A Proposed MIMIC Structural Equation Model for Assessing Factors Affecting Time to Degree—The Case of the Greek Tertiary System

Abstract

Using a structural equation modeling approach, this study has attempted to untangle the underlying pathways on how students’ demographics and pre-college characteristics that reflect academic preparation, combined with major factors formulated in the university environment, affect time to degree. It does so by developing and evaluating a conceptual framework whereupon time to degree is associated with specific observed or latent factors. A properly tailored Multiple Indicator Multiple Causes SEM was used for evaluating the hypotheses made on a sample of 1137 graduates which came from a Greek University of Social and Political Science, Athens, Greece. AMOS and LISREL packages were used for the analysis. The results reveal interesting direct and indirect relationships of the various predictor variables with time to degree. In particular, the great contribution of student performance and academic integration to time to graduation has been highlighted. However, the contribution of the pre-university features is also worthy of attention.

1. Introduction

The interest of researchers in the phenomenon of non-completion of studies by a substantial number of students is traced back to the early decades of the last century, especially in the USA, where the term student mortality was initially used to characterize it (Bean & Metzner, 1985). Over time and in parallel with the massification of education, similar phenomena, including, in general, any form of delay in obtaining a degree, occur in many countries or levels of education, while various terms have been used for their typification, which often reflect local circumstances. Student attrition, late graduation, dismissal, long time graduation, delayed completion, delayed potential graduation, dropout, stopout, voluntary withdrawal, and delayed college graduation are some of these terms (see, among others, Bean, 1980; DesJardins et al., 2006; Lassibille & Gómez, 2008; Pascarella & Terenzini, 1977). Overall, these terms or associated phenomena are linked and are usually determined according to the final outcome (or event) of studies when it is not identical to “on-time graduation”, i.e., graduation as provided by the rules of an educational system or institution.

The systematic study of these phenomena and, in particular, of the risk of such occurrences, of the length of time until their occurrence, and of the factors related to them has caused increasing interest in researchers, given that the number of students with such characteristics is somewhat increasing (Spady, 1970; Tinto, 1975; Bean, 1980; Belanger et al., 2002).

This interest has primarily to do with labor planning of human resources for higher education graduates at the nationwide or regional level. In addition, the corresponding research results are useful for educational policy design and decisions at an institutional level, since, in recent years, these issues have been used as performance indicators of the institutions, while they should also be of interest to potential graduates (Donald, 1999; Beer & Lawson, 2017).

While the questions investigated by researchers vary and may have a local character reflecting the graduation rules applying in the educational system of a country or in a particular educational institution or level of education, the main questions usually concern the distribution of the duration of studies whose final event could correspond to graduation at a foreseeable time, graduation at a later time (early or late), or dropping out from studies, as well as to the factors related to such occurrences.

Various theories, models, or approaches have been developed for understanding, explaining, and predicting the above phenomena. These approaches are categorized as psychological, sociological, organizational, and integrated. Within the psychological approaches, the “students’ involvement theory” proposed by Astin (1999) features prominently. The idea is that student persistence is conditional on the amount of physical and psychological energy that the student devotes to the academic experience. From the sociological perspective, Tinto’s student integration model is the most cited (Lassibille & Gómez, 2008). The model explores how the joint interactions of the academic and social systems may determine student persistence within an educational institution. Tinto’s view is that student retention is positively related to students’ academic and social commitment to graduation. From the organizational/economic perspectives, concepts usually met in human research management are used to handle students’ departure from university institutions. Thus, Bean (1980) extends approaches used for studying employees’ turnover and job satisfaction to the study of student attrition. In particular, Bean’s “student attrition model” explores the relationship between the university’s organizational structures and student retention, taking into account students’ perceived satisfaction with their studies. Finally, the integrated approaches combine elements of the aforementioned theories towards a better explanation and understanding of the problem. Thus, for example, Cabrera et al. (1993) showed that when sociological and organizational approaches are not considered as mutually exclusive but as complementary, they can better explain students’ attrition issues.

The empirical investigation of the above phenomena and, principally, the empirical confirmation of the theoretical models that have been proposed for their interpretation is based almost exclusively on statistical analysis. For this purpose, several statistical methods have been employed, including simple descriptive and multidimensional methods like survival analysis, and ordinary and logistic regression (see, among others, Peng et al., 2002; Zwick & Sklar, 2005). However, the latent nature of the variables involved and the direct and indirect relationships that have to be investigated in many cases require the use of most advanced methods like structural equations models for processing. Indeed, these methods are increasingly used in such applications and have proven particularly effective in investigating the distribution of the length of studies and in estimating the likelihood of different types of termination of studies as well as in analyzing the factors that cause or are related to these issues.

In the above context, Kalamatianou and McClean (2003) studied quantitatively, on the basis of nonparametric and parametric survival methods, a peculiar distribution of the time duration of studies which emerges in education systems where there is no upper time limit for the completion of studies. More specifically, in such systems, students who first enroll in a university department can formally graduate if they successfully fulfil a certain number of modules, while there is a fixed minimum duration of time for doing that. Graduation is not possible before this minimum time, so it is a threshold for graduation. However, students who fail to satisfy the modules condition in the minimum time can proceed until fulfilling these conditions by taking exams for an unlimited number of times and graduating later. Thus, there is a time threshold for graduation but not an upper time limit. In such cases, time to degree may last for a long time, and the corresponding distribution may have a long right tail that never reaches the time axis, allowing for a type of perpetual studentship. Greek universities exemplify this phenomenon, but it seems that the same stands in Finland’s and South Africa’s educational systems, while time to submission for British PhD students provides another example (Ziegel et al., 1998; DesJardins et al., 2006; Yue & Fu, 2017).

Considering the above, a reasonable question arises as to the factors that shape or are associated with this pattern of the distribution of time to degree. In the present study, there was an attempt to provide key answers to this question for the case of a Greek university.

To this end, first, a conceptual framework is established, which draws on the models of Tinto (1975) and Bean (1980), but it is tailored to cover the details of the present application. In summary, it includes hypotheses made on how students’ demographic and other pre-college characteristics, as well as factors formulated during their studies, affect time to degree. These hypotheses were formulated on the basis of previous findings and issues regarding studies’ progress in general. The particular hypothesized direct and indirect relationships are also indicated in this framework. Then, for testing the validity of this theoretical model in the data, a proposed Multiple Indicators Multiple Structural Equation Model was modified to meet the application’s requirements. The application was made by means of the LISREL and Analysis of Moment Structure (AMOS) packages.

The results demonstrate primarily the validity of the conceptual model and the suitability of its statistical SEM counterpart. They also provide useful information regarding the factors associated with the duration of studies for the university from which the data were derived.

The rest of this paper is organized as follows. In the next section, a brief literature review is provided, which motivates the choice of explanatory variables and the major hypotheses about the factors that are considered to be associated with time to degree. In Section 3, the conceptual framework is considered, regarding the mechanism by which the various factors or variables affect the duration of studies. Section 4 concerns the statistical analysis, and the particular SEM used to analyze the data and to evaluate the conceptual model. In Section 5, the results of the analysis are presented, along with comparisons with the findings of similar surveys, while in the last section, conclusions are drawn.

1.1. Factors Related to Time to Degree: Literature Review and Hypotheses

The literature reveals many factors or variables that affect, correlate, or cause the abovementioned phenomena, including the peculiar distribution of time to degree that appears in Greek reality, where the idea is that the two latter parts of the distribution, reflecting late graduation and possible perpetual studentship, can both be considered as a type of student attrition, while the latter can be considered as potential dropout as well.

Nevertheless, the factors influencing these phenomena can be categorized into two broad categories corresponding to student demographics and pre-college characteristics and to factors formulated during studies. Next, there is reference to the individual factors of each category, along with a brief overview of the corresponding empirical results guiding the working hypotheses of this research.

1.1.1. Student Demographics and Pre-College Characteristics

The literature reveals a convergence of views regarding the role of students’ demographic characteristics, such as gender, age, hometown location, family socioeconomic status, etc., as well as of their pre-college characteristics, such as secondary school grades, degree of goal commitments (i.e., how important it is to graduate from college), as factors affecting students’ persistence (e.g., Diaz Lema et al., 2024), attrition (e.g., Davidson et al., 2009), dropout (e.g., Berger & Braxton, 1998), or even time to degree (e.g., DesJardins et al., 1999).

Gender is almost always included in a survey regarding general issues of student progress; however, the results differ among surveys. Thus, some authors have found that female students tend to be more successful in academic work than male students, in the sense that they seem to persist more in getting their degree (Whiteman, 2004), they drop out their studies less often (Respondek et al., 2017), and they have a shorter duration of studies or are less likely towards late graduation than male students (Yue & Fu, 2017). This could be explained by differences in communication patterns and the sense of community between genders. Nevertheless, other researchers (Ekornes, 2022) found no correlation between gender and persistence (Whiteman, 2004) or dropout (Kocsis & Molnár, 2024).

Age-related effects in these phenomena have also been shown to differ among studies. Some researchers conclude that the age factor is directly related to dropout decisions (Nikolaidis et al., 2022; Kemp, 2002), in the sense that students who enter university at an older age are more likely to drop out. Instead, other studies have found that young students are significantly more likely to drop out than their older counterparts (Thies & Falk, 2023) or that the age factor has a limited yet significant effect on student attrition (Longwell-Grice & Longwell-Grice, 2007).

Early studies on the impact of students’ hometown–college distance have found that students from rural or out-of-state areas and from small towns had higher attrition rates. However, the corresponding results of later studies have reported no significant effect of this factor on students’ retention (Carnevale & Rose, 2013; Ishitani, 2003; Cipollone & Cingano, 2007).

Students’ socioeconomic status or that of their parents has also been found to be related to student progress. In particular, it seems that students belonging to disadvantaged or mid- to lower socioeconomic classes are more likely to prolong their studies on the ground that they have to discontinue their studies for some time to seek employment that secures them a living and then to re-enroll for continuing their studies (Dowd & Coury, 2006), but the latter is not always the case. This finding is in line with the conclusions of other researchers (Bynum, 2011a, 2011b), who, on the contrary, have noted that students who receive financial support from their families or from college assistance programs are more likely to stay in college to complete their degree. Indeed, the financial situation or, rather, the lack of financial funding that students can have from the institution or from their families forms what Bean (1980) calls “a student’s intent to leave college”.

Students’ educational background and academic preparation, such as final school mark and university access score, which are seen as pre-college features, are also generally proposed as determinants of dropout or attrition rates. The assumption is that the higher the performance students have at school or, in other words, the better the cognitive background they acquire, the more receptive they are to recruiting new academic knowledge in the university, leading to better academic progress (Donald, 1999; Carnevale & Rose, 2013; Belloc et al., 2011).

Students’ initial commitments, also considered as pre-college characteristics, are mentioned as affecting degree completion as well. According to Tinto (1975, 2006), they can be separated into goal and institutional commitments, representing, correspondingly, the degree to which a student is committed or motivated to get a university degree in general or to graduate from a specific university department. Initial commitments are involved in the study of dropout behavior under the assumption that students enrolled in the university with high-level initial goals are expected to persist in their degree completion.

All the above considerations allow the formulation of the following compound hypothesis.

Hypothesis H1: 

Students’ demographic and pre-college characteristics may have various direct or indirect effects on time to degree. In particular, it is expected that women students, students entering the university at a later age than the usual, those with a high academic background as well as high goals and institutional commitments, and those whose hometown is close to the university’s location tend to graduate faster than their fellow students with different characteristics. On the contrary, it is expected that students with a low family socioeconomic status tend to extend their time to degree.

1.1.2. Students’ Academic and Social Integration

Student integration reflects how students interact with both of the systems of the university environment, the academic (i.e., academic integration) and the social system (i.e., social integration). In other words, the term integration reflects how students are assimilated by the university environment, on the basis of their interactions within the campus environment, incorporating academic and social experiences into their perceptions and involvement behaviors (Lassibille & Gómez, 2008). Academic integration is defined as a student’s perceived academic involvement, while social integration is defined as the quality of a student’s relationships with both the peer group and the faculty members (Lassibille & Gómez, 2008; Donald, 1999). Academic integration consists of high-level instructional assimilation in the classroom (e.g., class participation), while social integration concerns informal peer group associations, extracurricular activities, and interaction with faculty members and administrators (Lassibille & Gómez, 2008; Donald, 1999). According to Bean and Metzner (1985) and other following researchers (Jones, 2010; Meeuwisse et al., 2010), these forms of integration are a central element for reducing the probability of dropout. However, Tinto (1975, 2006) regards social integration as an intermediate outcome variable, leading to greater academic integration and, therefore, to student retention or attrition reduction.

In the spirit of Tinto’s perceptions, the following hypothesis is considered.

Hypothesis H2: 

Students’ academic integration, by interacting with social integration (H2a), affects negatively time to degree, on the grounds that it contributes to decreasing it (H2b).

1.1.3. Academic Performance

There is ample evidence that students’ academic performance, often measured by the mean score of semester grades (an indicator of academic achievement), influences attrition (Bean & Metzner, 1985) or dropout rates (Lassibille & Gómez, 2008), in the sense that good performance motivates students to persist in their degree completion. According to (DesJardins et al., 1999, 2006), high semester grades lower the risk of dropout. In addition, it has been found that academic performance in the first semester of studies is especially important and predicts the chances of graduation and enrollment intensity in later semesters (Attewell et al., 2012). Moreover, it is reported that, relative to other factors, academic performance has a large direct effect on graduation and time to degree (Allen & Robbins, 2010; Yue & Fu, 2017).

Consequently, it is expected that the following hypothesis will be confirmed by the analysis.

Hypothesis H3: 

Academic performance is negatively related to time to degree, in the sense that higher performance contributes to reducing time to degree.

1.1.4. Institutional Image

Since the early theoretical models of the 1970s (Pascarella & Terenzini, 1977) and especially those of Spady (1970) and Tinto (1975), a more rigorous interdisciplinary approach has arisen which involves students’ academic and social integration as well as students’ motives and expectations regarding studies as predictors or determinants of degree completion. However, Bean (1980) and Bean and Metzner (1985), while developing their theoretical model of dropout behavior, suggested that researchers should consider the individuals’ expectations, motivational attributes, and satisfaction with studies as predictors, beyond background characteristics. Today, it is widely recognized that when students commence higher education, they bring with them not only prior knowledge and academic achievements but also an accumulation of motives, expectations and satisfaction which will lead them to shape a perception of the image of the university, a perceived image of the institution, that can be used as a predictive factor in degree completion (Belanger et al., 2002).

In terms of motives, prior research indicates that students who are intrinsically motivated, i.e., they are interested in their studies, tend to acquire a positive institutional image and achieve, consequently, their personal goals, and they actively engage in learning with the intention of attaining understanding and intellectual development (Donald, 1999). Similarly, expectations that students acquire during their studies regarding purchasing knowledge and professional rehabilitation, as well as confidence in their abilities, are positively associated with higher academic performance and lower attrition rates (Nakajima et al., 2012; Beer & Lawson, 2017). Finally, student satisfaction is often linked to re-enrolment behavior, in the sense that students who report satisfaction with the university’s services and programs are more likely to persist and graduate faster. Therefore, the following hypothesis is stated.

Hypothesis H4: 

The perceived image of the institution will negatively influence time to degree, decreasing the duration of studies.

1.1.5. External Factors

External factors usually refer to incidents that occur during a student’s life, but their cause lies outside of the university environment, for example work to cover living expenses, family issues, and physical or emotional challenges like one’s own or family member illness. These factors have been identified as predictors of student dropout rates (Beer & Lawson, 2017). In this context, students are often forced to seek employment to meet tuition fees, which affects their graduation time. Some authors highlight that being forced to pay for college expenses is the number one factor that leads college students to drop out (Arulampalam et al., 2004; Nakajima et al., 2012). Thus, some authors found that a number of students discontinue their studies for some time to seek employment, and then they re-enroll to continue their studies (Nakajima et al., 2012). However, this is not always the case, since some other researchers have found that students do not always return to finish their studies (Arulampalam et al., 2004). On the other hand, it is reported that students who receive financial support are more likely to stay in college to complete their degree (Bynum, 2011b).

Marital status is also a factor reinforcing student attrition, as students who get married while being in college have extra family responsibilities and thus are more likely to withdraw. This argument is supported by several studies (Stratton et al., 2007; Ge, 2011), while some others state that women are those who are most concerned about family responsibilities, which affects their decision to drop out of college more often than men (Astin, 1999; Bean & Metzner, 1985).

Finally, there is evidence that the occurrence of unexpected and rather sad events during the course of study is a reason for students to abandon their studies. In particular, this evidence shows that a significant number of students withdraw from college or university for family issues, such as death or illness of a close family member, or because the students themselves have been influenced by severe illness or a family fatality (Stratton et al., 2007; Ge, 2011).

Based on the above knowledge, the following hypothesis will be investigated.

Hypothesis H5: 

External factors are positively related to student attrition in the sense that their occurrence during studies contributes to delaying time to degree.

1.2. The Conceptual Framework

Considering all of the above, in this study, it is assumed that the distribution of time to degree of social science students is shaped or may be interpreted by observed and latent factors as indicated in the conceptual framework shown in Figure 1. Obviously, factors that either predate the entrance of students at university or are created during the studies as a result of students’ interaction with the university environment have been assumed.

The mechanism by which these factors impact students’ time to degree, i.e., direct and indirect relations, also indicated in Figure 2, is mainly inspired by Tinto’s (Tinto, 1975) and Bean’s (Bean, 1980) theoretical models, as well as by the techniques of Cabrera’s (Cabrera et al., 1993) model.

Based on Tinto’s (Tinto, 1975) model about student attrition, the university environment constitutes an autonomous social system with its own values and structures which, however, consists of two subsystems, the academic and social system, which are in constant interaction. This study has adopted this basic assumption.

According to Tinto (Tinto, 1975), students must integrate into both systems in order to persist and successfully complete their studies. On the contrary, lack of integration leads students to depart and drop out from their studies. Academic integration, a latent construct, reflects students’ academic performance and intellectual development, while social integration (also a latent issue) reflects students’ interaction with college society (peers and faculty). In this study, Tinto’s assumption is extended to the case of time to degree, namely that integration into the university’s social and academic systems lead to graduation at a time closer to the minimum required (i.e., the time threshold for graduation), while lack of integration leads to an extended duration of studies and to delayed graduation.

From Bean’s (Bean, 1980) model of student retention, the following assumption is made. A lack of satisfaction with studies, a lack or low level of financial funding during studies, a lack of emotional support and encouragement from the family or friends’ environment, and various external factors are causes generating phenomena related to students’ attrition. In this study, this assumption is adjusted as follows: satisfaction with studies, external encouragement, and the absence of exogenous factors during studies lead to accelerating studies, while the opposite cases contribute to extending the duration of studies.

Finally, the idea of combining Tinto’s and Bean’ assumptions into an integrated model for approaching the factors related to time to degree was adopted for this study, following Cabrera (Cabrera et al., 1993).

Summarizing the rationale behind the proposed conceptual framework regarding the factors that affect the time to degree, the following are noted. Students enter university carrying a number of academic and personal characteristics, which constitute a basic background, to proceed with their studies towards the final goal of getting their degree. Time to degree is affected by these features in the ways described in H1.

However, from the beginning and during the course of study, students come into contact with a new environment and encounter obligations and habits that, perhaps, are different from those they faced in their lives before. Therefore, students will have to compromise or understand the rules of operation of the new system they have chosen in order to achieve their goal of obtaining their degree. According to theory, all these constitute factors that coexist in the university environment and are directly linked to the achievement of the ultimate goal. These factors correspond to academic and social integration and to academic performance. The way in which these factors affect time to degree is described in H2 and H3, respectively. Beyond that, students obtain knowledge and experience about the various operations and activities of the university, while also making comparisons with other institutions, and develop feelings and attitudes towards their institution (perceived institutional image), which is directly related to the final goal, as indicated under H4. Finally, students, as human beings, during their studies, may cope with unexpectedly pleasant or unpleasant events that, despite not being related to the university environment, are directly related to the final goal and are almost always linked to the removal of its attainment (H5).

2. Materials and Methods

2.1. Sample and Variables

The data come from a sample of 1137 graduates from the Panteion University, which was selected from a population of 15,629 graduates enrolled in various departments of the university during the academic years 1983–1984 to 1999–2000 and they had at least two years more than the minimum required to graduate. Thus, the follow-up period is defined from the academic year 1983–1984, when the Panteion University started to operate as an university, until the academic year 2022–2023. Therefore, each of the participants had at least 22 years to graduate. After this period, if one of the students in the sample did not manage to graduate, he or she was considered a perpetual student. The sample was selected using finite population sampling techniques and it is a proportionally stratified sample based on gender and academic department. The collected information relates to the following variables.

The dependent variable: The ultimate outcome variable in this study is time to degree, denoted by $W$; that is, the time between the date of first enrolment in the university and the graduation date. Specifically, it was measured in months, according to the technique adopted by Kalamatianou and McClean (2003, pp. 314–315). Some information from this work facilitates an understanding of the determination of the values of this variable. Graduation from the Greek university institutions providing four-year curricula is possible three times a year just after the June, October, and February examination periods. As graduation takes place on the last day of the corresponding month and new student enrollments are possible during September of each academic year, for a new student, the first possible opportunity for graduation is after four years in June (following the first academic year). As a result, the minimum time for graduation equals 46 months. This constitutes a threshold for graduation, and students who also satisfy conditions concerning the successful fulfilment of a certain number of modules can graduate at exactly 46 months after the date of their first enrollment. However, students who fail to do that can proceed to the next examination period for an unlimited number of times until the conditions for the modules have been satisfied. As a consequence, the possible values of the outcome variable $W$ are 46, 46 + 4 i where i = 1, 2, 3, …. indicates the number of examination periods that will be needed until graduation beyond the earliest opportunity. In this way, the observed values of the variable $W$ in the data are 46, 50, 54, …, 278 months. Thus, $W$ is a quantitative discrete variable.

The independent variables: According to the conceptual model in Figure 2, two groups of exogenous variables, students’ pre-college characteristics, and factors formulated during studies are assumed to affect time to degree. Regarding the first group, 10 observed variables, denoted by $X_g,g=1,2,\dots ,10$, were considered. In particular, X₁, X₂, and X₃ correspond to students’ personal characteristics (i.e., gender, students’ age at the time of first enrolment, hometown location); X₄ and X₅ to students’ prior academic achievements (i.e., academic aptitude (SAT score) and high school grade point average—GPA); X₆–X₉ to students’ initial commitments (i.e., goal and institutional commitments); and X₁₀ to students’ socioeconomic background (i.e., parental socioeconomic status). For reasons of brevity, the full description of these variables, as well as of those mentioned below, and their values are given in

Table 1. Variables’ definitions and values. Descriptive statistics, factor loadings, and construct reliability (CR) (n = 1137).

Observed/Latent Variable	Values and Coding	Percentage/Average	Factor Loadings	CR
Pre-college characteristics
Students’ personal characteristics
X1: Students’ age when enrolled at university	Range 16.8–45.6	19.8 years
X2: Gender	1 male; 0, female	45.6% 54.4%
X3: Hometown location	1, Athens; 0, other	75.0% 24.5%
Students’ prior academic achievements
X4: Academic aptitude, SAT score (average score achieved at national exams for entering the university)	Scale [0–100]	67.27
X5: High school grade point average (GPA)	Scale [10–20]	17.1
Students’ initial commitments—goal and institutional commitment
X6: Goal commitment1	**	4.2
X7: Criteria of academic department of choice	1, personal choice; 0, other	48.9% 51.1%
X8: Pre-enrollment ranking of academic department of choice	1, scale 1–10; 0, scale 11-	60.0% 40.0%
X9: Desire to graduate from the particular department	1, yes; 0, no	90.8% 9.0%
Students’ socioeconomic status
X10: Parental socioeconomic status (SES)	1, low; 2, high	60.5% 39.5%
Factors formulated during studies
Academic performance
X11: Academic performance (average score of the first year of studies)	Scale [0–10]	6.2
${\mathsf{\eta }}_1:$Students’ academic integration				0.989
Y1: Class participation (to what extent did you participate in conversation with the whole class or exchanges between professors and students?)	**		0.359 *
Y2: Class attendance (in a normal week, how regularly did you attend classes?)	**		0.631 *
Y3: Time spent studying (to what extent did you study throughout the semester?)	**		0.447 *
${\mathsf{\eta }}_2:$Students’ social integration				0.984
Y4: Housing while attending college (for how much of your studies did you live at the university campus?)	**		−0.010
Y5: Interaction with peers (to what extent did you pursue hanging out with classmates?)	**		0.669 *
Y6: Participation in student elections (to what extent did you participate in student elections?)	**		0.343 *
Y7: Engagement in college political parties (to what extent did you engage in student political parties?)	**		0.157 *
Y8: Participating in extracurricular activities (to what extent did you participate to cultural and other university events?)	**		0.433 *
${\mathsf{\eta }}_3:$Institutional image				0.982
Y9: Prestige of the faculty (to what extent did you expect to gain prestige derived from the specific curriculum?)	**		0.326 *
Y10: Employability after graduation (to what extent did you expect to gain vocational rehabilitation and the skills required by the labor market from the specific curriculum?)	**		0.205
Y11: Usefulness of gained knowledge (to what extent did you expect to gain knowledge acquisition on the specific science from the specific curriculum?)	**		0.321 *
Y12: Satisfaction with the teaching faculty members	**		0.313 *
Y13: Parental influence/interest (to what extent were your parents interested in the progress of your studies?)	**		0.300 *
Y14: Satisfaction with class environment (e.g., class conditions like temperature, seating facility, noise conditions)	**		0.795 *
Y15: Satisfaction with curriculum	**		0.721 *
${\mathsf{\eta }}_4$: External factors				0.992
Y16: Work during studies	1, yes; 0, no		0.343 *
Y17: Work during studies on grounds of subsistence	1, yes; 0, no		0.233 *
Y18: Marriage during studies	1, yes; 0, no		0.644 *
Y19: Family responsibilities	1, yes; 0, no		0.423 *
Y20: Unforeseen events (sad events that took place during studies e.g., personal illness, illness or loss of a relative, loss of a job, etc.)	1, yes; 0, no		0.417 *

*** p < 0.1; ** p < 0.05; * p < 0.01.

together with the results of the descriptive and factor analysis. Regarding the second group, one observed variable, denoted by X₁₁, that represents academic performance (average score of the first year of studies) and four latent variables denoted by ${\eta }_i,$ (i = 1, 2, 3, 4), reflecting, respectively, students’ academic integration, social integration, institutional image, and external factors, were considered. For the development of these four latent constructs, 20 more indicator variables were measured. These variables are denoted as ${\mathrm{Y}}_k$, $k=$1, …, 20, and they are also fully described along with the corresponding latent variables they measure in Table 1. In particular Y₁, Y₂, and Y₃ correspond to students’ academic integration (i.e., class participation, class attendance, time spent studying); Y₄–Y₈ to students’ social integration (i.e., housing while attending college, interaction with peers, participation in student elections, engagement in college political parties, and participating in extracurricular activities); Y₉–Y₁₅ to institutional image (prestige of the faculty, employability after graduation, usefulness of gained knowledge, satisfaction with the teaching faculty members, parental influence/interest, satisfaction with class environment, satisfaction with curriculum); and, finally, Y₁₆–Y₂₀ correspond to external factors (i.e., work during studies, work during studies on grounds of subsistence, marriage during studies, family responsibilities, and unforeseen events that took place during studies).

2.2. The Statistical Methodology

The validity of the conceptual framework described above (Figure 1), in the Greek reality, is tested by means of a SEM that is precisely specified below on the basis of the collected data. What is commonly acknowledged as structural equation models or modeling (SEM) is a powerful procedure where different types of statistical methods are used to quantify how well a complex and sophisticated theoretical model or proposition of interest fits into reality. Theoretical models may hypothesize relationships among the observed variables, but also those sets of variables define constructs or latent factors which may also relate to each other. SEM indeed allows the analysis of the relationships between observed and unobserved or latent factors by combining key elements of regression, path, and factor analysis methods. This is often the case in sociological, psychological, and educational research, where SEM has shown an increasing popularity (Jöreskog & Moustaki, 2001).

The most common SEM is the well-known LISREL model (Linear Structural Equation Models), first introduced by Jöreskog and Moustaki (2001). It consists of two parts known as the structural part (or model) and the measurement part. In matrix form, the former is specified by Equation (1) and the latter by Equations (2) and (3)

$$\mathsf{\eta }=\mathsf{\alpha }+\mathsf{{\rm B}}\mathsf{\eta }+\mathsf{\Gamma }\mathsf{\xi }+\mathsf{\zeta }$$

(1)

$$\mathbf{x}={\mathsf{\alpha }}_{\mathrm{x}}+{\mathsf{\Lambda }}_{\mathrm{x}}\mathsf{\xi }+\mathsf{\delta }$$

(2)

$$\mathbf{y}={\mathsf{\alpha }}_{\mathrm{y}}{+\mathsf{\Lambda }}_{\mathbf{y}}\mathsf{\eta }+\mathsf{\epsilon }$$

(3)

where in (1), $\mathsf{\eta }$ and $\mathsf{\xi }$ are, respectively, the vectors of the endogenous (or dependent) and exogenous (or independent) latent variables; $\mathbf{B}$ is a matrix of the coefficients of the effects of endogenous on endogenous variables ($\eta$’s); $\mathsf{\Gamma }$ is a matrix of the coefficients of the effects of exogenous variables$(\xi$’s) on endogenous variables ($\eta$’s); $\mathsf{\zeta }$ is a vector of the residuals or errors; and $\mathsf{\alpha }$ is a vector of the intercept terms. In simple terms, the structural part of the LISREL model is a set of linear regression-type equations representing the relations of each dependent latent variable with the rest of the dependent latent variables ($\mathsf{\eta }$) and with a set of independent latent variables ($\mathsf{\xi })$.

Muthén (1984) first modified the measurement part of the LISREL model, presented above in Equations (2) and (3), by stacking vectors $\mathbf{x}$ and $\mathbf{y}$ into a single vector, denoted by ${\mathbf{y}}_j$, and the vectors $\mathsf{\eta }$ and $\mathsf{\xi }$ into a single vector, denoted by ${\mathsf{\eta }}_j$. In the same way, the error vectors, $\mathsf{\delta }$ and $\mathsf{\epsilon }$, were composed into vector ${\mathsf{\epsilon }}_j$, and the intercept vectors ${\mathsf{\alpha }}_{\mathrm{x}}$ and ${\mathsf{\alpha }}_y$ into vector ${\mathsf{\alpha }}_j$.

Thus, the structural and the measurement parts of Muthén’s modified SEM are specified by Equations (4) and (5), respectively

$${\mathsf{\eta }}_j={\mathsf{\alpha }}_j+\mathbf{B}{\mathsf{\eta }}_j+{\mathsf{\zeta }}_j$$

(4)

$$\mathbf{x}{\mathbf{y}}_j={\mathsf{\alpha }}_j+\mathsf{\Lambda }{\mathsf{\eta }}_j+{\mathsf{\epsilon }}_j$$

(5)

where, in (4), ${\mathsf{\eta }}_j$ is a vector of the latent variables (regardless of if they are endogenous or exogenous), $\mathsf{{\rm B}}$ is a matrix of the coefficients of the effects of the latent variables on latent variables (${\eta }_j$’s), ${\mathsf{\zeta }}_j$ is the vector of the residuals, and ${\mathsf{\alpha }}_j$ stands for the vector of the intercept terms. In simple terms, the structural part of Muthén’s modified SEM (Muthén, 1984) is a set of linear regression-type equations representing the relations between each latent variable and the rest of the latent variables (${\eta }_j$’s).

Next, Muthén (1984) extended the structural part described in Equation (4) by including an extra term, ${\mathsf{\Gamma }\mathbf{x}}_1$, for regressions of the latent variables (${\mathsf{\eta }}_j$) on a set of observed (continuous, ordinal, or dichotomous) covariates, presented in vector form as ${\mathbf{x}}_1$.Then Equation (4) is written as

$${\mathsf{\eta }}_j={\mathsf{\alpha }}_j+\mathbf{B}{\mathsf{\eta }}_j+{\mathsf{\Gamma }\mathbf{x}}_1+{\mathsf{\zeta }}_j$$

(6)

where $\mathsf{\Gamma }$ is a matrix of the coefficients of the observed covariates on the latent variables (${\eta }_j$’s). In simple terms, the structural part described in Equation (6) is a set of linear regression-type equations representing the relations of each dependent latent variable with the rest of the dependent latent variables and with a set of observed covariates (${\mathbf{x}}_1)$.

Later, Muthén and Muthén (Muthén & Muthén, 1998) extended the measurement part described in Equation (5) by adding the term $\mathsf{{\rm K}}{\mathbf{x}}_2$ for regressions of the observed responses on the observed covariates, as follows

$${\mathbf{y}}_j={\mathsf{\alpha }}_j+\mathsf{\Lambda }{\mathsf{\eta }}_j+\mathsf{{\rm K}}{\mathbf{x}}_2+{\mathsf{\epsilon }}_j$$

(7)

where ${\mathbf{x}}_2$ is a vector of the observed covariates (sometimes vectors ${\mathbf{x}}_2$ and ${\mathbf{x}}_1$ are the same), and $\mathsf{{\rm K}}$ is a matrix of the coefficients of the observed covariates on the indicator variables ($y_j$’s). Note that, usually, only a few of the elements of $\mathsf{{\rm K}}$ have non-zero elements; a zero element states that an indicator variable is not directly influenced by one or more $x$covariate. When $\mathsf{{\rm K}}$ has only zero elements, the measurement Equation (7) reduces to (5) (Muthén & Muthén, 1998).

The Proposed Baseline SEM for Time to Degree and Related Factors

It is considered that the SEM counterpart of our conceptual framework (Figure 1), from this point onwards, called the baseline SEM for time to degree and related factors, is a Muthen-type SEM, as it is described by (6) and (7). On the basis of the above-defined variables and notation (Table 1), this baseline SEM is particularly specified for the application in this study as follows:

$$\mathrm{The} \mathrm{structural} \mathrm{part}: W={\tilde{\mathsf{\alpha }}}_j+\tilde{\mathbf{B}}{\tilde{\mathsf{\eta }}}_j+\tilde{\mathsf{\Gamma }}\mathbf{x}+{\tilde{\mathsf{\zeta }}}_j$$

(8)

$$\mathrm{The} \mathrm{measurement} \mathrm{part}: {\tilde{\mathbf{Y}}}_j={\tilde{\mathsf{\nu }}}_j+\tilde{\mathsf{\Lambda }}{\tilde{\mathsf{\eta }}}_j+\tilde{\mathbf{K}}\mathbf{x}+{\tilde{\mathsf{\epsilon }}}_j$$

(9)

The structural part (1) depicts the relation of the dependent variable (time to degree) $W$ with the four latent variables and the eleven observed covariates; thus ${\tilde{\mathsf{\eta }}}_j=({\eta }_1, {\eta }_2, {\eta }_3$,${{ \eta }_4)}^{\prime }$ and $\mathbf{x}={(x_1,\dots x_{11})}^{\prime }$ represent, respectively, the vectors of the latent variables and observed covariates described in Table 1. $\tilde{\mathbf{B}}$ is a $1\times 4$ matrix of the regression coefficients of the effects of the four latent variables (${\tilde{\mathsf{\eta }}}_j$) on the dependnt variable $W.$ $\tilde{\mathsf{\Gamma }}$ is a $1\times 11$ matrix of regression coefficients of the effects of the 11 independent covariates ($\mathbf{x}$) on $W$, and, finally, ${\tilde{\mathsf{\alpha }}}_j$, and ${\tilde{\mathsf{\zeta }}}_j$ represent the intercept and error terms, respectively. In the current application, it is considered that ${\tilde{\mathsf{\alpha }}}_j=0$.

The measurement part (9) illustrates the relations of the 20 indicator variables (${\tilde{\mathbf{Y}}}_j)$ with the 4 latent variables and with the 11 observed covariates; hence ${\tilde{\mathbf{Y}}}_j$ is a $20\times 1$ column vector whose elements correspond to the indicator variables ${\mathrm{Y}}_1,{\mathrm{Y}}_2,\dots ,Y_{20}$, used to measure the 4 latent variables, ${\tilde{\mathsf{\eta }}}_j=({\eta }_1, {\eta }_2, {\eta }_3$,${{ \eta }_4)}^{\prime }$, and $\mathbf{x}={(x_1,\dots x_{11})}^{\prime }$ denotes the vector of the observed covariates. $\tilde{\mathsf{\Lambda }}$ is a $20\times 4$ matrix of factor loadings of ${\tilde{\mathbf{Y}}}_j$ on the four latent variables (${\tilde{\mathsf{\eta }}}_j$), $\widetilde{\mathsf{{\rm K}}}$ is the $20\times 11$ matrix of regression coefficients of the observed covariates on the indicators, and, finally, ${\tilde{\mathsf{\epsilon }}}_j$ and ${\tilde{\mathsf{\nu }}}_j$ are both $1\times 20$ vectors of the errors and intercepts, respectively. It should be noted that the vectors ${\mathbf{x}}_1$ and ${\mathbf{x}}_2$ of Muthén and Muthén (1998) in this application are the same, denoted by $\mathbf{x}$.

However, the structural part of the above baseline SEM can be used as a predictor model for time to degree. The analytic form of this model is written as follows

$$W=\sum _{i=1}^4{\beta }_i{\eta }_i+\sum _{i=1}^{11}{\gamma }_i{\mathrm{X}}_i$$

(10)

where ${\beta }_i$’s and ${\gamma }_i$’s denote the elements of $\tilde{\mathbf{B}}$ and $\tilde{\mathsf{\Gamma }}$, respectively.

Graphically, the baseline SEM for time to degree and related factors is summarized in Figure 2. Its validity for the data is tested in the following section.

3. Results

3.1. Results of the Descriptive and Factor Analysis

Besides the description of the variables used in the analysis, Table 1 provides the sample statistics as well as the results of the factor analysis (standardized factor loadings, sfs, and fit statistics) for all latent variables of this study. The sample of 1137 graduates is composed of 45.6% male and 54.5% female students, and their average age at the time of enrollment at the university was 19.8 years (STD = 3.24, range 16.8–45.6). The large majority of the sample members (75.0%) reported Athens (the capital of Greece) as their hometown, and 60.5% of the total stated that they have a middle or somewhat high parental socioeconomic status (SES).The sampled students appear to have been admitted to the university with rather good academic performance and prospects: the high school (Lyceum) grade point average (GPA) equals 17.1 on in the scale [10–20], the average scholastic aptitude test score (SAT) achieved in national exams for entering the university equals 67.27 on the scale [0–100], the average level of goal commitment is 4.2, and 90.8% of the sample stated their desire to graduate from the particular department of study. In addition, a little less than half of the graduates (48.9%) stated that the particular field of study was a personal choice, and for the 60% of the sample, the corresponding department of study was among the first 10 preferences in the pre-enrollment ranking list. Finally, the academic performance of the sample during their first year of study is rather modest, estimated as 6.2 on the scale [0–10].

The assessment of the four latent variables was made on the frame of the model (9) via confirmatory factor analysis on the basis of the students’ responses to the indicator variables Y₁, …, Y₂₀, (Table 1), using LISREL 8 software. Τhe results are presented in Table 1 and

Table 2. Fit measures for the structural model.

	$\mathbf{Chi} \mathbf{Square} \mathbf{Test} \left({\mathsf{\chi }}^2\right)$	Degrees of Freedom (df)	$\frac{{\mathsf{\chi }}^2}{\mathbf{d}\mathbf{f}}$	Tucker–Lewis Index (TLI)	Comparative Fixed Index (CFI)	Mean Squared Error of Approximation (RMSEA)
Measurement model	671.074	230	2.92	0.944	0.922	0.052
Final SEM	1044.262	373	2.79	0.919	0.960	0.068
Acceptable range			<3	>0.900	>0.900	<0.080

. All four latent variables achieved a good fit to the data (see Table 2) with construct reliability (CR) being within an acceptable range (CR ≥ 0.982, Table 1), and most of the considered indicator variables were found to be significant (p < 0.0001). In particular, it appears that students’ academic integration (${\mathit{\eta }}_1)$ is mostly defined by class attendance (Y₂, sfs = 0.631) and time spent studying (Y₃, sfs = 0.447), and less by class participation (Y₁, sfs = 0.359). Students’ social integration (${\mathit{\eta }}_2$) is mainly defined by interaction with peers (Y₅, sfs = 0.669), participation in extracurricular activities (Y₈, sfs = 0.433), and participation in student elections (Y₆, sfs = 0.343) and less by engagement in university political parties (Y₇, sfs = 0.157) and housing while attending college (Y₄, sfs = 0.010). Institutional image (${\mathit{\eta }}_3$) is largely defined by satisfaction with the class environment (Y₁₄, sfs = 0.795) and satisfaction with the curriculum (Y₁₅, sfs = 0.721), and less by the prestige of the faculty (Y₉, sfs = 0.326), usefulness of gained knowledge (Y₁₁, sfs = 0.321), satisfaction with the teaching faculty members (Y₁₂, sfs = 0.313), and parental influence (Y₁₃, sfs = 0.300), while employability after graduation (Y₁₀, sfs = 0.205) has the lowest effect. Finally, the fourth latent variable, external factors (${\mathit{\eta }}_4$), is defined mostly by the incident of marriage during studies (Y₁₈, sfs = 0.644), family responsibilities (Y₁₉, sfs = 0.423), and unforeseen events that took place during studies (Y₂₀, sfs = 0.417). Work during studies as well as work to cover living expenses have the lowest effect (Y₁₆, sfs = 0.343; Y₁₇, sfs = 0.233).

Construct reliability exceeds 0.6 (see Table 1), suggesting adequate reliability for each latent variable. Factor loading estimates are highly significant in all cases except for the one corresponding to the item Y₄ (housing while attending college), which is linked to latent variable ${\mathit{\eta }}_2$(students’ social integration) and was removed from the analysis.

3.2. SEM Results: Final SEM of Pathways to Time to Degree

The applicability of the structural part of the baseline model, Equation (1), to the data was made using AMOS software version 26. Initially, the significance of the parameters corresponding to the 11 covariates (independent observed variables), X₁–X_11, was checked (partial goodness of fit test). Three of them, namely hometown location (X₃, p = 0.685), goal commitment (X₆, p = 0.555), and desire to graduate from the particular department (X₉, p = 0.501), were not found to be significant and were removed from the analysis, which was repeated for the rest of the variables. The final model is presented in Figure 3. Results concerning the goodness of fit indicators are given in Table 2 and suggest the adequate fit of the final model to the data.

Table 3. Direct, indirect, and total standardized effects for the final structural equation model for the study of factors associated with student attrition.

Variable		Variable	Standardized Beta Coefficients (sbc)
			Total Effect	Direct Effect	Indirect Effect
Students’ pre-college characteristics
W: Time to degree	←	X1: Students’ age when enrolled at university	−0.07 (−1.33) *	−0.07 (0.56) *
		X2: Gender	0.02 (1.65) **		0.02 (1.16) *
		X4: Academic aptitude (SAT score)	−0.30 (−55.29) **	−0.17 (−31.72) **	−0.13 (−23.57) **
		X5: High school grade point average (GPA)	−0.10 (−3.91) *		−0.10 (−3.91) *
		X7: Criteria of academic department of choice	−0.06 (−6.29) **	−0.06 (−6.29) **
		X8: Pre-enrollment ranking of academic department of choice	0.16 (18.26) *	0.16 (18.26) *
Y17: Work during studies on grounds of subsistence	←	X10: Parental socioeconomic status (SES)	−0.17 (−0.12) *	−0.17 (−0.12) *
X8: Pre-enrollment ranking of academic department choice	←	X2: Gender	0.90 (0.09) *	0.90 (0.09) *
Factors formulated during studies—key factors for time to degree
W: Time to degree	←	${\mathit{\eta }}_1:$ Academic integration	−0.27 (−78.28) *	−0.27 (−78.28) *
		X11: Academic performance	−0.50 (−20.28) **	−0.50 (−20.28) **
		${\mathit{\eta }}_3:$ Institutional image	−0.10 (−30.72) **	−0.10 (−30.72) **
		${\mathit{\eta }}_4:$ External factors	0.265 (116.30) *	0.265 (116.30) *
${\mathit{\eta }}_1:$ Academic integration	←→	${\mathit{\eta }}_2$: Social integration	0.77 (0.33) *	0.77 (0.33) *
X11: Academic performance	←	X4: Academic aptitude, SAT score	0.25 (1.16) **	0.25 (1.16) **
		X5: High school grade point average (GPA)	0.19 (0.21) *	0.19 (0.21) *

Notes. Unstandardized coefficients are in parentheses, *** p < 0.1; ** p < 0.05; * p < 0.01.

displays the standardized and unstandardized total, direct, and indirect path coefficients and the p-values. All structural paths and correlations are significant at either the 0.05 or 0.001 level, providing evidence supporting most of the hypotheses made above towards the conclusion that certain pre-college features of graduates and factors developed during studies shape the distribution of time to degree. In particular, it is shown that factors formulated during studies, which are considered as key factors for time to degree in the conceptual framework above, have the highest impact, while the pre-college characteristics follow.

4. Discussion

Attempting a description of the results, in order of the significance of the factors, the following can be noted. Academic performance (average score of the first year of studies), the only observed independent variable among factors formulated during studies, proved to have the highest impact on time to degree (sbc = −0.50, p < 0.05). As expected by Hypothesis H3, academic performance has a negative effect, meaning that the higher the academic performance, the shorter the graduation time. This conclusion was also reached by Calcagno et al. (2007), DesJardins et al. (2006), regarding students’ persistence. In addition, as will be shown below, it turns out that academic performance also plays a role as a mediator between some of the pre-college characteristics of the graduates and time to degree.

Academic integration, interacting with social integration (sbs = 0.77, p < 0.001), engages the second highest direct negative effect on time to degree (sbc = −0.27, p < 0.001). Thus, Hypotheses H2a and H2b are supported, establishing that students who are socially and academically integrated into the university environment and interact with it tend to graduate faster. These results are in line with the theoretical assumptions of Tinto (1975) and Bean (1980), according to which, students who have an extensive and high-quality interaction with the institutional social and academic system are more likely to continue their enrollment at the university. They are also complying with the results of Pascarella and Chapman and of Cabrera, wherein the validity of Tinto’s model was investigated for the case of students’ withdrawal and persistence. Finally, these results are in agreement with the results of other authors (Berger & Braxton, 1998; Veenstra, 2008; Jones et al., 2012) who examined students’ retention process and their intent to re-enroll at the local university next semester, as well as with results of Astin (1999), who studied persistence related to student involvement.

As is claimed under Hypothesis H4, institutional image proved to have a direct negative effect on time to degree (sbc = −0.10, p < 0.05). This could mean that a positive perceived image of the institution, as it is formed on the basis of students’ motives, expectations, and satisfaction with studies, contributes to a shorter duration of studies. This result is consistent with those derived on the grounds of similar hypotheses included in early models regarding students’ attrition and college withdrawal (Spady, 1970; Tinto, 1975; Pascarella & Terenzini, 1977), as well as in more recent approaches (Kocsis & Molnár, 2024), where it is specifically discussed that when students commence higher education, they bring not only their prior knowledge and prior academic achievements but also an accumulation of motives, expectations, and satisfaction, which will form a perceived image of the institution that is associated with lower attrition rates.

Of the external factors, ($\eta 4$) is the only key factor that proved to have a direct positive effect on time to degree (sbc = 0.265, p < 0.001), consistent with Hypothesis H5. It is clear that the occurrence of unexpected sad events (e.g., illness) or other circumstances (e.g., need for work, family responsibilities) that are not related to the university environment tend to prolong the duration of studies.

This is a result that has also been shown in the early models for predicting retention (Spady, 1970; Tinto, 1975; Pascarella & Terenzini, 1977; Bean & Metzner, 1985), where the important role of external factors in student attrition was emphasized. However, similar conclusions come also from more recent studies, where the role of external factors, such as financial issues, employment, starting a family, and physical/emotional and family challenges, like personal or family illness, has been identified as a predictor of student dropout rates (Nakajima et al., 2012; Ge, 2011).

The results concerning the effect of pre-college characteristics on time to degree support the various individual aspects of Hypothesis H1 and reveal various direct and indirect relationships (see Table 3, Figure 3). Students’ age at the time of entering the university appears to have a negative direct effect on time to degree (sbc = −0.07, p < 0.001), meaning that older students tend to graduate faster. This finding coincides with the findings of Xenos et al. (2002) and Whiteman (2004), where student attrition rates in relation to first-year students’ age were discussed.

Consistent with previous results regarding academic success and graduation rates (Smith & Naylor, 2001; Kemp, 2002), no direct significant effect of gender on time to degree was found. However, an indirect effect was revealed through the pre-enrollment ranking of the academic department of choice (X₈, sbc = 0.09, p < 0.05). In this case, the positive value of the coefficient means that female students enrolled at the university may have higher institutional commitment than male students and, thus, they tend to graduate faster. An interpretation of this could be that female students consider higher education as a means of achieving specific goals (mainly vocational and skills acquisition), and this may activate them towards faster graduation than male students. This mediator role of gender has also been shown in the works of Astin (1999), Bean (1980), and Herzog (2005), regarding students’ attrition.

Parental socioeconomic status (SES, X₁₀) was found to be not directly related to time to degree, a conclusion which has also emerged in the works of Astin (1999), Smith and Naylor (2001), and DesJardins et al. (2006) on students’ persistence. This is an expected result for Greek data, given the willingness of the Greek parents, regardless of their socioeconomic background, for their children to receive university-level studies. However, it was found that parental SES has also an indirect effect on time to degree through the indicator variable (i.e., work during studies on grounds of living expenses) (Y_17, sbc = −0.17, p < 0.001). This means virtually that students whose families have a high socioeconomic level have a reduced need to work or are not forced to work during their studies, which is associated with faster graduation. This finding is in line with the findings of Ishitani (2003), Cipollone and Cingano (2007), Longwell-Grice and Longwell-Grice (2007), and Carnevale and Rose (2013) concerning university dropout, graduation rates, or withdrawal and persistence rates.

The analysis further showed that of the students’ prior academic achievements, academic aptitude (SAT score) has a direct effect on time to degree (sbc = −0.17, p < 0.05) as well as an indirect one (sbc = −0.13, p < 0.05), mediated by academic performance, which increases significantly (sbc = 0.25, p < 0.05). However, the influence of high school grade point average on time to degree is only indirect, mediated also by academic performance (sbc = 0.10, p < 0.001). In other words, students enrolled at the university with a high prior academic achievement are expected to have a shorter period of study. These students also achieve better academic performance (i.e., average scores by the end of the first year of studies) and, because of this, they tend to graduate faster. These results, which demonstrated a direct as well as an indirect effect between students’ educational background and academic performance, have been proposed as a determinant of students’ attrition rates (Smith & Naylor, 2001; Jones et al., 2012).

Finally, from the group of students’ institutional commitment variables, it was found that two them, criteria of the academic department of choice (X₇) and pre-enrollment ranking of the academic department of choice, had a direct significant effect on time to degree (sbc = −0.06, p < 0.05 and sbc = 0.16, p < 0.001, respectively). These results are in line with most studies, where it is supported that students who enrolled at university with a high level of institutional commitment are more likely to persist in their degree completion (Oram & Rogers, 2022; Zheng, 2024).

5. Conclusions

The aim of this research was to develop a comprehensive model that examines the various factors and their interactions influencing the unique distribution of time to degree completion in university studies. This distribution is characterized by a lower (minimum) time limit for degree completion but no upper (maximum) limit. The Greek context of higher education exemplifies this condition and motivated the interest for this work. Although the legislation governing Greek universities has been changed, since the beginning of 2000, with the biggest legislative changes occurring in 2014 and 2020, even today, and despite the changes in the legislation on the deletion of perpetual students, no provision of the laws providing for the deletion of perpetual students has been implemented. On the contrary, the new Minister of Education recently has re-legislated, and the law, which will be effective in 2025, once again provides for the deletion of eternal students, which, according to buildings, amount to about 335.000. In fact, the Minister forewarned the universities that refuse to proceed with the deletion of eternal students would be presented with a cut-off of funding. This fact indicates the importance that the stakeholders of the Greek public education system attach to the “de-registration” of students.

The work proceeds by developing a conceptual framework which, on the basis of well-known theories, assumptions, and research results (Tinto, 1975; Bean, 1980; Bean & Metzner, 1985; Cabrera et al., 1993), links various demographic and other pre-college characteristics of students, as well as factors that are formed during studies, with time until graduation. It also indicates possible individual relationships among the variables or, in other words, the underlying mechanism covering the logic of these relationships. Utilizing the Muthén and Muthén (Muthén & Muthén, 1998) SEM, a particularly refined SEM was developed for testing the hypotheses made and the hypothesized framework, in general, on data from a representative sample of 1137 students of the Athenian University of Social and Political Science.

The findings confirmed most of the research hypotheses, evidencing that time to degree is associated with students’ demographic and pre-college characteristics, but it is mostly shaped in accordance with students’ integration into the university environment, their perceived institutional image, and with the occurrence of unexpected external factors; even more so, it depends on the academic performance of the students. The foremost directions that have been confirmed are that higher academic performance, greater academic and social integration, a better positive perceived institutional image, and the absence of external factors make on-time graduation more likely. Some pathways that have been revealed through SEM, regarding the relations of students’ demographic and pre-college characteristics with time to degree, are also worth mentioning. Gender has only an indirect effect on time to degree through its relationship to institutional commitment which, when it is high, has a direct effect on reducing the duration of studies. However, the case is more apparent for female students, who exhibit higher institutional commitment than male students. However, students’ parental socioeconomic status has also an indirect effect on time to degree through the external factor reflecting work during studies to cover living expenses, whose appearance contributes to extended time to degree, if not to minimizing the probability of obtaining a degree. Finally, the importance of academic achievements before admission to university, high values of which contribute to reducing the duration of studies directly and indirectly, since they contribute to academic performance, is evidenced.

Limitation of the Study

However, although the above results add to the knowledge about the factors related to time to degree and delayed graduation, they also have limitations to be taken into account. One limitation has to do with the fact that the data of this study come from a single university that is particularly oriented towards social and political sciences. This may limit internal and external comparisons, although the results, as mentioned above, are generally consistent with those of other surveys. Despite the aforementioned limitations, however, the findings and conclusions of the present study have resonance throughout Greek higher education, as far as the duration of studies is concerned, as there is no upper limit to the duration of studies in all Greek public universities. In this article, we do not study the problems of Greek higher education in general, so there is a differentiation between the various faculties, for example, between theoretical sciences and technical sciences. Instead, we study a common denominator, that is, a common phenomenon that is observed throughout Greek higher education, which is none other than the problem of excessive duration of studies and the factors that create it.

Another limitation concerns the measurement of perceptions, where the students were asked to evaluate their views and feelings retrospectively. The case for measuring the variables on a longitudinal basis was not possible in this study.