Confirmatory Factor Analysis and Propensity to Cheat Scale Validation in the Ethiopian Public Higher Education Institutions

1.1 Background of the study

Academic dishonesty is a broad term that encapsulates various detrimental behaviors within educational environments, such as plagiarism and falsification of information. Although difficult to define precisely (Lambert et al., 2003), academic dishonesty involves a variety of damaging activities, such as plagiarism or making false excuses (Yazici et al., 2011). The concept of academic cheating was initially described by Rich (1984) as deceitful actions related to coursework, including cheating on tests, exams, and assignments, along with activities like obtaining exams illegally, plagiarizing content, manipulating data, and tampering with library resources.

According to Davis et al. (2009), academic cheating involves deceiving or defrauding another’s through misleading or dishonest actions. They emphasize that it entails students engaging in behaviors that trick instructors into believing that the academic work submitted is their original creation. Academic cheating can be defined as “intentionally unethical behaviour” (Von Dran et al., 2001, pp. 40) and “using deception (fraud) in academic work” (Cochran, 2016, pp. 814), both of which result in a breach of the defined rules and accepted standards, granting cheaters an unfair benefit over those who do not cheat (Dick et al., 2003). ensuring quality outcomes. In order to guarantee quality outcomes, student assessment should actually be seen as a complex and multifaceted activity that requires alignment, balance, and rigor (Joughin & Macdonald, 2004). Academic integrity is paramount to upholding the quality of education, and reducing academic fraud is essential for a thorough evaluation process.

Academic dishonesty by students, which this study defines as plagiarism and cheating on assignments and tests (Adesile et al., 2016; Ives et al., 2017; Ives and Nehrkorn, 2019; Jensen et al., 2002; Lin and Wen, 2007; Owunwanne et al., 2010), undermines the goals of higher education institutions. It compromises assessment validity since results cannot be generalized to represent trustworthy estimates of students’ grasp of course-work content (Bouville, 2010; Muñoz-García and Aviles-Herrera, 2014).

The topic of propensity to cheat (PTC) among students is a significant concern in higher education institutions. A student’s propensity is the likelihood that they will engage in a specific action or behavior. Propensity is defined in the literature as the tendency or predisposition to participate in a particular behavior, in this case, cheating. Cheating tendencies are the inclinations of an individual or a group of individuals to commit crimes prior to, during, or following an exam with the intention of unfairly benefiting from an unfair advantage over other students (Offor, 2009). The Propensity to Cheat (PTC) refers to the likelihood that students will engage in dishonest behaviors to gain unfair advantages in academic settings. This propensity is measured by the variety of self-reported cheating methods that students have employed, indicating their tendency towards such behaviors. PTC encompasses a range of unethical practices, including cheating on exams, plagiarizing assignments, falsifying research data, and misusing library resources. It highlights the broader issue of academic dishonesty, which not only includes cheating during exams but also extends to dishonest actions related to assignments, research work, and the misuse or theft of educational materials.

The tendency to cheat is closely related to academic dishonesty, which encompasses various unethical behaviors in educational settings. When individuals are more inclined to cheat, it often indicates a higher likelihood of engaging in dishonest practices that undermine the integrity of their academic efforts. Cheating can take many forms, such as plagiarism—where students copy another person’s work without giving proper credit—and falsification of information, which involves altering or fabricating data or sources to misrepresent one’s work or findings (McCabe and Trevino, 1996). For the objectives of this study, understanding this broader concept of cheating propensity is essential for developing effective strategies to prevent academic misconduct.

The study intended to validate the Propensity to Cheat (PTC) scale and examine its relationship with students’ Cumulative Grade Point Average (CGPA) and academic self-concept. The PTC serves as a measure of students’ likelihood to engage in academic dishonesty, and understanding its correlation with academic performance and self-perception is crucial for addressing issues of academic integrity.

Academic accomplishment is one of the most often utilized constructs in educational research and assessment in higher education. Academic achievement is usually measured by students’ grades and overall performance in their courses. CGPA is a commonly used measure of student’s academic performance in higher education, reflecting their overall academic achievement across all semesters. In this study, CGPA was used as an indicator of academic achievement.

Some academics believe that academic cheating is linked to a student’s GPA. The study highlighted the importance of academic achievement as a determinant of students’ propensity to cheat. For example, academic dishonesty was negatively related to students’ GPA. Those with higher GPAs, according to Tang and Zuo (1997), are less inclined to cheat than students with lower GPAs. The researchers also expected that GPA and self-esteem have an interaction. Students who have a high GPA and a high sense of self-worth are less prone to cheat.

To put it another way, students with a high GPA are less likely to cheat, whereas students with a low GPA are more likely to cheat (Baird, 1980; Lipson & McGavern, 1993; Olafson et al., 2013). However, McCabe et al. (2012) found evidence that students who looked to be particularly competitive at the top of the rankings also showed a higher likelihood of cheating. In this study, CGPA was used to assess the discriminant validity of the measurement of propensity toward academic misconduct.

Self-concept is a multidimensional construct with one general facet and numerous specific facets, one of which is “academic self-concept.” The multi-dimensional model of self-concept shows that, along with social, emotional, and physical self-concept, intellectual self is one of the essential elements of self that contribute to an individual’s global self-concept (Marsh, 2006). Furthermore, this multidimensional model of self-concept distinguishes between general self-concept, which encompasses cognitive, affective, and behavioural elements, and academic self-concept, which refers to a person’s view of their academic competence (Bandura, 1993).

Academic self-concept refers to students’ perceptions of their academic abilities. It plays a significant role in their learning and academic performance. Academic self-concept, according to Trautwein et al. (2006), is an individual’s assessment of their academic ability. Academic self-concept is described as students’ assessment of their competency, dedication to, involvement in, and interest in classroom work, according to a study by Liu et al. (2005).

According to some scholars, academic misbehavior is linked to university students’ academic self-concept. Self-concept has an impact on examination misconduct perception, according to a study that looked into the relationship between students’ academic self-concept and examination misconduct perception (Marsh, 1990; Anderman & Murdock, 2007). Students with low self-esteem are more likely to engage in risky behaviors, such as cheating on exams, according to Kehinde (2004), whereas achievement-related beliefs, such as academic self-concept, will almost certainly influence students to engage in academic cheating behaviors, according to Rinn et al. (2014). In this study, academic self-concept was evaluated to measure students’ confidence and motivation regarding their academic abilities. To assess the discriminant validity of the measurement concerning the propensity toward academic misconduct, we utilized an academic self-concept scale. This approach allowed us to examine the relationship between students’ self-perception of their academic capabilities and their tendencies toward engaging in dishonest academic behaviors. By ensuring that the academic self-concept scale is distinct from measures of academic misconduct, we aimed to establish a clear differentiation between positive self-perception in academics and the likelihood of misconduct.

In this study, the PTC measure was evaluated for convergent validity by examining the internal consistency of the indicators measuring the same construct. In Bagozzi’s (1981) definition of convergent validity, the emphasis is on the internal consistency of indicators measuring the same construct. Researchers utilized reliability measures as one of the criteria to assess convergent validity (Fornell & Larcker, 1981). However, they recognized that evaluating convergent validity just on the basis of reliability is insufficient. To further assess convergent validity, a measurement model was estimated in which all indicators were related to the constructs they were intended to measure and not directly related to constructs they were not intended to measure (Fornell & Larcker, 1981; Hair et al., 2009). All indicators converge well on their own construct when the predicted measurement model sufficiently matches the data. Since model fit does not ensure measurement quality, researchers have argued that an appropriate model fit is insufficient to support convergent validity (Fornell & Larcker, 1981). As a result, additional criteria have been proposed to ensure that the indicators truly measure the intended construct.

The first condition for discriminant validity is establishing convergent validity (Bagozzi, 1981). This means ensuring that a construct is sufficiently represented by its indicators.

The objective of this stage was to establish further evidence based on relationships to other variables, i.e., evidence pertaining to the construct validity of the instrument (Balkin, 2017). Construct validity is a vital aspect of validating a tool, as it includes establishing the theoretical link between the variables being measured and other relevant constructs (DeVellis, 2016). One of the five main forms of validity evidence that focuses on showing connections between assessment scores and crucial criteria is evidence based on linkages to other variables (AERA, APA, & NCME, 2014). The strength and direction of the association between theoretically pertinent constructs are determined using discriminant validity, a kind of construct validity (DeVellis, 2016). The degree to which there are insignificant correlations between measures of theoretically unrelated constructs is referred to as discriminant validity. Discriminant validity is demonstrated, in accordance with Gefen and Straub (2005), “when each measurement item weakly correlates with another construct other than those to which it is theoretically associated.”

1.2 Prevalence of academic dishonesty

Various studies indicate that academic dishonesty is more prevalent than ever, and actions need to be taken by universities to educate students as well as faculty members about academic integrity and ethical professionalism (Diego, 2017; Feday, 2017; Jepngetich et al., 2017; Mebratu, 2014; Nelson et al., 2012). According to studies, dishonest behavior that began in high school can continue in college, and similarly, college students who engage in academic dishonest behavior are more likely to engage in dishonest behavior outside of the classroom (Harding et al., 2004). This suggests that there is a chance that dishonest behavior will continue over time and in different circumstances.

Academic cheating is a widespread problem around the world, with high rates of cheating reported among students in many countries. In the United States, surveys have revealed that up to 86 percent of college students have engaged in dishonesty in the classroom (McCabe et al., 2006). Bernardi et al. (2008) conducted a study to compare the perspectives of students from industrialized countries on cheating behavior. They discovered that 51% of the people in the samples admitted to cheating. According to Yussof and Ismail (2018), high rates of Malaysian students have cheated, primarily in assignments and quizzes that carry a lower weight in the final grade and are subject to less supervision and punishment. In South Korea, Ahmed (2018) discovered that 65 percent of pupils misbehaved using electronic media, and 80 percent of students committed academic fraud.

In Africa, academic dishonesty is equally prevalent. According to Umar (2004), cheating on exams was almost a standard practice in Nigeria. According to Munachonga (2015), examination misconduct has become increasingly common in Ghana. This is mostly because candidates are afraid of failing, lack confidence, are lazy, don’t prepare well, and, most importantly, are unable to apply themselves to their studies. Munachonga (2015) noted that officials of the Test Council of Zambia, the nation’s supervisory organization for exams, have occasionally been involved in examination malpractices.

When compared to various other studies carried out worldwide, the prevalence rate found in Ethiopia is high. Even though there is little to no evidence on the subject, cheating has from time to time increased in Ethiopia (Tefera & Kinde, 2009). Chala (2021) discovered that, with an incidence a rate ranging from 53 to 96 percent, academic dishonesty is a severe problem impacting university students in Ethiopia. Research conducted locally has, in one form or another, verified the growing epidemic of academic cheating in Ethiopian education at all levels (Feday, 2017; Hailu, 2015; Mebratu, 2014; Tefera & Kinde, 2009; Wondifraw, 2021). Furthermore, according to the Ethiopian Education Development Roadmap (2018–30), students frequently cheat on tests, and teachers are afraid that aggressive student evaluation may lead to an inaccurate assessment of their work and inflated grades (MoE, 2018).

Academic dishonesty can be justified in various ways in an academic setting. The majority of the time, in Ethiopia, the behaviors take the form of plagiarism and exam cheating (papers and assignments). To mention a few results of research done in the Ethiopian setting, Feday (2017) and Mebratu (2016) found that plagiarism in written work and exam cheating are the most common forms of academic dishonesty. Furthermore, several lecturers have expressed concern about the increasing incidence of plagiarism, assignment cheating, and exam dishonesty in Ethiopia’s public universities. This suggests that these universities share the researchers’ concerns based on their observations and teaching experiences. Specifically, some students at Hawassa University and Wolkite University College of Science and Engineering have been found to engage in academic dishonesty, as reported by researchers with several years of teaching experience in these courses. Furthermore, some students oppose the growing practice of cheating on exams and plagiarism in coursework. Based on our observations, a major contributing factor to exam cheating is a student’s lack of academic proficiency in a particular area. In this sense, the chance of earning a high cumulative grade point average (CGPA) is increased by cheating. Closing the gap and validating a scale that might possess the required psychometric properties to specify measurement accuracy was, thus, the primary objective of this work.

1.3 Problem statements

Numerous measures have been developed to assess university students’ tendencies toward academic dishonesty, as highlighted in a review of previous studies on the subject. Academic cheating is a multidimensional issue that affects individuals globally (Imran & Nordin, 2013; Iberahim et al., 2013; Nazir & Aslam, 2010; Thomas, 2017; Tadesse & Getachew, 2010; Saidin & Isa, 2013; Yang et al., 2013). Among these measures, the Academic Practices Survey stands out as a two-dimensional construct that encompasses plagiarism—referring to written work—and cheating, which pertains to class tests and exams, as identified by Roig and DeTommaso (1995) and Ferrari (2005). Furthermore, Adesile et al. (2016) introduced a three-dimensional scale for measuring academic integrity that includes plagiarism, cheating, and research misconduct. Given these varying dimensions of academic dishonesty, this study focuses specifically on the dimensions of the Academic Practices Survey, examining cheating on tests, assignments, research work (plagiarism), and the mutilation or theft of library resources.

The main issue with measuring the PTC in environments where academic dishonesty is prevalent is its legitimacy. While a few studies have addressed the PTC, none have utilized a well-developed and validated questionnaire. Furthermore, aside from a brief mention of the reliability coefficient, little is known about the confirmatory component analysis (Adesile et al., 2016). As Imran and Nordin (2013) noted, the dimensionalities of most measures have not been carefully examined, and there is insufficient evidence to suggest that they possess meaningful psychometric properties. Additionally, discriminant validity—the ability of a scale to differentiate between the relevant concept and other related constructs—is lacking in previous scales. This failure to demonstrate differentiation has led to confusion in interpretation (Bashir & Bala, 2018). Researchers can address these challenges and recommend further examination into the PTC by assessing the validity and reliability of existing measures.

This study aimed to assess the construct validity of a locally developed Propensity to Cheat Scale (PCS) among science and engineering students, focusing particularly on its convergent and discriminant validity. Establishing a valid and culturally reliable PCS is essential for accurately measuring the constructs related to cheating behavior. To ensure that psychological and behavioral constructs are properly defined and evaluated, it is crucial for instruments to undergo rigorous validation and assessment (Hair et al., 2010). Additionally, Hair et al. emphasized the importance of continuously confirming the validity and unidimensionality of conceptual frameworks, even when using well-established instruments. In this context, the study sought to analyze the convergent and discriminant validity of the PCS specifically. Although a Multimethods Multitraits (MMMT) investigation was employed to determine these validity measures, there was insufficient evidence to substantiate the findings. Ultimately, this research contributes to the body of data supporting the construct validity of the PCS, highlighting the necessity of validating instruments that measure the constructs of interest.

This study arises from the lack of an instrument to measure the tendency for academic dishonesty in Ethiopian higher education institutions. Despite the researchers’ experience in this field, there has been no prior research on the psychometric qualities and validity of the PCS used by these institutions. To address this gap in the literature, the researchers aimed to evaluate the psychometric properties of the PCS within the context of Ethiopian higher education.

1.4 Research questions

The researchers concluded that it would be appropriate to validate the locally developed PCS in order to assess its convergent and discriminant validity. Two distinct research questions were formulated in order to achieve the research objectives:

2.1 Research design

According to Creswell and Plano (2007, p. 58), a research design is the “procedures for collecting, analysing, interpreting, and reporting data in research studies.” It is the overarching strategy for linking the theoretical research issues with relevant (and doable) empirical research. In other words, the study design determines how the necessary data will be collected, how it will be analysed, and how it will be used to address the research question (Grey, 2014). Three different types of study designs can be used: exploratory, descriptive, and explanatory (Robson, 2002). His classification system is based on the goal of the research field because each design has a distinct ultimate goal. To understand the relationship between students’ reported tendency to cheat and their academic self-concept and academic accomplishment, this study uses a quantitative approach and an explanatory research design. A correlational research methodology used for explanatory purposes allows the researcher to assess the degree of association between two or more variables. Additionally, using this style of research design enables the researcher to gather data all at once (Creswell, 2015).

2.2 Population of the study

The study focused on fourth-year science and engineering regular undergraduate students attending Ethiopia’s public universities. The researchers selected this specific group from the target population of first- to fifth-year science and engineering undergraduate students in the participating universities. The target population consisted of 6,524 students, including 4,901 males and 1,623 females.

2.3 Samples and sampling procedure

In this study, the sample selection procedure was carried out systematically. First, three public universities in Ethiopia—Hawassa University, Ambo University, and Wolkite University—were selected from a total of 50 using purposive sampling. These universities were chosen to represent both established and newer institutions.

Next, a purposive sampling strategy was applied to select the colleges within the participating universities, focusing on the Colleges of Natural and Computational Sciences and Engineering/Institute of Technology. The College of Sciences and Technology was included in the study due to its significant role in influencing the quality of education. Additionally, these science and engineering colleges were chosen because of a curriculum shift towards a focus on science and technology, aligned with the policy of a 70:30 graduate mix. The emphasis on these fields is connected to social transformations and economic development (Ministry of Science and Technology, 2015-2025). Furthermore, according to the Ethiopian Education Development Roadmap (2017-2030), the curriculum aims to enhance higher-order thinking skills by teaching science, technology, engineering, and mathematics (STEM) subjects at appropriate educational levels.

After selecting the colleges, 30 departments from the engineering college and 17 departments from the College of Natural and Computational Sciences were randomly chosen. This random selection ensured a diverse representation among the departments within the selected colleges. Finally, a simple random selection procedure was employed to choose students from these departments for the study, ensuring that the student sample accurately represented the larger student population within the universities. Of the 550 students invited to participate in the study, only 500 completed the questionnaire correctly. The universities, their respective colleges, and the study participants are summarized in Table 1.

The researcher converted Gomez’s (2013) Sloven’s Formula into a formula to compute the sample size for this study. The sample size was determined using the following formula. The following is how this formula is laid out:

Where, N is the target population size

n is the sample size and

e is the level of precision (acceptable margin of error at 5% (95% confidence level), P = .5 are assumed for equation

For this study the target population (N = 6524) then the required sample size is calculated as

When selecting a sample size for the study, several criteria need to be considered. These include the size of the population, the level of confidence, the margin of error (confidence interval), the number of variables used in the study, the statistical analysis technique to be employed, as well as time, money, and effort constraints. In this regard, Ellen (2014) asserts that a formula must be employed to account for the margin of error and confidence level when a sample is drawn from a population. Further, research claims that Sloven’s formula should be used when there is little information available about how a population will behave (such as in the case of this study polling college students to assess their opinions and likelihood to cheat online), other than its size (Ellen, 2014; Kavai, 2014; Tanty & Rahayu, 2014; Abun & Cajindos, 2012). This equation enables the researcher to accurately sample the desired population (Ellen, 2014). It is thought reasonable to select a sample size using Slovin’s formula, which was developed in 1960. This is true in particular when there is uncertainty regarding the behavior of the population (Isip, 2014).

Of the 550 students invited to participate in the study, only 500 completed the questionnaire correctly. The universities, their respective colleges, and the study participants are summarized in Table 1.

The validation of the Propensity to Cheat Scale (PCS) followed a structured approach, which included data collection, Confirmatory Factor Analysis (CFA), and validation. The timeline and sample size are illustrated in the workflow chart below.

2.4 Research instruments

The Students’ Perceived PCS was used to measure the students’ views of their propensity for academic misconduct. The measure has thirty items scored on a Likert-type scale with a maximum of five points, ranging from 1 (Strongly Disagree) to 5 (Strongly Agree).

The study focused on the latent construction of students’ propensity for academic misconduct in higher education settings in Ethiopia, examining four dimensions of cheating behavior: (1) cheating on tests and examinations, (2) cheating on assignments, (3) cheating on research work (plagiarism), and (4) theft and mutilation of library materials. A total of thirty items were included in the study, with the distribution as follows: six items measured cheating on exams, eight items assessed cheating on assignments, nine items evaluated cheating on research papers, and seven items addressed stealing and mutilating library materials. Higher scores on each subscale indicated a greater tendency to engage in academic misconduct. The internal consistency reliability of the subscales was established, yielding alpha values of .93 for cheating on exams, .85 for cheating on assignments, .90 for cheating on research work, and.96 for theft and mutilation of library resources.

This study employed a self-administered questionnaire for the measurement of PTC by four first-order factors: theft and mutilation of library materials, cheating on tests or exams, cheating on assignments, and cheating on research work. All of the dimensions and their subdomains have been constructed, and this study included the reliability and validity (convergent and discriminant validity) of the scale. Notion validity in this context refers to how well a scale’s items are suited to measuring a certain theoretical construct (DeVon et al., 2007; Kane, 2001). Discriminant and convergent validity are two fundamental features of construct validity, according to most arguments (Cronbach & Meehl, 1955; Campbell & Fiske, 1959; Pike, 1992, 2006). The construct validity of the PCS was then investigated in this study by examining both its convergent and discriminant validity.

To assess academic self-concept, Liu and Wang (2005) developed a scale consisting of 20 items. These items are designed to measure students’ confidence and motivation in their academic abilities within a university environment. Participants rated each item on a five-point Likert scale, ranging from strongly disagree (1) to stronglyagree (5). The internal consistency reliability of this scale is 0.85. The academic self-concept scale was used to assess discriminant validity of the measurement of propensity toward academic misconduct.

2.4.1 Validation of the measures

Following the pilot study, which established face validity, content validity, and reliability, further testing of various types of validity and reliability is required before fully implementing the measures and considering them during the PCS validation process. This study involves an adequate number of participants to ensure robustness.

To assess the construct validity and dimensional structure of the PCS, we employed both Exploratory Factor Analysis (EFA) and Confirmatory Factor Analysis (CFA). EFA was initially used to explore the underlying factor structure of the PCS. This involved identifying the number of factors that best represent the data and determining which items load significantly onto each factor. Items that did not load significantly on any factor or that exhibited cross-loading onto multiple factors were removed based on the EFA results.

Study 1 represents the initial phase of the research focused on developing and preliminarily validating the Propensity to Cheat Scale (PCS). This phase included an exploratory factor analysis (EFA) aimed at identifying the underlying factor structure related to students’ propensity to cheat.

The EFA was conducted using the Maximum Likelihood (ML) method for factor extraction, and oblique rotation was applied to permit correlated factors, which is particularly suitable for research involving human behavior. The analysis revealed a four-factor solution, utilizing Maximum Likelihood analysis with Oblimin and Kaiser Normalization rotation.

The identified factors had eigenvalues exceeding one, with the first factor accounting for the majority of variance. Specifically, Factor One had an eigenvalue of 13.42 and explained 46.28% of the variance. Factor Two had an eigenvalue of 2.17, contributing 7.46%; Factor Three had an eigenvalue of 2.02, accounting for 6.98%; and Factor Four had an eigenvalue of 1.22, explaining 4.20%. Collectively, these four factors accounted for an average of 64.92% of the total variance.

The details of the factor loadings and communalities are as follows: Factor One: Comprising seven items (MOL55-MOL61) with loadings ranging from .58 to .93 and communalities from .72 to .87. Factor Two: Comprised six items (COT5-COT11) with loadings between .61 and .81 and communalities from .60 to .73. Factor Three: Included seven items (COR45-COR52) with loadings from .51 to .71 and communalities ranging from .41 to .69. Finally, Factor Four: Consisted of six items (COA22-COA30) with loadings between .52 and .66 and communalities from .37 to .53.

Subsequently, CFA was conducted to test the hypothesized factor structure identified in the EFA. This step involved specifying a model based on the EFA findings and evaluating the model’s fit to the data. To assess model fit, we utilized goodness-of-fit indices such as the Comparative Fit Index (CFI) and the Root Mean Square Error of Approximation (RMSEA).

The final step in the validation process involved confirming that the model fits the data well and that the items reliably measure the intended construct. This assessment also included evaluating the internal consistency of the scale using Cronbach’s alpha. Through this comprehensive approach, we aimed to ensure that the PCS is a valid and reliable measure for our study.

In a validation study, 550 students from three public universities were given a questionnaire to determine their PTC and their perception of their academic self-concept. Out of the initial 550 participants, 500 students fully answered the PTC and academic self-concept items and were included in the analysis. Among these 500 students, there were 367 males and 133 females, with ages ranging from 18 to 40 years. Additionally, 50 students who partially answered the items were disqualified from the study.

Prior to data processing, the frequency distribution and the minimum and maximum scores for each item were used to verify that the database had been entered accurately. Additionally, each measure’s underlying assumptions for CFA were examined, and they were confirmed to be valid for the analysis.

2.5 Data collection procedures

In November 2022, data collection for this research began immediately following the enrolment of fourth-year students for the first semester of the academic year. The measures were translated into Amharic and back-translated by three translators from the English and psychology departments of Wolkite University to validate the instrument (PSC). Approval was obtained from each college dean before the commencement of data collection for this study. The college deans informed the department heads and instructors of the individual institutions about the study’s objectives. Subsequently, arrangements were made for the researchers and data collectors to recruit participants in the classroom. Teachers at the universities were tasked with informing and motivating students to participate in the study and scheduling times for the researchers to speak with them during regular class hours. Before delivering the questionnaires, the fourth-year students selected for the study provided their verbal informed consent.

The study began with a thorough briefing for the participants, which included the goal and objective of the research, the focus of the questionnaire, and the potential benefits of the study for society and individuals. The participants were then asked for their consent to participate voluntarily, with the assurance that they could withdraw at any time without consequences and that all information provided would be kept confidential. Following the briefing, the researchers distributed the Amharic revised version questionnaires to students at Hawassa University, Ambo University, and Wolkite University from November to January 2022 during regular class time.

2.6 Methods of data analysis

The data processing procedures began by analyzing incomplete questionnaire responses and discarding any that were not complete. The remaining questionnaires were then coded and entered into a computer for further analysis. The data was cleaned to remove any errors that may have occurred during the coding process. Additional checks were conducted to ensure the accuracy of the data entry and measurement scales. The data was then examined for outliers, normality, skewness, and kurtosis to determine if it was at a normal level.

Confirmatory Factor Analysis (CFA) is a type of measurement-related structural equation modeling (SEM) that is essential for validating the relationships between observed items and their corresponding latent factors (Brown, 2015). CFA provides metrics to evaluate how well a proposed theoretical model fits the collected data, allowing researchers to refine these relationships within the measurement model (Stevens, 2012). In the context of social and behavioral sciences, CFA is considered a critical tool for validation (Brown, 2015).

A CFA was conducted for this purpose in order to strengthen the validity of the suggested model of PTC and validate the 4-factor solution obtained from the EFA. Several model fit indices were used to assess the goodness of fit of the CFA model. Brown (2015) divided model fit indices into three categories as a result: “absolute fit, fit adjusting for model parsimony, and comparative or incremental fit” (p. 71). According to Brown (2015), absolute fit indices, which include the chi-square statistic (χ²) and the standardized root mean square residual (SRMR) indices, evaluate model fit at an absolute level. Collier (2020) offers a relative chi-square test, which is the chi-square value divided by the degrees of freedom (χ²/df ), as a better method to utilize because the chi-square statistic is more sensitive to sample size. According to Collier (2020), a relative chi-square value (χ²/df ) that is considered acceptable may range from 3 to 5. Any number below 3 denotes an excellent model fit, while values between 3 and 5 demonstrate an acceptable level. As a result, the first-order CFA model’s outcomes demonstrated that it had a chi-square statistic (χ² = 3.877, p = .000).

Parsimony correction indices, sometimes also referred to as absolute fit indices, are the second category of model fit indices (Brown, 2015). The root-mean-square error of approximation (RMSEA) is the most widely used and advised index in this area, according to Brown (2015). The RMSEA is referred to as an error of approximation index since it measures how well a model matches the population (Brown, 2015). The purpose of this study is not to determine whether the model applies to the population exactly. The RMSEA is hence “sensitive to the number of model parameters” but “relatively insensitive to sample size” (Brown, 2015, p. 71). Comparative fit indices, which are also referred to as incremental fit indices and assess the fit of a user-specified solution against a nested baseline model, are the third category of model fit indices, according to Brown’s (2015) taxonomy. The CFI and TLI are the most popular and suggested indices in this category, according to Brown (2015).

In this study, CFA was employed to verify both the measurement and structural models. Specifically, a confirmatory factor analysis was conducted to assess the applicability of the scale and to evaluate the convergent and discriminant validity of students’ propensity for academic fraud. The research utilized a sample of public university students in Ethiopia, marking the first investigation of its kind in this context. By examining the effectiveness of the Propensity for Cheating Scale (PCS) in measuring tendencies toward academic dishonesty among Ethiopian higher education students, this study aims to contribute to the existing body of knowledge and address a notable research gap in this area.

The analysis of the data was conducted using the Statistical Package for the Social Sciences (SPSS) and STATA 14. Initially, data were entered into SPSS version 25, where missing data were validated before being transferred to STATA version 14 for confirmatory factor analysis (CFA). Cronbach’s alpha was employed to assess the internal consistency and reliability of the constructs. Furthermore, to evaluate the construct validity of the newly developed Propensity to Cheat Scale (PCS), both convergent and discriminant validity were examined.

The CFA was specifically utilized to determine whether the sample data aligned with the established propensity to cheat (PTC) measurement model. This model’s factorial structure was initially identified through exploratory factor analysis (EFA) with a sample size of n = 550.

To assess the fit of the established PTC measurement model to the data, CFA was employed along with various metrics: composite reliability (CR), average variance extracted (AVE), maximum shared squared variance (MSV), and average shared squared variance (ASV). These metrics also served to address specific research questions, such as the extent to which the PCS demonstrates convergent validity. Additionally, the discriminant validity of the newly constructed PTC instrument was evaluated in relation to other validated factors, such as academic self-concept, using Pearson correlation analysis.

Data collected from participants were analyzed based on IBM SPSS 25 and STATA 14.

3.1 Confirmatory Factor Analysis (CFA) and overall construct validity

Before conducting Confirmatory Factor Analysis (CFA) on the PTC measure, an item analysis was performed. This included evaluating item means, inter-item correlations, corrected item-total correlations, and the alpha value if any item was eliminated. Inter-item correlations varied across subscales, with ranges as follows: 0.01 to 0.65 for plagiarism, 0.07 to 0.67 for cheating on tests, 0.11 to 0.71 for cheating on assignments, and 0.48 to 0.77 for library theft. Item-total correlation scores were satisfactory, falling between 0.34 and 0.61 for cheating on tests, 0.39 and 0.67 for assignments, 0.37 and 0.79 for plagiarism, and 0.68 and 0.82 for library theft.

Items were selected based on inter-item correlations above 0.30, item-total correlations greater than 0.50, and mean values centered, following guidelines from Field (2009). The determinant of the correlation matrix was above 0.0001, indicating no multicollinearity issues. An inter-correlation matrix showed sufficient coefficients among certain items, and Pearson’s correlation coefficients were calculated between factors.

Exploratory Factor Analysis (EFA) was performed to examine the underlying factor structure of the items. Items with high cross-loadings, low factor loadings, or those incorrectly loaded on the wrong factors were removed. After the EFA, the final scale consisted of 30 items. These items were further validated using CFA to ensure they accurately measured the intended constructs of the PTC measurement instrument.

Following the exploratory factor analysis conducted in Study 1, the CFA established the dataset’s factor structure. The exploration of factor structure—how variables are related and grouped based on inter-variable correlations—in the CFA confirms the factor structure that was extracted in the EFA (Brown, 2015). To evaluate model parameters and fit indices across the clustering of independent variables, a CFA was also conducted for all datasets to assess the general construct validity (Chuang, Weng, & Huang, 2015).

Both the first-order (measurement model) and the second-order (structural model) were used to validate the model. The findings, as shown in Figure 1 and Table 3, strongly support the measurement of PTC using four first-order factors: cheating on tests or examinations, cheating on assignments, cheating on research work, and theft and mutilation of library materials, with a total of 26 items based on the PSC. These items were designed to elicit responses from university students regarding how they perceived their academic cheating.

Figure 1. Output of the measurement model for Latent Variable “Four-Factor of PTC” and Observed Indicators.

This figure demonstrates the measurement model for the latent variable Propensity to Cheat (PTC) with four first-order factors: Cheating on Tests (COT), Cheating on Assignments (COA), Cheating on Research Work (COR), and theft and Mutilation of Library Materials (MOL). Each factor is represented by multiple observed indicators (items) that measure the students’ perceived cheating behaviors.

3.2 Testing the measurement model

The measurement model was put to the test, and then the structural model was examined using the standards outlined in the literature (see Hair et al., 2016). The measurement models assess the fit of the data to the model and the link between the latent variable and its indicators (Henseler et al., 2009). The first-order CFA model has been developed to measure the relationships between the four dimensions (cheating on tests, cheating on assignments, cheating on research work, and theft and mutilation of library materials) and the twenty-six associated items that fit the empirical data well. The measurement models for the unobservable (latent) variables were generated and validated using the CFA approach in this study. Because the observed variables of the PTC subscales were determined in advance based on a review of the literature and interviews with experts, a CFA was used to assess the link between cheating on tests, cheating on assignments, cheating on research work, and theft and mutilation of library materials and each of their observed variables in order to develop a measurement model that fit the empirical data well. Figure 1 below shows the measurement models for the four-factor PTC. Figure 1 is available in the data repository (Gizaw, 2024).

Multiple fit indices were employed to assess the appropriateness of the CFA model’s fit to the data, including the χ²/df ratio, Tucker Lewis Index (TLI), comparative fit index (CFI), and root mean square error of approximation (RMSEA). The goodness-of-fit indices of the CFA model for the PCS are shown in Table 2. Despite the fact that the chi-square statistic has a significant value (χ²/df = 3.877, p =.000), alternative fit indices are encouraging due to the highly sensitive nature of this statistic to large samples (Bentler & Bonett, 1980; Stevens, 2002). The CFA results demonstrate that the model had fit statistics such as RMSEA = 0.076, RMR = 0.045, TLI = 0.907, and CFI = 0.916. According to the literature, the chi-square value of less than 5 is acceptable, and less than 3 is good (Marsh & Hocevar, 1985). Hu and Bentler (1999) and Browne and Cudeck (1992) proposed (RMSEA less than.08, RMR less than .05, TLI and CFI greater than .90) as the recommended values for this fit statistic. Based on the indices obtained after CFA, the results indicated that each dimension’s factor model fit each dimension well. Finally, the coefficient of determination for the entire model is extremely high (CD = 1.000).

3.3 Testing the structural model

It is required to move on to the structural model after developing the measurement model. A structural model explains how the many constructs in a model relate to one another. The four measurement models become structural models in the second-order CFA model by linking them collectively or through PTC, as shown below. In order to validate the four-factor structure developed by the exploratory factor analysis (EFA), a second-order CFA was applied. The next phase in structural equation modeling (SEM) analysis is to examine the goodness of fit of the full structural equation model and determine if an acceptable full structural model is obtained after acquiring an acceptable measurement model that fits the empirical data. A set of goodness-of-fit analyses was carried out for this purpose.

Figure 2 is shows the factor distributions and values obtained from the second-order CFA model. Figure 2 available in the data repository (Gizaw, 2024).

Figure 2. Output of the Structural Model for the Students’ Propensity to Cheat.

This figure illustrates the structural model for the student’s propensity to cheat, validated using a second-order Confirmatory Factor Analysis (CFA). The model includes the four first-order factors (COT, COA, COR, and MOL) and their relationships with the overall latent variable PTC. The standardized factor loadings and the coefficients of determination (R² values) for each observed indicator are displayed, indicating the strength of the relationships between the latent variables and their respective indicators.

To examine the relationships, a structural equation model created with STATA 14 was utilized. A good fitting model is acceptable if the CMIN/df value is less than 5; the Tucker Lewis Index (TLI) (Tucker and Lewis, 1973); and the confirmatory fit index (CFI) (Bentler, 1990) is greater than 0.90 (Hair et al., 2010). In addition, an adequate fitting model was accepted if the standardized Root-Mean Square Residual (SRMR) estimated by STATA was less than 0.05 and the Root Mean Square Error of Approximation (RMSEA) was between 0.05 and 0.08 (Hair et al., 2010). The alternative fit indices are encouraging due to the sensitivity of the chi-square statistic to sample size: χ²/df = 3.966, CFI = 0.913, TLI = 0.904, RMSEA = 0.077 and RMR = 0.049. The chi-square statistic has a significant value, χ² (295) = 1170.069, p <0.00. In this regard, the second-order CFA model was suitable since all of the indices met the threshold values of the requirements and because their values were comparable to those of the first-order CFA model. Finally, the total model’s coefficient of determination (CD = 0.912) is very high.

The study used CFA to explore the underlying latent variable structure of the PCS from the same sample on which the EFA was administered in order to evaluate the goodness of fit of the four-factor structure produced from the EFA.

Table 3 and Figure 2 demonstrate that each factor loading was significant at p<.05. Additionally, the results of the standardized factor loading coefficients showed that the subscales measuring cheating on exams and tests ranged from .75 to .86, the subscales measuring cheating on assignments and research work from .59 to .72, and the subscale measuring theft and mutilation of library materials from .77 to .92.

Furthermore, the squared multiple correlation coefficients also provide the coefficient of determination (R2), which illustrates the extent to which a factor may account for the variance in an item. In this regard, the COT5 item has the highest R2 (.74), meaning that the latent variable test-cheating accounts for 74% of the variation in the COT5 item.

The model was able to explain 64 percent, 52 percent, 79 percent, and 79 percent, respectively, of the factors related to cheating on tests, cheating on assignments, cheating on research work, and theft and mutilation of library materials ( Table 3). The results showed that, after exam cheating, the two most important factors were theft of library items and exam cheating. Cheating on exams, assignments, research work, and mutilating library materials collectively account for 91% of the variance in the PTC, as indicated by the squared multiple correlation for the PTC of 0.91 (see Figure 2).

4.1 Convergent validity

The standardized factor loading, extracted average variance (AVE), and composite reliability (CR) tests were used to assess convergent validity. Numerous studies have proposed evaluating convergent validity by looking at the statistical significance of standardized factor loadings (e.g., Dunn et al., 1994). For instance, Hair et al. (2009) claimed that all standardized factor loadings should be at least 0.5 and, ideally, at least 0.7, while Stevens (2002) suggested that the value of a factor loading should be more than 0.4 for interpretation purposes. The factor loadings were all greater than 0.59, and some loadings were greater than 0.70 (see Table 3). Some studies have utilized the Fornell and Larcker (1981) criterion for evaluating convergent validity in addition to looking at the standardized factor loadings (for instance, Yu et al., 2021; Zahoor et al., 2022). According to Fornell and Larcker (1981), convergent validity is demonstrated when a latent construct explains at least half of the variance in the indicators it is related to. To indicate the average amount of variance that a construct explains in its indicators relative to the sum variance of its indicators, they suggested using the average variance extracted (AVE).

The value of AVE for the PTC subscale equals the sum of the average square of factor loadings across all of its indicators divided by the number of items. In order to determine whether the items logged under each facet or domain were estimating the same concept, composite reliability (CR) values of 0.7 or higher and average variance extracted (AVE) values of 0.5 or greater were employed (Hair et al., 2019). Convergent validity, according to Hair et al. (2009), is seen when the CR is greater than the AVE and the AVE is greater than 0.5.

According to the table, all of the AVE and the items’ total standardized factor loading were both greater than 0.59, which is a sign of good convergent validity (Hair, Sarstedt, Ringle, & Gudergan, 2017). For this reason, the value of each PTC subscale’s composite reliability was higher than the value of the average variance extracted (see Table 4). The cheating on research work and cheating on assignments variables’ AVE values (0.37 and 0.38) are below but around the suggested cutoff limit of 0.50 (Malhotra, 2019), as shown in Table 4. The construct has acceptable convergent validity if AVE is less than 0.5 but the composite reliability is greater than 0.6 (Fornell & Larcker, 1981). Exam cheating and library material theft value components show acceptable AVEs of 0.56 and 0.61, respectively, exceeding the 0.5 cutoff and confirming the convergence validity of their latent construct.

Convergent validity is also demonstrated by the fact that the Maximum Shared Variance for all three variables is less than the pertinent Average Variance Extracted. Our variables have a high level of internal consistency, as seen by the table’s composite reliability of all factors, which is greater than 0.70 (Gefen et al., 2000).

4.2 Discriminant validity of the propensity to cheat scale

Discriminant validity is conducted to ensure that the PCS measures a construct that is distinct from other constructs. Scales that measure theoretically unrelated constructs should, according to discriminant validity, have a low correlation (DeVellis, 2003; DeVon et al., 2007). Discriminant validity testing is done primarily to demonstrate how different an item or set of items is from others in this study. In other words, the objective of this study is to demonstrate low correlations between items measuring various constructs or variables. This study examined the association between the PCS and other established measures, such as the academic self-concept scale. By assessing these associations, the researchers purpose to reveal the unique contribution of the PCS in measuring PTC in academic settings.

The perceived propensity of students to cheat and their academic self-concept were examined using a bivariate correlational analysis. Bivariate correlation was also used to assess the link between academic achievement and students’ perceived PTC. For the purposes of data analysis, the continuous composite scores on all scales were regarded as interval levels.

Based on a review of the literature, it was predicted that there would be little link between the PTC and academic self-concept. Table 5 displays the relationships between the PTC subscales (MOL, COT, COR, and COA) and academic self-concept (ASC), with the bolded red items denoting discriminant validity. The correlation between the ASC and the subscales (MOL, COT, COR, and COA, respectively) was found to be lower than the correlation between the items of the same construct, such as the respective PTC subscales for MOL, COT, COR, and COA were.678,.741,.586,.520,.571, and.578. This indicates a strong link between items on the PTC subscale alone, but weak correlations between those same items and academic self-concept. To sum up, the results indicate that there is a low relationship between the PTC and academic self-concept, as expected.

Furthermore, research on cheating has shown that there is only a weak negative correlation (Cizek, 1999) between cheating and grade point average (GPA). Higher GPA students are less likely to report or cheat, but students with lower grades are more likely to do both. For this reason, Table 6 displays that the correlations between the students’ cumulative grade point averages (CGPA) and their PTC subscales (MOL, COT, COR, and COA) are lower compared to the correlations between the items of the same construct (PTC subscales). This indicates discriminant validity, which is represented by the red-bolded items, as the relationships between the CGPA and the PTC subscales are not as strong as the relationships within the PTC subscales themselves. Table 6 indicates that there is a lower negative relationship between the CGPA and the MOL, COT, COR, and COA subscales than there is between the items of the same construct, such as the PTC subscales, which have correlations of.678, .741, .586, .520, .571, and.578 (MOL, COT, COR, and COA, respectively).

In short, the PTC and academic self-concept showed a positive, although weak, relationship, indicating discriminant validity. In addition to this, the PCS’s discriminant validity is supported by the low, negative correlations between students’ academic achievement and PTC.

5.1 Confirmatory factor analysis and model fit

In this study, the models were assessed to determine if they accurately reflected the data using a number of model fit indices. The measurement model outputs of the first-order CFA model-based fit indices in this analysis were CFI = .916, TLI = 0.907, SRMR = .045, and RMSEA = .076; these values suggest that the model had a good fit to the observed data (χ²/df = 3.88, p = .000). In regard to this, the CFI value was 0.92 indicated a satisfactory fit, whereas CFI values are 0.90 and higher as considered acceptable (Kline, 2023; Thompson, 2004). RMSEA = .077, CFI = .913, RMR = .049, and TLI = .904 are the results of the second-order CFA model run on the scale structure consisting of four factors and 26 items (χ²/df = 3.966, p = .000). The PCS has a valid and reliable factor structure as a result of the structural model’s adequate fit to the sample data. Thus, both the structural and measurement models demonstrated good fit to the empirical data, indicating that the PCS has a valid and reliable factor structure.

5.2 Construct validity assessment of propensity to cheat scale

Discriminant validity and convergent validity were used to investigate construct validity. Convergent validity is usually examined first, followed by discriminant validity. According to Collier (2020), this type of construct validity describes the extent to which all indicators of a certain construct are measuring the same thing (construct) that they are supposed to measure. “Evidence that different indicators of theoretically similar or overlapping constructs are strongly interrelated” (Brown, 2015, p. 2) serves as the basis for this determination.

The term “convergent validity” refers to the closely connected variables inside a single component (Kaplan & Saccuzzo, 1993). The convergent validity was investigated using factor loading, average variance extracted (AVE), and composite reliability (CR) ( Hair et al., 2017). All factor loadings were higher than 0.59, as shown in the result section. The standard for an item to be included within a factor is having factor loads above .40 ( DeVellis, 2014; Hair, 2009). In addition, if the factor loading was greater than 0.5, it was considered acceptable (Hidayat et al., 2018). The AVE abbreviation stands for the “average of the squared standardized pattern coefficients for indicators that depend on the same factor but are specified to measure no other factors” (Kline, 2023, p. 313). All CR was greater than.85, while AVE was over 0.37.

Accordingly, Hair et al. (2010) proposed that convergent validity is achieved when composite reliability is more than average variance extracted (AVE), with both values exceeding .70 and .50, respectively. The AVE values for plagiarism and theft of library resources are 0.56 and 0.61, respectively, which are both higher than the cutoff value of 0.5 and indicate the convergence validity of their latent construct. However, the AVE values for plagiarism in research work and cheating on assignments are slightly below but near the recommended cutoff of 0.50, at 0.37 and 0.38 (Malhotra, 2019). The construct has adequate convergent validity if the mean variance is less than 0.5 and the composite reliability is greater than 0.6 (Fornell & Larcker, 1981). In light of this, all subscales in this study satisfied the convergent validity of the constructs (Fornell & Larcker, 1981). Therefore, these results demonstrate that the scale has achieved convergent validity.

The test of discriminant validity in this study showed that the PCS, academic self-concept, and CGPA are distinct constructs, as evidenced by the low correlations between them. As demonstrated by Gholami et al. (2013), discriminant validity is reached when a construct measures how many indicators are used to indicate only one construct while being genuinely unique from the other constructs. Two latent variables representing various theoretical concepts are statistically different when they have discriminant validity.

In this study, the correlations between the PCS, academic self-concept, and CGPA were tested. A low negative correlation between the CGPA and the PCS would be indicative of discriminant validity. Similarly, it was supposed that evidence of discriminant validity would come from a low correlation between academic self-concept and PCS. The correlation between the PTC subscale, academic self-concept, and CGPA is less than 0.12, as indicated in the results section, indicating that discriminant validity was demonstrated. The findings of this study align with previous research suggesting that Comas-Forgas and Sureda-Negre (2010) suggest that students who struggle academically are more likely to commit plagiarism. As one of the most significant types of PTC, plagiarism, this evidence can be used to demonstrate the validity of the discriminant. Similarly, studies have indicated that students who perform well academically or have higher cumulative grade point averages are less likely to plagiarize (Guo, 2011).

Based on the CFA’s findings, the updated versions of the PCS’s four subscales had high support for the scale’s reliability, discriminant validity, and convergent validity. This study provides evidence that the PCS is a valuable instrument for measuring different aspects of PTC.

Ethical issues

The study was conducted in compliance with pertinent standards and regulations and was approved by Hawassa University. After the PhD dissertation proposal was approved and ethical clearance was obtained from Hawassa University’s Office of Vice President for Research and Technology Transfer (VPRTT), the study was carried out. The School of Teacher Education granted approval in this regard on 09/02/2014, with reference number COE-REC/017/24. The data repository contains the ethical approval certificate (Gizaw, 2024).

Underlying data

OSF: Propensity to Cheat and Academic Self-concept. 10.17605/OSF.IO/7G9JN (Gizaw, 2024).

This project comprises the following:

Extended data

OSF: Data and materials for propensity to cheat scale.

This project comprises the following:

Data are accessible under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0)