. Thus, the scale defines a continuum of increasing levels of intensity in test anxiety, with a low score indicating a low level of test anxiety and a high score indicating a state of intense test anxiety. Also included in the questionnaires were four items measuring students’ perception of test difficulty. Examples of the items were: “Compared to other subjects, this statistics exam will be difficult (before)/was difficult (after)” and “I think this exam will be a difficult one for me (before)/This exam was a difficult one for me (after).” The internal consistency estimates for this scale were .75 and .79 for the before- and after-exam measures, respectively. In this study, worry, emotionality, and perceived test difficulty were represented by two indicators each, with each indicator consisting of two items (the first and third items were combined to create the first indicator, and the second and fourth items for the second indicator). The final exam scores given to the investigator from the instructors of this statistics course ranged from 32 to 99, with a possible maximum score of 100.

DATA ANALYSIS The structural equation modeling (SEM) was used to examine the hypothesized relationships among the constructs in the study. The hypothesized models were tested with the EQS program (Bentler, 1995) by imposing the structure of direct and indirect effects on the current data. First, the fit of a measurement model was tested to determine whether the observed variables (indicators of the latent constructs, worry, emotionality, and perceived difficulty measured on two occasions) were generated by the corresponding latent constructs. The overall fit and the regression paths were analyzed in this endeavor. Second, the two originally hypothesized models (the full SEM models; see Figures 1 and 2) were tested. The indexes of the goodness of fit between the hypothesized models and the data were examined to determine which of the two models better describes the data. Third, a modification process was applied to the selected model from the previous analysis, so that the model could be improved further not only to represent

438

LEARNlNGANDINDh'lDUALLJFFERENCES

V0LUME11.NUMBER4.1999

a good fit to the data but to describe adequately the meaningful relationships among the constructs. This post-hoc model fitting procedure, which was exploratory in nature, was adopted in this study in addition to the tests of the two hypothesized models, because differential temporal effects-the direct and indirect effects among the constructs assessed in two different occasions in a testing situation-have not been investigated. That is, although the theoretical framework in building the two models was partly based on previous theoretical views and empirical research, some relationships represented in the models have not been investigated. Thus, additional exploratory examinations were required to better understand the temporal relationships among test anxiety, test difficulty, and test performance. The evaluation of model adequacy was based on chi-square statistic, comparative fit index (CFI), incremental fit index (IFI), and inspection of the values of standardized residuals. In addition, the results of Lagrange Multiplier (LM) tests and Wald tests were used to determine malfitting parameters in the model modification process. 1 The investigator’s knowledge of the data and theoretical aspects (Joreskog, 1971) of research in test anxiety were also considered in evaluating the model adequacy. Examination of skewness and kurtosis (univariate and multivariate) indicated that maximum likelihood estimation is appropriate for this study.2 Table 1 presents the covariances and correlations among the 13 indicators examined in the study. The correlations among the indicators of worry, emotionality, and perceived test difficulty were all statistically significant, ps < .05. Correlations between the exam score and the other indicators were all negative as expected. Five of these indicators (Bwol, Beml, Bem2, Aeml, and Aem2) did not show statistically significant relationships with the exam score, while the rest (seven) did.

Covariance Variables 1. Bwoln

2. Bwo2 3. Bern1 4. Bern2 5. Bdiffl 6. Bdiff2 7. Adiffla 8. Adiff2 9. Awol 10. Awo2 11. Aeml 12. Aem 13. Perform

1 .61 .32 .lO .17 .15 .14 .35 .31 .17 .16 .12 .13 -.12

and Correlation

2

3

.50

.49 .68 .73 .53 .45 .42 .lO .18 .49 .42 .36 .34 ~.05

.55 .23 .26 .23 24 .24 .33 .24 .25 .20 .19 -.09

TABLE 1 Matrix of the 10 Variables Examined

4

5

6

7

8

9

10

11

12

23

.43 .72 .69 .70 .54 .54 .16 .24 .52 .54 .47 .46 -.06

.20 .27 .24 .24 .73 .64 .17 .22 .49 .50 .55 .55 -.04

.30 .44 .39 .40 .55 .79 .12 .17 .50 .53 .56 .58 p.04

.17 .30 .31 .27 .51 .41 .56 .36 .16 .18 .16 .16 -.13

.20 .37 .29 .29 .50 .54 .60 .60 .22 .24 .19 .23 -.lO

.58 .53 .50 .49 .24 .30 .32 .34 .87 .65 .56 .54 p.17

.40 .73 .54 .64 .20 .35 .29 .36 .60 .76 .53 .56 -.18

.48 .59 .64 .62 .14 .36 .29 .33 .62 .62 .75 .68 -.06

.35 .61 .54 .65 .20 .30 .30 .33 .55 .71 .74 .79 -.06

-.08 -.15 -.04 -.08 p.18 -.20 -.27 -.25 -.18 -.36 -.06 -.13 .47

Note: Covariances, correlations, and variances are presented in the lower left and upper right triangle and diagonal, respectively. wo = worry; em = emotionality; diff = perceived test difficulty; perform = performance on the final examination. ““8” indicates the data collectrd before the exam, and "A" indicates those collected after the exam.

TEMPORAL PATTERNSOFTESTANXIETY

439

RESULTS THE MEASUREMENT

MODEL

The measurement model specified six factors-worry, emotionality, and perceived test difficulty assessed immediately before and after the exam. In this model, each indicator was constrained to load only on the factor it was designated to measure, the residual terms for all indicators fixed to be uncorrelated, no equality constraints on the factor loadings were imposed, and the factor covariances were free to be estimated. The model fit the data reasonably well [x2(45, N = 208) = 171.15, p < .OOl, CFI = .935, IF1 = .936]. Variances (R2) in the indicators accounted for by their corresponding constructs were all significantly large, ranging from .69 to .94 (Mdn = .87). The factor correlations of the measurement model are presented in Table 2. A strong relationship was shown between worry before the test and perceived test difficulty during the exam, which was recalled immediately after the exam. Students’ perception of test difficulty before the exam was strongly related to emotionality aroused during the exam as indicated in the recall of their test anxiety after the exam. The correlation between worry before the exam and emotionality after the exam was the lowest. However, due to the exclusion of the test performance variable (not a construct) in this model, these correlations should be interpreted with caution. The relationships among these six constructs and of those with test performance were tested in the full structural models presented in the following section. THE STRUCTURAL

MODELS

To examine the goodness of fit of the two hypothesized models, the measurement model was respecified by imposing the structure of each model (see Figures

Factor Correlations

Factor

Before 1. Worry 2. Emotionality 3. Test difficulty After 4. Worry 5. Emotionality 6. Test difficulty

TABLE 2 for Worry, Emotionality, and Perceived Test Difficulty Examination Times 2

for Before and After

2

3

4

5

6

.69

.64 .90

.95 .66 .61

66 .89 .87

.58 .81 .90

-

.65

.60 .89

Note: N = 208. Before = before the final exam; After = after the final exam; Test difficulty = perceived test difficulty.

440

LEARNlNGANDINDII/IDUAL DIFFERENCES

VOLlJME11.NUMBER4.1999

1 and 2). The test performance variable was included in the structural models, and the structural regression paths that represent the direct and indirect relationships were imposed in place of factor covariances among constructs. The two initially hypothesized structural models represented only a marginal fit to the current data [x2(55, N = 208) = 301.05, p < ,001, CFI = 876, IF1 = 877 for Model 1, and x2(57, N = 208) = 301.53, p < .OOl, CFI = .877, IF1 = .878 for Model 21, suggesting evidence of misspecification of some parameters in the model. The fit indexes of the two models were almost the same, and the differences of these corresponding indexes of the two models were not significant. Of the two, Model 2 was considered a better fitting model because it was the more parsimonious model of the two (Byrne, 1994). The Wald test also suggested that the two free parameters in Model 1, which were not specified in the original Model 2, be dropped; that is, perceived test difficulty and emotionality before the exam did not have direct effects on test performance. Thus, Model 2 was selected for further analysis, which was conducted by respecifying some of the parameters. The LM and Wald tests were utilized in this process, along with close examinations of meaningfulness of the paths as suggested by the theories of relationships among the constructs. One structural regression coefficient in Model 2, the path from test performance to worry anxiety measured after the exam, was not statistically significant. This parameter was fixed to zero. The LM statistics identified two parameters that were not included in the earlier model as those contributing most to model misfit (worry construct before the exam had a direct effect on the two indicators of perceived test difficulty observed after the exam). Given findings of marginal fit of Model 2, the model was respecified taking into account the Wald and LM statistics and subjected to the goodness-of-fit test. The fit indexes of the modified model (Model 3) indicated that the model represented a good fit to the data [x2(56, N = 208) = 145.42, p < .OOl, CFI = .955, IF1 = .955]. All structural regression coefficients presented in Model 3 were statistically significant (see Figure 3). Significant direct effects of perceived test difficulty on worry and emotionality before the exam and on perceived test difficulty after the exam were observed. Worry before the exam had a direct effect on test performance and on the two indicators of perceived test difficulties. Test performance had a direct effect on perceived test difficulty and emotionality recalled after the exam. Emotionality before the exam did not have a direct effect on test performance. Test performance did not have an effect on worry recalled after the exam. However, both worry and emotionality recalled after the exam were directly influenced by perceived test difficulty. This model, along with the estimates of standardized regression coefficients, factor loadings, and residual variances and covariances is presented in Figure 3. The parameter estimates were all reasonable and statistically significant, except for the two factor loadings for recalled perception of test difficulty. Variances (X2) in worry and emotionality accounted for by perceived test difficulty before the exam were .16 and .72, respectively. Variance in the recall of perceived test difficulty attributed by test difficulty perceived before the exam and test performance was .87. Test performance on the final examination was directly affected by

441

TEMPORALPATTERNSOFTESTANXIETY

.i5

.ie

/

.41

BEM \-

05

.12

.15

FIGURE 3 The final model. (“B” indicates the data collected before the exam, and “A” indicates those collected after the exam. wo = worry; em = emotionality; diff = perceived test difficulty; perform = performance on the final examination.)

worry, with only 8% of the variance accounted for. A total of 82% of the variance in the recalled worry anxiety was accounted for by the recalled perception of test difficulty, and 83% of the variance in the recalled emotional@ was attributed by test performance and the recalled perception of test difficulty. To summarize, with the fit index of .955 for both CFI and IFI, the significant parameter estimates, and the parsimony and meaningfulness of the paths included in the model, Model 3 was considered a fairly good fit to the current data.

DISCUSSION The final model supported the proposed hypothesis regarding the relationships among students’ perceived test difficulty, worry, and emotionality observed immediately before and after the exam, and their relationships with test performance, with minor modifications of the relationships represented in the original Model 2. Students’ perception of test difficulty had a significant effect on the arousal of both worry and emotionality anxiety. That is, students who perceived the test more difficult showed higher worry and emotionality than did their peers who thought of the test less difficult. Kim and Rocklin (1994) also found that stu-

442

LEARNlNGAND/ND/V/LJUAf DIFFERENCES

V0LlJME11.NUMBER4.1999

dents’ test anxiety increased when they were attempting difficult items, but then decreased if they were attempting easy items. Thus, the findings of these two studies indicate that both perceived test difficulty and difficulty of the actual test items have significant impacts on students’ test anxiety. Temporal changes occurred in the degrees of these relationships. The effect sizes of the relationships changed across the time; whereas test difficulty perceived before and after the exam both had effects on worry and emotionality observed at the corresponding occasions, the effect sizes were larger in the afterexam parameters. That is, test difficulty perceived after the exam had a stronger influence on the arousal of worry (p = .91) and emotionality (p = .97) than that perceived before the exam (p = .40 for worry and p = .85 for emotionality). The findings indicate that although no direct feedback was provided on test performance during and immediately following the examination, test difficulty, worry, and emotionality perceived during the exam might have been based on the assessment of their own actual performance (i.e., feedback given to self based on the perceived level of performance). Accordingly, test difficulty perceived during the test (which was recalled after the test) had a significantly stronger direct effect on the arousal of worry and emotionality, compared to that measured before the exam. The correlations among the two worry and two emotionality factors indicated in the measurement model showed strong associations among them, suggesting the arousal of both anxiety components caused by the anxiety-provoking final exam situation. However, worry anxiety, but not emotionality, had a negative effect on students’ test performance. It has been evidenced in previous studies that worry is the test anxiety component that has a strong inverse relationship with performance (e.g., Hong, 1998; Kim & Rocklin, 1994; Morris et al., 1981; O’Neil & Abedi, 1992; Zeidner & Nevo, 1992). The current study again supports this contention. Perceived test difficulty did not have a direct effect on test performance, but instead it had an indirect effect through the mediation of worry. Thus, when students perceived the exam difficult, it aroused students’ worry, which in turn affected their test performance. Students’ performance on the test was not directly influenced by how they perceive the test difficulty, but worry aroused by perceiving the test as a difficult one influenced their performance on the test. As expected, test difficulty perceived before the exam had a direct effect on the same construct observed after the exam. Thus, students who perceived the test difficult at the beginning of the examination consistently thought that the test was difficult during the examination. This finding is not unlikely given that the direct feedback on test performance was not provided during the exam. Although some students could have changed their perception of test difficulty based on their self-perceived test performance, the pattern of change among students might have not been consistent enough to make a difference on the strength of the relationship between perceived test difficulty measured before and after the examination. Worry anxiety at the beginning of the test also had a strong direct effect on the two indicators of test difficulty perceived after the exam. That is, the two indicators had stronger relationships with the worry felt before the test, but not with

TEMPORALPATTERNSOFTESTANXIETY

443

their own latent construct, perceived test difficulty. In the analysis of the measurement model, the factor loadings between this construct and its indicators were highly significant (.91 and .88), but when the structure of the relationships among constructs was imposed, these loadings became nonsignificant, as compared to the loadings coming from the worry construct (factor loadings were .76 and .75). These findings might indicate that the worry component of test anxiety influence students’ perception of test difficulty throughout the testing period. This might indicate that students who are worried about the exam may be so because they are not well prepared for the test, which in turn will have them perceive during the exam that the test would be difficult. The effects of test performance on test anxiety and test difficulty should be interpreted with caution. Because worry, emotionality, and perceived test difficulty were measured immediately following the examination by asking students to recall what they thought during the test, test performance was located before these three constructs in the structural models. However, the test performance variable was the performance score graded after the exam. That is, due to the timing of the measurement, test performance may be conceived in part as reflecting students’ perception of their own performance on the test during the examination (as considered earlier in the discussion) and in part as the test performance represented by the test score. The test performance had a negative effect on test difficulty perceived after the exam but not directly on worry anxiety recalled after the exam. This finding is reasonable since students were not given direct feedback on their performance during the examination, and the recall of worry was based on their perceived level of achievement in the exam, but not on the actual exam score. Emotionality recalled after the exam and performance score had a statistically significant positive relationship, although the magnitude was small. Unlike the strong evidence of the negative relationship found between worry and test performance, the pattern of relationships of emotionality to performance has not been consistent, from no relationship to a weak negative relationship, and to a positive relationship. In a study by Kim and Rocklin (1994), emotionality during the exam did facilitate the performance of subjects who had high performance expectancy. This indicates that there can be situations during the exam (as opposed to before the exam) when emotionality does not hamper student performance, but rather facilitate it. The investigations into the relationship of emotionality to performance during the exam with other mediating variables are warranted. Understanding students’ perception of test difficulty is important because of its impact on the arousal of worry anxiety, which in turn has a negative effect on performance. Tallent-Runnels et al. (1994) found that among various study strategy and study behavior variables such as test strategies, selecting main ideas, and motivation, anxiety was the one that discriminated most between the high- and average-achieving students. Thus, understanding and treatment of student anxiety in efforts to alleviate it will likely result in improved performance. There are other important variables that might arouse anxiety in testing situations that are not tested in the current study. Variables such as item difficulty, self-efficacy, perceived values, and performance expectancy can be examined to determine their effects on test anxiety and test performance. In addition, whether a similar tempo-

444

LEARNINGAND /ND/V/DUAL DIFFERENCES

V0LUME11.NlJMBER4.1999

ral pattern would emerge with older (graduate students) and younger (elementary or secondary students) samples is a question worthy of further investigation.

NOTES 1. To evaluate the model adequacy, a number of fit indexes reported by the EQS program were used (Bentler, 1995). Although a nonsignificant chi-square statistic indicates, in most cases, that the hypothesized model fits the observed data well, because the chi-square statistic is often significant due to its sensitivity to departures from multivariate normality and to sample size (Jareskog & S&born, 1983), the model fit should be evaluated with additional indexes. Among the fit indexes, Bentler’s comparative fit index (CFl) and Bollen’s incremental fit index (IFI) (Bollen, 1989) were recommended by Bentler (1990) and Gerbing & Anderson (1993). The CFI ranges from 0 to 1, and a value greater than .90 indicates an acceptable fit to the data. The CFI and IF1 avoids the normed fit index’s underestimation of fit with small samples and the large sampling variability of the non-normed fit index (Bentler, 1992). In addition, results of Lagrange Multiplier (LM) tests and Wald tests provided by EQS were inspected in establishing a well-fitting model. The LM test evaluates the statistical necessity of the restrictions (fixed parameters), that is, whether some of the fixed parameters in the model could be freed, whereas the Wald test is used to evaluate whether some of the free parameters in the model could be restricted. In analyzing the results both univariate LM test and multivariate LM test were examined (Bentler, 1995). 2. Examination of the univariate statistics indicated that there were no severely skewed or kurtotic items (all smaller than ∣l∣, except for one skewness with -1.02). The normalized estimate of multivariate kurtosis indicated the kurtosis was positive, 3.92, indicating a small degree of kurtotic multivariate distributions. These results indicated that the distribution of the sample data was not substantially different from normal distribution. Thus, the present study estimated models with maximum likelihood estimation.

REFERENCES Anup, S. (1994). “Attentional skills training/cognitive modeling: Short term therapeutic cognitive interventions for test anxiety.“ Psychological Studies, 39, l-7. Benson, J. & D. L. Bandalos. (1992). “Second-order confirmatory factor analysis of the Reactions to Tests scale with cross-validation.” Multivariate Behavioral Research, 27,459487. Benson, J. & E. Tippets. (1990). “Confirmatory factor analysis of the test anxiety inventory.” Pp. 149-156 in Cross cultural anxiety (Vol. 4), edited by C. Spielberger & R. Dias-Guererro. New York: Hemisphere/Taylor-Francis. Bentler, P. M. (1990). “Comparative fit indexes in structural models.” Psychological Bulletin, 107,238-246. Bentler, P. M. (1992). “On the fit of models to covariances and methodology to the B&etin.” Psychological Bulletin, 112,400-404.

TEMPORAL PATTERNSOFTESTANXIETY

445

Bentler, P. M. (1995). EQS Structural Equations Program Manual. Encino, CA: Multivariate Software. Bollen, K. A. (1989). “A new incremental fit index for general structural equation models.” Sociological Methods and Research, 17,303-316. Byrne, B. M. (1994). Structural Equation Modeling with EQS and EQS/Windows: Basic Concepts, Applications, and Programming. Thousand Oaks, CA: Sage. Calvo, M. G., M. W. Eysenck, P. M. Ramos, & A. Jimenez. (1994). “Compensatory reading strategies in test anxiety.” Anxiety, Stress and Coping: An International Journal, 7,99-116. Gerbing, D. W. &J. C. Anderson. (1993). “Monte Carlo evaluations of goodness-of-fit indices for structural equation models.” Pp. 40-65 in Testing structural equation models, edited by K. A. Bollen &J. S. Long. Newbury Park, CA: Sage. Gierl, M. & W. T. Rogers. (1996). “A confirmatory factor analysis of the Test Anxiety Inventory using Canadian high school students.” Educational and Psychological Measurement, 56,315-324. Hembree, R. (1988). “Correlates, causes, effects and treatment of test anxiety.“ Review of Educational Research, 58,47-77. Hodapp, V. & J. Benson. (1997). “The multidimensionality of test anxiety: A test of different models.” Anxiety, Stress, and Coping, 20,219-244. Holroyd, K. A. (1978). “Effectiveness of an ‘attribution therapy’ manipulation with test anxiety.“ Behavior Therapy, 9,526-534. Holroyed, K. A., T. Westbrook, M. Wolf, & E. Badhom. (1978). “Performance, cognition, and physiological responding in test anxiety.“ Journal ofAbnormal Psychology, 87,442-451. Hong, E. (1998). “Differential stability of individual differences in state and trait test anxiety.“ Learning and Individual Differences, 10,51-69. JGreskog, K. G. (1971). “Simultaneous factor analysis in several populations.” Psyckometrika, 57,239-251. Jiireskog, K. G. & D. Siirbom. (1983). LISREL V: Analysis of linear structural relationships by the method ofmaximum likelihood. Chicago: International Educational Services. Kim, S. H. & T. Rocklin. (1994). “The temporal patterns of worry and emotionality and their differential effects on test performance.” Anxiety, Stress and Coping: An International Journal, 7,117-130. Kivimaki, M. (1995). “Test anxiety, below-capacity performance, and poor test performance: Intrasubject approach with violin students.” Personality and lndividual Dijjferences, 18,47-55. Liebert, R. M. & L. W. Morris. (1967). “Cognitive and emotional components of test anxiety: A distinction and some initial data.” Psychological Reports, 20,975-978. Morris, L. W., M. A. Davis, & C. H. Hutchings. (1981). “Cognitive and emotional components of anxiety: Literature review and a revised worry-emotionality scale.“ Journal of Educational Psychology, 73,541-555. Morris, L. W. & W. B. Engle. (1981). “Assessing various coping strategies and their effects on test performance and anxiety.” Journal of Clinical Psychology, 37,165-171. Morris, L. W. & R. S. Fulmer. (1976). “Test anxiety (worry and emotionality) changes during academic testing as a function of feedback and test importance.“ Journal of Educational Psychology, 68,817~824. Morris, L. W. & R. M. Liebert. (1969). “The effects of anxiety on timed and untimed intelligence tests: Another look.“ Journal of Consulting and Clinical psychology, 33,240-244. Nasser, F. & T. Takahashi. (1996, April). An application of confirmatory factor analysis with item parcels for testing the structure of test anxiety among Israeli-Arab high school students. Paper presented at the annual meeting of National Council on Measurement in Education, New York.

446

LEARNING AND INDIVIDUAL DIFFERENCES

VOLUME

11. NUMBER

4.1999

Nicaise, M. (1995). “Treating Test Anxiety. A Review of Three Approaches.” Teacher Educution and Practice, 22,65-81. O’Neil, H. F. & J. Abedi. (1992). “Japanese children’s trait and state worry and emotionality in a high-stakes testing environment.“ Anxiety, Stress and Coping: An International Journal, 5,253-267. O/Neil, H. F. & T. Fukumura. (1992). “Relationship of worry and emotional&y to test performance in the Juku environment.” Anxiety, Stress and Coping: An International journal, 5,241-251. Richardson, F. C., H. F. O’Neil Jr., S. Whitmore, & W. A. Judd. (1977). “Factor analysis of the test anxiety scale and evidence concerning the components of test anxiety.” Journal of Consulting and Clinical Psychology, 45, 704-705. Sapp, M. (1993). Test anxiety: Applied research, assessment, and treatment intervention. Lanham, MD: University Press of America. Sapp, M. (1996). “Three treatments for reducing the worry and emotionality components of test anxiety with undergraduate and graduate college students: Cognitive-behavioral hypnosis, relaxation therapy, and supportive counseling. Journal of College Student Development, 37, 79-87. Sarason, I. G. (1972). “Experimental approaches to test anxiety: Attention and the uses of information.” Pp. 381403 In Anxiety: Cuvrent trends in theory and research (Vol. 2), edited by C. D. Spielberger. New York: Academic Press. Sarason, I. G. (1984). “Stress, anxiety, and cognitive interference: Reactions to tests.” Journal of Personality and Social Psychology, 46,929-938. Schwarzer, R. (1984). “Worry and emotionality as separate components in test anxiety.” International Review of Applied Psychology, 33,205-220. Seipp, B. (1991). “Anxiety and academic performance: A meta-analysis of findings.” Anxiety Research, 4,2741. Smith, C. A. & L. W. Morris. (1976). “Effects of stimulative and sedative music on two components of test anxiety.” Psychological Reports, 38,1187-1193. Spielberger, C. D. (1980). Test Anxiety Inventory; Preliminary Professional Manual. Palo Alto, CA: Consulting Psychologists Press. Sud, S. & S. Sharma. (1990). “The differential effect of attentional skills training on worry and emotionality components of test anxiety.” Psychological Studies, 35,143-150. Tallent-Runnels, M. K., A. Olivarez Jr., A. C. C. Lotven, S. K. Walsh, A. Gray, & T. R. Irons. (1994). “A comparison of learning and study strategies of gifted and average-ability junior high students.” Journal for the Education of the Gifted, 17, 143-160. Williams, J. E. (1996). “Gender-related worry and emotionality test anxiety for high-achieving students. Psychology in the Schools, 33,159-62. Wine, J. D. (1971). “Test anxiety and the direction of attention.” Psychological Bulletin, 76, 92-104. Wine, J. D. (1980). “Cognitive-attentional theory of test anxiety.” Pp. 349-385 In Test anxiety: Theory, research, and application, edited by I. G. Sarason. Hillsdale, NJ: Erlbaum. Zeidner, M. & B. Nevo. (1992). “Test anxiety in examinees in a college admission testing situation: Incidence, dimensionality, and cognitive correlates.” Pp. 288-303 in Advances in test anxiety reseurch (Vol. 7), edited by H. M. van der Ploeg, R. Schwarzer, C. D. Spielberger (series eds.), K. A. Hagtvet, & T. B. Johnsen (vol. eds.). Amsterdam/ Lisse, The Netherlands: Swets & Zeitlinger. Zimmer, J. W. & D. J. Hocevar. (1994). “Effects of massed versus distributed practice of test taking on achievement and test anxiety.” Psychological Reports, 74,915-919. Zimmer, J., D. Hocevar, P. Bachelor, & D. L. Meinke. (1992). “An analysis of the Sarason (1984) four-factor conceptualization of test anxiety.” Pp. 103-113 in Advances in test

447

TEMPORAL PATTERNSOFTESTANXIETY

anxiety research (Vol. 7), edited by H. M. van der Ploeg, R. Schwarzer, C. D. Spielberger (series eds.), K. A. Hagtvet, & T. B. Johnsen (vol. eds.). Amsterdam/Lisse, The Netherlands: Swets & Zeitlinger.

APPENDIX

Questionnaire Items Used Before and After the Final Examination

worry

Emotionality

Before theExam

After theExam

Thinking about my grade in the course will

Thinking about my grade in the course interfered with my work on the test. During the test I got so nervous that I forgot the facts I really knew. I was concerned about what would happen if I did poorly. I thought about how important the test was for me. I felt very jittery when I was taking test. I felt very panicky when I took the test. While taking test, 1 had an uneasy, upset feeling. During the test, I felt very tense.

interfere with my work on the test. I am so nervous that I think I am forgetting facts I really know. I am concerned about what would happen if I do poorly. I am thinking about how important the test is for me. I feel very jittery. I feel very panicky. I have an uneasy, upset feeling. I feel very tense.