In educational psychology, learning motivation has received considerable attention from researchers. Motivational theories are primarily adopted in studies that use social cognitive frameworks and that regard learning motivation as the product of the cognitive assessment of learners. Researchers have begun focusing more on academic emotions; however, studies on test emotions are greatly lacking. Test emotions are essential to a student’s learning process and have a profound effect on learning effectiveness. However, most studies on test emotions still focus on test anxiety, and studies on other types of test emotions are relatively rare. Therefore, this study explored various types of test emotions and analyzed their variable components.

According to Pekrun et al. (2002), emotions are caused by factors such as genetic disposition, physiological processes, and cognitive appraisal. Academic emotions may contain both trait and state components. Regarding the state and trait components of test emotions, Pekrun et al. (2004) defined test emotions as subjective emotions that are associated with exams and may be experienced by learners before, during, or after an exam. As long as the emotion related to the test is experienced, the predisposition of a learner to experiencing it habitually is called a trait test emotion; if a specific time point is emphasized (e.g., before, during, or after the test), the experienced test emotion is a state test emotion. Therefore, test emotions, like most psychological variables, may be affected by both internal personal and external environmental forces, meaning that they have both trait and state components (Cherng, 2013–2015). Strong empirical evidence on the composition of state and trait components of test emotions is still lacking. This study was based on the premise that test emotions experienced at specific time points are more dependent on the environment than are academic emotions.

In motivational research, studies that have explored the interactive effects of context and personal traits have mostly been experimental studies that performed analyses of variance (Deinzer et al., 1995). However, such person-centered analytical methods cannot clearly establish whether certain psychological attributes contain more trait components or more state characteristics. With the development of structural equation modeling techniques, various models proposed by latent state–trait theory (LST) allow researchers to estimate the effect of the environmental context on measurements of psychological variables in nonexperimental correlational studies.

LST, developed by Steyer et al. (1992) and Steyer and Partchev (2001), is based on the estimated variation ratio of observed variables and aims to reveal the trait or state components a psychological attribute contains (Keith et al., 2003). LST is based on classical test theory. According to classical test theory, the observation variable is equal to the sum of the true score and the error variable. Steyer et al. asserted that true scores can be used to describe individuals in a given context. Thus, true score variables are considered latent state variables, which can be decomposed into “individual,” “situational,” and “individual and contextual interaction” components. Individual components are called latent trait variables. By estimating the three components, the consistency coefficient and occasion specificity can be calculated. Consistency is defined by the ratio of the variation of the underlying trait variable to the variation of the observed variable. Occasion specificity is defined by variation of the observed variable caused by the situation or the interaction effect of the situation and the individual (Cherng, 2013–2015).

According to LST, consistency and occasion specificity must be measured at least twice to determine the estimations of the psychological variables. Moreover, at least two scales must be measured . Studies have shown that if the number of occasions is greater than three and the interval between occasions is 6 months to 1 year, then the estimated consistency and occasion specificity are relatively stable (Cherng, 2013–2015; Deinzer et al., 1995; Steyer et al., 1992). Therefore, most studies conducted on the basis of LST theory have used longitudinal research designs.

This study adopted a longitudinal research design to analyze the trait and state components of test emotions on the basis of LST theory and compared the consistency and occasion specificity of test emotions. Four latent state–trait models were constructed, including the multistate (MS) model, multistate multimethod (MSMM) model, single-trait multistate (STMS) model, and single-trait multistate multimethod (STMSMM) model. A two-year longitudinal study was conducted, and 403 participants in seventh grade were recruited to complete the test emotions scale in three waves each semester. The data were analyzed using the competitive model approach to compare the fitness of the four models.

The results revealed that most test emotions (hope, anxiety, and boredom before the test; enjoyment, anxiety, shame, and hopelessness during the test; relief, anger, and hopelessness after the test) fit the observed data when the MSMM model was applied. A few test emotions (enjoyment and hopelessness before the test and shame after the test) had the best fit when applying the MS model. Most test emotions of the middle school students were dominated by state components.

The STMS model of anger during the test revealed both trait and state components, and the consistency was less than the occasion specificity. Although anger is affected by traits, it is still primarily determined by the state. Some test emotions (e.g., hope during the test and enjoyment and pride after the test) of middle school students have both trait and state components and were partially affected by the method . However, further analysis indicated that the method-specific values of emotions were low; therefore, the results of the odd-numbered and even-numbered questions of the test emotion scales used were not significantly different.

The results of this study confirmed that some test emotions, like academic emotions, can be determined by both state and trait components, which is consistent with the hypothesis of Pekrun and colleagues (Pekrun et al., 2002) that “the academic emotions are jointly determined by individual genetic predisposition, physiological functioning and cognitive assessment.” Most test emotions are determined by the current state of an exam. The results are also consistent with the prediction that test emotions refer to the various subjective emotions that learners experience when taking a test (Pekrun et al., 2004). Emotions are triggered by a specific event (i.e., an exam) and are therefore more state-dependent than are academic emotions. Based on the findings of this study, suggestions are provided for future research.