How the Entry Profiles and Early Study Habits Are Related to First-Year Academic Performance in Engineering Programs

Abstract

This paper explores how the entry profiles of engineering students are related to their academic performance during the first year of university in a sample of 255 first-year engineering students (77 females and 178 males) at a university in Northeast Mexico. The predictors used were the high school grade point average (HSGPA), Scholastic Aptitude Test (SAT) results, the first admission test, and a Spanish adaptation of the Survey of Study Habits and Attitudes Test (SSHA) from Brown and Holtzman. The SSHA adaptation was tested for internal consistency reliabilities via Cronbach’s alpha globally (0.92) and for the following categories: delay avoidance (DA: 0.79), work methods (WM: 0.81), teacher approval (TA: 0.89), and educational acceptance (EA: 0.74). The results were compared with those of other studies to validate their consistency. To assess the different entry profiles between high- and low-achieving students, we performed a Kruskal–Wallis test and found significant differences (p < 0.001) between both profiles for all variables. We then measured the relationships between the variables and academic success by constructing a correlation table, where HSGPA, SAT, and DA showed the highest correlations: 0.61, 0.40, and 0.36, respectively. With these outcomes, a predictive model via a logistic regression ($R^2=0.52$) was built to forecast first year academic performance in the specific context.

1. Introduction

In response to Sustainable Development Goal (SDG) 4: Quality Education of UNESCO [1], this work aims to contribute to SDG 4.3 (equal access to higher education) by increasing and expanding retention rates in higher education. Worldwide, according to the Organization for Economic Cooperation and Development (OECD) [2], the dropout rate is about 24%. In Latin America and the Caribbean, the gross enrolment rate in higher education increased to 54%, with a 22% dropout rate [3,4]. In Mexico, only 21.6% of the population has higher education [5], with the national dropout rate being 8.2% [6]. Nuevo Leon is one of the states with the lowest dropout rate at only 0.5%, which is well below the national percentage [7]. This has decreased in recent years due to the constant strategies applied in universities for this purpose. For this reason, an easy-to-apply tool is proposed so that administrators and teachers can effectively distinguish low-achieving and high-achieving students. To create strategies that allow these students to successfully carry out their university careers, it is important to pay attention early in a student’s first year to implement the necessary strategies to avoid negative experiences in their university life [8] and to eliminate school dropouts.

The first year of university constitutes a critical period that significantly influences a successful or irregular career and school dropout [9]. Academic success during this time is considered a multidimensional concept that includes successful results in various areas; it is crucial because the students’ experiences in it are the foundation of their future psychological wellbeing and academic paths [10].

Tinto [11] found that approximately 75% of the students who left their university studies did so during the first two years of the degree. This finding helps us determine that students in their first year of university have the highest probability of dropping out, and this period is considered a critical period for achieving academic success and ensuring retention. The author also mentioned that the first year is considered an adaptation phase.

The grade point average (GPA) obtained during the first semester in university is the main determinant of early dropouts [12]. In addition, some authors [13] emphasized that achieving high grades in science courses, such as physics, chemistry, and calculus, is directly related to retention predictions in engineering programs.

It has also been observed that academic skills, confidence, time management [14], independent learning [15], and emotional factors [16] are generally related to the the student’s ability to adapt to university [17,18].

For this reason, assessing the factors involved in academic success during the first year is essential when acquiring data from students. These can serve as a great support for the development of actions aimed at support and timely intervention for those students who require it, such that successful retention is ensured during the first year, leading to academic success and, therefore, to an increase in the graduation rates of students in the field of engineering.

Aim and Structure of the Study

The objective of this research was to discover how the entry profiles of engineering students are related to their academic performance during the first year of university, which can lead to the design of appropriate care and early intervention strategies and an increase in retention, thus ensuring higher graduation rates.

For this task, we chose the Survey of Study Habits and Attitudes (SSHA) developed by Brown and Holtzman [19], which was applied at the beginning of the semester. This allowed us to measure categories such as: (a) delay avoidance (DA), which is focused on measuring the degree of students’ procrastination, that is, their ability to manage time and perform tasks on time;

(b) work methods (WM) for levels of effective study skills;

(c) teacher approval (TA) for a student’s opinions about a teacher’s behavior and methods; and (d) educational acceptance (EA) for measuring a student’s educational objectives.

We also analyzed the variables of study habits (SH) and study attitudes (SA), which are defined as follows:

$$\mathrm{SH}=\mathrm{DA}+\mathrm{WM},$$

(1)

$$\mathrm{SA}=\mathrm{TA}+\mathrm{EA}.$$

(2)

Additionally, we included information on admission metrics:

-

SAT as an entry requirement; students take a Spanish version of the SAT that was developed by the College Board [20] to determine their mathematical and verbal abilities;

-

High School Grade Point Average (HSGPA) to have a measure of the student’s previous academic performance.

We measured academic performance in the first year using the grades in a differential calculus course as an output variable. Based on these grades, we categorized the students as high achievers (HAs) and low achievers (LAs). More details can be found in Section 2.4. Afterwards we analyzed how the other variables are related to this outcome and could possibly predict it.

To resolve if there were significant differences among students with different performances in the first year, we studied the students’ profiles in both categories (HAs and LAs) estimated with the variables of the SSHA, HSGPA, and their results on the SAT test, as shown graphically in Figure 1. In Section 2, we present the details of the materials used, the characteristics of the sample, and our statistical analyses. In Section 3, we show the results obtained from our research, which we discuss in Section 4, and we end with our conclusions in Section 5.

2. Materials and Methods

This study aimed to determine how the entry profiles of engineering students are related to their academic performance during the first year of university so that early attention strategies can be designed and retention can be increased and, particularly, in order to provide a simple prediction model that academic administrators can implement. Therefore, the research questions (RQs) were:

RQ1.

How different are the entry profiles and early study habits of HAs and LAs?

RQ2.

How is the first-year academic performance related to the entry profile and early study habits?

RQ3.

How well can admission metrics and early study habits measurements predict student performance in the first year of engineering programs?

2.1. Participants and Application Procedure

This study involved the participation of 255 first-year engineering students (77 females and 178 males) from a private university in Northeast Mexico in the fall of 2019. For the sample, we selected the whole cohort of students enrolled in the differential calculus course and, later, in the integral calculus course.

2.2. Questionnaires

2.2.1. The Survey of Study Habits and Attitudes Test (SSHA)

Brown and Holtzman developed the SSHA and designed it for use as a screening and diagnostic tool, a teaching aid, and a research tool [19]. It is one of the most widely used inventories, and it is focused on distinguishing between levels of study habits, skills, and attitudes. The SSHA was originally designed to provide a single score that measures effectiveness in academic achievement. Since it measures achievement through the habits and attitudes of students, its scale, with 100 items, is also considered an accurate predictor [21] of a student’s academic outcomes.

The original version of the test was developed in English many decades ago; thus, we made a translation and a revised version in Spanish to adapt it to the modern Mexican context. We have included it in Appendix A. One semester prior to when this study was performed, we conducted a pilot test to determine its validity through Cronbach’s Alpha with a group of 62 students enrolled in an engineering program at the same university. As a result, we found the values of Cronbach’s Alpha—presented in

Table 1. Cronbach’s Alpha of the revised and Spanish version of the SSHA that we used in this study. We have included it in Appendix A.

Category	$\mathit{\alpha }$	Consistency
SSHA	0.92	Excellent
DA	0.79	Acceptable
WM	0.81	Good
TA	0.89	Good
EA	0.74	Acceptable

—which confirmed the validity of the test to be used in this work. According to the criteria of George and Mallery [22] for Cronbach’s Alpha, we can see in Table 1 that the results obtained in the pilot test ranged from acceptable to excellent in individual constructs, while Cronbach’s Alpha for the general test resulted in an excellent coefficient. For this reason, it can be concluded that the update of the SSHA survey for the research carried out is reliable and valid.

2.2.2. The Scholastic Aptitude Test (SAT)

The Scholastic Aptitude Test was developed by the College Board [20], and it measures what a student is capable of achieving with the knowledge acquired in their years of study [23,24,25,26,27,28] and assesses the academic potential of the student in the continuation of university studies.

SAT scores [29] are reported on the College Board scale. The total score is a number between 400 and 1600. Any score above 1050 is above average and indicates that the student will be accepted by many colleges [30]. There are two steps for designating scores:

• One determines the adjusted scores on the verbal and math sections by counting the number of exercises answered correctly and subtracting the fraction of exercises answered incorrectly;
• One then translates the adjusted scores to the corresponding ones on the College Board scale.

Currently, some private universities in Mexico apply the SAT to determine the admission profiles of their students.

2.3. Percent Grade and Grade Point Average Scores

In general, for students, academic achievement is an essential factor in their university careers. This motivates most institutions to request a given level of academic performance and a threshold score in their grades to allow students to continue their studies in their first year and the remaining years.

Commonly, universities focus on measuring success with numerical grades and semester-by-semester course accreditation.

Table 2. Comparison of GPA scales.

Percent Grade	4.0 Scale	Letter Grade
97–100	4.0	A+
93–96	4.0	A
90–92	3.7	A−
87–89	3.3	B+
83–86	3.0	B
80–82	2.7	B−
77–79	2.3	C+
73–76	2.0	C
70–72	1.7	C−
67–69	1.3	D+
65–66	1.0	D
60–64	1.0	D−
59 or less	0.0	F

shows the equivalence of the grade scales that are mainly used by universities according to their context. For example, Mexico primarily uses a percent scale.

The high-school GPA (HSGPA) is the final average obtained in high school upon graduation and is measured on the same scale as the university average.

2.4. Output Variable of Academic Achievement (ACH)—LAs and HAs Classification

Budny et al. [31] analyzed how student retention decreases with lower first-year science and math course grades.

Table 3. GPA courses vs. retention rate.

			LAs			HAs
Course GPA	0–1.0	1.0–1.5	1.5–2.0	2.0–2.5	2.5–3.0	3.0–3.5	3.5–4.0
Retention rate	7.0	18.2	30.6	47.4	62.0	77.0	87.0

shows the relationship between the retention rate of students and the grades obtained by students in their courses, such as calculus, chemistry, and physics.

Students who obtain an average above 2.5 on the 4.0 scale have improved academic performance rates in engineering, exceeding a 50% retention rate. We could see that a high average was obtained by students in the first year.

Private universities use the percent scale, for which an average higher than 2.5 would be equivalent to a grade higher than 80. Based on this, in order to measure academic performance, a categorical variable, ACH, was defined by using the final grade in the differential calculus course, and the threshold grade of 80 was used as the separation between LAs and HAs:

$$\mathrm{ACH}=\left\{\begin{array}{ccc}\mathrm{HAs}\hfill & if& \mathrm{GPA}\ge 80.\\ \\ \mathrm{LAs}\hfill & if& \mathrm{GPA}<80.\end{array}\right.$$

(3)

2.5. Statistical Analysis and Data Processing

For data analysis, we used the statistical software R [32]. For the SSHA adaptation, internal consistency reliabilities were tested by using Cronbach’s Alpha globally and for each of the categories. The correlations of the HSGPA, SAT, and SSHA variables with academic performance were measured. With these results, we propose a predictive model through logistic regression that can be used to predict academic performance in the first year. We also verified the AIC, Nagelkerke $R^2$, accuracy, sensitivity, and specificity.

3. Results

3.1. RQ1: Kruskal–Wallis Analysis

To address RQ1, we performed a Kruskal–Wallis test to resolve the discriminatory capability of the variables and, thus, be able to distinguish between high-achievers (HAs) and low-achievers (LAs), with a sample of 136 HAs and 116 LAs.

As shown by the Kruskal–Wallis test (

Table 4. Kruskal–Wallis test.

	K	DF	p	${\mathit{\eta }}_{\mathit{H}}$
HSGPA	99.6	1	1.86 × 10−23	0.390 (large)
SAT	37.5	1	8.91 × 10−10	0.144 (large)
DA	33.4	1	7.44 × 10−9	0.128 (moderate)
WM	27.8	1	1.37 × 10−7	0.106 (moderate)
TA	15.7	1	7.56 × 10−5	0.0580 (small)
EA	22.2	1	2.51 × 10−6	0.0836 (moderate)
SH	35.8	1	2.21 × 10−9	0.137 (moderate)
SA	19.7	1	9.14 × 10−6	0.0738 (moderate)

), all the variables were significant in separating the LAs and HAs. The variables that showed a greater size effect were HSGPA (${\eta }_H$ = 0.390, large) and SAT (${\eta }_H$ = 0.144, large). Those that had a moderate size effect were DA (${\eta }_H$ = 0.128, moderate), WM (${\eta }_H$ = 0.106, moderate), and EA (${\eta }_H$ = 0.0836, moderate), and TA showed a small size effect (${\eta }_H$ = 0.0580, small). Therefore, we selected HSGPA, SAT, and DA for our main model.

Analyzing the means (

Table 5. Means and standard deviations of entry profiles for low achievers and high achievers.

		HAs			LAs
	n	Mean	sd	n	Mean	sd
DA	136	23.17	8.49	119	16.65	8.37
WM	136	27.10	8.66	119	20.96	8.58
TA	136	25.29	9.65	119	20.49	9.37
EA	136	26.07	8.88	119	20.82	8.93
SH	136	50.26	15.67	119	37.61	14.96
SA	136	51.36	18.10	119	41.30	17.97

) of the profiles of the HAs and LAs in terms of the variables of the SSHA test, DA was the variable with the lowest mean values, with 23.17 (HA) and 16.65 (LA), followed by TA, where HAs had a higher mean score (25.29) compared to that of LAs (20.49) and EA, with a mean of 26.07 for HAs and 20.82 for LAs. WM was the variable with the highest mean values: 27.10 for HAs and 20.96 for LAs. The SH and SA composite variables showed means of 50.26 and 51.36 for HAs and 37.61 and 41.30 for LAs, respectively.

The results indicate a significant difference between HAs and LAs for all tested variables. Based on this, we claim that the SSHA, HSGPA, and the updated and revised version of the SAT are pertinent as predictive instruments for students’ academic performance. These results allowed us to partially answer RQ1, with evidence that HAs have significantly better attitudes, study habits, and SAT scores than those of LAs.

3.2. RQ2: Correlation Analysis

To answer RQ3, we performed a correlation test.

Table 6. Predictors and response correlation analysis.

	ACH	HSGPA	SAT	DA	WM	EA	TA
ACH	1	0.61 *	0.40 *	0.36 *	0.34 *	0.28	0.24
HSGPA		1	0.55	0.37	0.42	0.32	0.26
SAT			1	0.06	0.24	0.08	0.09
DA				1	0.66	0.53	0.49
WM					1	0.49	0.51
EA						1	0.92
TA							1

* High Correlation.

shows the correlations between all the variables considered in this analysis. For this purpose, HAs = 1 and LAs = 0. Correlations were classified according to [33]. Scores between 0.3 and 0.6 in the social sciences correspond to a moderate to strong correlation. The variables that were most correlated with academic performance were HSGPA, SAT, DA, and WM (0.61, 0.40, 0.36, and 0.34, respectively).

These results allowed us to answer RQ3, with evidence that the variable with the greatest correlation was HSGPA.

3.3. RQ3: Model Analysis

Based on the results obtained above, to determine the best model for predicting academic performance assembled with the variables studied in this work, we used Akaike’s information criterion (AIC) [34]. The model that included the most variables was Model 2, which included the independent variables HSGPA, SAT, and SH = DA + WM. In each of the other models, one or more of the variables were omitted to match the criteria for the eligibility of each university. The models were organized according to the AIC value. We present our results in

Table 7. Comparison of the Akaike information criterion (AIC) and Nagelkerke $R^2$ between models.

	Model	AIC	Nagelkerke ${\mathit{R}}^2$
1	HSGPA + SAT + DA	234.71	0.52
2	HSGPA + SAT + SH *	237.49	0.51
3	HSGPA + DA	238.39	0.50
4	HSGPA + SAT	242.43	0.49
5	HSGPA	243.52	0.48
6	SAT + DA	277.11	0.36

* SH = DA + WM.

; a lower AIC value corresponds to greater compatibility of the model with the data. We also included the Nagelkerke $R^2$ for each model. Both findings imply that using a combination of HSGPA, SAT, and DA would be the most suitable academic predictor.

We can see that, as the value of the AIC increased, the value of $R^2$ decreased, which was consistent with the choice of Model 1 as the best predictor. Model 1 (HSGPA + SAT + DA) had the lowest AIC with a value of 234.71 and the highest $R^2$, which was $0.52$. Although Model 2 had all four of the variables that had the highest correlations with the output variable (HSGPA and SAT + DA + WM), the AIC value increased to 237.49 and the $R^2$ decreased to 0.51, so we do not recommend including WM in the final model. In the case of reducing the variables analyzed in Models 3 to 6, we could see that they were still good predictors, since the value of $R^2$ ranged from 0.36 to 0.50.

Table 8. Comparison among the models. (*) stands for moderate, (**) medium, and (***) large significance.

	Model	Variable	$\mathit{\beta }$	SE	p-Value	VIF
1	HSGPA + SAT + DA	(Intercept)	0.10	0.17	0.54
		HSGPA	1.45	0.25	6.44 × 10−9 ***	1.13
		SAT	0.49	0.21	0.02 *	1.16
		DA	0.57	0.19	0.002 **	1.09
2	HSGPA + SAT + SH	(Intercept)	0.10	0.16	0.54
		HSGPA	1.47	0.25	3.97 × 10−9 ***	1.14
		SAT	0.41	0.20	0.04 *	1.11
		SH	0.49	0.19	0.01 **	1.06
3	HSGPA + DA	(Intercept)	0.12	0.16	0.46
		HSGPA	1.68	0.23	5.33 × 10−13 ***	1.02
		DA	0.47	0.18	0.008 **	1.02
4	HSGPA + SAT	(Intercept)	0.10	0.16	0.55
		HSGPA	1.67	0.24	4.74 × 10−12 ***	1.09
		SAT	0.34	0.20	0.08	1.09
5	HSGPA	(Intercept)	0.11	0.16	0.51
		HSGPA	1.82	0.23	2.08 × 10−15 ***	1.00
6	SAT + DA	(Intercept)	0.13	0.15	0.37
		SAT	1.07	0.18	4.05 × 10−9 ***	1.07
		DA	0.93	0.17	3.34 × 10−8 ***	1.07

presents the statistics for each model.

Table 8 shows the models that used standardized variables, since it was then possible to directly observe the variables that had the greatest impact on the models. We could observe in Models 1–5 that HSGPA generally had the highest $\beta$ coefficient and the lowest p-value, so it was more impactful. In the case of Model 6, when we did not use HSGPA, the aptitude test was the variable with the greatest impact on the model. The variance inflation factor (VIF) in all of the models was close to 1, so we can assume that the correlation between the predictor variables in the model was not large enough to require attention.

The accuracy, specificity, and sensitivity of the prediction of academic performance according to the predictors of each model are shown in

Table 9. Specificity, sensitivity, and accuracy of each model.

	Model 1					Model 2
	Observed	Predicted		% Correct		Observed	Predicted		% Correct
		0	1				0	1
Success	0 (LA)	88	31	Specificity	0.74	0	88	31	Specificity	0.74
	1 (HA)	22	114	Sensitivity	0.84	1	24	112	Sensitivity	0.82
				Accuracy	0.79				Accuracy	0.78
	Model 3					Model 4
	Observed	Predicted		% Correct		Observed	Predicted		% Correct
		0	1				0	1
Success	0	87	32	Specificity	0.73	0	89	30	Specificity	0.74
	1	25	111	Sensitivity	0.82	1	26	110	Sensitivity	0.81
				Accuracy	0.78				Accuracy	0.78
	Model 5					Model 6
	Observed	Predicted		% Correct		Observed	Predicted		% Correct
		0	1				0	1
Success	0	90	29	Specificity	0.76	0	82	37	Specificity	0.69
	1	24	112	Sensitivity	0.82	1	32	104	Sensitivity	0.76
				Accuracy	0.79				Accuracy	0.73

.

Table 9 presents the prediction accuracies of the above models. The prediction rates of LAs as (0) were 74%, 74%, 73%, 75%, 76%, and 69% for each model, respectively. The prediction rates of HAs as (1) were 84%, 82%, 82%, 81%, 82%, and 76%, respectively. Finally, the proportions of predictions identical to the current category among all of the data were 79% for Models 1 and 5, 78% for Models 2, 3, and 4, and 73% for Model 6. Model 1, which had greater sensitivity, allowed us to better identify HAs, as it was the most useful model for student selectivity. Where only the HSGPA was known, we could observe that Model 5 was useful for early warning models, as it had a higher specificity. Model 1 allowed us to further answer RQ3 by showing that the HSGPA, academic aptitudes (SAT), and study habits (SSHA) could be used to predict the academic performance of students in a first-year college program, as shown graphically in Figure 2. It was observed that the other models were still useful.

4. Discussion

The main objective of this study was to observe how the entry profiles of engineering students are related to their first-year academic performance. For this purpose, their HSGPA SAT scores and their entry Delay Avoidance DA were considered. First, the profiles of the students were observed to see if there were significant differences between HAs and LAs. Afterward, correlations between variables were performed to analyze which were more related to academic performance. Finally, regression models were used to observe the predictive power of the entry profile variables. Based on the study results, we observed that HAs and LAs have significantly different entry profiles in terms of HSGPA, SAT, and DA. Moreover, the observed differences in Delay Avoidance are consistent with studies undertaken in similar contexts

Table 10. SSHA validation.

		HAs			LAs
	This Work	Aquino (2011)	Consistent	This Work	Aquino (2011)	Consistent
DA	21.7–24.6	16.8–23.6	Yes	15.1–18.2	14.9–16.4	Yes
WM	25.6–28.6	22.6–31.9	Yes	19.4–22.5	16.2–17.9	No (Small difference)
TA	23.7–26.9	17.2–24	Yes	18.8–22.2	14.3–15.7	No (Small difference)
EA	24.6–27.6	23–31.7	Yes	19.2–22.4	16.7–18.4	No (Small difference)
SH	47.6–52.9	39.8–55.1	Yes	34.9–40.3	31.2–34.2	No (Small difference)
SA	48.3–54.4	40.7–55.2	Yes	38.1–44.5	31.2–34	No (Small difference)

[35]; therefore, measuring the student’s study habits on their entrance to the university provides valuable information about their expected academic performance.

We compared the results of a similar study performed by Aquino (2011) [35] with our results. Table 5 presents the standard deviations and mean values of entry profiles for low and high achievers in both studies. It can be seen that there were no significant differences between the results obtained in both studies; furthermore, we computed a more accurate comparison based on the confidence intervals for the mean value at 95% for each of the variables. Our results can be seen in Table 10. We found that the mean confidence intervals for all variables were consistent for HAs, but only those of DA were consistent for LAs. Moreover, because the DA intervals were always consistent, one may infer that the observed confidence intervals of DA (or a given DA threshold) may be used to classify students as HAs or LAs according to their entry profiles. Nevertheless, one may consider that both the study by Aquino and this study were conducted in similar contexts.

For comparison,

Table 11. Comparison of correlations with other studies.

Study	Variable Output	Academic Variables				Non-Academic Variables
	Performance Level	Previous Performance	Skills	Study Habits		Attitudinal
Our Study	HAs or LAs	HSGPA 0.61	SAT 0.40	DA 0.36	WM 0.34	EA 0.28	TA 0.24
Alhadabi et al. [36]	GPA			Mastery Goals 0.26	Self-efficacy 0.22	Consistency of interest 0.19
DeBerard et al. [23]	GPA	HSGPA 0.67	SAT 0.30			Mental health −0.01
Allen et al. [10]	GPA	HSGPA 0.51	ACT 0.44			Commitment to College −0.04
Olani [24]	GPA	HSGPA 0.41	Entrance exam score 0.35	Skills and Efficiency 0.17		Motivation 0.11
Van der Zanden et al. [37]	GPA	HSGPA 0.62		academic adjustment 0.44		Social adjustment −0.09

presents our results and those of previous studies with similar models and variables, where the correlation between academic performance in a student’s first year and the student’s entry profile was considered. Since the studies used different statistical methods, we compared the correlations between the output variable, i.e., academic performance, and the input variables of the students, i.e., their entry profiles.

In this study, as in those presented by other authors (Table 11), a high tendency was clearly seen in the correlation between the level of academic performance and the variable of previous performance. Similar to the other studies, the HSGPA showed the highest correlation. In our case, the high school average was 0.41–0.67. The second highest correlation was with the SAT or entrance exam (0.30–0.44). In turn, a correlation was also observed with the academic variables focused on study habits (0.17–0.44), which are tools that students need to develop to be successful in their studies. Finally, attitudinal variables showed considerable differences between studies, and there were very low correlations between academic performance (r < 0.3) and the output variable. This was expected, since attitudinal variables are difficult to measure, so using them as academic predictors is not recommended. In addition to the above, it was found, in the present study, as in the others analyzed, that the attitudinal variables did not present a significant correlation (0.04–0.28); for this reason, the category corresponding to the attitudes will not be taken into account as part of the final model concerning the output variable. For comparison, we list the variables of this work with those of other studies in

Table 12. Summary of the comparison of the studies.

Refs.	Model	Variables	Beta	SE	t	F	R	R2	R2 Change
This Work								0.52
		HSGPA	0.23	0.04 (p ∼6 × 10−9
		SAT	0.003	0.001 (p ∼0.02)
		DA	0.06	0.02 (p = 0.002)
Krumrei-Mancuso et al. (2013) [38]	Psychosocial Factors Predicting First-Year College Student Success (M3)						0.69	478	0.023 (p < 0.01)
		Gender	−0.1
		Ethnicity	0.6
		1st semester GPA	0.62 (p < 0.1)
		Academic Self-Efficacy	0.17 (p < 0.1)
		Organization and Attention to Study	0.07
		Stress and Time Press	−0.06
		Involvement with College Activity	−0.04
		Emotional Satisfaction with Academics	−0.08
		Class Communication	−0.01
T. Coyle et al. (2011) [39]	Regressions of GPA on SAT and g (multiple-aptitude test)					41.98 (p < 0.01)		0.15
		SAT	0.35		6.89 (p < 0.01)
		g	0.03		0.61
		SATxg	0.03		1.80 (p < 0.1)
E. Cohn et al. (2004) [27]	Partial regression coefficients and t-ratios (in parentheses). Dependent variable: college GPA	Intercept	−0.041 (−0.23)			55.70 (p < 0.1)		0.39
		Percent Rank	0.005 * (2.85)
		High School GPA	0.231 * (4.03)
		SAT	0.0017 * (10.02)
		White female	0.163 * (3.21)
		Nonwhite male	−0.006 (−0.08)
		Nonwhite female	0.029 (0.46)
Reynolds and Weigand (2010) [40]	Predicting GPA							0.08
		Self-efficacy	−0.08	−0.05 (0.08)
		Academic resilience	0.49 (p < 0.01)	0.56 (0.17)
	Predicting GPA							0.09	0.01
		University environment	0.12
	Predicting GPA							0.09	0.01
		IMO	−0.11	−0.06 (0.06)
	Predicting GPA							0.1	0.00
		EMO	−0.07	−0.05 (0.08)
	Predicting GPA							0.12	0.01
		Amotivation	−0.12	−0.08 (0.06)
	Predicting GPA							0.22	0.01
		Race	−0.34 p < 0.001	−0.61 (0.15)
Wolfe and Johnson (1995) [28]	All Forward Multiple Regression Analyses							0.38 (p < 0.01)
	JPI scales	High school averages	0.43	p < 0.01					0.19
		Organization	0.27	p < 0.01					0.07
		SAT total	0.26	p < 0.01					0.05
		Infrequency	−0.16	p < 0.05					0.03
		Risk taking	−0.17	p < 0.05					0.02
		Innovation	0.16	p < 0.05					0.02
	Big 3							0.35 (p < 0.01)
		High school average	0.43	p < 0.01					0.19
		Control	0.32	p < 0.01					0.09
		SAT total	0.24	p < 0.01					0.05
		Well-being	0.13	p < 0.05					0.02
	Big5							0.32 (p < 0.01)
		High school average	0.43	p < 0.01					0.19
		Conscientiousness	0.31	p < 0.01					0.09
		SAT total	0.23	p < 0.01					0.04
	Other predictors							0.32 (p < 0.01)
		High school average	0.43	p < 0.01					0.19
		Self-efficacy	0.27	p < 0.01					0.08
		SAT total	0.21	p < 0.01					0.03
		Attendance	0.16	p < 0.05					0.02
	All predictors							.35 (p < 0.01)
		High school average	0.43	p < 0.01					0.19
		Control (from Big 3)	0.32	p < 0.01					0.09
		SAT total	0.24	p < 0.01					0.05
		Infrequency (from JPI)	−0.14	p < 0.05					0.02

. The variables in the predictive models, such as the HSGPA, are frequently found to be significant variables for predictive models. As shown in the table, when models include the HSGPA and SAT, the beta values are below 0.62 and 0.35, respectively. Other variables, such as gender, intrinsic and extrinsic motivation, and resilience, have also been reviewed, as can be seen in Table 12.

The HSGPA and SAT were proven to be the best predictors of first-year academic performance and, therefore, student retention. This is in agreement with the literature [41,42,43,44]. Nevertheless, ref. [45] states that, from previous studies, GPA may be the most promising predictor, but it is sensitive to the particular institution of origin. Regarding the SAT, students with scores lower than 890 had the lowest retention rate, i.e., 63.8%. Similarly, of the students who had an HSGPA of C- or lower, only 40.0% returned for their second year of college [42]. This allows us to confirm that our model is a good predictor of academic performance of engineering in the first year of university students in similar contexts.

Limitations

Regarding the limitations of our research, contextual considerations should be made. There are factors that could have affected academic dropout that were not analyzed, such as those related to the socioeconomic levels of the students (given that a private university was taken as the object of study) and external personal and familial factors. Other factors that were not measured but could also be considered in future studies include the levels of motivation of the students, the reasons for which they chose to study engineering, and whether they had a financial scholarship at the university (see, for instance, [27,28,38,39,40]; for a systematic review, see [46] and the references therein).

Note that these results may not be entirely transferable for all contexts. For instance, [47] found that study attitudes and habits were not significant in predicting high school academic success. Moreover, a study of engineering studies in a different context [48] did not observe a significant effect of HSGPA on students’ academic performance. Therefore, the present outcomes should be taken with caution on the particularities of each case, even their origin institution [45].

5. Conclusions

In this work, we studied a sample of 255 first-year engineering students at a university in Northeast Mexico. Our aim was to determine the factors that influenced their academic performance in their first year of university so as to determine appropriate academic strategies and provide evidence for a simple prediction model that academic administrators can use.

By studying academic data, such as the SAT, HSGPA, study habits, and study attitudes, and by using a Kruskal–Wallis test, we found that high achievers had more positive attitudes, more effective study habits, and significantly higher SAT scores than those of LAs.

Furthermore, the results of our correlations showed that the HSGPA was strongly related to first-year academic performance, which allowed us to conclude that, by focusing on the HSGPA, SAT, and DA variables (or any other procrastination variable), which had proven predictive power with respect to student’s academic performance, certain variables can be used to forecast students’ academic performance in a first-year university program through the students’ admission profiles. While measuring procrastination sometimes requires the application of additional surveys, using only the HSGPA and SAT still provides a good approximation, and most institutions have such data available.

The SSHA survey was confirmed to be an excellent tool for measuring study habits and attitudes. We have provided an updated and reviewed Spanish version of this test.

Finally, we recommend the following:

• This study should be adapted to the context of each institution depending on their needs and experiences, and corresponding intervention actions should be taken for students whose academic performance is unfavorable or could be improved so as to ensure long-term academic performance;
• It is recommended that each institution choose an instrument that helps them determine students’ academic skills. In the present study, an adaptation of the SSHA test was selected given the context of the institution;
• Taking into account all of the above, an accurate classification system that allows for the implementation of support strategies focused on students’ needs can be developed to support the development of students’ academic performance, thus ensuring higher retention rates and much higher graduation rates;
• We also recommend considering each student’s personal threshold so that they can achieve academic performance in a healthy manner.

Thus, in this work, we have shown that a model with a few variables is sufficient to produce a robust predictive model. A natural enhancement of this study would be to extend it into other disciplines, particularly those with different gender proportions, in order to explore any relevant demographic effects. Such broader investigations, which are beyond the scope of the present work, would help to explore the wider transferability of these findings.