The Temporal Dynamics of Procrastination and its Impact on Academic Performance: The Case of a Task-oriented Programming Course

Procrastination is one of the most common problems among students causing mental health issues, low motivation, and poor academic performance. The emergence of Learning Management Systems has made it possible to accurately pinpoint when procrastination takes place due to the unobtrusive collection of fine-grained data from students. However, most studies that analyze procrastination regard it as an intrinsic characteristic of students. In this work, we study procrastination as a process that unfolds with time. We use sequence analysis to map the evolution of students' procrastination behavior over a task-oriented programming course. Our findings evince that students' procrastination is not constant throughout a course and that students who tend to procrastinate are those who achieve worse academic performance.


INTRODUCTION
Procrastination can be defined as the act of postponing a task that is essential for accomplishing a particular goal [20].When it is carried out by students it is referred to as academic procrastination [46].Several researchers have investigated the consequences of procrastination on students, finding that it is a common problem that persists over the years [19,29].In 1977, Ellis and Knaus estimated that 95% of students engaged in procrastination at some point [11].A similar percentage (80-95%) was reported by O'Brien in 2002 [44].More recent studies, such as the one conducted by He in 2017 at the University of Bristol [15], continued to observe this trend, with 97% of students being affected by procrastination, and 86% admitting it.The general belief is that procrastination has negative effects on students [22], although this is not always the case.Chu and Choi introduced the concept of positive procrastinators [8].They distinguished between active (or positive) procrastinators, as the ones who consciously choose to procrastinate without affecting their performance; and passive procrastinators, as the ones who do not choose to procrastinate actively.Passive procrastinators tend to experience blockage, stress, and a negative impact on their outcomes [17,22].Procrastination often leads to poor academic performance, as tasks are rushed or submitted late.This, in turn, can increase stress and anxiety as deadlines approach and assignments pile up [7,28,53].Furthermore, procrastination can negatively impact students' motivation for failing to meet their goals [55].It can also lead to an unhealthy dependence on motivation and urgency, which ultimately limits the development of effective time management and planning skills [26].
Regarding the causes of procrastination in education, several factors can be identified.Lack of motivation and difficulty in determining goals are two major causes [23,49].Additionally, a lack of self-discipline and the inability to manage time also lead to procrastination [14,34].Other causes are perfectionism [55] and low self-confidence [42], both related to the fear of failure.Similarly, excessive self-confidence has also been identified as a cause of procrastination [41].Digital distractions, such as constant access to electronic devices and social media, can also tempt students to postpone their academic tasks [25,27].Students are not the only ones responsible for their own procrastination.Poor supervision by teachers also encourages it [48].Bad course schedules, poorly dimensioned tasks, and unrealistic submission deadlines also have an impact on student procrastination [6].
In this study, we analyze the impact of procrastination on students enrolled in a higher education programming course.In contrast to previous research that categorizes students according to their level of procrastination [16,43,51], our approach will focus on examining students' evolution throughout the multiple assignments of the course.Our aim is to investigate how the timing submission behavior of students changes over time.To achieve this, we classify the assignments submitted by the students into levels of procrastination.Subsequently, we build a sequence of students' procrastination levels throughout all the assignments in the course.We study the evolution of students' procrastination behavior and its impact on their academic performance.We use sequence analysis methods and data provided by an automated assessment tool with the aim of answering the following research questions: • RQ1: What levels of student procrastination can be identified from students' assignment submissions?• RQ2: How does the submission procrastination behavior of students evolve during the course?• RQ3: What is the impact of procrastination on the academic performance of students?
The article is structured as follows.In Section 2 we present a review of the literature on academic procrastination and its effect on students.In Section 3 we explain the methodology followed in our work and in Section 4 we analyze the results obtained.Lastly, we discuss the results and present the conclusions and future work of our study.

RELATED WORK
A key aspect of studying the effect of procrastination on student performance is how procrastination is measured.Different methods have been used to measure procrastination.In particular, self-report through surveys has been a prominent choice, such as the Procrastination Assessment Scale -Students (PASS) method [42], or the Tuckman's procrastination scale [45].Recent research studies continue to employ these methodologies, as demonstrated in the work of Li et al. [24], who examined the relationship between physical activity and procrastination in students.
However, some authors, such as Tan and Samavedham [43], have pointed out limitations on these scales, highlighting their adaptation to specific theoretical frameworks and potential inaccuracies due to the reliance on student responses.They propose an alternative approach that relies on Learning Management Systems (LMSs) to monitor student activities and analytically measure procrastination.These technologies allow to capture data in a passive way, without disturbing students or teachers [32].Objectivity is another advantage of digital trace log data over self-reports, which may suffer from recall inaccuracies [54] and might reflect students' intentions rather than their actual behavior [13].
There are studies where procrastination has yielded positive results in terms of academic performance [5,39,43].However, the majority of research points to a negative impact of procrastination on students [29,35].Programming courses follow this trend, accentuated by the fact that these courses are based on incremental knowledge with learning modules that are not independent [7,28,53].Some research suggests intervention mechanisms to mitigate procrastination [35].Martin et al. [28] identified email notifications as an effective intervention mechanism in programming courses.Pereira and Díaz [30] implemented a chatbot that helps students organize their tasks.Kazerouni et al. [18] analyzed student events during coding laboratories (e.g., number of code executions, number of test passes) to predict student performance and procrastination.
The rise of educational technology, such as LMSs, has enabled the study of procrastination more objectively, from an analytical point of view [31,43].Authors have made advances in procrastination prevention [40,52], detection [2,16], assessing its effects [50], and implementing intervention mechanisms to reduce its negative impact [18,28,30].However, few studies analyze the students' evolution throughout the course in terms of procrastination.Akçapınar and Kokoçb did examine student progression over Operating Systems courses [3,21].The course they analyzed was based on theoretical questions and small problems that were unrelated to each other.They found that students' behavior patterns remain relatively consistent over time.However, there are currently no studies that analyze the evolution of procrastination in practical courses where the knowledge acquired in one assignment is necessary for the subsequent ones.

Sample and course description
The context of this study was a third-year course that is part of the Telecommunications Bachelor's Degree program at the Universidad Politécnica de Madrid.This course focused on web development, including the study of the HTTP protocol, front-end technologies (HTML, CSS, JavaScript), back-end technologies (Node.jsand Express), and SQL databases.The course instruction was conducted face-to-face and had a duration of 115-135 working hours, equivalent to 4.5 European Credit Transfer and Accumulation System (ECTS) credits, spread over four months.The course was divided into ten learning modules.Each module included videos, slides with the theoretical content, and additional information such as references to textbooks or solved exercises.Each module was accompanied by a programming assignment based on an unfinished piece of software that students had to complete before a deadline, or before the final exam but in this case with a penalty of 30% of the grade.Assignments were divided into tasks, as it has been shown that breaking down a complex task into simpler tasks facilitates the programming skill development of students [10].The work of Shaffer and Kazerouni [40] reveals that milestone projects help to reduce procrastination among students.The learning materials, assignment instructions, and course announcements were made available through the Moodle LMS.Students' assignments were automatically scored by the students using an automated assessment tool.Students could also participate in a collaborative forum in which other students and instructors addressed questions related to the course and assignments.
A total of 311 students were enrolled in the course, but only 288 students participated in the course, comprising 216 males (75%) and 72 females (25%).The criterion for determining participation in the course is whether they submitted at least one assignment throughout the entire course or attended the exam.The course evaluation consisted of a final exam with practical questions about the content of the ten modules and the ten assignments.The exam represents 70% of the overall grade.The minimum pass grade for the exam was set at 4 out of 10 points.The remaining 30% of the grade was distributed among ten practical assignments.To pass the course, students were required to obtain a minimum of 4 out of 10 points for each practical assignment.Late submissions incurred a 30% grade deduction.To pass the course, the final grade had to be higher than 5 out of 10 points.

Automated assessment tool
Throughout the course, a student-centered automated assessment tool was used [4].This tool, upon request of the student, executes a series of validations on the assignments and calculates a grade based on the degree of compliance with the criteria defined by the teaching team.In addition to the grade, the student receives feedback outlining both correct and incorrect aspects of their work.Students can submit their assignments using this tool, which uploads it to the Moodle LMS.The submission can be done as many times as they want while the assignment is open for full credit, and during an additional period after it closes until the final exam with a 30% penalty.The LMS registers the final grade of the assignment and the automated assessment tool records the student's identifier, course identifier, assignment identifier, the timestamp when the student evaluated their assignments using the tool, and the grade they obtained.The teaching team has access to all the data that the tool registers during the course.They can monitor the progression of submissions and the grades achieved by students in real time.

Data preparation
The data used for analysis was sourced from two primary data streams: the Moodle LMS and the automated assessment tool.Moodle provided information on assignments (start date, end date, and description), each student's final grade for the practical assignment, and the final exam grade.The automated assessment tool supplied data on each attempt made by the student and the resulting grade for each attempt.Students gave their consent for the collection of their information and its use for research purposes.Data from both sources were collected anonymized for the experiment.Using the original data, we calculated the following variables: the amount of time the assignment remained open (  ), the relative timestamp when the student used the automated assessment tool to grade and obtain feedback on the assignment for the first time (  ), and the relative timestamp when the student completed and submitted the assignment (  ).It is worth noting that since students could submit assignments even after the official deadline (with corresponding penalties), the values of   and   could exceed one.

Sequence analysis of submission procrastination
The purpose of this study is to analyze how procrastination in completing assignments impacts student performance.Instead of treating procrastination as an inherent trait of the student, we will first investigate procrastination at the assignment level (RQ1), and then at the course level, by investigating each student's procrastination evolution throughout the course (RQ2) and how this trajectory affects their performance (RQ3), which is assessed at the end of the course with an exam.The analysis is divided into three steps: capturing states from variables (e.g., procrastination level from submission time), modeling each student as a sequence of states (e.g., sequence of students' procrastination level per assignment), and finally analyzing longitudinal behaviors of students' procrastination sequences (e.g., longitudinal or within-student entropy) and its impact on the students' academic performance.This method has been used to analyze the evolution of students over time in various fields, such as student engagement [37] or learning strategies [36].Each state in the sequence can have a fixed temporal duration (e.g., days, weeks, years) or can be delimited by time-ordered events (e.g., sessions, assignments, etc.).
Taking a similar approach, Tan and Samavedham [43] studied procrastination using LMS data and sequence analysis.Their research focuses on a 35-day course in which students are required to watch videos related to the course content before its end date.They group students according to their level of procrastination and the intensity with which they complete tasks, resulting in each student belonging to one group.Our research analyzes the evolution of the student throughout the course, which may involve transitioning through several different procrastination states.
The method applied to assess students' procrastination evolution consists of the following three phases: (1) Determination of procrastination states from submission timestamps (RQ1).In the literature, there are various proposals for grouping students according to their level of procrastination.Martin et al. [28] grouped students based on whether they completed the assignment before the last day, on the last day, or after the deadline.Agnihotri et al. [1] categorized students as procrastinators if they started the assignment after 75% of the students had already started.Their research detected a decline in student performance beyond this point.This approach has the limitation that there is always a fixed number of procrastinators (25% of the students) regardless of whether student submissions are widely distributed over time or not.Tan et al. [43] grouped students by the amount of time dedicated per day to complete the assignment.With this information, they evaluated students' procrastination and their consistency.The works of Hooshyar et al. [16], and Yang et al. [51] proposed applying clustering techniques to group students based on their level of procrastination.Yang et al. conclude that K-Means is a suitable algorithm in the presence of a small number of features and linear support vector machine when there is a large number of features.In the programming course examined in this study, the students were grouped according to the relative timestamp when they completed the programming assignment (  ).First, students who did not submit the assignment on time (  > 1) were separated.The rest of the students were clustered using the R library Ckmeans.1d.dp for univariate clustering problems [47].The number of clusters was chosen based on a significant drop in the Bayesian Information Criterion (BIC), identifying the elbow as the point beyond which the BIC value does not decrease by more than 10% of the total when adding more clusters, and a high average silhouette grade with a value above 0.6.
(2) Building the sequences.A sequence object was constructed from students' ordered procrastination states (assignments 1-10) using the TraMineR library [12].From the procrastination state sequence, it is possible to analyze the individual progression of a student throughout the course and represent an overall view of the evolution of procrastination.
(3) Analyzing the sequences.In the last step of our analysis we studied the sequence properties that allow to quantify students' evolution throughout the course (RQ2).We first calculated the between-student entropy.Between-student entropy allows to assess the variability of procrastination states among students for each assignment.The between-student entropy will be zero for an assignment when all students are grouped in the same procrastination state.Conversely, entropy will be at its maximum (i.e., 1) when the variability of states among students is highest for that assignment.Between-student entropy allows the identification of predictable assignments (those with low entropy across assignments) and unpredictable assignments (those with high entropy across assignments).We also calculated the within-entropy (also called longitudinal entropy).Within-entropy measures the variability of each student's states over time.It is minimal (i.e., 0) when a student has the same procrastination state in all assignments, and maximal (i.e., 1) when the variability of states for the same student is highest.The within-student entropy allows us to gauge the degree of variability in states for each student across assignments.Works like that of Saqr et al. [38] propose using both within-entropy and between-entropy to characterize the sequences.We also calculated other properties to study each sequence.Ritschard [33] proposes different sequence properties such as precarity to measure the negative stability of a sequence; volatility to measure the variability of states and transitions within a sequence; or integrative potential to measure the ability to achieve a positive state.
Finally, we analyzed the effect of students' procrastination sequences and their relationship with their academic performance (RQ3).We calculated the correlation and partial correlation between the sequence properties mentioned above and the exam grades.We performed Spearman's correlation as the distributions did not meet the assumptions of normality.We follow the Dancey and Ready guidelines to determine the Spearmans' correlation ranges for positive and negative relationships with a significant level of p-value < 0.05: | | < 0.  [9].Lastly, we fitted a linear regression model to investigate the impact of the sequence integrative potential on achievement.
Figure 1 summarizes the implementation of the methodology applied.It shows three different states of procrastination (step 1) but in practice, the number of clusters may vary.

(1) Determination of procrastination states (RQ1)
The first phase of the analysis involves identifying different levels of procrastination per assignment.A total of 288 students participated in the study and 2,749 submissions were registered.On average, there were 9.5 submissions per student, slightly less than the expected 10 tasks per student (one task per module) since some students did not submit all the assignments.Specifically, 131 assignments were not submitted.To identify procrastination levels, submissions that were delivered late or not delivered (n = 567, 19.7%) were separated from those submitted on time (n = 2,313, 80.3%).Among those submitted on time, clustering was performed using the K-means method for univariate problems.The best combination (based on BIC and a high average silhouette grade) resulted in two clusters delimited by the midpoint   = 0.656 with a silhouette grade of 0.66.As a result, three possible states were obtained: • Early birds (n = 866, 30.1%):Assignments submitted before 65.6% of the time passed between the assignment's release and its deadline.• Regular (n = 1,447; 50.2%):Assignments submitted after 65.6% of the time passed between the assignment's release and its deadline.• Delayed (n = 567, 19.7%): Assignments not submitted or submitted after the deadline.

(2) Building the sequences
The next step involves constructing the procrastination state sequence for each student throughout the 10 assignments.Figure 2 illustrates the evolution of early birds, regular, and delayed students throughout the course.The sequence distribution plot shows that in the initial assignments, the majority of the students are early birds (circa 70%), followed by regular students (circa 25%), and with very few delayed (less than 5%).As the course progresses, the number of early birds decreases, while the number of regular and delayed increases.A peak of delayed students was reached in the seventh assignment.From then onward, the number of early birds begins to rise again, while the number of delayed decreases, and the number of regular remains constant.

(3) Analyzing the sequences (RQ2)
Once the state sequences for each student have been obtained, we studied the between-student entropy and the within-student entropy (Figure 3 and Figure 4).Between-student entropy has a high value throughout the course with an increasing trend (mean = 0.84, median = 0.85, SD = 0.30).It reaches its minimum in assignment 1 (entropy = 0.69) and its maximum value in assignment 9 (entropy = 0.98).As the course progresses, it becomes more difficult to predict the students' state, as their variability increases.At the beginning of the course (assignment 1), the majority of students (n = 200, 69.4%) are grouped in the early birds state, while at the end of the course (assignment 9), students are distributed among early birds (n = 80, 27.8%), regulars (n = 126, 43.7%), and delayed (n = 82, 28.5%) states.
Figure 4 shows a summary of the longitudinal properties of the sequences.For the calculation of the precarity index and the integrative potential, we selected delayed as the negative state and early birds as the positive state [33].The within-entropy is quite high (mean = 0.59, median = 0.61, SD = 0.24).This indicates that the variability of states within each student is high and difficult to predict.Regarding volatility (mean = 0.49, median = 0.47, SD = 0.21), three groups of students can be distinguished.Students with very low volatility (around 0), low-medium (around 0.35), and medium-high (around 0.7).Those with a volatility of 0 are the ones who do not have transitions between states, low-medium  We also analyzed the transitions between assignments (Figure 5).Staying in the same state is the most frequent transition (n = 1,741; 67.2%).We observe fluctuations in extreme states, that is, early birds and delayed.The most frequent transitions are between early bird to regular (n = 298, 11.5%) followed by regular to early bird (n = 200, 7.7%), regular to delayed (n = 175, 6.8%), and delayed to regular (n = 149, 5.7%).Transitions between the extreme states are very infrequent (n = 29, 1.1%).Results reveal that is more common to move from a positive state to a negative state (n = 496, 19.1%) than from a negative state to a positive one (n = 355, 13.7%).).To answer RQ3, we analyzed the correlation and partial correlations between the students' final exam grades and the properties of the sequences obtained by each student.
Results of Spearmans' correlation revealed a strong negative relationship ( = −0.40,p < .001)with the precarity index.That is, students who spend a lot of time in the negative state (delayed) obtain lower exam grades.There is a moderate positive relationship ( = 0.38, p < .001) between the exam grades and the integrative potential.It means that students who tend to achieve and stay in the positive state (early bird) get better grades.Results also show a weak negative relationship ( = −0.24,p < .001) between the grades and the longitudinal entropy.That is, students with high variability of procrastination states (i.e., not consistent students) get worse grades.Figure 6 represents the partial correlations between each pair of variables (i.e., the correlations after controlling for the correlation with the rest of the variables in the network).It shows that grades have a positive partial correlation with integration (the ability to assume a positive state e.g., begin early bird, sustain it, and end in it).Grades also had a negative relationship with entropy (a measure of instability) and precarity (a measure of complexity).
Volatility and proportion of positive potential did not exhibit a direct association with grades but had a very strong positive correlation with entropy and the integrative index, respectively.Based on the previous results, we fitted a linear regression model using Ordinary Least Squares (OLS) to model the final exam grade as a function of the integration index.We obtained an intercept value of 5.04 (SE = 0.13, t = 38.93,p < .001)and an integrative coefficient of 2.32 (SE = 0.32, t = 7.27, p < .001).The model is statistically significant with a moderate proportion of variance (R-squared = 0.17, F(1, 255) = 52.8,p < .001,adjusted R-squared = 0.17).The integration index has a statistically significant and positive effect on the final exam grade ( = 2.32, 95% CI [1.69, 2.95], t = 38.93,p < .001).

DISCUSSION OF RESULTS AND CONCLUSIONS
This study demonstrates that students' procrastination is not consistent over time.In fact, most students start as early birds (n = 200, 69.4% in the first assignment of the course).As the course progresses, the number of early birds decreases, the number of regular students increases, and toward the end of the course, the number of delayed rises.Longitudinal entropy, and volatility show that students change their behavior as the course progresses.In fact, only 18 students (6.25%) are consistent (i.e., they remain within the same state during the whole course).
The results reveal that students with the capacity to reach and stay early birds, obtain better exam grades.On the contrary, students who spend a high proportion of time in the delayed state and students who experience a lot of transitions among states get worse exam grades.From the results, we can conclude that tending to procrastinate behaviors have a negative effect on the students' academic performance.These findings are consistent with previous research that associates procrastination with poor student performance [7,28,53], in contrast to other studies that identify a positive effect [5,39,43].The structure of the course could be a reason for this result.Tan and Samavedham's course [43] involved only two assignments (one midterm and one final), allowing for greater flexibility for students.In our analysis, assignments are distributed throughout the course with fixed submission dates, and students need the knowledge from previous assignments to be able to complete the subsequent ones.In their work, Koko et al. [21] concluded that students do not change their procrastination habits during the semester.However, our investigation reveals that although the most common behavior is to remain in the same state between assignments, it is uncommon for students to remain in the same state throughout the whole course.
Among the limitations of this case study is the limited number of students.In future versions, the automated assessment tool will be enhanced, and new variables will be included in the research, such as metrics regarding access to course materials, the number of changes between two assignments, questionnaires made to the students to combine classical scales with LA methodologies, etc.Furthermore, the experiment will be extended to other subjects and courses to observe if similar patterns and states are observed, aiming to avoid restricting the experiment to a single case study.
Future lines of research could involve extending the automated assessment tool to monitor students' progress throughout the course.Early detection systems could be implemented that alert students identified as potentially tending-to-delayed.We can analyze if interventions can have a positive effect by reducing the number of delayed students.

Figure 1 :Figure 2 :
Figure 1: Steps of the method for determining sequences of students' procrastination evolution behavior and studying their academic impact

Figure 6 :
Figure 6: Spearmans' partial correlations among sequences properties and final exam grades.