Learning to E-Cheat: A Criminological Test of Internet Facilitated Academic Cheating

Abstract

An increasing problem of great concern for academic institutions is the pervasiveness of cheating among students. Further compounding this problem is advancements in technology that have created new ways for students to engage in cheating. Despite a growing interest in technology facilitated cheating, little is known about why students may employ electronic resources to cheat. However, Akers' social learning theory offers one plausible explanation. Surveys were collected from a sample of 534 college students at a large southeastern university in order to quantify the prevalence and frequency of Internet facilitated cheating. These surveys allowed for an exploration of factors associated with this form of cheating and a comparison between what we refer to as E-cheating and traditional forms of cheating. Results indicate that approximately 40% of students have engaged in some form of E-cheating in the last year. Social learning variables emerge as the strongest predictors of both the occurrence and frequency of E-cheating while self-control and strain variables have little effect. An exploration of the relationship between E-cheating and similar technology free cheating behaviors suggests that there is significant overlap, but that some students do “specialize” in E-cheating or technology free cheating. We conclude by offering suggestions for teaching strategies, course and assignment design, and testing that will best limit E-cheating.

Notes

1. Allison (2000) recommends using listwise deletion when it will eliminate less than 15% of the total cases in a model. As only 2% were missing data relevant to this study, we utilized this technique.

2. We compare the demographic makeup of our sample to that of the college that houses the criminology program since similar data were not readily available for the criminology program.

3. While some students may be differentially able to commit advanced forms of computer crime, the types of online cheating explored in this study do not require advanced skills. As the sample was collected from a state’s flagship institute, all of the students are assumed to have basic web-browsing and word processing skills.

4. In order to be more comprehensive in our analysis, we also created a scale from these 10 items and all 20 cheating items and estimated models identical to those created for E-cheating and the parallel behaviors for each. Results were congruent with the E-cheating models and parallel behaviors models.

5. As Agnew (1992) has argued that strain operates through negative emotions, situational specific measures of anger and sadness were created in order to test for mediation should strain variables have significant coefficients in the final model for any of the cheating outcomes. As strain was insignificant in these models, it would have been futile to check for mediation so we have refrained from discussing those measures here.

6. We feel that inclusion of the diagnosed with a learning disability control is important as the university allows for special testing procedures that these students may utilize (such as testing in a more spacious and quiet room as opposed to the classroom and having 1.5 times the length of time allowed to other students). Since we are unaware of the effect of these procedures and no item assessed whether students with disabilities utilize this option, we included this measure as a control. We choose to exclude GPA as a control as it may have been directly affected by the dependent variable. Those who reported cheating in the prior semester would likely have had their GPA affected by this behavior (current GPA was recorded as opposed to GPA prior to the start of the current academic year) and GPA would therefore be inappropriate to include into models. The strain educational aspirations variable does assess subjective dissatisfaction with previous academic performance and in a way accomplishes the same goal as introducing GPA into the models.

7. Since we cannot determine which form of cheating preceded the other from the available data, we can only report whether or not individuals engaged in both the Internet facilitated behavior and its analog, and cannot make any assumptions about one behavior affecting the other. We examine this topic as a direction for future research in the discussion section.

8. These numbers are lower than most estimates of academic dishonesty at universities in the literature. The authors believe this can be partially attributed to the sample population. The survey was administered at the flagship university of the state with substantially higher SAT scores than other institutions.

9. Though individual item response analyses are not the purpose of the present study, we report this information as a guide which may help faculty by alerting them to the forms of cheating which are most common among students in criminology courses. For the sake of parsimony and since different cheating behaviors may be differing deviant ways of reaching the same goal, we do not explore factors related to individual cheating types within the regression analysis, and instead use the composite measure. We do, however, return to single item comparisons in Table 4. For more information about measurement of academically dishonest behavior and specific forms of cheating, we refer the reader to Passow et al. (2006) and Akbulut et al. (2008).

10. We choose not to discuss effects in terms of odds ratio since the units of the theoretical constructs do not have a clear intuitive meaning. We instead focus on the direction and significance of each coefficient (for similar reasons coefficients are presented in the tables rather than odds ratios). In addition, there seems to be no logical reason to compare effect sizes within a theory since each variable is a piece of the same theory. Such comparisons are therefore avoided.

11. Zero inflated models were also estimated since a large portion reported no cheating. Results were substantively equivalent.

12. In order to be comprehensive, models were estimated using a scale of the 10 items that were neither Internet facilitated or the mirror of one of these behaviors as the dependent variable. Results were substantively equivalent to the models predicting the parallel behaviors. Similarly, a scale of all 20 cheating items produced congruent results.

13. Since the measures are cross-sectional and time ordering cannot be ascertained it would be inappropriate to make any arguments stating that one potentially caused the other or including one as a predictor of the other in the previous regression models. We simply examine the degree of overlap which may speak to the degree of specialization students have in their academically dishonest behaviors. Though there is substantial overlap between online and analogous, sufficient differences exist to warrant examining separate regression models as was previously done.

14. For example: Storing information in a calculator. This is a behavior that utilizes technology, but not the Internet.

15. We remind the reader that we examined separate measures for differential expectations of social reinforcement and differential expectations of academic reinforcement rather than a single differential reinforcement measure so five social learning measures were evaluated as opposed to four.