cheating

High School Cheating: Within A Context of Competitive Goal Structures

A Paper presentation to MWERA: October, 2002

Rich Hofmann & Larry Sherman

MiamiUniversity

ABSTRACT

Construct validity is established for a two-scale attitude instrument, Rationalizing Cheatingand Intolerance for Cheating, using 169 suburban high school students, 92% of whom self-report as cheaters. Initial results are curious as students report a general lack of endorsement for Rationalizing Cheating but do not endorse Intolerance for Cheating. These findings are discussed within a context of a competitive goal structure. It is concluded that attitude and tolerance for cheating (on homework as well as on tests) is socially constructed and is not associated with general honesty or dishonesty – rather it is moderated by goal structure.

Running Head: Cheating

Rich Hofmann & Larry Sherman

School of Education

350 McGuffey Hall

MiamiUniversity

Oxford, OH45056

We wish to acknowledge the help of Marci Nichols, Keith Brackenridge and Tom Moffitt with early data gathering.

Historically, academic dishonesty (cheating) has been a topic of considerable research. Burton (1963) suggested that cheating had more to do with the particular situation than with the general honesty or dishonesty of an individual. Woolfolk (1998) has likewise stated that "Most students will cheat if the pressure to perform well is great and the chances of being caught are slim." (p. 89). Schab (1980), surveying high school students, listed three reasons for cheating including 1) laziness about studying, 2) fear of failure, and 3) parental pressure for good grades. Thus, certain learning environments are indicated as competitive and are viewed as strongly related to cheating and academic survival.

The Johnson's (Johnson & Johnson 1975) have detailed three learning environments which they describe as goal structures including 1) Cooperative, 2) Competitive and 3) Individualistic. It is the competitive goal structure that serves as the basis for this study. Goal structures are the ways students relate to others who are also working toward a particular goal. This is an interpersonal factor. Kohn (1991) has developed a rationale suggesting the likely hood of cheating to occur in the context of a competitive goal structure, as well as the likely hood of altruistic helping behaviors to occur in a cooperative goal structure. Johnson, et. al., (1975) define a competitive goal structure as one where students believe they will reach their goal if and only if other students do not reach their goal. Woolfolk (1998) describes Learning Goals as the personal intentions to improve abilities and understand, no matter how performance suffers. She goes on to describe Performance Goals as when a personal intention is only to seem competent or perform well in the eyes of others. Under the conditions of a competitive goal structure, performance goals will be encouraged and cheating may be more likely to occur. Thus, the motivation to cheat is directly associated with the competitive goal structure.

The purpose of this study was to discuss high school students‘ perceptions of cheating in two different contexts, specifically with regard to rationalizing cheating and tolerating cheating on homework and tests.

Unlike many of the more recent studies on cheating this study did not solicit faculty input or provide a definition of cheating for the respondents (Evans and Craig 1990; Graham, Monday, O’Brien and Steffen 1994). Furthermore, it is based on high school students as opposed to college students. Interest centered about the structure of attitude toward cheating and its relation to certain demographic variables and self-reported cheating behavior.

Methods
Subjects

The 169 suburban high school students in used in this study came from three different school districts and represented three different grade levels: grade 10, 39%; grade 11, 41%; grade 12, 14%; no grade level provided, 5%. The sample is summarized in Table 41 by gender, grade point average, sports involvement, college versus non-college bound, weekly work commitment external to school. The sample is also described with regard to self-report information regarding cheating activity in Table 1. As part of the data gathering agreement we agreed to neither report nor analyze the data by school districts.

The Instrument

The intention of this study was to develop an attitude instrument assessing high school student attitudes toward various rationales for cheating and consequences for those who cheat. A 15 statement Likert scale with a four point response scale ((strongly disagree, disagree, agree, strongly agree) was developed. The statements of the instrument deal with student endorsement of rationales and consequences for cheating on homework and on tests. The content of the statements was derived from a number of studies that discussed reasons for cheating (Singhal & Johnson 1979; Arkin, Cooper, & Kolditz 1980; Houser 1982; Houston 1983;Stevens 1984;Kibler & Patterson 1988;Cochran & Celebreese 1990;Evans & Craig 1990;Johnson, Brown, & Christopherson 1990; Jendrek 1992;Roberts & Toombs 1993; Roberts & Toombs 1993;Fishbein 1994; Lvotsky 1994; Rost & Wild 1994;DeBruyn 1995). The statements are reported in Tables 2 and 3.

Data Gathering

Data were gathered from the students anonymously within classes. Students were informed of the intention of the instrument and it was noted verbally and in writing on the instrument that responding to the instrument was voluntary and that a respondent could decide not to respond either before or after seeing the instrument or while completing the instrument. A very small number chose not to respond.

The demographic variables used to describe the sample are demographic variables previously identified as being related to cheating: gender; college bound versus non-college bound; having an outside job and the degree of involvement in the outside job (i.e. greater than or less than or equal to 15 hours a week;); general GPA reported as A, B, C, and D.

We intentionally included the demographic questions within the instrument statements so that we could somewhat “privately“ request information about respondent cheating behavior. Students were asked four specific questions each requiring a self-report on some aspect of cheating. They were asked if they first cheated on a test. They were asked if they first cheated on homework. They were asked if they typically cheat on tests. They were asked if they typically cheat on homework. It was felt that a better self-report response rate would occur if the questions were embedded in the instrument.

Analyses

Cheaters and When They Cheat

First Cheat. The two questions requesting information about first time cheating were combined. There were 155 responses to these two yes-no questions. There were 15 respondents who did not respond to either question. It was assumed that these 15 respondents did not remember when they first cheated. It was assumed that respondents who selected ”no” for both questions were clearly non-cheaters. The most frequent response was first cheating on homework, 65%, followed by first cheating on a test, 18%, never cheating, 9%, not recollecting when they first cheated 9%.

The two question specifically requesting information about cheating on homework and cheating on tests were also yes-no questions. These questions asked the respondent if he or she cheated on tests and the second asked about cheating on homework. There were respondents who refused to provide a response for either of these two questions, approximately 6%. Approximately 7% self-reported as only homework cheaters. Those responding “no” to both questions are assumed to be non-cheaters, approximately 8%. Approximately 3% self-reported as only test cheaters. Approximately 77% self-reported as cheating on both tests and homework. Based on this breakdown Approximately 86% of the 169 respondents self-report themselves as cheating and only 8% self-report as non-cheaters. In a later analysis it will be demonstrated that the 6% of the respondents not responding have attitudes most like those respondents who self-report as cheating on both tests and homework, suggesting that approximately 92% of this sample cheats.

A chi-square test of independence between test cheating and first cheat was significant, [(2) =20.57; p< .001], Seventy-three percent of the respondents who indicated that they first cheated on homework also indicated that they cheat on both homework and tests. This is more than might be expected by chance. Sixteen percent of the respondents indicated that they first cheated on tests and also indicated that they cheat on both homework and tests, fewer than expected. Thus, on the basis of this self-report information it is reasonable to assume that most students first cheat on homework and then, eventually, cheat on tests.

Table 1 about here

In Table 1 the demographic groups are briefly described by frequency counts and then analyzed within group for differences in self-identification as cheaters. The non-respondents are not included in these analyses. The large percentage of self-reported cheaters makes the cell sizes associated with the non-cheaters quite small. Because of the smallness of the non-cheaters a significance level of .20 was adopted for interpreting the chi-squares reported in Table 1. Gender and GPA define statistically significant relationships with cheating. Males tend to cheat more on homework than expected and cheat more frequently than females. For GPA there are fewer homework cheaters and fewer test cheaters than expected in the high achievement category, the A category, There is also a greater frequency of homework cheating than expected in the C category of GPA. For Play Sports there are fewer homework only cheaters than expected in the group of students playing sports and more homework only cheating in the group of students not playing sports.

Although not significant, Work hours is also related to cheating. Neither Work, being College bound, nor Class membership appear to be related to cheating. A lack of relationship with cheating does not mean that there is no cheating, rather it means that there are no statistically significant differences in frequency of cheating between categories within each demographic variable.

Establishing Factor Validity

The data were initially factor analyzed using principal components analysis, with a root curve analysis to determine the number of factors, followed with a transformation to oblique simple structure (Feldman, Gagon, Hofmann and Simpson 1994). The resulting exploratory analysis served as an initial hypothesized target for a confirmatory factor analysis. This initial target solution was analyzed using a two step confirmatory maximum likelihood factor analysis based on Bentler’s (1991 – year 2000 confirmatory algorithms version 5.7b) EQS and Hofmann’s (1995) psychometric iterative procedure for refining a targeted hypothesized independent cluster solution.

The 15 statements defined two correlated factors. The first factor was defined by eight statements and the second factor by seven statements. The associated confirmatory statistics suggest that the hypothesized independent cluster solution is statistically adequate. The resulting confirmatory solution defined a Bentler-Bonett nonnormed fit index of .92 and a comparative fit index of .93, both within the accepted range of .90 to 1.00 (Bentler 1991 1980). The Jöreskog index (ratio of chi-square to its associated degrees of freedom, [(89) = 153.01; p< .001) of 1.72 is within the acceptable range of 2 or less (Jöreskog & Sörbom 1978). The average absolute standardized residual associated with the fit of the confirmatory solution is .04, also indicative of a good fit (Bentler 1991). These statistics suggest that the two-factor solution is an excellent non-chance fit to the data defined by the 15-attitudinal statements.

The eight statements defining factor one, as represented in Table 2, address various rationales as ”okay” for cheating on homework and cheating on tests. The structure loadings range from a low of .54the least definitive of those statements associated with the factor, to a high of .79, for the most definitive of those statements associated the factor. This factor represents attitude toward cheating rationales and is named Rationalizing Cheating. The coefficient alpha for the eight statements is .87, suggesting a very homogeneous collection of statements.

Table 2. Factor 1 Attitude toward Rationalizing Cheating.

The seven statements defining factor two, see Table 3., address personal feelings about cheating, more specifically an intolerance for cheating. The structure values for the definitive statements range from a low of .58, the least definitive, to a high of .71 for the most definitive statement. This factor represents attitude toward intolerance for cheating and is named Intolerance For Cheating. The coefficient alpha for these seven statements is .83, suggesting that these seven statements also define a very homogeneous collection of statements.

Table 3. Factor 2 - Attitude toward Intolerance For Cheating.

The average sample response for each statement is also reported in Tables 2 & 3. The statement responses were scaled from 1 to 4, making 2.5 a theoretical midpoint. Any statement mean below 2.5 suggests that generally the respondents did not agree with the associated statement while a statement mean equal to or greater than 2.5 suggests a general agreement with the statement. With the exception of two statements on factor 2 all statements means are below 2.5.

The overall mean score for the statements defining factor 1, Rationalizing Cheating, is 2.19. This suggests that a large majority of the respondents do not endorse the cheating rationales. Similarly, for factor 2, Intolerance For Cheating, the overall statement mean is 2.26, also below 2.50. This too suggests that a large majority of the respondents do not endorse the Intolerance For Cheating statements. However, the higher mean for Intolerance For Cheating suggests that the respondents are relatively more favorable toward Personal Responsibility than toward Rationalizing Cheating.

The confirmatory factor analysis estimated the correlation between the two factors to be -.76, a high negative correlation. The actual empirical correlation between the two factors is -.66. These negative correlations suggest that those who have the relatively strongest positive endorsements with the Rationalizing Cheating tend to have the relatively strongest negative endorsements with Intolerance For Cheating and vice-versa, as would be expected logically.

Demographic Variables and Average Scores on Factor Scales

Utilizing the statement responses for the variables associated with each factor scale scores were computed for each respondent as their average statement response. Two scale scores were computed for each respondent, one for Rationalizing Cheating and one for Intolerance For Cheating. A score above 2.5 should be considered as representing a general endorsement of a factor while a score below 2.5 should be considered representing general disagreement with the factor. The demographic variables associated with this study are listed in the first column of Table 4. The second column of the table lists the categories associated with each of the demographic variables. Subsequent columns are associated with the means and standard deviations of the groups for the two factors, Rationalizing Cheating and Intolerance For Cheating.

Table 4.Means and standard deviations and ANOVAs on attitude factors by level within demographic variable

As reported in Table 4 there are significant attitudinal differences between those respondents who: cheat and those who don’t; those who work more than 15 hours a week and those who don’t; are males and those who are female; generally are A level students and those who are not. The only mean scores in this table (excluding those who didn’t respond to the cheating questions) that are in the agree range are associated with the D students who endorse cheating behaviors, and with those students who have not cheated who provide a strong endorsement for assuming an intolerance for cheating on tests and on homework.

In order to establish construct validity for the two attitude scales it was important for them to discriminate between cheaters and non-cheaters, which was demonstrated.

A fifth category associated with cheating, but not included with the ANOVA is “didn’t respond”. These were the respondents who did not respond to the two questions regarding present cheating activity. A comparison of their mean responses to those of the other categories associated with cheating suggests that they are most like the respondents who cheat on both tests and homework. It is also interesting to note that their means are very similar to those of the D students. Of the nine non-responding students four were in the A grade category and three were in the D grade category. Four of the nine indicated they did not have a job while five of them did not respond to the outside job question. Five of the non-respondents were females and three were males. Three indicated they were college bound and four indicated that they were not college bound.

Brief Summary & Discussion

The following statements provide a summary of the analysis findings.

• Cheating is so pervasive, 95% of the males and 87% of the females, that it is somewhat problematic to look for differential cheating rates by categories within demographic variables. It is not possible, on the basis of this study, to conclude that there is a single demographic category where there is not widespread cheating.