An Investigation Into the Psychometric Properties of the Test Anxiety Measure for College Students

A Need for a New Measure of Test Anxiety for College Students

Tests are a source of stress for many college students (Zeidner, 2007) and there is a need for new measures of test anxiety that are based on theory and current research in the field (von der Embse, Mata, Segool, & Scott, 2014) and are designed specifically for U.S. college students, as these individuals experience different challenges than students in elementary and secondary schools (Nelson, Lindstrom, & Foels, 2015). Most existing measures of test anxiety used with U.S. college students are more than 20 years old (P. A. Lowe, Lee, & DeRuyck, 2004) and/or are not multidimensional (Friedman & Bendas-Jacob, 1997; Zeidner, 1998). Researchers concur that the test anxiety construct is complex, it consists of many different dimensions, and it has been difficult to encapsulate due to its complexity (Zeidner, 1998). Moreover, developmental differences in symptoms of test anxiety have been found for different age groups (P. A. Lowe, Grumbein, & Raad, 2011; Wren & Benson, 2004), supporting the need for different measures of test anxiety for different age groups. Furthermore, there is also a need for a new measure that assesses the facilitating aspects of anxiety (P. A. Lowe et al., 2008; Zeidner, 1998) among U.S. college students.

Different Conceptualizations of the Test Anxiety Construct: Past and Present

The unidimensional model of test anxiety was the dominate model in the field during the early years (Sarason & Mandler, 1952). In contrast, multidimensional models (P. A. Lowe, 2014; Zeidner, 1998) and higher order models (Benson & El-Zahhar, 1994; P. A. Lowe & Lee, 2008; P. A. Lowe et al., 2011) have become more prevalent in recent years based on advancements in theory and research in the field.

Different researchers have conceptualized the construct of test anxiety in different ways based on the zeitgeist of the time (Zeidner, 1998). During the early years, test anxiety was viewed as a measure of drive. Sarason and Mandler (1952) theorized that students experience task-directed drives and learned anxiety drives. Those students who are test-anxious experience stronger learned anxiety drives, resulting in task-irrelevant behaviors (e.g., avoidant behaviors) and poorer test performance. Similarly, Nottelmann and Hill (1977) stated that test-anxious students experience off-task behaviors (e.g., distracted, restive, and fidgety behaviors) in evaluative situations. The work of P. A. Lowe (2014), P. A. Lowe et al. (2011), and Wren and Benson (2004) support a behavioral component of test anxiety, as these researchers developed test anxiety measures for children or adolescents that included a behavioral facet.

Alpert and Haber introduced their bidirectional theory to the field of test anxiety in 1960. According to these theorists, there are two types of anxiety (i.e., facilitating anxiety and debilitating [test] anxiety). Facilitating anxiety and test anxiety represent two different drive (i.e., arousal) states in evaluative situations. Facilitating anxiety motivates individuals to perform well in testing situations. Individuals with facilitating anxiety demonstrate task-related behaviors, such as focusing on an exam while taking the test and preparing for an exam in advance. In contrast, test anxiety interferes with students’ performance. Individuals with test anxiety do not perform well or do poorly on tests. These individuals exhibit task-irrelevant thoughts and behaviors, such as avoiding studying for an exam in advance or experiencing self-deprecatory thoughts that distract them while they take a test. Thus, facilitating anxiety increases or improves students’ academic performance and test anxiety decreases or hinders individuals’ academic performance (Zeidner, 1998). Rafferty, Smith, and Ptacek (1997) found in their study with a sample of undergraduate college students that high levels of facilitating anxiety were related to high examination performance and high examination preparation for tests and that high levels of test anxiety were associated with poor test performance.

Alpert and Haber (1960) stated that facilitating anxiety and test anxiety are independent constructs and they should be assessed independently. These two types of anxiety can be represented on a dimension as independent end points (Zeidner, 1998). Individuals may possess high or low levels of both facilitating anxiety and test anxiety or high levels of one type of anxiety, but not of the other (Alpert & Haber, 1960).

Liebert and Morris (1967) theorized that test anxiety consists of a worry and an emotionality (now referred to as physiological hyperarousal; see Joiner et al., 1999) component. The worry component involves students’ worries about failing and the consequences of failing tests, whereas the emotionality component is the autonomic reactions students experience in testing situations (Liebert & Morris, 1967). Spielberger’s (1980) work supports these two dimensions of test anxiety. He developed the Test Anxiety Inventory (TAI), one of the most popular measures of test anxiety for high school and college students, and his measure consisted of a worry and an emotionality component.

Wine (1971) introduced the cognitive interference model of test anxiety. Wine stated that test-anxious students are unable to focus on an exam because they divide their attention between performing the task, such as taking an exam, and task-irrelevant (e.g., self-deprecatory) thoughts. Because they spend so much time attending to their task-irrelevant thoughts, students are not able to focus on the task at hand, resulting in poorer test performance. Like Wine, Friedman and Bendas-Jacob’s (1997) model of test anxiety includes a cognitive interference (i.e., a cognitive obstruction) component along with two other dimensions, tenseness and social derogation. Tenseness is similar to the emotionality component found in Liebert and Morris’s (1967) model of test anxiety and social derogation is students’ concerns of disparagement they will experience from significant others if they do not do well on tests. Likewise, P. A. Lowe (2014) and P. A. Lowe et al. (2011) found support for a social derogation dimension, referred to as social concerns, in their four-factor and five-factor models of test anxiety developed for children and adolescents.

Theory and research has shown that social derogation/social concerns, cognitive interference, emotionality/physiological hyperarousal, task-irrelevant behaviors, and worry are purported to be different dimensions of the test anxiety construct. Thus, these facets along with facilitating anxiety were used as guidelines in the development of items for a new measure of test anxiety designed specifically for U.S. college students.

Test Anxiety and Gender Differences

Gender differences have been found among college students on measures of test anxiety, with females reporting higher levels of test anxiety than males (e.g., Hembree, 1988; Seipp & Schwarzer, 1996). Seipp and Schwartz found these gender differences to be about a third of a standard deviation in the United States and abroad.

The research on gender differences in facilitating anxiety have been more equivocal with some studies reporting higher levels of facilitating anxiety for males than for females (Couch, Garber, & Turner, 1983; Hembree, 1988). Couch et al. (1983) and Hembree (1988) examined gender differences on Facilitating Anxiety scales in samples of ninth- to 12th-grade students as well as college students and found that males reported higher levels of facilitating anxiety than females. In contrast, P. A. Lowe and Lee (2008) found no difference in facilitating anxiety between males and females.

Relations Between Test Anxiety and Other Variables

Spielberger, Anton, and Bedell (1976) conceptualized test anxiety as a situation-specific form of trait anxiety and Reynolds (1985) stated that manifest anxiety stemmed from a trait theory of general anxiety. Therefore, the scores of measures of test anxiety and manifest or general anxiety should have positive correlations with each other. Likewise, facilitating anxiety scores should positively correlate with self-reported grade point averages (GPAs) because facilitating anxiety is reported to enhance academic performance (Alpert & Haber, 1960). In Hembree’s (1988) meta-analysis of nine studies with 1,664 college students, he reported small to large correlations between facilitating anxiety scores and GPAs. In contrast, test anxiety scores and student self-reported GPAs should negatively correlate with each other. Hembree reported a small mean correlation of −.12 between the scores of a test anxiety measure (i.e., the Test Anxiety Questionnaire [TAQ]; Sarason & Mandler, 1952) and college student GPA. Furthermore, P. A. Lowe et al. (2011) and P. A. Lowe (2014) found negligible to small correlations between the scores of test anxiety measures and lie scale scores, but these findings were reported with samples of elementary (rs = −.14 to .00) and middle school and high school (rs = −.13 to −.04) students.

The purpose of the present study was to develop a new, multidimensional measure of test anxiety for college students and to examine its psychometric properties. Various models were tested, including a two-factor model (i.e., a Facilitating Anxiety dimension and a Test Anxiety dimension), a five-factor model (i.e., a Cognitive, Physiological Hyperarousal, Social Concerns, Task-Irrelevant Behaviors, and Facilitating Anxiety dimension), a six-factor model (i.e., a Cognitive Interference, Physiological Hyperarousal, Social Concerns, Task-Irrelevant Behaviors, Worry, and Facilitating Anxiety dimension), and a six-factor model (same six factors in the six-factor model) with a higher order factor. The six-factor model and the six-factor model with a higher order factor were tested because researchers have not been able to capture the multiple dimensions of the test anxiety construct (Zeidner, 1998) and multidimensional and higher order models have become more prevalent in recent years due to the advancements in theory and research in the field of test anxiety. The five-factor model was examined because researchers (e.g., Benson & El-Zahhar, 1994; P. A. Lowe et al., 2011; Wren & Benson, 2004) have found developmental differences in the test anxiety construct, with younger versus older individuals less able to differentiate items that assess cognitive interference symptoms from worry symptoms. The two-factor model was also examined because according to Alpert and Haber’s (1960) bidirectional theory, there are two types of anxiety, facilitating and test anxiety, and the author of the current study had developed sets of items purported to capture both of these dimensions.

The hypotheses for the present study were (a) the six-factor model would provide the best fitting model to the data for the Test Anxiety Measure for College Students (TAM-C); (b) the TAM-C scores would have adequate internal consistency reliabilities; (c) gender differences would be found on the TAM-C scales, with females reporting higher levels than males on all of the scales, with the exception of the Facilitating Anxiety scale where males would report higher levels than females; (d) the TAM-C scale scores, excluding the Facilitating Anxiety scale scores, would have significantly higher correlations with the total scores of a general anxiety measure, the scores of a Test Anxiety scale, and the scores of a similar dimension (where applicable) of general anxiety than its correlations with lie scale scores; and (e) the facilitating anxiety scores would have small to large correlations with college student self-reported GPA.

Method

Procedures

Item development

The author, who has more than 15 years of experience in the development of tests and specifically in the development of measures of test anxiety, reviewed the test anxiety literature and based on the review wrote 60 items purported to measure the cognitive interference, physiological hyperarousal, social concerns, task-irrelevant behaviors, and worry dimensions of the test anxiety construct along with items assessing facilitating anxiety. In addition, the author wrote items for the TAM-C to reflect the time (i.e., before, during, or after a test) when symptoms of test anxiety are reported to occur (see Rafferty et al., 1997). Once the items were written, they were reviewed by three measurement experts and 18 college students. (The targeted age group for the TAM-C was 18- to 26-year-old college students, so the author had two 18-year-olds, two 19-year-olds, two 20-year-olds, two 21-year-olds, two 22-year-olds, two 23-year-olds, two 24-year-olds, two 25-year-olds, and two 26-year-olds review the items.) The measurement experts and the students were requested to review the items to determine whether the items were readable and aligned well with their respective dimensions. They also reviewed the items for their clarity and the possibility that some of the items might be repetitive. Five items were rewritten because they lacked in clarity. A total of eight items were deleted because of their redundancy with other items. The final version of the TAM-C consisted of 52 items and these items were administered to the college students using a web-based survey. Examples of the final items appearing on the different dimensions of the TAM-C are shown in Table 1.

OPEN IN VIEWER

Table 1.

Examples of the Items on Each Factor of the Test Anxiety Measure for College Students.

Item number	Item	Factor
31	I play with my hair during a test.	I
42	I stay home on days when I am scheduled to take a test.	I
28	My heart pounds in my chest while I take a test.	II
45	I experience shallow breathing right before a test.	II
21	I get nervous before I have to take a test.	III
46	I continue to worry about a test though it is over.	III
20	I have a hard time paying attention when I take a test.	IV
52	I have difficulty thinking clearly while I take a test.	IV
33	I worry that the other students in my class will laugh at me if I do poorly on a test.	V
41	I worry that my instructor will be upset with me if I do poorly on a test.	V
36	I focus better on a test when I feel slightly anxious.	VI
39	I feel I am more prepared when I am slightly anxious before I take a test.	VI

Note. Factor I = Task-Irrelevant Behaviors; Factor II = Physiological Hyperarousal; Factor III = Worry; Factor IV = Cognitive Interference; Factor V = Social Concerns; Factor VI = Facilitating Anxiety.

Administration

College students were recruited for the study via Qualtrics, a web-based survey. The students, who were registered with Qualtrics, received notification and a link to the survey. The students gave their consent by reading the information statement and then clicking the forward button on the survey. After this, the students completed demographic information requested, including age, gender, ethnicity, year in college, major, name of college, location of college, self-reported GPA, and region of the country where they resided. Then they completed two questionnaires, the AMAS-C and the TAM-C. Two items were also included on the survey to help determine whether students were paying attention to the items when completing the survey. The content of the two items asked students to give a yes or no response to getting perfect scores on all tests ever taken and to completing all tests ever taken in less than 1 min. Those individuals who endorsed either of these two items were not to be included in the sample. None of the college students endorsed either of these two items. Once the students completed the survey, they received a small amount of money for their participation in the study. The survey took an average of 8 min to complete. The current study was approved by the author’s institutional research review board.

Results

The sample was split in half (i.e., an odd-even split) and half of the sample (n = 360) was used to scale the measure and the other half of the sample (n = 360) was used to validate the factor structure. Imputation of the data was not needed, as there was no missing data. SPSS, Version 22 (IBM, 2013) and Mplus, Version 7.11 (Muthén & Muthén, 1998-2013) were used to perform the different analyses.

Reynolds and Kamphaus’s (2015) criteria were used to eliminate poor items on the TAM-C. The criteria used included (a) low item-scale correlations; (b) high modification index values suggesting that an item loaded not only on its own dimension, but on one or more other dimensions; (c) low item loadings; and (d) clinical judgment. The elimination of poor items involved a two-step process. First, the item-scale correlations for each item on each scale were examined. Those items with the lowest item-scale correlations in comparison with the item-scale correlations for the others items on each scale were eliminated from the TAM-C as long as the content of each of those items did not make a unique contribution to the scale and would be worth keeping to increase the representativeness of the item content on that scale or domain. (It should be noted that no items with low item-scale correlations and earmarked for elimination were retained because of the unique contribution their content would make to their respective scale.) Then those items that were retained were placed on one factor only and the factors underwent confirmatory factor analysis (CFA). Those items with high modification index values in comparison with the modification index values of the other items (clinical judgment was used in this process) and items with low factor loadings (i.e., λ < .40) with their respective dimension were identified and eliminated from the measure (see Brown, 2015). Based on the aforementioned criteria, nine items were eliminated (i.e., five items were eliminated due to low item-scale correlations, four items were deleted due to high modification index values, and no items were eliminated due to low factor loadings). Following the elimination of these items, CFAs were performed on the remaining 43 items with the second half of the college student sample to determine the best fitting model for the TAM-C.

Internal evidence of validity

A two-factor model, a five-factor model, a six-factor model, and a six-factor model with a higher order factor were examined. Because simulated studies support treating responses on a 4-point Likert-type scale as categorical variables (Finney & DiStefano, 2013), the robust weighted least squares (WLSMV) estimator was used to estimate the model parameters. CFAs conducted with the WLSMV estimator yield an adjusted chi-square estimate that provides a more accurate test statistic when data do not follow univariate and multivariate normal distributions (Curran, West, & Finch, 1996). Guidelines used to indicate good model fit included a comparative fit index (CFI) value and a Tucker–Lewis Index (TLI) value of close to .95, a root mean square error of approximation (RMSEA) of <.06, and a nonsignificant WLSMV χ² and guidelines used to indicate an adequate model fit included CFI and TLI values ≥.90, an RMSEA value of ≤.08, and a nonsignificant WLSMV χ² (Hu & Bentler, 1999). To determine whether one model was considered to be better than the other models, the better model needed to have an adequate or good model fit and a meaningfully better model fit than the alternative models (i.e., a change in the CFI, ΔCFI, value > .01 and a change in the RMSEA, ΔRMSEA, value > .015; see Canivez, Watkins, & Dombrowski, 2016; Chen, 2007; Cheung & Rensvold, 2002). In addition, chi-square difference test results were also considered for nested models.

Summary of the fit indices for the four different models are presented in Table 2. The two-factor model had a poor model fit, the five-factor model had an adequate model fit, and the six-factor model and the six-factor model with a higher order factor produced a good model fit, according to Hu and Bentler’s (1999) guidelines. Using the ΔCFI > .01 and ΔRMSEA > .015 criteria, the six-factor model had a superior model fit when compared with the two-factor model and the five-factor model, but the six-factor model was not superior to the six-factor model with a higher order factor. Because the six-factor model and the six-factor model with a higher order factor are nested models and theory and research supports both multidimensional and higher order models in the field of test anxiety, a chi-square difference test was conducted and it was found to be statistically significant, ΔWLSMV χ²(9) = 50.732, p < .001, with the six-factor (target) model providing a better fit to the data than the six-factor model with a higher order factor. Therefore, the six-factor model was selected as the best fitting model for the TAM-C. Standardized coefficients for the six-factor model are shown in Table 3. Interfactor standardized coefficients ranged from .76 to .86.

OPEN IN VIEWER

Table 2.

Summary of the Fit Indices for the Two-Factor Model, Five-Factor Model, Six-Factor Model, and Six-Factor Model With a Higher Order Factor for the Test Anxiety Measure for College Students (n = 360).

Model	WLSMV χ2	df	CFI	TLI	RMSEA [90% CI]
Two-factor	2,941.336****	859	.872	.865	.082 [.079, .085]
Five-factor	2,026.526****	850	.928	.923	.062 [.059, .065]
Six-factor	1,496.550****	845	.960	.957	.046 [.042, .050]
Six-factor with a higher order factor	1,575.862****	854	.956	.953	.048 [.045, .052]

Note. WLSMV χ² = robust mean- and variance-adjusted chi-square; df = degrees of freedom; CFI = comparative fit index; TLI = Tucker–Lewis Index; RMSEA = root mean square error of approximation; CI = confidence interval.

***

*p < .0001.

OPEN IN VIEWER

Table 3.

Standardized Factor Coefficients for the Six-Factor Model for the Test Anxiety Measure for College Students.

Standardized factor coefficients
Item No.	I	II	III	IV	V	VI
26	.73
24	.71
40	.70
13	.65
31	.54
42	.51
28		.84
45		.83
32		.81
16		.79
49		.72
37		.72
11		.69
21			.86
18			.85
29			.84
4			.83
51			.75
46			.75
9			.70
38			.69
52				.89
20				.86
43				.85
47				.83
35				.80
15				.79
27				.77
8				.72
33					.91
25					.90
41					.84
50					.82
3					.79
44					.79
10					.74
36						.86
39						.85
17						.81
30						.78
19						.77
6						.73
7						.63

Note. Factor I = Task-Irrelevant Behaviors; Factor II = Physiological Hyperarousal; Factor III = Worry; Factor IV = Cognitive Interference; Factor V = Social Concerns; Factor VI = Facilitating Anxiety.

Group differences

Means and standard deviations for males and females were computed on each TAM-C scale (see Table 4). Six independent t tests with Bonferroni corrections were then performed. The p values were set at .008. The results of these t tests indicated that females had significantly higher levels of Cognitive Interference, Social Concerns, Task-Irrelevant Behaviors, and Worry than males, with small to medium effect sizes (ds = .22-.58).

OPEN IN VIEWER

Table 4.

Means and Standard Deviations for Male and Female College Students on the Test Anxiety Measure for College Students Scales and Effect Sizes (d).

Scale	Male		Female		d
	M	SD	M	SD
Cognitive Interference	17.23*	5.34	18.44*	5.57	.22
Physiological Hyperarousal	11.96	4.41	12.72	4.63	.17
Social Concerns	13.20*	5.29	14.40*	5.57	.22
Task-Irrelevant Behaviors	12.06*	3.75	12.94*	4.00	.23
Worry	20.45*	5.86	23.82*	5.75	.58
Facilitating Anxiety	14.32	4.39	14.03	4.06	.07

***

p < .008 based on Bonferroni corrections.

External evidence of validity

Pearson r correlation coefficients were computed between the scores of the TAM-C and the scores of the AMAS-C and self-reported GPAs (see Table 5). Green and Salkind’s (2014) suggested criteria (.10 = small correlation, .30 = medium correlation, and .50 = large correlation) were used to interpret the magnitude of the correlation coefficients.

OPEN IN VIEWER

Table 5.

Correlations Between the TAM-C Scores, the AMAS-C Scores, and Student Self-Reported GPA.

Scales	TAM-C
	1	2	3	4	5	6
AMAS-C
Physiological Anxiety	.47**	.56**	.42**	.46**	.48**	.08*
Social Concerns/Stress	.41**	.35**	.46**	.37**	.42**	.02
Test Anxiety	.53**	.52**	.51**	.43**	.72**	.07
Worry/Oversensitivity	.35**	.30**	.41**	.32**	.57**	.07*
Total Anxiety	.53**	.52**	.53**	.47**	.68**	.07*
Lie	.03	.10**	.06	.02	.00	.05
Self-reported GPA	−.11**	.07	.02	−.07	.02	.12**

Note. 1 = Cognitive Interference; 2 = Physiological Hyperarousal; 3 = Social Concerns; 4 = Task-Irrelevant Behaviors; 5 = Worry; 6 = Facilitating Anxiety. GPA = college student self-reported grade point average; significance is based on the correlation being significantly different from zero in this table. TAM-C = Test Anxiety Measure for College Students; AMAS-C = Adult Manifest Anxiety Scale–College Version.

*

p < .05. **p < .01.

Medium to large positive correlations (rs = .30-.72) between the TAM-C Cognitive Interference, Physiological Hyperarousal, Social Concerns, Task-Irrelevant Behaviors, and Worry scores and the scores of the AMAS-C Anxiety scale and subscales were found. It is interesting to note that the scores of the TAM-C Cognitive Interference, Physiological Hyperarousal, Social Concerns, Task-Irrelevant Behaviors, and Worry scales had their highest correlations with the scores of the AMAS-C Total Anxiety scale, AMAS-C Test Anxiety scale, and a similar dimension where applicable on the AMAS-C. These findings provide support for the convergent evidence of validity for the TAM-C scores. In contrast, the TAM-C test anxiety scores had negligible to small correlations (rs = .00-.10) with the AMAS-C Lie scale scores. These findings provide support for the discriminant evidence of validity for the TAM-C test anxiety scores.

Z tests were computed between the correlations of the TAM-C scores, except for the TAM-C Facilitating Anxiety scores, and the AMAS-C Total Anxiety and AMAS-C Test Anxiety scores and scores of a similar dimension where applicable on the AMAS-C and the correlations of the TAM-C scores, except for the TAM-C Facilitating Anxiety scores, and the AMAS-C Lie scores. The p values were set at .0038. The results of the Z tests indicated that the correlations between TAM-C test anxiety scores and the AMAS-C Total and AMAS-C Test Anxiety scores and scores of a similar dimension where applicable on the AMAS-C were significantly higher than the correlations between the TAM-C test anxiety scores and the AMAS-C Lie scores (Zs ≥ 8.50, ps < .0001). ¹

Correlations were also computed between the TAM-C scores and college student self-reported GPAs. A small positive correlation of .12 was found between the TAM-C Facilitating Anxiety scores and college student self-reported GPAs. This result is in line with Hembree’s (1988) meta-analysis findings of small to large correlations between the scores of measures of facilitating anxiety and students’ GPAs. This finding provides support for the convergent evidence of validity of the TAM-C Facilitating Anxiety scores. Negligible to small correlations were reported between the TAM-C Cognitive Interference, Physiological Hyperarousal, Social Concerns, Task-Irrelevant Behaviors, and Worry scores and college student self-reported GPA. Correlations ranged from −.11 to .02.

Discussion

Overall, the majority of the findings of the present study provide support for the psychometric properties of the TAM-C scores. CFA results indicated that both the six-factor model and the six-factor model with a higher order factor provided a good model fit to the data. However, a chi-square difference test indicated that the six-factor provided a better fit to the data than a six-factor model with a higher order factor. Advancements in the field of test anxiety (i.e., in research and theory) support multidimensional models of test anxiety (P. A. Lowe, 2014; Wren & Benson, 2004; Zeidner, 1998). Moreover, the internal consistency reliability estimates for the TAM-C scores and their 95% CIs were considered adequate for research purposes (see Nunnally & Bernstein, 1994). In addition, females reported higher levels of test anxiety than males on the TAM-C Cognitive Interference, Social Concerns, Task-Irrelevant Behaviors, and Worry scales. These findings are in agreement with the literature, with females reporting higher levels of test anxiety than males (Hembree, 1988; Seipp & Schwarzer, 1996). Furthermore, evidence for the convergent and discriminant validity of the TAM-C scores was found.

No gender difference was found on the TAM-C Facilitating Anxiety scale in the present study. This finding is not in agreement with Couch et al.’s (1983) and Hembree’s (1988) results of males reporting higher levels of facilitating anxiety than females among secondary and college students. However, the current finding is in agreement with the result from P. A. Lowe and Lee’s (2008) study that examined factorial invariance and observed score differences across gender among 696, fourth- to 12th-grade students. Lowe and Lee found no gender difference on their Performance Enhancement (i.e., Facilitating) Anxiety scale among elementary and secondary students. Likewise, no gender difference was found on the TAM-C Physiological Hyperarousal scale. This finding was somewhat surprising but is in agreement with the result reported in P. A. Lowe and Ang’s (2012) study of test anxiety, but with a younger age group.

The direction and magnitude of some of the correlations between the TAM-C test anxiety scores and college student self-reported GPAs found in the current study were somewhat unexpected based on Hembree’s (1988) meta-analysis. However, Alpert and Haber (1960) stated that occasionally nonsignificant and positive associations between test anxiety and academic performance have been reported when the anxiety present among a group of individuals motivates students’ performance rather than hindering it. Another possibility for the correlations reported between the TAM-C test anxiety scores and students’ self-reported GPAs in the current study is that one or more variables (e.g., cognitive ability) may be moderating the relationship between test anxiety and self-reported GPA. Future research is needed to explore this issue in more detail.

The TAM-C is a comprehensive measure of test anxiety and it consists of six different dimensions. The most popular test anxiety measure used with college students now is the TAI. The TAI was developed in the United States more than 35 years ago and it has old norms and old items, consists only of two dimensions of test anxiety, and does not encapsulate the complexity of the test anxiety construct. The TAM-C may be able to fill this gap. Moreover, once additional research is conducted on the TAM-C’s psychometric properties and if the findings support the reliability and validity of the scores of the measure and norms are established, then future research should investigate whether elevations on specific dimensions of the TAM-C can be directly linked with specific interventions to assist mental health professionals in their work with college students and reduce these college students’ test anxiety. Students with elevated scores on the TAM-C Cognitive Interference, Physiological Hyperarousal, Social Concerns, Task-Irrelevant Behaviors, or Worry scales may benefit from attentional training, relaxation training, interpersonal/social skills training, behavior modification, or cognitive-behavioral therapy, respectively (Huberty & Dick, 2006; Zeidner, 1998). In addition, the TAM-C, a new multidimensional measure of test anxiety based on theory and current research, is unique as it also includes a Facilitating Anxiety scale. Alpert and Haber (1960) found that facilitating anxiety and test anxiety together served as significantly better predictors of college students’ GPA than when these predictors were used separately. Therefore, the TAM-C may be invaluable to mental health professionals working with college students in helping to identifying those students who are risk for school dropout. Once identified, the most appropriate intervention(s) for these individual students can be selected and implemented to reduce their test anxiety and improve their academic performance. Improvement in college students’ academic performance may result in higher retention and graduate rates, thus possibly increasing the likelihood that these students may have better career opportunities, the possibility of higher wages, and more lifestyle choices in the future (Stelnicki, Nordstokke, & Saklofske, 2015).

There are several limitations associated with the current study. The sample used in the present study was a sample of convenience. Samples of convenience may limit the generalizability of the findings of a research study. Another limitation was that the sample consisted of a larger percentage of females and students from the northeastern region of the United States and a smaller percentage of Caucasians and students from the southern region of the United States in comparison with the most recent U.S. Census data (National Center for Education Statistics, 2015). Future research should be conducted with a random, stratified sample to determine whether the findings reported in the current study are replicable with a more representative sample of U.S. undergraduate students. A third limitation is the possible use of self-reported GPA in the current study. The use of self-reported GPA in research studies may produce less valid results. The less than perfect relationship between self-reported GPAs and actual grades may be due to systematic biases, random errors, or both. It is possible if systematic biases exist, they may result in spurious relationships between students’ self-reported GPAs and other variables (Kuncel et al., 2005). Kuncel and colleagues suggest that research findings that include the use self-reported GPAs, especially from students who have lower actual GPAs or students who are lower in cognitive ability may yield less reliable results. Although the authors still believe self-reported GPAs can be used in research studies, self-reported GPAs should be interpreted with caution. The authors also suggest that the use of students’ transcripts in addition to self-reported GPAs can result in greater confidence in research findings. Therefore, future studies that include self-reported GPA as a variable of interest should also include student transcripts if possible. Other future research directions for the TAM-C include measurement invariance studies across different ethnic groups, countries, and test-anxious and non-test-anxious college students. (A measurement invariance study across gender has been conducted with a sample of undergraduate students on the TAM-C and strong invariance appeared tenable; P. A. Lowe, Zhu, Liu, Crouse, & Mendez, 2016.)

Although additional research is needed to examine the psychometric properties of the TAM-C, the majority of the findings of the present study support the reliability and validity of the TAM-C scores. The TAM-C appears to be a promising, new measure designed to assess test anxiety in the college student population. Test anxiety is ubiquitous on college campuses (Tatum et al., 2006) and a measure, such as the TAM-C, may be helpful in identifying those students who are test-anxious, so that mental health professionals can help these students manage their test anxiety.