2.1 Importance of CT Skills

CT is defined as the ability to understand and solve problems using concepts that come from computer science [18]. CT has emerged as a crucial skill set for success across various academic disciplines in higher education. Cabrera et al. [19, 20] highlight that introducing CT at the preuniversity level not only enhances students' understanding of programming and computer science but also cultivates critical thinking skills applicable to a wide range of fields [21-23].

Although traditionally associated with STEM fields [24], the relevance of CT has significantly broadened. Killen et al. [25] demonstrate that CT skills enhance problem-solving abilities not only in technical disciplines but also in nontechnical majors, such as business and social sciences [26]. Additionally, [27] found that early exposure to CT activities boosts overall academic performance and increases student engagement across a wide range of university programs [28].

CT fosters essential skills such as decomposition, pattern recognition, abstraction, and algorithmic thinking, which are now acknowledged as crucial in various fields, from the humanities to the sciences. For example, [29] reported significant improvements in English writing, while [30] observed enhancements in mathematical skills among students who participated in CT training.

In this sense, it is clear that integrating CT into preuniversity education prepares students not just for technology-focused careers but equips them with a versatile mental toolkit applicable to any academic discipline as state [31] in their study.

2.2 The University in Analysis

In 2024, the preuniversity course at the Universidad Estatal de Milagro (UNEMI)¹ stands out as one of the main gateways to higher education in Ecuador, particularly through its online learning platform. With approximately 4500 students choosing online education and more than 9000 studying in blended and face-to-face modalities, this course becomes a crucial starting point for thousands of young people seeking university admission. Located in Milagro, an area with significant rural challenges and economic disadvantages, UNEMI's preuniversity course offers a vital opportunity for students to overcome geographical and socioeconomic barriers.

The admission process at UNEMI begins with the Higher Education Access Exam (EAES), which evaluates the aptitudes and knowledge necessary for entry into higher education. Based on their exam results, applicants can apply to their chosen degree programs. Once assigned a place, admitted students enter the preuniversity course. A key component of the preuniversity course is the focus on CT, which is taught across all UNEMI degree programs due to its relevance in developing essential skills in today's digital environment. This course equips students with crucial problem-solving skills, preparing them for university life where technology and logical thinking are critical for success in various academic fields.

The teaching of CT promotes competencies such as logical analysis, structured problem-solving, and algorithm development—tools that prove useful across various areas of knowledge. These skills enable students to approach complex problems efficiently and structurally, regardless of their academic discipline. The decision to teach CT across all degree programs is justified by the growing demand for digital competencies in all employment sectors.

2.3 Related Work

Numerous studies have investigated the role of CT in preuniversity education, focusing on both the development of CT skills and the effectiveness of various instructional strategies. For instance, [1] present a conceptual model for integrating CT into educational curricula, emphasizing its importance for students' future success in STEM fields. Similarly, [3] explore the challenges faced by students in acquiring CT skills and highlight the critical role of effective pedagogy in preuniversity settings. While these studies provide insights into the importance of CT in foundational courses, they tend to focus more on general instructional methods rather than on comparing the specific impact of different teaching modalities (online, blended, or face-to-face) on student performance. This leaves a gap in understanding how diverse learning environments affect the development of CT skills, particularly in multimodal preuniversity courses.

Existing research on multi instructional modalities education has explored the effectiveness of online and blended learning for enhancing CT skills. For example, [7] examine how gamification within blended learning environments can promote student engagement and improve outcomes in CT courses. Li et al. [8] investigated how online learning impacts CT development variably; effectiveness depends on student age, course integration, and duration. Challenges include reduced interaction and motivation maintenance. The article of [32] highlights the integration of CT in preuniversity classrooms through blended learning, using Scratch, Alice, and innovative educational platforms. However, while these studies provide valuable insights into specific aspects of instructional design, few focus on directly comparing academic performance across different modalities in a unified setting. This gap is particularly evident in Latin American settings, where the adoption and evaluation of different modalities learning in preuniversity CT courses are still in the early stages of exploration and analysis.

3.1 Study Design

This study used a comparative cross-sectional design to analyze the academic performance of students in CT across three modalities of preuniversity courses (online, blended, and face-to-face) at the UNEMI, Ecuador. The study period covered from January 2023 to March 2023.

3.2 Data Collection

Data were collected from the university's academic records system (SGA) for all students enrolled in the CT preuniversity course across the three modalities. The dataset included sociodemographic variables (age, gender, marital status, health status, household head status, housing type, and familiar help) and academic variables (admission score, assessment scores N1-N4, exam score, final grade, course outcome, and performance level). The description of each variable is shown in Table 1. All data were anonymized to protect student privacy.

3.3 Statistical Analysis

Descriptive statistics were calculated for all variables, including means and standard deviations for continuous variables, and frequencies and percentages for categorical variables. Cross-tabulations and chi-square tests were performed to determine significant relationships between categorical variables and modalities.

To answer RQ1, we conducted a Kruskal–Wallis test and subsequent post hoc analyses to compare final grades (fail/good/very good/excellent) across the three modalities. We also used ordinal regression to examine the relationship between performance level and modality. For RQ2, we applied machine learning (ML) with feature selection to identify factors influencing academic performance and constructed a decision tree to predict course outcomes (pass/fail).

3.4 Data Analysis

All statistical analyses were performed using R version 4.4.0. The significance level was set at $$$ \alpha =0.05 $$$ for all tests. To perform cleaning, and transformation, initially, irrelevant columns and rows with missing values were removed. The data were filtered to exclude entries based on criteria such as final grade being at least 1. Feature engineering was performed by converting variables to appropriate types and creating new categorical variables. Continuous variables were then categorized into bins, such as age and exam scores, to facilitate analysis.

For feature selection and modeling, the dataset was split into a training set and a test set, with 75% of the data being used for training and 25% for testing. Feature selection involved assessing the importance of variables through ad-hoc techniques. The selected features were then used to build predictive models, which were evaluated using performance metrics such as accuracy and precision on both the train and test data.

Additionally, a multiple linear regression model was applied to examine the relationship between students' final grades (dependent variable) and a set of sociodemographic characteristics. The independent variables included age, gender, marital status, health status, household head status, housing type, and familiar financial help. Instructional modality (FTF, BL, and OL) was considered as a grouping variable. That is, separate regression models were estimated for each instructional modality to assess whether the influence of sociodemographic factors on academic performance varied across modalities. This modeling approach allowed for controlling potential confounding factors and estimating the individual contribution of each predictor to students' academic outcomes.

4.1 Sociodemographic and Academic Information

A total of 12,164 participants were included in this study. Of these, 6,277 (51.60%) were FTF students, 1,677 (13.79%) were BL students, and 4,210 (34.62%) were OL students. Tables 2 and 3 presents descriptive statistics and $$$ {\chi}^2 $$$ tests across these three educational modalities, revealing significant differences in various demographic and academic characteristics ($$$ p<0.001 $$$ for most comparisons).

The age distribution shows significant differences across the three modalities ($$$ {\chi}^2=1742.3 $$$, $$$ p<0.001 $$$). A large proportion of students aged 16–20 prefer FTF format (72.90%), while fewer students in this age group opt for BL (45.40%) and OL (40.00%). Students aged 21–24 are more evenly distributed, with 15.00% in FTF, 25.80% in BL, and 15.90% in OL. Older students (29 years or more) predominantly choose OL (30.50%), indicating that online education may be a necessity rather than a preference for those balancing work and family responsibilities. Additionally, online learners are more likely to be household heads (31.30%) and married (9.70%), reinforcing the idea that their choice may be driven by necessity rather than preference. There are significant gender differences across modalities ($$$ {\chi}^2=610.49 $$$, $$$ p<0.001 $$$). Female students dominate the BL group (89.50%) compared to FTF (59.80%) and OL (73.30%). In contrast, male students are more prevalent in FTF (40.20%) and OL (26.70%) than in BL (10.50%). Marital status also varies significantly by modality ($$$ {\chi}^2=390.47 $$$, $$$ p<0.001 $$$). The highest proportion of single students is found in FTF (97.00%) and BL (91.30%), whereas OL students are less likely to be single (86.90%). OL students have higher rates of marriage (9.70%) and common-law partnerships (2.00%) than those in FTF and BL. Health status shows slight but significant differences across modalities ($$$ {\chi}^2=16.26 $$$, $$$ p=0.002 $$$). While the majority of students report good or excellent health, OL students tend to report better health (55.20% good, 30.00% excellent) compared to their BL and FTF counterparts. OL students are significantly more likely to be household heads (31.30%) than those in BL (19.60%) and FTF (12.30%) ($$$ {\chi}^2=570.61 $$$, $$$ p<0.001 $$$), which could be associated with the higher age profile and independence of OL learners. Regarding housing type, OL students are more likely to rent their homes (30.40%) compared to FTF (21.70%) and BL (22.90%) ($$$ {\chi}^2=138.31 $$$, $$$ p<0.001 $$$). However, a significant portion of students across all modalities live in family-provided or owned homes. Family support is more common among FTF students (78.20%), followed by BL (69.20%) and OL (57.60%) ($$$ {\chi}^2=506.93 $$$, $$$ p<0.001 $$$). This suggests that OL students may rely less on familial support than their FTF and BL peers. Admission scores differ significantly across modalities ($$$ {\chi}^2=1246.20 $$$, $$$ p<0.001 $$$). The OL group has the highest proportion of students with scores between 801 and 900 (60.00%) and the highest percentage of students scoring above 901 (26.80%). Conversely, FTF students are more common in the 701–800 range (25.70%).

In terms of academic information, significant differences are observed in all unit tests and exams. For the first unit test (N1), FTF students have the highest performance, with 86.50% scoring above 5, compared to 80.80% in BL and 88.90% in OL ($$$ {\chi}^2=68.14 $$$, $$$ p<0.001 $$$). In the second unit test (N2), OL students outperform both FTF and BL students, with 83.60% scoring above 5, while the BL group shows a decline (73.20%) ($$$ {\chi}^2=88.76 $$$, $$$ p<0.001 $$$). This trend continues in the third unit test (N3), where OL students maintain strong performance (82.90%) compared to FTF (73.10%) and BL (68.90%) ($$$ {\chi}^2=185.60 $$$, $$$ p<0.001 $$$). In the fourth unit test (N4), OL students continue to outperform the other groups, with 80.00% scoring above 5 ($$$ {\chi}^2=237.99 $$$, $$$ p<0.001 $$$). Exam scores also differ significantly across modalities ($$$ {\chi}^2=1128.20 $$$, $$$ p<0.001 $$$), with OL students showing more variability and a higher percentage scoring in the lowest range (0–15 points). Conversely, FTF students are more concentrated in the mid and upper ranges.

In terms of course outcomes, FTF students have the highest approval rate (71.60%), followed by BL (51.10%) and OL (37.60%) ($$$ {\chi}^2=1216.00 $$$, $$$ p<0.001 $$$). OL students also have the highest failure rate (62.40%). The performance level analysis shows that FTF and OL students have similar proportions of students achieving “excellent” or “very good” levels (14.20% and 14.30%, respectively), while BL students show much lower levels of high performance (5.40%).

The failure rate is significantly higher in OL (62.40%) than in BL (48.90%) and FTF (28.40%). Finally, the analysis of final grades reveals that FTF students achieved the highest mean grade ($$$ 63.67\pm 20.24 $$$), outperforming both BL ($$$ 56.58\pm 21.06 $$$) and OL students ($$$ 56.11\pm 22.38 $$$). While FTF instruction appears to be associated with higher academic performance, recent studies suggest that the impact of instructional modality is limited. The study of [33] found that performance differences between online and in-person learning were not always significant. These findings emphasize that academic success is shaped by various factors beyond modality, including technological access, adaptation to pedagogical strategies, and admission scores.

4.2 Academic Performance Differences Between Instructional Modalities of CT Courses (RQ1)

As illustrated in Figure 1, the FTF modality exhibits a slightly higher median compared to the other two, accompanied by a narrower interquartile range, suggesting less variability in grades. Conversely, the BL and OL modalities demonstrate greater dispersion in the data, with some students attaining exceptionally high grades and others achieving considerably lower scores. These visual disparities corroborate the findings of the Kruskal–Wallis test, indicating that the FTF modality might be associated with superior average academic performance.

To further understand the magnitude of these differences, the effect size was assessed using epsilon squared ($$$ {\epsilon}^2 $$$). The calculated $$$ {\epsilon}^2 $$$ value was 0.0374, indicating that approximately $$$ 3.74\% $$$ of the variability in final grades can be attributed to the differences among the educational modalities. Although the Kruskal–Wallis test shows statistically significant differences, this small $$$ {\epsilon}^2 $$$ value suggests that the practical impact of these differences is modest.

To determine which specific pairs of modalities differ, Dunn's post hoc pairwise comparisons were performed, providing detailed insights into the exact differences in performance between the modalities. The results of Dunn test are shown in Table 4.

Table 4 indicates that there are significant differences in academic performance between FTF and BL modalities, as well as between FTF and OL, with both comparisons showing very significant $$$ p $$$-values ($$$ p<0.001 $$$). However, no significant difference was found between BL and OL, as indicated by a nonsignificant $$$ p $$$-value (0.739).

The model generates separate equations for each cumulative probability, except for the highest category (Excellent). Thereby:

The model assumes proportional odds, meaning that the effect of the predictors ($$$ {\beta}_1 $$$ and $$$ {\beta}_2 $$$) is the same across all cumulative logits, while only the intercepts ($$$ {\alpha}_1 $$$, $$$ {\alpha}_2 $$$, $$$ {\alpha}_3 $$$) differ. This allows us to interpret the impact of educational modalities on performance levels consistently across the entire range of the ordinal outcome.

The ordinal regression analysis yielded significant results ($$$ p<0.001 $$$ for all coefficients). Specifically, when comparing to the FTF modality, BL Students had lower odds of achieving higher performance levels ($$$ {\beta}_1=-0.8062 $$$). This suggests that students in BL are less likely to progress to “Good”, “Very Good”, or “Excellent” compared to FTF students. In the case of OL students, they had even lower odds of achieving higher performance levels ($$$ {\beta}_2=-1.1377 $$$). This indicates that students in OL were significantly less likely to progress compared to FTF students. The model identified three significant thresholds: $$$ Fail\mid Good $$$, $$$ Good\mid Verygood $$$, and $$$ VeryGood\mid Excellent $$$. These thresholds represent the odds of moving from one performance level to the next. For the case of Fail|Good, the negative threshold ($$$ {\alpha}_1=-0.7742 $$$) suggests that FTF students were more likely to achieve Good or better performance than Fail. Meanwhile, $$$ Good\mid Verygood $$$ and Very $$$ Good\mid Excellent $$$, the positive and increasing thresholds ($$$ {\alpha}_2=1.5198 $$$ and $$$ {\alpha}_3=2.8620 $$$) suggest that it becomes progressively more difficult to achieve higher performance levels, even for FTF students.

Finally, the odds ratios obtained from the model are observed in Figure 2.

The bar chart in Figure 2 illustrates the probability (odds ratios) of achieving different performance levels across three educational modalities. The x-axis represents the performance levels, ranging from “Fail” to “Excellent”, whereas the y-axis shows the probability (odds ratio) for each level and modality. The “Fail” category, the odds ratios were 0.316 for FTF, 0.508 for BL, and 0.590 for OL. This indicates that students in the BL modality had approximately 60.76% higher odds of failing, and students in the OL modality had approximately 86.71% higher odds of failing, compared to students in the FTF modality. The “Good” performance level, the odds ratios were 0.505 for FTF, 0.403 for BL, and 0.345 for OL. This suggests that students in the BL modality had approximately 20.20% less odds of achieving a “Good” performance, and students in the OL modality had approximately 31.68% less odds, compared to students in the FTF modality. In the “Very Good” category, the odds ratios were 0.125 for FTF, 0.064 for BL, and 0.048 for OL. This indicates that students in the BL modality had approximately 48.80% less odds of achieving a “Very Good” performance, and students in the OL modality had approximately 61.60% less odds, compared to students in the FTF modality. In the “Excellent” performance level, the odds ratios were 0.054 for FTF, 0.025 for BL, and 0.018 for OL. This suggests that students in the BL modality had approximately 53.70% less odds of achieving an “Excellent” performance, and students in the OL modality had approximately 66.67% less odds, compared to students in the FTF modality.

4.3 Identified Factors Affecting Performance Between Instructional Modalities of CT Courses (RQ2)

We conducted multiple linear regression analyses separately for each instructional modality, using the final course grade as the dependent variable. The aim was to assess how sociodemographic variables (used as controls) influenced academic performance within each modality and to determine whether these effects varied across instructional formats.

The multiple regression results, summarized in Table 5, revealed that the effects of sociodemographic factors on students' final grades vary across instructional modalities.

The model for the FTF group was statistically significant ($$$ p<0.001 $$$), although it explained only a small portion of the variance in final grades ($$$ {R}_{adj}^2=0.230 $$$). Older students tended to perform worse, with each additional year of age predicting a 0.36-point decrease in final grade ($$$ \beta =-0.36 $$$, $$$ p<0.001 $$$). Female students scored slightly lower than male students ($$$ \beta =-1.11 $$$, $$$ p=0.033 $$$). Reporting “Regular” health was associated with better performance compared to “Good” health ($$$ \beta =2.43 $$$, $$$ p=0.001 $$$), while being the head of household was negatively associated with performance ($$$ \beta =-2.78 $$$, $$$ p=0.002 $$$). Receiving family help showed a positive relationship with academic outcomes ($$$ \beta =2.58 $$$, $$$ p<0.001 $$$). The model for the BL group was also statistically significant ($$$ p=0.002 $$$), though its explanatory power remained very limited ($$$ {R}_{adj}^2=0.011 $$$). Among all variables, gender showed the strongest association: female students had lower grades compared to males ($$$ \beta =-7.03 $$$, $$$ p<0.001 $$$). Additionally, being the head of household was related to slightly lower performance ($$$ \beta =-3.26 $$$, $$$ p=0.028 $$$). Other demographic factors had no statistically significant effect. The model for the OL group reached statistical significance as well ($$$ p=0.001 $$$), but like the others, explained only a minimal portion of the variance in final grades ($$$ {R}_{adj}^2=0.005 $$$). Female students again showed lower performance than their male peers ($$$ \beta =-2.40 $$$, $$$ p=0.003 $$$). Reporting “Excellent” health (vs. “Good”) was linked to a slight decrease in performance ($$$ \beta =-1.64 $$$, $$$ p=0.036 $$$), and receiving financial support from family was positively associated with academic outcomes ($$$ \beta =1.68 $$$, $$$ p=0.045 $$$).

Table 6 summarizes the statistically and marginally significant predictors, providing a clear interpretation of how each factor influences performance across instructional modalities. The results reveal that demographic factors affect academic success differently depending on the learning format (OL, BL, and FTF). By analyzing each modality separately, we could better capture these variations and understand how individual and contextual factors shape outcomes in preuniversity CT courses.

To compare academic performance (using the course outcome as dependent variable) across the three modalities, ML techniques for feature selection were applied. Figure 3 presents the results of the Boruta algorithm [34], which has been used in a bunch of studies [35-37] to identify the most relevant features affecting academic performance. Here, the boxplots show the importance of various factors for each modality. The y-axis represents the importance score, while the x-axis lists different features. The color-coding of the boxes indicates the significance of each feature. Green boxes represent confirmed important features. Red boxes represent confirmed unimportant features. Purple boxes represent tentative features. White boxes represent reference features (shadowMin, shadowMean, shadowMax) used by the Boruta algorithm to determine importance thresholds.

In Figure 3A, the most important factors affecting performance in the FTF modality are age, marital status, household head, and familiar help. These green boxes have the highest importance scores. Admission score is also important but to a lesser degree. Meanwhile, Figure 3B shows the BL modality, admission score, and gender are the most crucial factors. Interestingly, age, which was highly important in the FTF modality, does not have important significant in the BL context. Finally, in the OL modality (Figure 3C), Admission score emerges as the most critical factor by a significant margin. Gender follows in importance. Other factors do not have impact on performance.

Building on insights from our feature importance analysis, we created a decision tree using the four most relevant variables across different modalities to predict course outcomes. Figure 4 shows the resulting tree, which has an accuracy and precision of 0.70. This means it correctly predicts pass/fail outcomes 70% of the time. A precision of 0.70 also indicates that 70% of predicted approvals are correct, but in 30% of cases, a student predicted to pass may actually fail.

In Figure 4, the tree primarily splits on admission score and gender, revealing this variables as key factors influencing the student performance. Students with admission scores between 801 and 900 have the highest probability of approval (40%), suggesting a strong correlation between high entrance scores and preuniversity course success. For other score ranges, gender emerges as a determining factor, with male students showing a 6% approval rate. Interestingly, female students with scores outside the 801–900 range show a 54% probability of not passing the course, indicating a potential area for targeted support. These results highlight the complex interplay between admission scores, gender, and academic performance across different instructional modalities, providing valuable insights for educators and administrators to tailor their approaches and support systems.

5.1 Performance Differences Between Instructional Modalities of CT Courses

The Kruskal–Wallis test revealed statistically significant differences in academic performance across FTF, BL, and OL modalities ($$$ p<0.001 $$$). However, the effect size ($$$ {\epsilon}^2=0.0374 $$$) suggests that the practical significance is between small and medium. This finding aligns with the work conducted by [38], which found only modest advantages for blended learning over FTF instruction. Our results extend this understanding to the preuniversity context, highlighting the varied nature of modality effects in this critical transitional period.

The analysis of predicted probabilities (odds-ratios) for performance levels in the modalities provides a more detailed picture of student outcomes. With a 31.6% probability of failure and a 50.5% probability of achieving a “Good” grade, the FTF modality appears to offer a balanced distribution of outcomes. This distribution may reflect the benefits of direct interaction and immediate feedback in traditional classroom settings, as highlighted by [39] in their study on CT integration in preuniversity programs. However, these findings contrast with those of [40], who found that OL courses were associated with lower grades and higher dropout rates. The discrepancy suggests that the effectiveness of different modalities may vary depending on the specific context, student population, and course design.

5.2 Factors Influencing Student Performance in CT

We conducted multiple linear regression analyses separately for each instructional modality, using the final course grade as the dependent variable. The goal was to examine how sociodemographic characteristics influenced academic performance and whether these effects varied by modality. While the models' low $$$ {R}^2 $$$ values indicate that demographic variables alone explain only a modest portion of the variance in student outcomes, this finding is consistent with previous educational research [41]. Academic success depends on a complex interplay of psychological, environmental, and instructional factors [42], many of which lie beyond the scope of demographic attributes.

The feature importance analysis using the Boruta algorithm revealed distinct patterns of influential factors across the three modalities. In the FTF modality, demographic factors such as age, marital status, and household responsibilities emerged as crucial. This finding aligns with the work of [43], who emphasized the importance of noncognitive factors in predicting academic success in traditional higher education settings.

In contrast, in the BL and OL modalities, admission scores and gender were the most important factors. The varying importance of demographic factors across modalities reflects the complex interplay between students' background and their chosen learning environment. In BL and OL settings, where students may have more flexibility, previous academic performance (e.g., admission scores) and gender become more prominent factors. This does not necessarily imply that personal circumstances are less influential in digital learning environments, but rather that these factors interact differently in the context of the chosen modality.

The significance of admission scores in these modalities is similar to findings from [44], which highlight the importance of previous knowledge to academic performance. Additionally, the varying importance of these factors across modalities supports the work of [45], which found that self-regulation skills are especially critical in online learning environments. Our results suggest that the skills required for success may vary significantly depending on the instructional modality, emphasizing the need for personalized support strategies.

The decision tree model, with an accuracy and precision of 0.70, provides a practical tool for predicting student outcomes based on admission scores and gender. The model's performance underscores the complex interplay between these factors and academic success. Notably, the gender-based disparities in approval rates, particularly for students with lower admission scores, raise important questions about equity in educational outcomes.

The results from the multiple linear regression, Boruta analysis, and decision tree show that different factors influence student performance depending on the learning modality. In FTF courses, the regression results emphasize the influence of age, gender, health status, family responsibilities, and financial support. These personal and social variables appear to exert a stronger impact in face-to-face settings, consistent with findings from previous studies [46-48]. In BL, the regression revealed that being a household head and gender significantly affected final grades, while Boruta and the decision tree highlighted admission scores and gender as key predictors. This suggests that both personal burdens and previous academic preparation shape hybrid learning outcomes. For OL, regression analysis identified significant effects from gender, health status, family responsibilities, and financial support. Notably, female students performed worse across all modalities, aligning with the findings of [49], who reported gender disparities in online course performance. Our findings extend this evidence to preuniversity contexts, suggesting persistent gender-based performance gaps regardless of format, though with varying intensity. The lower success rates for female students with admission scores outside the 801–900 range reinforce concerns raised by [50] about the underrepresentation of women in STEM-related pathways. These results suggest that targeted interventions such as mentorship or additional academic support could help bridge the gender gap in performance.

5.3 Limitations of the Study

While this study gives useful insights into the impact of different teaching methods (online, blended, and face-to-face) on academic performance in CT courses, there are some limitations to consider.

First, although we found significant differences between FTF and both BL and OL methods, we did not fully examine how differences in teaching quality might have affected the results. Things like how engaged the instructors were, the design of the courses, or how easy it was for students to access technology could have varied between the methods and may have influenced the results. Future studies should try to control for these factors to focus on the teaching method itself.

Second, this study was done at only one university, the Universidad Estatal de Milagro in Ecuador. This limits how much we can apply the findings to other universities or countries that have different education systems or student populations. Future research should include more universities and more diverse groups of students to make the results more general.

Third, while our analysis accounted for key demographic factors (age, gender, admission scores), unmeasured variables—such as previous technological experience, familiarity with digital learning platforms, and individual learning styles—may have influenced outcomes, particularly in online and blended modalities.

Finally, this study was cross-sectional, meaning we only looked at students' performance at one point in time. This limits our ability to say that the teaching method caused the differences in performance. A longitudinal study, following students over time, would give a clearer picture of how different teaching methods affect learning in the long term.