As previously mentioned, temporal changes in expectancy-value beliefs regarding a class or a course of study have been empirically assessed in terms of developmental changes (e.g., Wigfield, 1994) or after performance feedback (e.g., Benden and Lauermann, 2022) rather than changes of study motivation for a specific study task related to the proximity of a temporal landmark. Still, considering this research is likely informative for two reasons: First, levels of more general expectancy-value beliefs regarding a class or a study domain are likely correlated with expectancy-value beliefs for specific study tasks (Hulleman et al., 2016; Dietrich et al., 2019). For example, it seems more likely that a student with high motivation for a class or their studies in general at the end of the semester reports high study motivation for a specific study task at that time than a student with low motivation for a class. Indeed, it seems plausible that students’ task-specific motivation various around their class- or domain-specific motivation. Secondly, many longitudinal studies assessed changes in expectancy and value over one full semester (e.g., Dresel and Grassinger, 2013; Perez et al., 2019). Hence, the end of the assessment period often coincides with relevant “high stakes” deadlines at the end of a semester, although such deadlines can already appear during a semester (e.g., Muis and Edwards, 2009; Benden and Lauermann, 2022). It thus seems plausible that motivational effects of exams are to some degree reflected in these studies.

However, studies that directly assessed the development of students’ achievement motivation over one semester almost unanimously found a significant decline in both expectancy and value beliefs over time, regardless of differences in measurement specificity and operationalization (Zusho et al., 2003; Dresel and Grassinger, 2013; Johnson et al., 2014; Sonnert et al., 2015; Kosovich et al., 2017; Kaplan et al., 2018; Robinson et al., 2019; Benden and Lauermann, 2022, with the exception of high performing students; for descriptive declines, see Perez et al., 2019; see Grays, 2013, for statistically non-significant changes). Decreases in expectancy and value beliefs over time are also well-documented among school children (for a recent meta-analysis, see Scherrer and Preckel, 2019). Even considering the methodological caveats mentioned above, these findings seem to be contrary to the predictions of the temporal discounting models, warranting further investigation.

The decline in class- or domain-specific expectancy and value beliefs over time becomes even somewhat more surprising when considering the changes in the time students spend studying (i.e., the quantity of study activities) over time: In concordance with theoretical predictions, research that directly assessed students’ study time in relation to a temporal landmark consistently found that students’ study activities increase toward exam dates at the end of the semester. These increases are sometimes dramatic and referred to as “cramming” (Schouwenburg and Groenewoud, 2001) or “deadline rush” (Steel and Weinhardt, 2018). Indeed, Taraban et al. (1999) investigated the pattern of study behavior based on self-report measures and data trails for three weeks and reported that students studied mainly just before exams (see also Blasiman et al., 2017). The increase in study efforts toward a temporal landmark is fairly universal, even if the exact shape of the trajectory of study quantity varies somewhat in the literature, presumably depending on the time period covered (Schouwenburg and Groenewoud, 2001). For example, Liborius et al. (2019) reported that students’ learning time trajectories over one full semester were best described by a quadratic function. In a reanalysis of the data used by Taraban et al. (1999) and König and Kleinmann (2005) found that the increase in study activities followed a hyperbolic increase in a different study over ten weeks, which was also reported by Dewitte and Schouwenburg (2002). Similarly, Capelle et al. (2022) reported that university students’ probability to engage in study activities increased exponentially over two weeks before exams. This unambigious constellation of findings suggests that temporal landmarks can be considered “strong” situations in the sense that they elicit fairly similar behavior (i.e., increasing their study quantity) among most individuals (Mischel, 1977; Fleeson, 2007; Peetz and Wilson, 2013; Dweck, 2017). In the present research, we therefore expected students’ study quantity to increase as an exam comes closer in time.

Most motivation theories distinguish in some way between the quantity of motivation (i.e., how much a student is motivated overall) and the quality of motivation (i.e., why a student is motivated to pursue a specific study activity; e.g., Grund, 2013; Hofer and Fries, 2016; Senko, 2016). These qualitative differences pertain to different incentives (i.e., the value component as opposed to the expectancy component in an expectancy-value framework; Bong, 2001; Grund, 2013; Hulleman et al., 2016). For the purpose of this study, we consider qualitative differences of motivation in terms of the fundamental duality of approach and avoidance motivation (Elliot and Thrash, 2001; Harackiewicz et al., 2002; Elliot, 2006; Elliot and Friedman, 2007; Feltman and Elliot, 2012). Specifically, we consider the distinction between performance approach and performance avoidance goals (Elliot and Harackiewicz, 1996; Elliot, 1999): If a student studies because they hope to gain something such as demonstrating their knowledge by outperforming others, they are motivated by a performance approach goal. If a student studies because they want to avoid negative consequences such as failing an exam, they are motivated by a performance avoidance goal (Elliot, 2006; Elliot and Friedman, 2007; Bjørnebekk and Gjesme, 2009; Feltman and Elliot, 2012).¹ We considered a focus on performance approach and performance avoidance goals appropriate, as they are valid and widely used constructs to assess the approach-avoidance duality in educational psychology (e.g., Elliot and Church, 1997; Cury et al., 2006; Senko et al., 2011; Scott et al., 2015) and seem particularly well-suited in the context of an upcoming exam as a “high-stakes” performance test (Harackiewicz et al., 2002; Cole and Osterlind, 2008; Darnon et al., 2009).

Previous research has consistently linked more general (i.e., trait-like regarding an overall study course or domain; Barron and Harackiewicz, 2003) performance-avoidance goals with worse performance- and wellbeing-outcomes (Murayama et al., 2011; Senko et al., 2011; for overviews, see DeShon and Gillespie, 2005; Wigfield and Cambria, 2010). In contrast, performance approach goals are mostly associated with better performance-outcomes and, in some cases, wellbeing-outcomes (Harackiewicz et al., 1998; Senko et al., 2011; Urdan and Kaplan, 2020; for meta-analyses, see Hulleman et al., 2010; Huang, 2012), but not uniformly so (e.g., Midgley et al., 2001; Senko, 2016; Miller et al., 2021), likely due to differences in operationalizations (Hulleman et al., 2010; Senko et al., 2011; Senko and Dawson, 2017).

Available research linking class-specific goals and performance such as final grades among college students has found that performance approach goals are positively related to class-performance, while performance avoidance goals are negatively liked to class performance (Elliot and Church, 1997; Church et al., 2001; Elliot and McGregor, 2001; Baranik et al., 2010). The same pattern can be found regarding associations between task-specific goals and task-performance (Elliot, 1999; Zusho et al., 2005; Pekrun et al., 2009; Baranik et al., 2010). Indeed, a meta-analysis of experimentally induced goals found that approach goals (including performance approach goals) were causally linked with better performance in laboratory settings (Van Yperen et al., 2015). Moreover, task-specific performance approach goals have been positively linked to positive achievement emotions and task interest, while task-avoidance goals are positively linked to negative achievement emotions and negatively linked to task interest (Zusho et al., 2005; Pekrun et al., 2009). By and large, the associations of performance approach and avoidance goals and relevant educational outcomes are consistent, regardless of level of specificity (Huang, 2012; Linnenbrink-Garcia et al., 2012; cf. Richey et al., 2018).

By considering performance approach and avoidance goals in addition to expectancy and attainment value, we focus on combining perspectives on motivation rather than considering other commonly used subfacets of task value (i.e., intrinsic and utility value as well as costs, e.g., Barron and Hulleman, 2015; Eccles and Wigfield, 2020), although these subfacets also consider different reasons why a student might study. We had several reasons for this decision: First and most fundamentally, considering the duality of approach and avoidance goals promises to be a conceptually meaningful addition to the duality of expectancy and value dimensions of motivation (Elliot and Covington, 2001; Elliot, 2006, 2023; Liem et al., 2008; Hulleman et al., 2016). Secondly, the positively-valenced subfacets of value are often highly correlated and generally positively related to educational outcomes, as outlined above (e.g., Eccles and Wigfield, 1995; Jacobs et al., 2002; Kosovich et al., 2015). As such, we considered it most promising to focus on expectancy and attainment value (i.e., subjective importance) as apparently valid measures of motivation quantity. By contrast, performance approach and avoidance goals are generally related to distinct educational outcomes, suggesting them to be a more meaningful measure of motivation quality (Harackiewicz et al., 2002; Murayama et al., 2011). Finally, the approach-avoidance distinction allowed us to make differential theoretical predictions on how these are discounted over time by drawing on older conflict theories.

As previously mentioned, Temporal Motivation Theory recognizes different effects of approach and avoidance incentives without making specific predictions regarding their trajectories (Steel and König, 2006; Steel and Weinhardt, 2018). However, older conflict theories (Miller, 1944; Lewin, 1946; Losco and Epstein, 1974; Epstein, 1978; Shelley, 1994; for overviews, see Trope and Liberman, 2003; Steel and König, 2006) predict that approach and avoidance incentives develop differently over time: “The strength of avoidance increases more rapidly with nearness than does that of approach.” (Dollard and Miller, 1950, p. 352). Put differently, the discounting function of avoidance incentives is assumed to be steeper than that of approach incentives (Trope and Liberman, 2003). Whenever a temporal landmark is ambivalent (i.e., characterized both by approach and avoidance incentives), it is thus predicted that approach incentives are adopted to a relatively higher degree when the landmark is further away and avoidance incentives are increasingly adopted to a relatively higher degree as it comes closer in time (Trope and Liberman, 2003). In the present case of an exam and performance goals, this would mean that the goal to perform better than others is more likely to be salient when the exam is farther away and the goal to avoid failure becomes more salient as the exam draws closer.

Direct empirical evidence for differential discounting functions of approach and avoidance incentives relies on operationalizing them in terms of gains and losses of monetary rewards, respectively. While Mogilner et al. (2007) report evidence that avoidance goals are indeed discounted more steeply, other studies found evidence to the contrary, i.e., that losses are discounted less steeply than gains (Loewenstein, 1987; Benzion et al., 1989; for a review, see Frederick et al., 2002).

Studies that investigated changes in university students’ performance approach and avoidance goals longitudinally over actual passing time are rare and show similarly contradictory results: Task-specific performance approach and avoidance goals were found to increase, decrease, and remain stable between different tasks both when they were task-specific (e.g., exams or writing assignments; Fryer and Elliot, 2007; Muis and Edwards, 2009; Han, 2016) and course-specific (Senko and Harackiewicz, 2005; Jagacinski et al., 2010). Overall, however, the adoption of performance avoidance goals appears to be relatively more likely to increase over time and the adoption of performance approach goals more likely to be either stable or decline (Senko and Harackiewicz, 2005; Fryer and Elliot, 2007; Muis and Edwards, 2009). Similar patterns have been found for trait-like goal orientations (Dresel and Grassinger, 2013; for an overview among school children, see Scherrer and Preckel, 2019). The available evidence thus suggests a pattern of changes that is broadly compatible with the predictions of conflict theories. However, changes in goal adoption were mostly considered between different exams or assignments rather than with reference to an upcoming temporal landmark (e.g., Senko and Harackiewicz, 2005; Fryer and Elliot, 2007; Muis and Edwards, 2009). As such, these relative changes in goal adoption that were found likely reflect effects of direct performance feedback rather than effects of temporal proximity (Senko and Harackiewicz, 2005; Benden and Lauermann, 2022) and can thus inform the present research only to a limited degree.

As the present research is, to the best of our knowledge, the first to address potential changes among performance approach and avoidance goals as a relevant exam comes closer, and against the backdrop of the inconclusive state of the empirical evidence just outlined, we base our predictions on changes in motivation quality mainly on the theoretical arguments made by conflict theories (e.g., Trope and Liberman, 2003). We now consider the quality of study activities.

Study time can be spent pursuing qualitatively different activities, which are commonly categorized into learning strategies (e.g., Pintrich et al., 1993). Learning strategies thus describe how students study. Different types of learning strategies have been proposed and investigated in the context of self-regulated learning (e.g., Pintrich et al., 1993; Pintrich, 2004). We focus on cognitive learning strategies because they are concerned with the actual processing of learning material. Moreover, the distinction between surface and deep learning strategies (also referred to as elaboration strategies) captures the idea of different qualities of study activities particularly well (e.g., Gow and Kember, 1990; Biggs et al., 2001; Duncan and McKeachie, 2005; Justicia et al., 2008). Surface strategies refer to learning activities such as memorizing disjointed pieces of information over a short time period, typically with the goal of reproducing them in order to pass an exam (Gow and Kember, 1990; Entwistle, 1998). In contrast, deep learning strategies refer to learning activities such as coming up with new examples and finding connections between different concepts, typically with the goal of gaining a deep and long-term understanding of the respective study content (e.g., Vermunt, 1998; Zlatović et al., 2015). While the use of deep learning strategies instead of surface learning strategies is sometimes associated with better learning outcomes (e.g., Duncan and McKeachie, 2005; cf. Zusho et al., 2003), the adaptive choice of learning strategies suitable for a specific task is likely more relevant for learning outcomes than the use of any single strategy (Purdie and Hattie, 1999; Credé and Phillips, 2011; Hattie and Donoghue, 2016; Stark, 2019). Moreover, both the quantity (i.e., expectancy and value; Pintrich et al., 1993; Berger and Karabenick, 2011; Credé and Phillips, 2011) and the quality of students’ motivation (i.e., the degree to which students adopt performance approach or avoidance goals) are related to their use of different learning strategies (Elliot, 1999; Elliot and McGregor, 2001; Liem et al., 2008; for a review, see Senko et al., 2011).

There is, again, only little empirical research on how students’ use of learning strategies changes over actual passing time that has so far produced ambiguous findings. Studies by Enders and Weinzierl (2017) over one semester and Naujoks and Händel (2020) over nine weeks found that students tend to use less deep learning strategies and more surface strategies as an exam comes closer in time. In contrast, Zusho et al. (2003) found that cognitive surface learning strategies decreased over 15 weeks, while deep learning strategies both increased (organization) and decreased (elaboration) and that trajectories varied with students’ achievement level.

Again, the current state of research warrants no clear expectations regarding changes in the use of surface and deep learning strategies as a temporal landmark draws closer. Theoretically, if one assumes that study quantity increases, as we expected in the current study, one might also expect the usage of both surface and deep learning strategies to increase (Busato et al., 1998; Vermunt, 1998; Berger and Karabenick, 2011). On the other hand, performance-avoidance goal adoption, which we assume to increase as an exam comes closer, is associated with the usage of surface learning strategies (Elliot, 1999; Senko et al., 2011). Accordingly, one might expect only surface learning strategies to increase as an exam draws closer. Given that the available empirical evidence both supports and contradicts such theoretical considerations, we assumed that students’ usage of learning strategies likely changes as an exam comes closer but had no clear expectation regarding the direction of change.

Our final theoretical approach to investigate motivational changes over time is the perspective of multiple goals. One assumption in much of the research on study motivation and learning strategies is that students are not necessarily motivated for only one thing at a time (e.g., Atkinson and Birch, 1970; Boekaerts and Corno, 2005; Eccles, 2009). Importantly, this assumption implies that high study motivation regarding a study activity by itself (“absolute study motivation”) is a necessary but not sufficient condition to actually engage in studying. Rather, study motivation has to be higher or stronger than the motivation to pursue other activities: A student might refrain from studying not because they lack study motivation but because they are more motivated to pursue a different activity (e.g., Atkinson and Birch, 1970; Eccles, 2009; Dweck, 2017; Eccles and Wigfield, 2020). The Theory of Motivational Action Conflicts (TMAC; Schmid et al., 2007; Fries et al., 2008; Hofer and Fries, 2016) focuses on situations in which students are similarly motivated for two or more mutually exclusive activities at once. According to the TMAC, such situations are characterized by motivational conflicts that can persist even when a decision for one of these activities has been made (Grund et al., 2015a; Brassler et al., 2016). Within an expectancy-value framework, an omitted activity (i.e., the activity that is not pursued) can be considered an opportunity cost (Eccles, 2005; Barron and Hulleman, 2015; Hofer and Fries, 2016). The experience of motivational conflicts can thus be seen as an indicator of the combined latent motivational tendencies for an activity relative to its alternatives (Grund and Fries, 2012; Grund, 2013). For example, motivational conflicts become more likely if a student is similarly motivated for different activities at once, whereas they are less likely if they are distinctly more motivated for one activity (Capelle et al., 2022). Indeed, the TMAC explicitly assumes that the configuration of both expectancy and value as well as approach and avoidance incentives for each activity influence the experience of motivational conflicts in a particular moment (Hofer and Fries, 2016). Based on these configurations, motivational conflicts can be distinguished into want and should conflicts. Experiencing a want-conflict is characterized as the desire to pursue a more enjoyable activity (e.g., “I would prefer to read an interesting book”), thus indicating that the competing (i.e., not pursued) activity is perceived more positively than the focal activity one is actually pursuing (e.g., Grund et al., 2015b). In contrast, a should-conflict is characterized by the belief that one should do something else (“I should tend to my studies”), thus indicating that the competing activity is perceived as more negatively (Grund et al., 2015b). The experience of either want- or should-conflicts can thus be seen as an indicator of both the latent motivational strength as well as valence of students’ motivational tendencies for the activities they pursue relative to their omitted (i.e., available but not pursued) action tendencies. As such, motivational conflicts provide additional insights into students’ latent motivation and their changes beyond directly measured motivational constructs pertaining to a specific (study) activity.

The trajectories of motivational conflicts over time have rarely been investigated so far. Taking them into account could, however, be a valuable addition to the absolute measures of study motivation traditionally considered: If motivational conflicts are an indicator for the relative magnitude of motivational tendencies and if an exam that comes closer in time affects one of these tendencies (i.e., study motivation) systematically, this should be seen in changes in the nature, frequency, and intensity of students’ motivational conflict experiences. The frequency of motivational conflicts has indeed been shown to decrease as an exam comes closer in time, while the intensity of the motivational conflicts that still occurred increased (Capelle et al., 2022). The authors interpret these findings as indicators that students’ underlying study motivation increases relative to students’ motivation for other activities. As the study does not distinguish between want- and should-conflicts, these findings do not allow conclusions regarding the valence of students’ relative motivation. Still, the trajectories of overall motivational conflicts they found are compatible with theoretical predictions suggested by temporal discounting and the empirical findings on increasing study quantity over time (Ainslie, 1992; Hofer and Fries, 2016). In order to extend these results and to utilize motivational conflicts as indicators of possible changes in latent motivation, we consider changes in the experience of both want- and should conflicts in the present research.

Expectancy-value beliefs (motivation quantity) were assessed using one item for expectancy (“I think I can do well on the exam.”). This operationalization was deemed adequate as the most common conceptualizations of expectancy (i.e., success expectancy and ability beliefs) are routinely assessed as one factor (Hulleman et al., 2016; Eccles and Wigfield, 2020). Similarly, attainment value was assessed with one item (“It is important to me to do well in the exam.”; cf. Battle, 1965; Acee et al., 2012; but see Eccles and Wigfield, 2020). Note that this operationalization has also been referred to as importance value (e.g., Crandall et al., 1962; Cole and Osterlind, 2008; Dietrich et al., 2013). As mentioned earlier, focusing on the attainment subfacet was deemed appropriate given the purpose of the present research (e.g., Perez et al., 2019). Moreover, we assumed attainment value to be reasonably representative of overall task value given its close conceptual and factorial relationship with the other subfacets of task value (e.g., Eccles and Wigfield, 1995; Jacobs et al., 2002; Durik et al., 2006; Barron and Hulleman, 2015; Lauermann et al., 2017; Parrisius et al., 2022) and utility value in particular (Zusho et al., 2003; Eccles, 2005; Musu-Gillette et al., 2015; Parrisius et al., 2022).

Similarly, in order to assess motivation quality along the performance approach-avoidance distinction (e.g., Elliot, 1999, 2006), we applied one item aiming to capture performance approach motivation (“It is important for me to do better than the other students.”) and one item capturing performance avoidance motivation (“My fear of doing poorly on the exam is what motivates me.”). All motivational variables were assessed using a six-point Likert scale on which students could indicate their agreement with the respective statements (from “does not apply at all” to “is completely true”). Note that all questions are phrased with respect to the upcoming exam as the relevant temporal landmark. Note further that single-items are commonly used in repeated measure designs (with eight measurement points in our case) in order to increase participant retention (e.g., Wanous et al., 1997) and often achieve high validity (e.g., Loo, 2002; Bergkvist and Rossiter, 2007; Allen et al., 2022).

Learning time invested for the exam (study “quantity”) was assessed using the question “Approximately how much time did you spend in the last 7 days in total on the preparation and the follow-up of the lecture and the exam [lecture name], respectively, apart from accompanying seminars/exercises?.” Students could respond by indicating the number of hours they had studied during the past week. In order to assess learning strategy use (i.e., study quality) in the present study context, we adapted items from the subfacets “elaboration” of the Berlin Reading Strategy Inventory (BLSI, McElvany and Richter, 2009). Specifically, four items were rephrased so that they refer to deep learning with respect to the approaching exam (example item: “I tried to find my own examples that fit the material.”). In addition, surface learning strategies were operationalized by two items reflecting a repetitive and outcome- rather than understanding-oriented approach (example item: “I tried to memorize everything if possible.”) to learning. Both strategy complexes were treated as manifest constructs by forming the average of the responses to the four or two items, respectively. As for the motivational variables, all students could indicate their agreement with these statements on a six-point Likert scale. Note that students were only queried regarding their study strategy use when they had indicated that they had studied one hour or more (i.e., when their response to the study quantity question was >0). Consequently, we recorded fewer measurement points for learning strategies than for the other constructs (cf. Table 2).

Following Grund et al. (2015a), participants indicated their experience of motivational conflicts over the past seven days by answering the question “Within the last 7 days, to what extent did you feel, overall, that you rather wanted to do something other than what you actually did?” to indicate want-conflicts and “…that you should rather be doing something other than what you actually did?” to indicate should-conflicts on a six-point Likert scale from “not at all” to “very much.”

In order to account for potential differences between groups of students, we considered four control variables in an additional modeling step (see below). Exams were written on two separate dates: The didacts exam (education science) and the statistics exam (psychology) were written three days after the last measurement point while the introduction to psychology (education science) exam was written ten days later (cf. Table 1). We consequently considered the exam date as a control variable (0 = earlier, 1 = later; Mehta and West, 2000). Additionally, we considered study course (0 = education science, 1 = psychology). Note that taken together, the dummy variables for exam date and study course distinguish all three lectures and their respective exams in our sample (e.g., Didactics: exam date = 0; study course = 0; cf. Table 1). Additionally, we considered students’ age and gender (0 = male, 1 = female).

We used unconditional parallel process models, that is, first order latent growth curve models, to statistically model the trajectories of our focal variables among all students in our sample (e.g., Geiser, 2012; see Kosovich et al., 2017, and Robinson et al., 2019, for similar approaches). In order to model the passage of time, the eight measurement points were used as factor weights for each measurement point, starting at zero, thus setting the model intercepts at the beginning of the eight-week time period covered (Geiser, 2012). We reasoned that setting intercepts at the beginning of the assessment period to be most informative as it allowed us to consider the correlation of students’ initial motivation and study behavior with changes over time. We tested a parallel process model concerned with motivation (four variables), study activities (three variables), and motivational conflicts (two variables) separately. In order to account for potential differences in growth factors between individuals, we subsequently regressed all growth factors in each model on the four control variables (conditional parallel process models). Statistical significance tests were one-sided for directed hypotheses (i.e., H2 and H3) and two-sided for all undirected hypotheses as well as the control variables in the conditional growth models.

Participants answered 81.5% of all weekly survey invitations (626 out of 776). We plotted the missing patterns for all variables and visually screened them for systematic dropout over time. We could not find any apparent evidence of systematic dropout. In case of the motivational variables, the average percentage of missings per measurement point was 19.2% and varied between a minimum of 13.4% on the first measurement point and a maximum of 22.7% on the third measurement point. In case of the variables concerned with learning strategies, which students were only asked about if they indicated at least one hour of studying in the previous week, 64.3% of measurement points were answered (499 out of 776). Here, the average of missings per measurement point was 32.8% and varied between a minimum of 25.8% on the last two measurement points and a maximum of 61.9% on the third measurement point. The relatively high missing rates on the third measurement point can likely be attributed to the fact that it occurred during holidays between Christmas and New Year, which occurred between measurement points two and four. Missings were handled using the Full Information Maximum Likelihood (FIML) method. Note that the final sample size (N = 96 individuals with 626 and 776 measurement points, respectively) is well within the recommended range of sample sizes assumed to yield reliable results (Muthén and Curran, 1997; Maas and Hox, 2005; Curran et al., 2010).

We first modeled the trajectories of the four motivational variables. The results are summarized in Table 3. Model-implied trajectories are shown in Figure 1. The linear growth factors of expectancy (p = 0.031) and attainment value (p = 0.015) beliefs are statistically significant and indicate a decline in both constructs over time for the average student. We therefore retain H1. The slope of performance approach motivation is not statistically significant (p = 0.118) but its estimated coefficient pointed toward an average decline. In contrast, performance avoidance motivation increases statistically significantly (p = 0.016) over time. We thus partially retain H2. All growth factors are characterized by statistically significant variance between individuals. Moreover, the slopes of expectancy and value (p = 0.008), expectancy and performance approach motivation (p = 0.012), and attainment value and performance approach motivation (p = 0.002) are statistically significantly and positively correlated, indicating that their decline occurred in tandem for many students. Moreover, the slopes of performance-approach and performance avoidance were positively correlated (p = 0.048), indicating that students whose performance avoidance motivation increases more steeply, performance approach motivation decreases less steeply or even increases over time.

The second model is concerned with study activities. Results are summarized in Table 4 and model-implied trajectories are shown in Figures 2, 3, respectively. The growth factor for study quantity indicates that study time increased significantly (p < 0.001). Specifically, the result indicates that the average student in our sample studied approximately 1 h and 30 min per week for the exam at the beginning of the sampling period and around 6 h and 40 min per week at the last measurement point. We therefore retain H3. With regard to learning strategies, results indicate that both surface learning strategy usage (p < 0.001) and deep learning strategy usage (p < 0.001) increase over time. However, the growth of surface learning strategies was substantially larger than the growth of deep learning strategies, indicating that students tended to use more surface learning strategies than deep learning strategies as the exam comes closer. We therefore retain H4.

With regard to the correlations of growth factors, results indicate that the slopes of surface and deep learning strategies are positively correlated (p = 0.004). Interestingly, the intercept and slope of both learning strategies (surface: p = 0.012; deep: p < 0.001) are negatively correlated, indicating that students who used a learning strategy less at the beginning of the sampling period tended to increase the use of that strategy more rapidly as the exam approached. Likewise, while the intercepts of deep and shallow learning strategy use are positively correlated with each other (p < 0.001), they are each negatively correlated with the slope growth factors of the respective other construct. This means that students who tended to use more deep learning strategies at the beginning of the sampling period tended to increase their use of shallow learning strategies less steeply (and vice versa). Moreover, variance of all growth factors except for the study time intercept was statistically significant, indicating that students’ study time was similarly low at the beginning of the sampling period.

The third model complements the previous analyses by considering the changes of want and should conflicts over the time period. Model results are summarized in Table 5 and model-implied trajectories are shown in Figure 4. Results indicate that want conflicts increased over time (p = 0.002). In contrast, should-conflicts decreased over time (p = 0.007). We therefore retain H5.

Interestingly, the intercepts of want and should conflicts were strongly positively correlated (p < 0.001), indicating that students who experienced more want conflicts at the beginning of the sampling period also experienced more should conflicts and vice versa. Moreover, results indicate statistically significant variance between students for all growth factors except the slope for should conflicts, indicating that the decline in should conflicts over the sampling period was fairly universal among students.

In order to account for potential differences in growth factors between students, we introduced the control variables age, study course, gender, and exam date (early vs. late) into the three unconditional models.

With regard to motivation, results show that the intercepts of both attainment value (b = 1.089, SE = 0.255, p < 0.001) and performance approach motivation (b = 0.834, SE = 0.419, p = 0.047) varied with study course affiliation. Specifically, both were higher for psychology students at the beginning of the assessment period. Moreover, the linear slope growth factor of expectancy varied significantly with age (b = −0.018, SE = 0.006, p < 0.001). Specifically, the younger students are, the more their expectancy declines as the exam comes closer. Finally, the intercept of performance-avoidance motivation varied with age (b = 0.098, SE = 0.044, p = 0.027). Specifically, the older students are, the higher were their initial performance avoidance motivation at the beginning of the sampling period.

With regard to study activities, results show that the linear slope growth of study quantity varied with age (b = −0.084, SE = 0.028, p = 0.003), indicating that younger students increased their study activities less rapidly than older students. The study quantity slope also varied with study course affiliation (b = 1.108, SE = 0.219, p < 0.001), indicating that psychology students increased their study efforts more rapidly over time. With regard to learning strategies, results show that the intercept of shallow learning strategies varied with study course affiliation (b = −0.600, SE = 0.028, p = 0.049), indicating that psychology students used fewer shallow learning strategies at the beginning of the assessment period. Moreover, results show that age predicted variance of deep learning strategies both with regard to intercepts (b = 0.115, SE = 0.036, p = 0.001) and slopes (b = −0.016, SE = 0.005, p = 0.002). Specifically, younger students used more deep learning strategies at the beginning of the assessment period but increased their use less rapidly.

With regard to conflict experience, results show that the intercepts of should conflicts varied with gender (b = 0.712, SE = 0.305, p = 0.020), indicating that female students experienced more should conflicts at the beginning of the assessment period. Moreover, the linear slopes of should conflicts varied with study course affiliation (b = 0.137, SE = 0.056, p = 0.015), indicating that should conflicts decreased less and even increased among psychology students.

Having conducted the main analyses, we noticed that the model fit was not very good for the models concerned with study activities (cf. Table 4). This is not very surprising given the high number of parallel processes and that the Christmas holidays were included in the sampling period between measurement points two and four, which is likely to interfere with any study routine, as we discuss below. Still, we were interested in how the linear trajectories we investigated compared to nonlinear growth models. Specifically, we were interested if nonlinear models might provide more nuanced insights into how our central variables developed over time for future research. We thus compared the linear growth models of single variables (rather than the full unconditional parallel process models we reported above) to models including a quadratic and cubic growth factor based on common relative fit indices (AIC and BIC) while considering absolute indices of model fit (see Table 6). We found that linear models fit best for performance avoidance motivation, surface and deep learning strategies as well as want and should conflicts. We further found that quadratic models best fit for expectancy, expectancy, and performance approach motivation. However, quadratic growth factors were not statistically significant for value and performance approach motivation. Best fit with an additional cubic growth factor was found for study quantity. While these additional growth factors can provide additional nuance, the central messages that can be derived from the linear models stay the same (see Figures 5, 6 for model-implied trajectories of the non-linear models).

One way to make sense of our findings is to consider more closely the different contributions expectancy and value might have at energizing study behavior at different phases of goal pursuit (Liberman and Förster, 2008; Vancouver et al., 2008; Steel and Weinhardt, 2018). According to this line of thought, high levels of expectancy and value are not necessarily positively related to study activities, depending on the time point in the goal pursuit process. Different stages in goal pursuit have long been proposed by different theories (e.g., Lewin et al., 1944; Heckhausen and Gollwitzer, 1987; Vancouver et al., 2008; for an overview, see Steel and Weinhardt, 2018). While different goal phases are not the focus of the present research, earlier stages of goal pursuit likely correspond to greater temporal distances from a temporal landmark while later stages likely correspond to smaller temporal distances. As such, they are worth considering to make sense of our results. A major difference that has been discussed in the literature is between goal choice (e.g., deciding to take an exam) and actual goal pursuit (e.g., investing time and resources into studying), which will guide our discussion.

The idea of different stages of goal pursuit is especially applicable to the expectancy component, as the relationship of expectancy beliefs and study behavior may change at different stages of goal pursuit: In contrast to the generally positive association of expectancy beliefs and performance, some research has found negative relationships of expectancy and performance or effort in various contexts (Vancouver et al., 2001; cf.: Bandura and Locke, 2003; for an overview, see Vancouver et al., 2008). A possible solution to these seemingly disparate findings is the assumption of a nonmonotonic relationship of expectancy and effort, depending on the stage of the goal process (Kukla, 1972; Vancouver et al., 2008). According to this line of thought, expectancy beliefs are important to determine whether someone decides to pursue or abandon a goal early in the goal pursuit process (i.e., the goal choice phase): if someone has high expectancy beliefs with regard to a sufficiently valuable goal (e.g., passing an exam), then they are likely to pursue the goal (i.e., start studying), while someone with low expectancy beliefs deems a goal unachievable and thus likely abandons it (Kukla, 1972; Gollwitzer, 1996; Vancouver et al., 2008). Indeed, it has been argued that an “optimism bias,” the tendency to underestimate the effort required to achieve a certain goal, facilitates goal commitment at this stage (Sharot, 2011; Steel and Weinhardt, 2018). Nonmonotic models further assume that during actual goal pursuit, lower expectancies predict greater (study) effort (Vancouver et al., 2008; Steel and Weinhardt, 2018). Steel and Weinhardt (2018), referencing Bandura (1997) suggest: “in a classroom setting, where individuals need to prepare for an upcoming exam, low rather than high self-efficacy may be preferred” (p. 11). Self-efficacy describes an individual’s perceived ability to perform a specific task and thus similar but not equal to most measures of expectancy beliefs (Hulleman et al., 2016). According to this line of thought, the decline in expectancy beliefs and increase in study behavior over time we and others found might thus reflect high expectancy beliefs at the stage of goal choice at the beginning of the sampling period when not much studying takes place. Decreasing expectancies during goal pursuit might then reflect, for example, that students realize how much they still have left to study. This realization would, however, indeed be necessary to energize increasing study efforts. As mentioned, this idea is compatible with the interpretation offered above that students tend to study out of fear close to the exams (and as indicated by increasing performance avoidance motivation), which we will discuss in more detail below.

Contrary to this line of reasoning, however, Liberman and Förster (2008) describe the “‘cold feet’ phenomenon,” according to which a task may seem subjectively more difficult and less controllable (hence lowering expectancy) as it comes closer, leading to decreased motivation (see also Gilovich et al., 1993; Savitsky et al., 1998). One approach that might help to reconcile these somewhat contradictory assumptions regarding expectancy and its relationship with study effort over time is to consider different sub-facets of the expectancy construct than are typically considered in educational expectancy-value research. Indeed, Liberman and Förster (2008), largely reflecting Heckhausen and Rheinberg’s (1980) “extended cognitive motivation model“, distinguish between four different facets of expectancy that may change differently as a deadline approaches (for similar considerations see Liberman and Trope, 1998): (i) subjective task difficulty, which may either remain constant or increase (as per the “cold feet” phenomenon mentioned above), (ii) sufficiency, describing the “probability of passing [an exam] given studying” (p. 518), thus entailing both self-efficacy (e.g., Bandura, 1997) and controllability (e.g., Rotter, 1966), (iii) necessity, describing “the likelihood of achieving the goal without the action” (Liberman and Förster, 2008, p. 518), and (iv) probability, reflecting the uncontrollable subjective probability of some event to occur, which seems less applicable to an exam. The authors describe different situations in which each of these types of expectancy may either increase, remain constant or decrease as an exam comes closer under different circumstances (Liberman and Förster, 2008, p. 524), and are thus not readily applicable to explain previous findings. Nonetheless, the distinctions seem promising to gain more fine-grained insights into the trajectory of expectancy and its relation to studying in future research. For example, our operationalization of expectancy in the present study mainly reflects what Liberman and Förster (2008) call “sufficiency.” Future research could consider these subfacets of expectancy (possibly in addition to the more commonly used success expectancy and ability beliefs) in order to examine their changes and differential contribution to study behavior over time. At the same time, there is a risk that the relationship of motivational beliefs and study activities becomes theoretically arbitrary if predictions are not sufficiently precise and thus theoretically dissatisfying. It therefore seems necessary to formulate and to test exact “boundary conditions” for when which sub-facets of expectancy are predicted to be positively or negatively related to actual study behavior.

With regard to value, temporal discounting predicts that subjective value of studying (in our case, attainment or importance value) should increase as an exam comes closer in time and thus motivate studying (Ainslie, 1992; Schouwenburg and Groenewoud, 2001; Frederick et al., 2002; Steel and König, 2006). Why then do we, as have others, observe decreasing (attainment) value and increasing study activities as an exam comes closer? Again, the idea of different stages of goal pursuit seems helpful. In the following, we discuss two lines of thought: (i) changes in motivation quality as the valence of the value component and (ii) explicit devaluation as a protective mechanism against exam-related anxiety.

In addition to the mechanisms suggested above, it is conceivable that the role of motivational expectancy and value beliefs in energizing study behavior changes in other ways as well. Specifically, we discuss the potential roles of (i) relative motivation and of (ii) urgency or pull effects close to a deadline.

First, various authors (e.g., Atkinson, 1957; Hofer and Fries, 2016; Achtziger and Gollwitzer, 2018; Capelle et al., 2022) suggested that not “absolute” study motivation (as measured by most instruments) is the crucial factor determining study behavior but rather study motivation relative to other activities. The mechanism underlying the different trajectories of motivation and studying could then be that while absolute study motivation declines, it becomes stronger relative to other action alternatives. For example, students could primarily study to avoid experiencing should-conflicts with regard to studying: Other activities might simply become less attractive because they are increasingly “tainted” by students’ bad conscience for not studying (Capelle et al., 2022). Note that such an assumption is consistent with the idea that study behavior is mainly driven by performance avoidance motivation as well as the trajectories of both want- and should-conflicts we find (cf. Table 6).

Secondly, and related to the multiple action alternative perspective, Zhu et al. (2018) described a “mere urgency effect” in consumer choice research: Under spurious task urgency, people tend to work on tasks that are urgent, even if that means forgoing tasks with objectively better payoff. The authors explain this phenomenon with an attentional shift away from the value of a task outcome toward an impending deadline. Applied to the present context, this could mean that the value component is “decoupled” from energizing behavior close to the deadline. Indeed, in research on student motivation, Capelle et al. (2022) speculated that an upcoming deadline creates a situational pull which “takes over” the energization of study behavior from motivational beliefs: Like an object caught by the gravitational field of another object, individuals are thus almost passively “pulled towards” working on a task that is related to a deadline. While both concepts could explain a situative exception from temporal discounting (i.e., a decoupling of subjective task value and behavior), they are distinct in that the mere urgency effect is theoretically concerned with situations in which individuals may act against their priorities, while the situational pull is mainly concerned with a shift in the (subjective) source of behavior energization (which also sets it apart from Field Theory, e.g., Lewin, 1938). It thus seems interesting to consider the relationship of study motivation and study activities in terms of relative study motivation as well as the possibility of either a mere urgency effect or a situational pull.

We have so far discussed changes in motivation and study behavior as trajectories of the average student over time, which is reflected in our growth curve models. However, as we also found substantial variance in these trajectories as well as some significant predictors of between-person differences in how motivation and study behavior changes over time. It thus makes sense to briefly address potential between-person moderators in the future in order to increase the predictive power of growth models related to motivation and deadlines.