Trends of Social Anxiety in University Students of Pakistan Post-COVID-19 Lockdown: A Healthcare Analytics Perspective

Abstract

This paper disseminates our research findings that we conducted on university students in the years 2021, 2022, and 2023, with the year 2021 taken as the base year. Our research mined and excavated a concealed prevalence of social anxiety as an important and crucial facet of study anxiety in the university students of Pakistan. Using the Liebowitz Social Anxiety Scale (LSAS), we found a significant increase in the social anxiety level among university students in the past three years after the COVID-19 lockdown. Our data showed that the ‘very severe anxiety’ level soared up to 52.94% in the year 2023 from just 5.98% in the year 2021, showing a net increase of 47.06%. Statistical analyses demonstrate noteworthy differences in the overall social anxiety levels among the students, reaching significance at the 5% level and a discernable upward trend in the social anxiety levels as study anxiety. We also employed predictive analytics, including binary classifiers and generalized linear models with a 95% confidence interval, to identify individuals at risk. This study highlights a dynamic shift with escalating social anxiety levels among the university students and thus emphasizing its awareness, which is significantly important for the timely intervention, potentially preventing symptom escalation and improving outcomes.

1. Introduction

Study anxiety is a psychological phenomenon characterized by a student’s apprehension and fear of academic failure [1]. It manifests as a depressive attitude towards one’s studies, leading to emotional inefficacy. Avoidance of studying or testing situations, procrastination, irritability, and defiant behavior are some of the observable expressions of emotional inefficacy because of study anxiety [2]. Current research indicates that social anxiety and test anxiety are two facets of study anxiety that commonly distress students [3]. However, there are similarities, but they also have divergent individualities and influences on students’ lives and academic performances. Social anxiety, also known as social phobia, is a distress of social situations where, in particular, a student feels that they would be judged, evaluated, or scrutinized by others [4]. This could lead to fear of being embarrassed and humiliated in academic social situations including classrooms; during lectures, presentations, and project displays; and during assessments etc. Cognitive symptoms of social anxiety include excessive worry about social interactions, identify-loss situation among peers, and fear of negative judgement. Ultimately, this avoidance of necessary academic interactions causes increased risks of academic under achievements. In this way, social anxiety could also behave as an important factor causing test anxiety. Test anxiety is a form of recital anxiety where an individual practices extreme distress and anxiety during examinations/assessment and evaluation situations, which can interfere with their ability to perform well [5].

Social anxiety and test anxiety can often be interrelated. A student with social anxiety might also experience test anxiety due to the fear of being judged by peers or teachers based on their performance. Conversely, chronic test anxiety can contribute to social anxiety if a student feels embarrassed about their academic performance in front of others.

Both social anxiety and test anxiety significantly impact students’ academic performance and overall well-being. Understanding the trends and analytics of them can help educators, parents, and mental healthcare professionals provide targeted support to lessen these anxieties, fostering a more positive and productive educational experience for students.

Data analytics has become increasingly critical in the domain of healthcare sciences [6]. Researchers in [7] explored how big data analytics (BDA) facilitates the implementation of risk management (RM) practices, leading to improved quality of healthcare services (QoHS). They also examined the indirect impact of BDA on QoHS through the application of RM practices. In the context of our research work, this converges to addressing social anxiety issues among the students. University students have many study anxiety problems, and research suggests that social anxiety is a significant factor leading to it. Data analytics can be appropriately utilized for understanding the growing prevalence of social anxiety amongst students.

In this paper, we have determined collective triggers and predisposing factors that have led to social anxiety among the university students of Pakistan. Some important features that we have utilized include identifying the patterns and risk factors associated with social anxiety trends over a period of time. With objective of early detection, we also performed predictive analytics to identify individuals at risk of developing social anxiety. Timely detection enables appropriate intervention, potentially preventing the escalation of symptoms and improving the victim’s outcomes.

2. Related Work

Extensive research has delineated study anxiety into two primary categories, namely, (1) test anxiety and (2) social anxiety. Test anxiety pertains to a student’s fear and apprehension regarding summative assessments such as tests, quizzes, and examinations. This anxiety is typically transient, arising only during the assessment phase with normalcy returning post-assessment. Despite its transient nature, research emphasizes the significance of test anxiety that it is one of many important reasons of the academic decline of the numerous students and subsequently impacting their future careers [8,9,10]. In contrast, social anxiety is a student’s apprehension about socializing within the academic environment, be it with peers or friends, during co-curricular activities, from instructors, or during presentations. This behavior causes students to miss class lectures and group assignments/discussions, and in extreme cases, skipping the university’s attendance altogether. Although researchers have studied social anxiety in the context of study-related anxiety, its integration into university instructional practices remains insufficient. There has not yet been research on how to incorporate strategies to avoid social anxiety in the curriculum design of courses. Failure to address these anxiety factors can have detrimental consequences, potentially compromising the overall personality of an individual, thereby affecting the product which a university is supposed to develop for the betterment of society. Consequently, a comprehensive approach for finding of the prevalence of social anxiety and understanding its trends and patterns are imperative for fostering a conducive learning environment and nurturing well-rounded individuals.

In [11], research conducted on 770 students of University Malaysia, Pahang, elucidated factors causing social anxiety in the students. These included hostel mates; cultural diversity; and, most importantly, identity crises. The research also found that these factors significantly affected the academic performance of the students, especially in the courses requiring mental effort.

A comprehensive review in [12] investigated anxiety issues among university students of Saudi Arabia, Indonesia, Korea, and Thailand. This study was conducted on students’ awareness of anxiety moods. Researchers found that family liabilities, financial issues, cultural shock, studying a new language, curriculum difficulty, and identity loss were the key contributing factors in the study anxiety of the students. All these factors can be classified as social anxiety factors. The authors also recommended creating a supportive learning environment to help students stay mentally healthy and happy. This would protect them from anxiety and prevent negative attitudes towards their studies.

In [13] similar research was conducted on 206 Bangladeshi university students. The research showed that 82.5% of the sample was experiencing slight to risky anxiety. The main reasons contributing to the anxiety accordingly were family size, gender, accommodation issues, and lack of internet facilities for learning.

In [14], a cross-sectional study was conducted at Aga Khan University on a sample of 283 students from their medical school, school of nursing, and midwifery and dental hygiene program. All of them displayed higher than cutoff levels of stress and anxiety. In another study, researchers conducted a systematic review and meta-analysis to investigate the prevalence of depression symptoms among university students in Pakistan [15]. They utilized databases such as PubMed, Web of Science, PsycINFO, and Google Scholar, spanning from 15 January to 30 January 2020, and included cross-sectional and longitudinal studies available until 31 December 2019. The prevalence of depression symptoms was found to be 42.66% at 5% level of significance. Although there are some studies indicating research on study anxiety among students from Pakistan, they only encompass test anxiety not social anxiety in particular. Thus, social anxiety needs probing in this demographic, because most of the anxiety among the students is being associated with test anxiety and hence an important other factor (social anxiety) is not being critically considered.

In [16], researchers conducted a bibliographic analysis of adolescent social anxiety across 15 variables, covering the period of 2020–2021 in peer-reviewed journals. The aim was to provide a comprehensive summary of adolescent social anxiety in relation to academic/school achievement, performance, self-concept, self-esteem, self-efficacy, self-attributions, goals, attachment, adjustment, engagement, refusal, absenteeism, anxiety, learning strategies, and self-regulated learning. However, in addition to performing a meta-analysis on existing published results, there is a need of analyzing social anxiety on original data.

In [17], the researchers investigated the interaction between two psychological variables—social anxiety (the fear of negative evaluation by others) and academic self-efficacy (confidence in overcoming academic challenges)—and student perceptions of evidence-based instructional practices (EBIPs) in relation to their final grades in a STEM-related course on students in a community college with a sample of size 227. The sample size is not adequate and does not involve a study of changing pattern of social anxiety over a period of time.

In [18], the authors utilized an alternative scale derived from the Liebowitz Social Anxiety Scale (LSAS) [19,20] on a sample of 300 university students from the Punjab province of Pakistan and conducted exploratory factor analysis (EFA). Consequently, the primary aim of the research was to assess the reliability and validity of the new scale rather than using it to measure social anxiety with an already validated scale.

A study aimed at exploring the frequency of depression, anxiety, and stress among university students in Sialkot, Pakistan, was conducted in [21]. Data were gathered using a survey method from three universities in Sialkot, employing a simple random sampling technique. The sample included 500 university students, and the DASS-21 scale was used to measure levels of depression, anxiety, and stress. The findings revealed that 75% of students experienced depression, 88.4% experienced anxiety, and 84.4% experienced stress. The results were generalized to study anxiety without distinguishing it as either social anxiety or test anxiety.

The literature study suggested that there is a dire need to mine and navigate existence of social anxiety over a period of time in Pakistani university students to determine its risk of existence and predictability by taking a larger sample specifically for this region and demography.

3. Materials and Methods

3.1. Problem Definition

The prevalence of social anxiety among university students has become an increasingly critical issue, particularly in the aftermath of the COVID-19 pandemic [22]. Social anxiety, characterized by intense fear and avoidance of social interactions, significantly impacts students’ academic performance, mental health, and overall well-being. Despite its importance, there is a lack of comprehensive research on the specific triggers, demographic influences, and long-term trends of social anxiety within this population. Existing studies have not adequately utilized data analytics on original data over a span of time to identify varying patterns of social anxiety and thus help to identify at-risk individuals and predict future occurrences effectively. This research aims to fill this gap by systematically investigating the prevalence, patterns, and risk factors of social anxiety among university students, with a particular focus on the post-pandemic context. By employing robust data analytics and predictive modeling, this research study seeks to be useful in providing early detection mechanisms and inform targeted interventions to mitigate the adverse effects of social anxiety on university students of Pakistan.

3.2. Research Hypothesis

The foundation of our research problem statement arose from the observations in the classroom. In 2021, following the conclusion of the COVID-19 lockdown, students resumed their university attendance after nearly a year of remote and home learning. It became evident that students exhibited a greater degree of reservation in classroom activities—being less participative in the discussions, more prone to idleness, and lacking enthusiasm for regular class attendance. The primary factor contributing to this behavioral shift was their preference for the relaxed and convenient nature of home-based study over a more structured university schedule. Many had ventured into freelancing roles across different digital platforms, finding convenient ways to earn income from the comfort of their homes. Learning through YouTube^® and monetizing content became prevalent. While these opportunities appeared positive, they inadvertently sowed seeds of social anxiety among our students. To validate this observation, in 2021, we conducted our initial research by gathering data from 361 university students in Pakistan. They were asked to complete the LSAS self-assessment scale to gauge the level of social anxiety among them. The findings indicated level of social anxiety ranging from low to moderate risk, as illustrated in Figure 1.

These initial findings validated the existence of social anxiety among students in the early stages. The subsequent logical step was to interpret these observations through a pertinent theoretical framework. As documented in [22], the pandemic emerged as a global stressor, and when coupled with the repercussions of social isolation and financial crises, it intensified the experience of social anxiety among the students. Our review of existing literature indicated that poverty, identity loss, and cultural shocks were prevalent and influential factors contributing to social anxiety in adults. The confinement to home for studying, learning, and working during the lockdown produced cultural shock, making it difficult for individuals to readjust to society after COVID-19 restrictions were lifted. Although students returned to their schools and universities after the lockdown, in-home learning opportunities from home, growing financial crises, and freelancing earning avenues via mobile and desktop would confine students more within their home rather than interacting and also learning with peers in the universities. Building upon this theory, we formulated our null $(H_0)$ and experimental hypotheses $\left(H_1\right)$ based on the earlier observations as follows:

H₀. 

Social anxiety is expected to remain unchanged in the coming years.

H₁. 

The prevelance of social ansiety is expected to rise in the coming years.

3.3. Methodology

This research was conducted in three distinct phases to investigate the prevalence and trends of social anxiety among university students in Pakistan. The methodology is outlined as follows:

3.3.1. Phase 1: Preliminary Observations and Hypothesis Formulation (2021)

Initial observations were made to understand the context and factors contributing to social anxiety among university students. Based on these observations, a hypothesis ($H_1$) was formulated to guide the research.

3.3.2. Phase 2: Data Collection and Instrumentation (2021–2023)

The Liebowitz Social Anxiety Scale (LSAS) was used to measure social anxiety levels. The Liebowitz Social Anxiety Scale (LSAS) is designed to measure the range and severity of social anxiety symptoms. It was developed by Dr. Michael Liebowitz and is widely used for both clinical and research purposes. The LSAS consists of 24 items, each describing a social situation or performance scenario. The 24 questions/items are divided into two sub scales; 13 items relate to social interaction situations and 11 items relate to performance situations. There are two ratings for each item: fear and avoidance. In fear, for each situation, the individual rates the amount of fear they experience on a four-point scale as 0 = None, 1 = Mild, 2 = Moderate, and 3 = Severe. In avoidance, the individual rates how often they avoid the situation on a four-point scale as, 0 = Never, 1 = Occasionally, 2 = Often, and 3 = Usually. The scores for fear and avoidance are summed separately for the social interaction and performance subscales and then combined to give a total score. The total score ranges from 0 to 144, with higher scores indicating greater social anxiety. Scores are interpreted using specific thresholds to categorize the severity of social anxiety. Commonly used threshold are 0–54 as minimal or no social anxiety, 55–65 as mild social anxiety, 66–80 as moderate social anxiety, 81–95 as marked social anxiety, 96-110 as severe social anxiety, and 111–144 as very severe social anxiety. Additionally, demographic questions regarding gender, domain of study, and the students’ anxiety history were also included. In 2021, baseline data were collected using the LSAS scale and demographic questionnaire. In 2022, the same instrument was employed to gather data, allowing for the observation of any emerging trends or patterns. In 2023, final data collection was performed, ensuring consistency in the instrument used across all three years. This approach was aimed at observing the varying trends and patterns in social anxiety among university students and to validate our hypotheses. Data were collected from urban university students of Pakistan, ensuring that any student responding to the LSAS scale did not have any diagnosed anxiety disorder to prevent bias in the study.

The variables considered for data analysis included social anxiety levels (measured by LSAS), along with gender, domain of study, and students’ anxiety history. The variables considered during data analysis are listed in

Table 1. List of variables used in data analysis.

Variable Name	Comments
Gender	Gender of students. It is a categorical variable with categories males, females, and prefer not to mention.
Domain	Domain of study. It is a categorical variable with categories Engineering and Computer Science, Business and Economics, Medicine and Health Sciences, and Media Sciences.
History	Anxiety history. It is a categorical variable with categories yes and no.
Age	Age group of students. It is categorical variables with 4 age groups, namely, 18–26, 27–30, 30–40, and greater than 40.
24 variables of LSAS scale [14]
Anxiety value	Anxiety value calculated as sum of LSAS scale variables for each university student who responded to the LSAS scale. It is a continuous/scaled variable.
Anxiety level	It is a categorical variable. It showed category/level of anxiety according to 6 anxiety categories defined in the LSAS scale: no anxiety, mild anxiety, moderate anxiety, marked anxiety, severe anxiety, and very severe anxiety.
Year	It is a categorical variable with three categories. year 2021, year 2022, and year 2023.

Data were collected from urban university students in Pakistan. To prevent bias, only students without any diagnosed anxiety disorder participated in the study.

.

3.3.3. Phase 3: Data Analysis and Hypothesis Testing

The collected data from 2021, 2022, and 2023 were analyzed to identify patterns and trends in social anxiety among the students. Statistical methods were used to test the formulated hypothesis ($H_1$) and validate the observations. The analysis aimed to determine the prevalence of social anxiety and its correlation with the demographic variables. We performed descriptive analytics, inferential analytics, and predictive modeling on the collected data.

Descriptive analytics was performed to summarize the basic features of the data, providing simple summaries about the sample and measures. Inferential Statistics was used to make inferences about the population based on sample data, including hypothesis testing using appropriate statistical tests.

To observe changes and patterns in social anxiety levels over the three years, we performed trend analysis.

Predictive modeling was also accomplished to identify individuals at risk of developing social anxiety using models such as binary classifiers and generalized linear models (GLM) at a 95% confidence interval.

The statistical tests and procedures we conducted included data descriptive and predictive analytics. The former involved utilizing data visualization using bar graphs, which helped in finding changing trends using slopes or average rate of change values. This was performed over a wide span of data we had obtained from 2021 to 2023. Line graphs very effectively illustrated the upwards and downwards trends for the different social anxiety levels over the specified period of time, showing no cyclic patterns. The magnitude of steepness of the line indicated fast and slow changes. A steeper slope represents a faster rate of change, while a gentler slope indicates a slower rate. Line graphs also helped in comparing different datasets for each year on the same plot.

In predictive analytics, we have worked two folds, (1) we have modeled a binary classifier using logistic regression and (2) we performed ANOVA [23] and contrast analysis to infer difference between the population distinguished by three time spans of 2021, 2022, and 2023.

The logistic regression utilized sigmoid function in the output layer to scale the linear regression into a range of 0 to 1. In the regression model, the numeric value of anxiety (that was obtained from LSAS) was used as a predictor, and its slope was obtained using negative gradient of binary cross entropy as the cost function. Binary cross entropy (BCE) is a loss function we used in the binary classification task where the goal was to distinguish between two classes. It quantified the differences between the predicted probabilities and the actual binary labels (0 or 1). BCE is also known as log loss. The two classes we used as output were presence of anxiety and absence of anxiety. The cutoff value used for distinguishing between two classes was 0.5.

We also modeled the analysis of variances (ANOVA) statistical test for comparing anxiety values among the three time spans of 2021, 2022, and 2023, in which 2021 was used as controlled input. The ANOVA test was followed by contrast analysis to compare particular groups. Both ANOVA and contrast were performed at 95% confidence intervals.

We also conducted a chi-squared test at the 5% level of significance to find the significant difference between time and anxiety level.

Our research methodology for the investigation of social anxiety as a significant factor can be graphically shown as in Figure 2.

3.3.4. Ethical Considerations

Participants were informed about the study’s purpose and provided their consent before participating. Participants’ data were kept confidential and used solely for research purposes. We ensured that participants did not have any pre-diagnosed anxiety disorders to avoid bias.

4. Results

4.1. Patterns in the Variations of Social Anxiety Level and Descriptive Statistics

As this study spans over time, we gathered data on three separate occasions. The initial data collection took place in 2021, shortly after the lockdown ended, with the results depicted in Figure 1. Subsequent data collections were conducted in the following years, 2022 and 2023, with the results depicted in Figure 3 and Figure 4.

As is evident, there was a notable change in social anxiety levels of the students in 2022 in contrast to 2021 (initial observations). Moreover, this change was even more pronounced in 2023. The summary of these descriptive statistics is presented in

Table 2. Anxiety levels of 2022 vs. 2021.

No Anxiety
2021 (A)	2022 (B)	Change in Anxiety Level (B-A)	Remarks on Change in Anxiety
24.66%	14.55%	−10.11%	Decreasing slope indicating reduction in no anxiety from 2021 to 2022
Mild Anxiety
2021 (A)	2022 (B)	Change in Anxiety Level (B-A)	Remarks on Change in Anxiety
29.92%	27.27%	−2.65%	Decreasing slope indicating reduction in mild anxiety from 2021 to 2022
Moderate Anxiety
2021 (A)	2022 (B)	Change in Anxiety Level (B-A)	Remarks on Change in Anxiety
26.04%	34.18%	8.14%	Increasing slope indicating increase in moderate anxiety from 2021 to 2022
Marked Anxiety
2021 (A)	2022 (B)	Change in Anxiety Level (B-A)	Remarks on Change in Anxiety
18.01%	18.18%	0.17%	Increasing slope indicating increase in marked anxiety from 2021 to 2022
Severe Anxiety
2021 (A)	2022 (B)	Change in Anxiety Level (B-A)	Remarks on Change in Anxiety
1.11%	3.27%	2.16%	Increasing slope indicating increase in severe anxiety from 2021 to 2022
Very Severe Anxiety
2021 (A)	2022 (B)	Change in Anxiety Level (B-A)	Remarks on Change in Anxiety
0.28%	2.55%	2.27%	Increasing slope indicating increase in very severe anxiety from 2021 to 2022

and

Table 3. Anxiety levels 2023 vs. 2021.

No Anxiety
2021 (A)	2023 (B)	Change in Anxiety Level (B-A)	Remarks on Change on Anxiety
24.66%	0.54%	−26.12%	Decreasing slope indicating reduction in no anxiety from 2021 to 2023
Mild Anxiety
2021 (A)	2023 (B)	Change in Anxiety Level (B-A)	Remarks on Change on Anxiety
29.92%	1.90%	−28.08%	Decreasing slope indicating reduction in mild anxiety from 2021 to 2023
Moderate Anxiety
2021 (A)	2023 (B)	Change in Anxiety Level (B-A)	Remarks on Change on Anxiety
26.04%	32.88%	6.84%	Increasing slope indicating increase in moderate anxiety from 2021 to 2023
Marked Anxiety
2021 (A)	2023 (B)	Change in Anxiety Level (B-A)	Remarks on Change on Anxiety
18.01%	39.95%	21.94%	Increasing slope indicating increase in marked anxiety from 2021 to 2023
Severe Anxiety
2021 (A)	2023 (B)	Change in Anxiety Level (B-A)	Remarks on Change on Anxiety
1.11%	22.28%	21.17%	Increasing slope indicating increase in severe anxiety from 2021 to 2023
Very Severe Anxiety
2021 (A)	2023 (B)	Change in Anxiety Level (B-A)	Remarks on Change on Anxiety
0.28%	2.45%	2.27%	Increasing slope indicating increase in very severe anxiety from 2021 to 2023

for the comparisons of anxiety levels in 2022 and 2023 with the initial observations in 2021. The change in anxiety levels were obtained as slope variations. They are calculated in Table 2 and Table 3 as average rate of change between the specified anxiety levels.

Taking 2021 as the base year, ‘No Anxiety’ and ‘Mild Anxiety’ levels are decreased in 2022 and 2023. The decrease is high in 2023 compared to 2022. However, ‘Moderate Anxiety’, ‘Marked Anxiety’, ‘Severe Anxiety’ and ‘Very Severe Anxiety’ levels are increased. This increase is high in 2023 as compared to in 2022.

Each social anxiety level, i.e., ‘no anxiety’, ‘mild anxiety’, ‘moderate anxiety’, ‘marked anxiety’, ‘severe anxiety’, and ‘very severe anxiety’, was also analyzed explicitly to observe their changing patterns within its specific category overall in the three years. The results are shown in Figure 5, Figure 6, Figure 7, Figure 8, Figure 9 and Figure 10.

As evident from the above results, there was a significant decrease in ‘no anxiety’ and ‘mild anxiety’ levels. However, there was a strong increase in the other higher levels of social anxiety of the students. Results are summarized in

Table 4. Average rate of change in anxiety levels.

Average Rate of Change of No Anxiety vs. 2021
2022	2023
−37.31%	−66.41%
Average Rate of Change of Mild Anxiety vs. 2021
2022	2023
−17.37%	−53.16%
Average Rate of Change of Moderate Anxiety vs. 2021
2022	2023
0%	8.74%
Average Rate of Change of Marked Anxiety vs. 2021
2022	2023
−5.73%	31.3%
Average Rate of Change of Severe Anxiety vs. 2021
2022	2023
5.26%	82.11%
Average Rate of Change of Very Severe Anxiety vs. 2021
2022	2023
35.3%	47.06%

.

The findings presented in Table 4, indicate an overall decline in instances of ‘no anxiety’ and ‘mild anxiety’ cases. But this decline, in turn, led to an uptick in cases of ‘moderate anxiety’, ‘marked anxiety’, ‘severe anxiety’, and ‘very severe anxiety’. The significant increases of 31.3%, 82.11%, and 47.06% in marked, severe, and very severe anxiety, respectively, are noteworthy. This shift in the anxiety levels among students is graphically represented in the area chart shown in Figure 11. It is an area graph where area is used as a metric to compare amount of anxiety in different years.

4.2. Binary Classifier to Predict Anxiety

Developing a statistical model to predict a student’s anxiety level is a valuable approach. By utilizing the obtained data, a binary classifier can be created easily. We established a cutoff or threshold such that all cases from marked anxiety onwards to very severe anxiety are classified as Class 1, while cases below marked anxiety down to no anxiety are classified as Class 0. Using a maximum likelihood based cross-entropy cost function, we constructed a binary logistic model for predicting anxiety class, where Class 1 represents anxiety is present and Class 0 signifies anxiety is absent. This logistic model (using sigmoid function) is probabilistic with a cutoff at 0.5, and is shown in (1):

$$P\left(Anxiety\right)=\frac{1}{1+e^{-\left(25.220\times AnxietyValue-1626.90\right)}}$$

(1)

In (1), only one predictor is used, i.e., Anxiety Value. This is the sum of all the 24 variables from LSAS scale used. The probabilistic model made in (1) can only be implied as descriptive statistics. Modeling was accomplished using SPSS. The summary of the confusion matrix and modeling are shown in

Table 5. Confusion matrix showing accuracy.

Observed		Predicted
		Binary Level		Percentage Correct
		0	1
Binary level	0	630	0	100.0
	1	0	374	100.0
Overall percentage				100.0

and

Table 6. Binary logistic model.

Variables in the Equation
	B	df	Sig.	Exp(B)
Anxiety value	25.220	1	0.876	89,727,635,741.200
Constant	−1626.900	1	0.876	0.000

.

4.3. Predictive Statistical Inference

All our data analysis and findings thus far indicate a clear distinction in the anxiety levels among undergraduate students from our three samples taken from three independent populations, namely, 2021, 2022, and 2023. However, for hypothesis testing, inferential statistical analysis is entailed to initially assess the presence of differences among the populations and subsequently determine the direction of these differences.

To examine the existence of differences among the populations, we conducted an F-test using one-way analysis of variances or ANOVA at 5% level of significance. The choice of an F-test over a t-test is motivated by the need to avoid family-wise error rate, i.e., inflation of Type 1 error. Additionally, to infer the direction of change among the samples from the three populations planned contrasts were performed at 5% level of significance. Results of F-test are presented in

Table 7. F-Test using one-way ANOVA.

	Sum of Squares	df	Mean Square	F	Sig.
Between groups	121,461.838	2	60,730.919	196.203	0.000
Within groups	309,839.918	1001	309.530
Total	431,301.756	1003

. Results of ANOVA indicate that it is significant at 5% level of significance which is interpreted as there is significant difference in the populations.

To execute planned contrasts, we performed two distinct comparisons. The first contrast involved the control group, specifically examining anxiety variations in 2021 versus the combined variations in 2022 and 2023. The second contrast focused on the variations in anxiety values between the years 2022 and 2023. A schematic representation of this contrast test is illustrated in the block diagram presented in Figure 12. Contrast coding used for each contrast is shown in

Table 8. Orthogonal contrasts.

Group	Contrast 1	Contrast 2	Product
Control	−2	0	0
Year 2022	+1	−1	−1
Year 2023	+1	+1	+1
Total	0	0	0

in accordance with the principles of orthogonal contrast coding [23]. The model was obtained by performing planned contrasts on SPSS, the results of which are shown in

Table 9. Chi-square test results.

	Value	df	Asymptotic Significance (Two-Sided)
Pearson chi-squared	326.637	10	0.000
Likelihood ratio	388.813	10	0.000
N of valid cases	1004

and model in (2). As indicated from the results, both the contrasts are significant at 5% level of significance. Positive slope values further show that there is rise in anxiety values each year.

$$Anxiety Value=5.624 \left(Contrast 1\right)+9.232\left(Contrast 2\right)+55.921\pm 17.5934$$

(2)

A chi-squared test of independence conducted at 5% level of significance between the categorical variable ‘Anxiety Level’ and categorical variable ‘Year’ confirmed that there was an association between year and anxiety levels.

5. Discussion

Inferential statistics works by uncovering the things that were unobservable before. Our research and data have revealed a notable upward trend in the levels of social anxiety among university students in Pakistan. Social anxiety is a significant contributor to study anxiety in students due to many correlating factors. Marked ones include fear of evaluation, problems in group settings, perceived expectations, negative self-talk, avoidance behaviors, impact on concentration and focus, etc. The LSAS scale uses these factors in evaluating social anxiety levels. Various other factors contributed to increased social anxiety trends, with a prominent factor being the increased prevalence of remote work, a trend that became entrenched during the extensive COVID-19 lockdown. The shift towards a work-from-home culture led to the heightened awareness of freelancing opportunities, contributing to a societal shift from the information age to the social media age. While the accessibility of online learning from inexpensive resources became convenient, it also alleviated concerns about attending traditional universities and lectures. Moreover, the flexibility provided by remote work allowed students to engage in freelancing activities from the comfort of their homes, enabling them to earn income concurrently. Despite the initial positive aspects, our research indicates that this transition, as reflected in the increasing social anxiety levels among students, has raised concerns about their mental health and overall well-being. This is a significant contribution of our research and study. We have attempted to data mine the hidden factor of social anxiety in the university students and also its increasing trend. Previous study anxiety was only considered to be because of test anxiety (i.e., fear of assessments and exams). However, our research contributes the idea that social anxiety is another significant factor contributing to study anxiety in students. Hence, just like a test-anxious student cannot be certain in performing good in assessments, so would be the case of a socially anxious student.

Our contribution in methodological innovation has helped in extracting new insights and predictions with meaningful features for a real-world problem using real-world self-collected data. Teaching pedagogy can play a critical role in dealing with study cum social anxiety in the students [24,25,26,27,28,29].

6. Conclusions

In this research, we conducted a longitudinal study to observe and analyze the anxiety patterns among students over three consecutive years: 2021, 2022, and 2023. The primary goal was to understand how anxiety levels fluctuated during this period and to identify any significant trends or patterns that emerged. Our findings revealed notable changes in the distribution of anxiety levels among students. Specifically, we observed a significant decrease in the proportion of students experiencing ‘no anxiety’ and ‘mild anxiety’. From 2021 to 2023, the percentage of students with ‘no anxiety’ decreased by 66.41%, and those with ‘mild anxiety’ decreased by 53.16%. This decline indicates a shift towards higher levels of anxiety among the student population.

Conversely, we identified a substantial increase in higher anxiety levels. ‘moderate anxiety’ saw an increase of 8.74%, ‘marked anxiety’ increased by 31.3%, ‘severe anxiety’ surged by 82.11%, and ‘very severe anxiety’ rose by 47.06% from 2021 to 2023. These increases suggest that a growing number of students are experiencing more severe forms of anxiety over the years. To classify and analyze the levels of anxiety, we developed a binary logistic regression model. This model served as a classifier to accurately categorize the anxiety levels based on the data collected. The use of binary logistic regression allowed us to account for various factors and provided a robust method for analyzing the data.

We performed statistical inference at a 5% level of significance, ensuring that our findings were statistically valid and reliable. The results of our analysis indicated that the differences in anxiety levels across the years were significant and could be generalized to the larger student population. This means that the trends we observed are not limited to our sample but reflect broader patterns within the student body. To further validate our findings, we conducted a planned contrast analysis. This analysis confirmed that anxiety levels in 2022 and 2023 were significantly higher than in 2021. Additionally, it showed that anxiety levels in 2023 were greater than those in 2022. This step-by-step increase highlights a worrying trend of escalating anxiety among students over the observed period. The association between anxiety levels and the year was also found to be significant. This finding underscores the importance of temporal factors in understanding anxiety patterns. It suggests that external factors specific to each year, such as academic pressures, socio-economic changes, and global events like the COVID-19 pandemic, may have contributed to the observed trends.

All our statistical analyses were performed with a 95% confidence interval, ensuring a high level of precision and reliability in our results. The use of a 95% confidence interval means that we can be reasonably sure that the true values of our findings lie within this range. The implications of our study are significant for educators, policymakers, and mental health professionals. The marked increase in higher anxiety levels suggests a growing mental health crisis among students, necessitating urgent interventions and support mechanisms. Schools and universities need to implement comprehensive mental health programs to address this rising anxiety trend and provide students with the necessary resources and support to manage their mental health effectively. Moreover, our study highlights the importance of continuous monitoring and assessment of student mental health. Regular surveys and studies should be conducted to keep track of anxiety and other mental health issues, enabling timely and targeted interventions. By understanding the evolving patterns of student anxiety, institutions can better tailor their support services to meet the needs of their students.

In conclusion, our longitudinal study provides valuable insights into the changing anxiety patterns among students from 2021 to 2023. The significant decrease in lower anxiety levels and the corresponding increase in higher anxiety levels underscore the need for urgent attention and intervention. By leveraging statistical models and robust analysis methods, we have provided a comprehensive understanding of the trends and their implications, paving the way for more effective mental health strategies in educational settings.