Which Components of Test Anxiety Predict University Dropout?

Abstract

As test anxiety has evolved conceptually, identifying specific components contributing to educational success is essential. This study is the first to examine how different components of test anxiety are related to university dropout. Hungarian university students were recruited through the university’s website and asked to complete a series of online questionnaires at three different points over two years to monitor test anxiety levels and potential dropout or graduation during this period. Of the 98 students who completed assessments at all time points, by the final measurement, 69 had already either graduated or dropped out of their studies. Test anxiety was measured using the multidimensional TAM-C-SF (Test Anxiety Measure for College Students—Short Form). Study dropout was defined as leaving a program before graduation. Task-irrelevant behaviors—a component of test anxiety that includes restless and avoidance behaviors—were significantly associated with dropout. Higher values of cognitive interference were also significantly associated with a higher likelihood of dropout. Task-irrelevant behaviors and cognitive interference might play a key role in academic persistence among university students. These findings highlight the importance of a multidimensional approach to assessing test anxiety and suggest interventional techniques that may help diminish these factors to support students in succeeding in their studies.

1. Introduction

As modern life is heavily oriented around intellectual competence, this often gives rise to anxiety related to academic achievement. Test anxiety, a form of performance anxiety, is a situation-specific type of anxiety that often occurs when an individual perceives an assessment (e.g., a test or exam) as a threat to their abilities, self-esteem, or future prospects. It involves a mix of cognitive (worry, intrusive thoughts), affective (e.g., fear, nervousness), physiological (symptoms related to physiological hyperarousal), and behavioural (avoidance, procrastination, task-irrelevant actions) components. Test anxiety can negatively impact a student’s academic performance, persistence, and well-being (Thomas et al., 2018) and affects approximately 15–30% of students (Goonan, 2003). It is often linked to academic underperformance (Rana & Mahmood, 2010), which has been explained by various theories. While some linked test anxiety to insufficient knowledge, abilities or skills (Everson et al., 1989), others suggested that anxiety disrupts the information recall process (Liebert & Morris, 1967). More recent models offer refined explanations, incorporating both models, viewing test anxiety as a cyclical process driven by perceptions of threat and self-evaluation. For example, according to the Self-Referent Executive Processing Model (Zeidner & Matthews, 2005), test anxiety develops and persists through interactions among three systems: deficits in executive self-regulation that impair effective attention control and coping, negative self-beliefs that increase vulnerability in evaluative settings, and maladaptive situational reactions such as intrusive worry or excessive psychological arousal. These systems reinforce each other, creating a self-perpetuating cycle that can weaken academic achievement and sustain anxiety over time.

The same process may cause an individual to feel unable to face the challenges required in an academic environment, which can lead to extended study periods or even abandoning a study program altogether. Dropout is defined as the abandonment of a study program before graduation, when such withdrawal is for a period long enough to rule out the possibility of a student’s return, as proposed by Vidal et al. (2022). Huntley et al. (2019), in their meta-analysis and systematic review of test anxiety programs for university students, state that “high-test-anxious students are more likely to drop out or repeat a year of study,” referencing the study by Schaefer et al. (2007). While part of this statement is accurate, Schaefer et al. (2007) found that, among a large group of students at the University of Cologne, those with high test anxiety were more likely to report at least one semester of study delay, with many being long-term students who delayed by over two semesters. However, their retrospective data regarding actual dropout—based on students who had previously left a program—showed that a higher percentage of students with low test anxiety (21%) had left school compared to those with high test anxiety (9%). Therefore, this evidence does not clearly establish a prospective link between test anxiety and dropout, and when examined retrospectively, the relationship between these two variables can be the opposite of what is often stated in the literature.

Theories connecting test anxiety to dropout are grounded in the understanding that anxiety may result in avoidant behavior. A significant aspect of students’ identity stems from being perceived as academically competent; therefore, a fear of failure or being judged as unintelligent are indices of social-evaluative threat (Dickerson & Kemeny, 2004), predisposing individuals to react to evaluative situations as threatening. This can result in avoidant tendencies, such as performance-avoidance goals, where individuals aim to avoid demonstrating incompetence, in contrast to performance-approach goals, which focus on receiving recognition and showcasing abilities. Performance-avoidance goals are linked to dropout and are associated with minimal effort behaviors, termed “avoidance of academic work goals” (Vidal et al., 2022). Avoidant coping mechanisms, such as focusing on irrelevant stimuli during exams, behavioral disengagement, and procrastination, are theorized to exacerbate test anxiety and create a vicious cycle (Weiner & Carton, 2012). On the other hand, avoidant intentions that are related to test anxiety may not always lead to avoidant behavior. For example, Lenski et al. (2024) found that while test anxiety influenced the intention to participate in the first exam, contrary to expectations, it did not significantly predict exam postponement. However, in the study conducted by Vidal et al. (2022), test anxiety was assessed concurrently with academic outcomes as well as various internal and external factors. The findings indicated that test anxiety served as a significant predictor of dropout in STEM higher education, alongside determinants such as prior academic performance, intrinsic motivation, self-efficacy, and avoidance-related goals and behaviors. Notably, in their research, avoidant behaviors were measured separately from the construct of test anxiety through measuring performance-avoidance goals.

As a result, there is still very limited evidence about how test anxiety predicts dropout rates in higher education. Evidence for the relationship between test anxiety and dropout is complex, not only in how it may contribute to dropout but also in which components of test anxiety could be linked to it, as the concept of test anxiety has evolved significantly over the past decades. However, most studies measuring test anxiety examine it in a uni- or bidimensional way (including Vidal et al., 2022). The upcoming section will examine the evolution of the concept of test anxiety and then discuss its key components previously associated with academic success.

The Development of the Concept of Test Anxiety

The construct of test anxiety has undergone significant development in recent decades. In the earliest models, test anxiety was considered a unidimensional phenomenon (Sarason & Mandler, 1952), and in the following period, it was described along the two components proposed by Liebert and Morris (1967)—worry and emotionality. The Test Anxiety Inventory (TAI), which is still widely used today and separates cognitive worries and physical-emotional symptoms, was developed to measure these two factors (Spielberger, 1980).

However, later research revealed some limitations of this model and the test anxiety research based on it, especially its susceptibility to jingle-jangle fallacies (Putwain et al., 2021). The term “jangle” refers to the practice of applying different labels to what is essentially the same construct—for example, referring to the affective-physiological component of anxiety with “emotionality”, “tenseness”, “autonomic reactions”. Conversely, “jingle” describes the situation where the same label is used to denote multiple, distinct processes, such as the term “Worry” being used to describe various aspects of test anxiety (e.g., sometimes it was defined just as thoughts about failure, other times it also included consequences of failure, self-confidence or even social anxiety). Therefore, research on test anxiety has been hindered by inconsistencies, making it difficult to compare results across studies. To address these issues, Putwain et al. (2021) developed the Multidimensional Test Anxiety Scale (MTAS), expanding the two original components into two-two subscales: Worry, Cognitive Interference, Tension and Physiological Indicators. This four-factor model advanced previous work by breaking down the simple cognitive–emotional dichotomy and clearly distinguishing anxiety aspects that may differently impact performance. However, it overlooked other factors that could provide a more nuanced understanding of how test anxiety manifests and affects students. Putwain’s approach was primarily theory- and content validation–driven, excluding the behavioral component because observable behaviors, in his view, are too ambiguous to reliably indicate anxiety. In contrast, Lowe (2021) adopted a biopsychosocial perspective, which may better reflect students’ real experiences and help identify specific problematic anxiety components. She also emphasized the need for age-appropriate measures of test anxiety and developed a tool specifically for college students. A six-factor model of test anxiety was developed, including a 43-item measuring instrument called the Test Anxiety Measure for College Students (TAM-C), which was also validated through its shortened 24-item version (TAM-C-SF). The TAM-C-SF assesses test anxiety across six subdimensions: it includes a facilitating anxiety scale and five performance-inhibiting test anxiety scales. These anxiety components are: cognitive interference; worry, physiological hyperarousal, social concerns; task-irrelevant behaviors, facilitating anxiety.

The “worry” scale assesses anxious thoughts about not doing well on tests and the negative outcomes related to failure (e.g., “I continue to worry about a test though it is over”). The “cognitive interference” scale measures how often intrusive, distracting thoughts occur (e.g., “I have a hard time thinking clearly when I take a test”). The “psychophysiological hyperarousal” scale evaluates the intensity of physiological symptoms (e.g., “My heart pounds in my chest while I take a test”). The “social concerns” scale focuses on fears about social judgment related to poor performance (e.g., “I worry that my instructor will be upset with me if I do poorly on a test”). The task-irrelevant behaviors scale records avoidant and non-performance-related actions (e.g., “I stay home on days when I am scheduled to take a test”). “Facilitating anxiety” measures how much anxiety acts as a motivating, performance-enhancing factor rather than a paralyzing one (e.g., “I feel I am more prepared when I am slightly anxious before I take a test”). By developing the TAM-C-SF, Lowe and colleagues addressed the shortcomings of previous instruments: this multidimensional questionnaire has good psychometric indicators and provides a comprehensive picture of the characteristic components of test anxiety. The practical benefit of this approach is that it helps to identify which anxiety components are causing problems for a given student, so that interventions can be designed more specifically.

Previous research highlights that different components of test anxiety may be linked to academic success and persistence in different degrees, with cognitive components (e.g., frequent and repetitive negative thoughts, preoccupation with potential negative consequences of underperformance) often being linked as key predictors of dropout intention (e.g., Sperduto et al., 2024). Roos et al. (2021) emphasize that the importance of cognitive interference lies in intrusive thoughts depleting cognitive resources necessary for academic performance. According to the latest test anxiety measures by Putwain et al. (2021) and Lowe (2021), this construct is captured by the Cognitive Interference scale of multidimensional measures. As previously discussed, regarding university dropout, avoidant behaviors may also constitute a significant factor—now integrated into the test anxiety construct through task-irrelevant behaviors as measured by Lowe’s assessment—since they are theorized to intensify test anxiety and perpetuate a vicious cycle (Weiner & Carton, 2012).

However, most research has relied on the Test Anxiety Inventory, which only broadly identifies two factors. As a result, it remains uncertain how various dimensions of test anxiety influence academic performance when measured in a nuanced, multidimensional manner. Measuring test anxiety in a multidimensional and age-appropriate way, such as with the Test Anxiety Measure for College Students, is helpful not only for understanding how different parts relate to academic results but also for creating targeted interventions to address specific aspects of test anxiety. To our knowledge, this is the first study to examine the relationship between multidimensional test anxiety and university dropout, whereas previous research has primarily focused on early indicators and intentions to drop out (e.g., Lenski et al., 2024) or unidimensional test anxiety measures (e.g., Vidal et al., 2022). By exploring these relationships, this research aims to shed light on the psychological mechanisms influencing academic persistence, potentially informing more effective, targeted support programs.

Based on the literature reviewed and the goals of this study, the following hypotheses were proposed:

H1. 

Total test anxiety, as measured by the total score of the Test Anxiety Measure for College Students—Short Form (TAM-C-SF) across the academic period, will be associated with university dropout, with higher levels predicting increased dropout risk, as suggested by previous research on the impact of test anxiety on academic persistence (Krispenz et al., 2019; Rana & Mahmood, 2010; Vidal et al., 2022).

H2. 

The association between test anxiety and university dropout will be reflected in the cognitive interference and task-irrelevant behaviour subscales of the TAM-C-SF, such that higher scores on these components will be associated with increased dropout risk (Roos et al., 2021; Sperduto et al., 2024; Weiner & Carton, 2012).

2. Materials and Methods

The study protocol received approval from the university’s ethical review board (2018/330). Participants were recruited in February 2021 through the official website of the Hungarian Eötvös Loránd University to participate in a study examining test anxiety. Recruitment was open to all study fields via an online application, which included providing informed consent and agreeing to the anonymized publication of results. Subsequently, a virtual group meeting was conducted by our team of psychologists to further inform participants about the study process and address inquiries. Participants then completed initial assessments, which consisted of a series of online questionnaires, including the Test Anxiety Measure for College Students—Short Form (TAM-C-SF), depression, stress and anxiety measured by the Depression, Anxiety, and Stress Scale (DASS-21, Osman et al., 2012) and the State and Trait Anxiety Inventory (STAI, Spielberger et al., 1970). In Spring 2021, participants were randomly assigned to either an intervention group receiving an 8-week-long digital mental health intervention targeting test anxiety or a waitlist control group, with short-term outcomes of the program reported in Csirmaz et al. (2023). Outcomes were measured at three points over two years: at the start of 2021, three months later (which coincided with the end of the intervention for the experimental group and before the exam period), and at the two-year follow-up (which coincided with a time after graduation for most participants). Contact was maintained solely via email. At the final assessment, participants also reported on their study status, including whether they had completed or dropped out, the number of semesters postponed, and their final diploma grades. All procedures were conducted online without offering incentives.

The primary outcome of this study was the Test Anxiety Measure for College Students-Short Form (TAM-C-SF; Lowe, 2021). The TAM-C-SF is the short form of the TAM-C, a 24-item self-report questionnaire designed to measure test anxiety across six dimensions: Social Concerns, Worry, Cognitive Interference, Task Irrelevant Behaviors, Psychophysiological Hyperarousal, and Facilitating Anxiety. Items are rated on a four-point Likert scale, with higher scores indicating greater levels of test anxiety. In the current sample, the TAM-C-SF demonstrated an internal consistency with Cronbach’s alpha values ranging from 0.76 to 0.90 across different time points, suggesting the measure has a good balance between internal consistency and item diversity.

In order to distinguish the relationship between test anxiety, broader negative affect (depression, trait anxiety) and university dropout, we conducted sensitivity analyses including the Depression, Anxiety, and Stress Scale (DASS-21, Osman et al., 2012) and the State and Trait Anxiety Inventory (STAI, Spielberger et al., 1970). Depression, Anxiety, and Stress Scale (DASS-21): The DASS-21 is a 21-item Likert scale questionnaire that measures three negative emotional states: depression, anxiety, and stress (Osman et al., 2012). Items are rated from 0 (“Never applies to me”) to 3 (“Almost always applies to me”). Each subscale has seven items, and the DASS-21 provides both subscale scores and a total score. In our sample, the DASS showed good internal consistency with Cronbach’s alphas ranging from α = 0.86 to 0.92 across different time points. State and Trait Anxiety Inventory (STAI): The STAI is a 40-item self-report questionnaire designed to assess both state anxiety (an acute response; STAI-S) and trait anxiety (a tendency to respond to situations with state anxiety; STAI-T; Spielberger et al., 1970). Items are rated on a four-point Likert scale, from 1 (“Never applies to me”) to 4 (“Almost always applies to me”). Internal consistency for the STAI in this study ranged from α = 0.78 to 0.85 across measurement points.

2.1. Dropout

The conceptualization of dropout in the present study was based on the definition: the abandonment of a study program before graduation, when such withdrawal is for a period long enough to rule out the possibility of a student’s return, as proposed by Vidal et al. (2022).

2.2. Sample Size Calculation

A priori sample size analysis for a binary logistic regression was conducted using G*Power 3.1 (Faul et al., 2009). The analysis was based on prior findings showing that university dropouts scored approximately 0.30 SD higher on test anxiety compared to completers (Vidal et al., 2022). This standardized mean difference translates to an odds ratio (OR) of approximately 1.70 per 1 SD increase in test anxiety (Chinn, 2000). Assuming a two-tailed α = 0.05, power (1 − β) = 0.80, and an expected dropout rate of 25–30%, the required total sample size was estimated at N ≈ 140–160. Given the longitudinal design and anticipated attrition over two years, we aimed to recruit around 180 participants at the beginning of the study. 178 students were recruited. 127 students filled out the second measurement point, and ultimately, 98 students provided complete data across all time points, including the final measurement two years later. This leads the current analyses to be slightly underpowered to detect especially small effects, increasing the chance of false negatives, and should be viewed as exploratory, offering preliminary evidence to inform future research with larger samples. This also limited us to using only one predictor at a time, so we concentrated on simple analyses to facilitate understanding of the subject.

2.3. Statistical Analysis Plan

All analyses were conducted using IBM SPSS Statistics (Version 29). Descriptive statistics and internal consistency estimates (Cronbach’s α) were calculated for all outcome measures at each measurement point. To determine whether test anxiety predicted dropout, binary logistic regression analyses were conducted with academic status at follow-up as the dependent variable (0 = dropout, 1 = graduation). Academic status was assessed at time point 3; however, this time point did not necessarily represent a definitive academic endpoint for all participants. Although many students had either completed their studies or dropped out by this stage, some remained actively enrolled. Continued enrollment did not necessarily indicate academic delay or poorer academic functioning but often reflected differences in program start dates or program duration. Therefore, primary models focused on resolved academic outcomes (dropout versus graduation), while additional multinomial models, including still-enrolled participants, were conducted as sensitivity analyses. The primary models tested the predictive value of total test anxiety and its hypothesized subcomponents. Mean scores across time point 1 and time point 2 were used for total test anxiety, cognitive interference, and task-irrelevant behaviors. These averaged scores were treated as continuous predictors; thus, a one-unit increase reflects a one-point increase in the mean score across the two measurement points. As further exploratory analyses, hierarchical logistic regression analyses were performed. Baseline scores were entered in the first block, followed by change scores (time point 2 − time point 1) in the second block to evaluate whether changes in anxiety components predicted academic outcomes beyond baseline levels. Additionally, multinomial logistic regression analyses were conducted as sensitivity analyses to examine whether anxiety variables differentiated between students who completed their studies, dropped out, or remained enrolled during follow-up. All statistical tests were two-tailed with α set at 0.05.

3. Results

In total, 98 students completed the online questionnaires at all three time points. Out of these 98 participants, 69 people reported at the last time point that either they graduated from the program or had dropped out of it, while the rest were still currently enrolled. A total of 44 students reported graduation, and 25 had dropped out. The currently still enrolled (N = 29) participants mainly included those still within the intended length of their studies (N = 25) and 4 participants who had already delayed their graduation by at least one semester. In Hungary, final diploma grades are classified on a five-point scale. Of those who finished their studies, 4 received “excellent with distinction,” 29 received “excellent,” 17 received “good,” and one received “satisfactory” and one received “pass”. The number of missed semesters ranged from 0 to 2, with a 0.32 mean for those who dropped out later (SD = 0.56), and a 0.18 mean for those who did not (SD = 0.54). The mean age of participants was 22.27, SD = 3.77. Participants were primarily identifying as female (N = 85), while N = 14 identified as male and N = 2 as other. Most participants (N = 80) were enrolled in non-stem disciplines like psychology, education, or sociology, while 21 were focused on STEM fields such as biology, chemistry, physics, or informatics, indicating that our sample was a mixed sample, predominantly comprising non-STEM students.

Table 1. TAM-C-SF Means and SDs across Timepoints.

Measure	Timepoint 1 (M ± SD)	Timepoint 2 (M ± SD)	Timepoint 3 (M ± SD)
Total Test Anxiety	54.86; 12.54	53.65; 12.13	51.79; 15.29
Cognitive Interference	9.59; 3.17	9.29; 3.13	10.41; 2.68
Task-Irrelevant Behaviors	9.32; 2.77	9.13; 2.80	7.13; 2.70
Worry	12.16; 2.87	11.93; 2.93	7.79; 3.22
Physiological Hyperarousal	8.34; 3.00	8.08; 3.11	8.05; 3.07
Social Concerns	7.67; 3.98	7.25; 3.83	9.81; 3.20
Facilitating Anxiety	7.78; 2.84	8.11; 2.80	8.16; 2.63

Note. Higher scores indicate greater test anxiety. Total test anxiety scores are the sum of all six subscales, with facilitating anxiety reverse-coded, and can range up to 96. Each subscale consists of 4 items, with a maximum score of 16 per subscale.

shows the mean and standard deviation of the different test anxiety scores at the three measurement points, broken down into components, and the overall mean, including all components.

3.1. Baseline Differences by Follow-Up Retention

To investigate potential differences in attrition, Mann–Whitney U tests compared baseline measures of participants who provided data at all three time points with those lost to follow-up by time point 3. The comparisons included trait anxiety (STAI), depression (DASS), and all test anxiety scales. No significant baseline differences were found between retained participants and those who dropped out. Trait anxiety showed no significant difference between groups, U = 3791.00, z = 1.33, p = 0.182, nor did depression, U = 3478.00, z = 0.32, p = 0.752. Similarly, no significant differences were observed across any test anxiety components, including social concerns, U = 3901.00, z = 1.72, p = 0.085; worry, U = 3908.50, z = 1.73, p = 0.084; cognitive interference, U = 3817.00, z = 1.43, p = 0.154; physiological hyperarousal, U = 3727.50, z = 1.13, p = 0.257; task-irrelevant behaviors, U = 3483.50, z = 0.34, p = 0.738; and facilitating anxiety, U = 3029.50, z = −1.15, p = 0.250. Overall, these results suggest that attrition was not systematically related to baseline trait anxiety, test anxiety, or depression within the measured sample.

3.2. Test Anxiety and Dropout

Regarding our first hypothesis, a binary logistic regression analysis was conducted to determine whether total test anxiety throughout the academic period predicted graduation versus dropout at time point 3. Total test anxiety was calculated as the mean total test anxiety score across time point 1 and time point 2. Dropout versus graduation served as the dependent variable (0 = dropout, 1 = graduation). The overall model was not statistically significant, χ²(1) = 2.10, p = 0.147, suggesting that total test anxiety did not significantly differentiate between students who graduated and those who dropped out. The model’s explanatory power was low (Cox & Snell R² = 0.030; Nagelkerke R² = 0.041). The Hosmer–Lemeshow goodness-of-fit test was non-significant, χ²(8) = 2.56, p = 0.959. Parameter estimates showed that mean total test anxiety was not a significant predictor of graduation (B = −0.029, SE = 0.020, Wald = 2.04, p = 0.154, OR = 0.97, 95% CI [0.93, 1.01]). Overall, levels of total test anxiety across the academic period did not significantly predict graduation versus dropout. As previously noted, false negatives cannot be ruled out, as the final sample size may have been underpowered to detect small effects.

To assess whether individual TAMC-SF components—cognitive interference and task-irrelevant behaviors—predicted graduation as outlined in the second hypothesis, separate binary logistic regression analyses were conducted for each subscale. Task-irrelevant behaviors were measured as the mean task-irrelevant behaviour score across time points 1 and 2. Academic status at follow-up served as the dependent variable (0 = dropout, 1 = graduation). The overall model was statistically significant, χ²(1) = 7.20, p = 0.007, indicating that chronic task-irrelevant behavior significantly distinguished between students who graduated and those who dropped out. The model’s explanatory power was modest (Cox & Snell R² = 0.099; Nagelkerke R² = 0.136). The Hosmer–Lemeshow test was non-significant, χ²(8) = 2.15, p = 0.976, suggesting an adequate fit. Parameter estimates showed that task-irrelevant behaviors were a significant predictor of graduation (B = −0.280, SE = 0.112, Wald = 6.27, p = 0.012, OR = 0.76, 95% CI [0.61, 0.94]). Specifically, higher levels of task-irrelevant behavior were linked to lower odds of graduation and a higher likelihood of dropout. Each one-point increase in the mean task-irrelevant behavior score was associated with a 24% decrease in the odds of graduation. The model correctly classified 66.7% of cases, with higher accuracy in identifying students who graduated than those who dropped out: specificity 32%, sensitivity 86.4%.

A similar binary logistic regression was performed with mean cognitive interference (across time points 1 and 2) as the predictor. The overall model was statistically significant, χ²(1) = 4.32, p = 0.038, indicating that mean cognitive interference significantly distinguished between students who graduated and those who dropped out. Model explanatory power was modest (Cox & Snell R² = 0.061; Nagelkerke R² = 0.083). The Hosmer–Lemeshow goodness-of-fit test was non-significant, χ²(7) = 3.81, p = 0.801, suggesting the model fit well. Parameter estimates showed that mean cognitive interference was a significant predictor of graduation (B = −0.168, SE = 0.083, Wald = 4.08, p = 0.043, OR = 0.85, 95% CI [0.72, 0.99]). Specifically, higher mean cognitive interference was linked to lower odds of graduation (and thus a higher chance of dropout). Each one-point increase in cognitive interference was tied to a 15% decrease in the odds of graduation. The model accurately classified 68.1% of cases, with higher accuracy in identifying graduates than dropouts: specificity 28%, sensitivity 90.9%. Although total test anxiety did not predict graduation or dropout in our sample, two components—task-irrelevant behaviors and cognitive interference—were significant predictors. Below, further exploratory and sensitivity analyses of these variables are detailed.

3.3. Exploratory Analyses

Exploratory hierarchical logistic regression analyses were conducted to examine whether baseline cognitive interference and subsequent change (T2 − T1) jointly predicted graduation versus dropout status, and whether change provided additional predictive value beyond baseline. Baseline cognitive interference and changes in cognitive interference (time point 2 − time point 1) during the academic period were used as predictors in a hierarchical logistic regression analysis. Baseline cognitive interference was entered in Block 1, followed by the change in cognitive interference (time point 2 − time point 1) in Block 2. Baseline cognitive interference alone did not significantly predict graduation versus dropout, χ²(1) = 2.67, p = 0.102. The addition of change in cognitive interference resulted in a trend-level improvement in model fit, Δχ²(1) = 3.76, p = 0.052. The overall two-predictor model was statistically significant, χ²(2) = 6.43, p = 0.040, explaining a modest proportion of variance (Cox & Snell R² = 0.099; Nagelkerke R² = 0.134).Within the final model, baseline cognitive interference emerged as a significant predictor of graduation B = −0.198, SE = 0.093, Wald = 4.57, p = 0.033, OR = 0.82, 95% CI [0.68, 0.98], indicating that higher baseline cognitive interference was associated with lower odds of graduation. Change in cognitive interference did not reach statistical significance, B = −0.317, SE = 0.170, Wald = 3.49, p = 0.062, OR = 0.73, 95% CI [0.52, 1.02], although the direction of the effect suggested that increases in cognitive interference over time were associated with reduced likelihood of graduation. The Hosmer–Lemeshow goodness-of-fit test was non-significant, χ²(8) = 6.82, p = 0.557, indicating adequate model fit. Overall classification accuracy was 69.4%, with higher accuracy for identifying completers than dropouts: 34.8% specificity and 89.7% sensitivity. Baseline cognitive interference did not significantly predict graduation in the unadjusted model; however, when the change in cognitive interference (T2 − T1) was included, it became a significant unique predictor, indicating a conditional association within the two-predictor framework. The effect of change alone only approached statistical significance and should be interpreted cautiously due to the exploratory nature of the analysis and limited statistical power.

A similar analysis was conducted for baseline task-irrelevant behaviors and change in task-irrelevant behaviors during the academic period (time point 2 − time point 1). The model, including baseline task-irrelevant behaviors was statistically significant, χ²(1) = 7.32, p = 0.007, indicating that baseline task-irrelevant behaviours significantly distinguished between students who completed their studies and those who dropped out. Model explanatory power was modest (Cox & Snell R² = 0.111; Nagelkerke R² = 0.152). The Hosmer–Lemeshow goodness-of-fit test was non-significant, χ²(8) = 2.14, p = 0.977, suggesting adequate model fit. Parameter estimates indicated that baseline task-irrelevant behaviors significantly predicted graduation, B = −0.252, SE = 0.099, Wald = 6.45, p = 0.011, OR = 0.78, 95% CI [0.64, 0.94]. Higher baseline task-irrelevant behavior scores were associated with lower odds of graduation (and thus a higher likelihood of dropout). The model correctly classified 67.7% of cases, with higher accuracy for identifying completers than dropouts. Adding change in task-irrelevant behaviors did not significantly improve model fit, Δχ²(1) = 1.48, p = 0.225. The overall two-predictor model nevertheless remained statistically significant, χ²(2) = 8.80, p = 0.012, explaining a modest proportion of variance in study graduation. Within the final model, baseline task-irrelevant behaviours remained a significant predictor, B = −0.328, SE = 0.122, Wald = 7.18, p = 0.007, OR = 0.72, 95% CI [0.57, 0.92]. In contrast, change in task-irrelevant behaviours was not a significant predictor of graduation, B = −0.188, SE = 0.155, Wald = 1.48, p = 0.225, OR = 0.83, 95% CI [0.61, 1.12]. Model fit indices showed adequate fit (Hosmer–Lemeshow χ²(8) = 4.77, p = 0.782), and the model correctly classified 66.1% of cases: specificity 34.8%, sensitivity 84.6%.

Taken together, the exploratory analyses revealed distinct patterns across components. For cognitive interference, baseline levels alone were not significant, but incorporating change over time improved the model fit and produced a significant combined model. Conversely, task-irrelevant behaviors were primarily driven by baseline levels, with no significant incremental contribution from the change. Given the limited sample size and exploratory nature of these analyses, these findings should be interpreted strictly as preliminary.

3.4. Sensitivity Analyses: Controlling for Trait Anxiety

To determine whether the observed dropout reflected broader vulnerability, sensitivity analyses were conducted by controlling separately for baseline trait anxiety and depression in hierarchical binary logistic regressions. In all models, the results remained consistent. Task-irrelevant behaviors continued to significantly predict decreased odds of graduation (OR = 0.74, 95% CI [0.59, 0.94], p = 0.014), and cognitive interference remained an important predictor (OR = 0.85, 95% CI [0.72, 1.00], p = 0.043), while trait anxiety was not significantly linked to the likelihood of graduation in any model (all ps > 0.70). Including trait anxiety did not substantially improve model fit or affect effect sizes. A similar outcome was observed when controlling for depression scores: logistic regression showed that higher levels of cognitive interference significantly predicted lower odds of graduation (OR = 0.82, 95% CI [0.68, 0.99], p = 0.043), even after accounting for depressive symptoms. Depressive symptoms were not a significant predictor (p = 0.570). Full model statistics are provided in the Supplementary Materials.

3.5. Sensitivity Analyses: Intervention Timing and Ongoing Enrollment

A multinomial logistic regression was performed as a sensitivity analysis to test robustness with the inclusion of participants still enrolled at Time Point 3 and to consider intervention timing, which is inherent in the waitlist-controlled design. The overall pattern of results closely matched those of the primary binary models: task-irrelevant behaviors significantly predicted dropout compared to graduation (OR = 1.32, 95% CI [1.06, 1.65], p = 0.012), but not graduation compared to still enrolled. Intervention timing was not linked to study status. A second multinomial model, including cognitive interference as the predictor, showed a partly similar pattern. Higher cognitive interference was associated with higher odds of dropout relative to graduation (but not still enrolled to graduation) at the individual predictor level (OR = 1.18, 95% CI [1.00, 1.40], p = 0.048); however, the overall model was not statistically significant. Intervention timing again showed no association with study status. Full model statistics are provided in the Supplementary Materials.

4. Discussion

Although our final sample size was too small to reliably detect small effects—raising the chance of false negatives—this may explain why our first hypothesis, that total test anxiety scores are related to study dropout, was not statistically significant. It is possible that some components of test anxiety are more influential on dropout than the overall scores; however, further research is needed to investigate this. Nonetheless, our analyses still uncovered important insights within the multidimensional perspective of test anxiety. Task-irrelevant behaviors, a new subscale included in Lowe’s (2021) contemporary model that captures avoidant behaviors often reported alongside test anxiety, were significantly linked to an increased likelihood of dropout. This supports earlier research highlighting the importance of avoidance in educational outcomes (Krispenz et al., 2019; Vidal et al., 2022). Additionally, cognitive interference—characterized by intrusive, focus-disrupting thoughts—was also significantly linked to a higher likelihood of dropout. Other research has recognized cognitive interference as a major predictor of students’ intentions to drop out (Rana & Mahmood, 2010; Roos et al., 2021; Sperduto et al., 2024), and our results align with this evidence. Furthermore, we propose that behavioral disruptions may represent another important dimension to consider in academic persistence. This slight discrepancy in the results of previous studies—and how they compare to ours—likely arises from differences in measurement methods. While Roos et al. (2021) used a single-item measure of test anxiety, Sperduto et al. (2024) and Rana and Mahmood (2010) employed multidimensional scales that did not separately measure cognitive interference and task-irrelevant behaviors. These findings underscore the importance of employing multidimensional and age-appropriate instruments, such as the TAM-C or the Multidimensional Test Anxiety Scale (Putwain et al., 2021), which can distinguish among the dimensions of test anxiety. This approach not only enhances the understanding of the construct but also provides insight into practical interventions tailored to specific subscales. For example, Lowe (2021) suggest implementing interventions like attentional training, relaxation techniques, social skills development, behavior modification, and cognitive-behavioral therapy to specifically target high scores in different TAM-C dimensions: cognitive interference, physiological hyperarousal, social concerns, task-irrelevant behaviors, and worry, respectively. Since task-irrelevant behaviors emerged as the most significant dimension, these findings tentatively suggest that interventions may benefit from considering this factor, focusing on strategies to help participants reduce avoidant behaviors and the causes underlying them.

While larger studies and further investigation of mediating and moderating factors are necessary to validate our findings, the multidimensional analysis of test anxiety indicates that future interventions might focus on specific aspects to enhance academic persistence and success. As for task-irrelevant behaviors, programs targeting procrastination and avoidant coping show promise in addressing behaviors beyond task completion. For example, Krispenz et al. (2019) used the inquiry-based stress reduction (IBSR) method, where students explored test-related fears through cognitive inquiry, leading to reduced test anxiety and procrastination by increasing self-efficacy. Training programs involving cognitive behavioral therapy (CBT) and behavioral activation—using structured action plans, time frames, and goal setting—may be similarly effective (Rozental et al., 2018; Van Eerde & Klingsieck, 2018). Additionally, behavioral activation and learning habit restructuring involve setting small-step goals, using timed focus blocks (e.g., 25–30 min), rewarding progress, managing stimuli, and exposure to exam scenarios (such as mock exams and timed tasks). These CBT and behavioral techniques have been shown to reduce anxiety and enhance performance by meta-analyses conducted on test anxiety interventions (Hembree, 1988; Ergene, 2003; Huntley et al., 2019). Additionally, structured action planning and “if-then” implementation intentions can help prevent procrastination. Meta-analytic results of cognitive-behavioral interventions also support the usefulness of these techniques (Rozental et al., 2018; Van Eerde & Klingsieck, 2018; Krispenz et al., 2019). Cognitive interference, i.e., intrusive, distracting thoughts, may also be targeted. Attention training that enhances attentional switching, selective focus, and mental “switching” skills can be effective in managing such thoughts. These include, for example, sound source tracking tasks, rapid focus shifts, or “thought-switching” exercises. Mindfulness-based attentional control and cognitive defusion are other helpful techniques: brief, daily attentional anchors (e.g., breath counting), “thought = not fact” defusion exercises, or scheduled worry periods that let distracting thoughts “park” may help lower anxiety (Hembree, 1988; Huntley et al., 2019). This can be supplemented with cognitive restructuring, which detects automatic thoughts that create interference and develops their rational reappraisal. Simultaneously, a clear formulation of the “next step” aids in re-establishing task-focused behavior (Ergene, 2003; Huntley et al., 2019).

Digital implementation of these techniques may help even more students combat test anxiety effectively. Just-in-time adaptive interventions, which aim to provide the right type and amount of support at the right time by adjusting to an individual’s changing internal and contextual states—as described by Nahum-Shani et al. (2018)—can remind students of typical procrastination triggers through push notifications or incorporate short, 2–3 min micro-exercises (such as IBSR prompts, defusion, or attention resets) to help manage the intense workload during exam periods. Adaptive content distribution can allow a digital intervention to automatically prioritize the related modules based on the screening scores of a student on a multidimensional test anxiety measure. In addition, visual feedback on progress—such as displaying subscale results in the form of curves or a self-efficacy scale—strengthens the sense of control and has a motivational effect on students (Krispenz et al., 2019).

The results of this study offer preliminary support for the multidimensional concept of test anxiety, which could help guide future research aimed at improving practical approaches for effective psychological interventions to prevent university dropout.

The multifactorial nature of academic success should also be considered, as it results from a complex interaction of variables such as motivation, prior academic achievement, life circumstances, and institutional support (Vidal et al., 2022). Although we performed sensitivity analyses controlling for trait anxiety and depressive symptoms, the study did not include a wider set of covariates (e.g., gender, baseline academic performance, field of study, dropout timing) in fully adjusted multivariable models. This was due both to limited statistical power and, in some cases, the unavailability of systematically collected data. Therefore, while our findings indicate that specific multidimensional components of test anxiety are linked to graduation status beyond general distress, the possibility that these effects partially reflect broader academic disengagement, vulnerability or other underlying factors cannot be ruled out. Future research with larger samples should examine these associations within comprehensive predictive models.

However, it is important to emphasize that dropping out does not always mean a negative outcome. According to some research, leaving higher education is often more of an adaptive decision or self-fulfillment than a sign of failure. For example, Schaefer et al. (2007) showed that students with lower test anxiety were in some cases even more likely to discontinue their studies, which the authors interpreted as a positive indicator of psychological well-being and self-awareness, since in many cases students recognize when they can better fulfill their needs or goals elsewhere. Similarly, Tinto (1993); and Bean and Metzner (1985) also report on the possibility of “voluntary dropout” in their models, when the completion of certain studies is the result of a rational, informed choice and an adaptive decision—for example, due to the choice of a course of study that is more in line with personal interests, a promising job opportunity, starting a business or even family reasons. In line with this, the European review by Heublein (2014) points out that among the reasons for student dropout, there are often so-called “positive pull factors”, which also represent a more promising, more satisfying opportunity for the given individual. Dropout cannot, therefore, be interpreted exclusively as a deficit, as it can also be a potentially constructive path of development, which can be interpreted depending on individual motivations and later life paths. At the same time, it is important to emphasize that the successful completion of studies in itself remains a fundamental indicator of academic competence and perseverance, which in the long term opens up numerous opportunities in the labor market and social advancement. Future research is essential to confirm and build upon our findings, providing a stronger empirical foundation for understanding. Moreover, in addition to test anxiety programs, interventions that strengthen self-efficacy, social support, and the perceived value of coursework might be critical, as they equip students to navigate barriers more effectively and bolster persistence (Chemers et al., 2001; Fan & Williams, 2010).

Limitations and Future Directions

A key limitation of this study, apart from its exploratory nature due to the underpowered final sample size, is the gender composition, which was predominantly female (85%). This precludes analysis of gender differences. Female students report higher levels of test anxiety and avoidance related to test anxiety than males, which can impair their academic performance to a greater extent (Karelis et al., 2020; Yusefzadeh et al., 2019; Liew et al., 2014). Hence, as obtaining a degree has significant long-term benefits, it can lead to significant disruptions in one’s future prospects if they fail to do so. In OECD (Organisation for Economic Co-operation and Development) countries, adults with a college/university degree earn on average ~39% more than those with a secondary education, while those with a master’s or doctoral degree have an average wage advantage of ~83% (OECD, 2025). It is therefore important that women receive adequate support to manage their test anxiety, as if it prevents them from completing their studies, it can put them at a long-term economic disadvantage. While the focus on a vulnerable demographic provides valuable insights, future research should aim to include more diverse populations to enhance generalizability.

Additionally, the reliance on self-report measures may introduce response biases, such as social desirability or recall bias, potentially distorting the true levels of anxiety. Future studies should consider adding physiological measures alongside questionnaires. Our study would have benefited from an additional qualitative approach, where students would elaborate on their specific reasons for dropping out.

Furthermore, additional factors and mechanisms of study dropout should be explored. Incorporating variables such as self-efficacy, perfectionism subtypes, and emotional regulation difficulties could provide valuable insights into how they mediate or moderate the relationship between anxiety and academic persistence. The operational definition of dropout, which classifies participants as either graduating or withdrawing from their study programs, may oversimplify the complexity of academic persistence. Including more detailed dropout patterns, such as delayed graduation or temporary stops, as well as gathering data on when dropouts occur, could offer a more complete understanding of the topic. Although we collected data on study interruption and final grades, the sample sizes in those cases were considered too small to include in the regression analyses; therefore, only descriptive data were reported.

Finally, this study did not explore the role of potential neurodivergent traits, such as sustained attention difficulties or executive functioning deficits, which may underlie task-irrelevant behaviors or cognitive interference in certain cases. Future research could investigate how these factors intersect with test anxiety and study persistence, particularly among students with neurodivergent profiles. Addressing these limitations would help to build a more nuanced understanding of the factors influencing academic success and inform the development of tailored interventions.

5. Conclusions

This study emphasizes the importance of using multidimensional measures of test anxiety and creating targeted interventions to address its different components. It may be helpful for interventions targeting test anxiety to assist students in recognizing and controlling task-irrelevant behaviors, cognitive interference and their root causes, thereby supporting their academic performance persistence. Customizing interventions to individual needs based on anxiety levels and focusing on the specific dimensions of test anxiety that a student finds challenging could further increase their effectiveness.