# «A CROSS-CULTURAL STUDY OF THE RELATIONSHIPS BETWEEN EPISTEMOLOGICAL BELIEFS AND MORAL JUDGMENT AS A PSYCHOLOGICAL FOUNDATION FOR MORAL EDUCATION by ...»

24; α =.63) Working on a problem with no quick solution is a waste of time. (.76) If you haven’t understood a chapter the first time through, going back over it won’t help. (.66) If you don’t learn something quickly, you won’t ever learn it. (.58) Factor 2: Omniscient Authority (Eigenvalue = 2.26; α =.65) People should always obey the law. (.73) When someone in authority tells me what to do, I usually do it. (.72) Children should be allowed to question their parents’ authority. (.56) People who question authority are trouble makers. (.48) Parents should teach their children all there is to know about life. (.46) Factor 3: Certain Knowledge (Eigenvalue = 2.02; α =.63) Absolute moral truth does not exist. (.74) Truth means different things to different people. (.64) What is true today will be true tomorrow. (.61) The moral rules I live by apply to everyone. (.42) Sometimes there are no right answers to life’s big problems. (.34) Factor 4: Simple Knowledge (Eigenvalue = 1.85; α =.44) Too many theories just complicate things. (.65) The best ideas are often the most simple. (.62) Things are simpler than most professors would have you believe. (.46) Factor 5: Innate Ability (Eigenvalue = 1.53; α =.49) Some people will never be smart no matter how hard they work. (.75) Really smart students don’t have to work as hard to do well in school. (.51) How well you do in school depends on how smart you are. (.47)

each of the five factors described earlier. All items but item 15 loaded unambiguously on only one factor and were related directly to the construct in question. Item 15 (i.e., How well you do in school depends on how smart you are.) with a loading of.47 on the innate ability factor also loaded on the quick learning factor (.43). In this study, item 15 was interpreted as an innate ability factor that led to a relatively high item-to-factor loading and corresponded to the factor structure of the EBI and EQ. In conclusion, analysis of the items using the larger sample data indicated that the Korean translation of the EBI was suitable for further use.

With the U.S. sample, the EBI was analyzed using the same factor-analytic procedures used to analyze the Korean translation of the EBI. This analysis yielded eight factors with eigenvalues greater than one that explained 63.5 % of the total sample variation. The first five observed factors corresponded to the five epistemological factors described by Schommer (1990) and Bendixen, Schraw, and Dunkle (1998; Schraw et al., 2002). These factors explained 44.4% of the total sample variation. The factor structure, loadings, eigenvalues, and values of coefficient alpha for each of the five factors are reported in Table 3.5.

All items but item 10 loaded unambiguously on only one factor and were related directly to the construct in question. Item 10 (i.e., Too many theories just complicate things.) with a loading of.69 on the simple knowledge factor also loaded on the quick learning factor (.31). In this study, item 10 was interpreted as a simple knowledge factor that led to a relatively high item-to-factor loading and corresponded to the factor structure of the EBI and EQ. The internal consistency using coefficient α was equal to.69 for

authority, and.53 for innate ability. The overall alpha for the EBI was equal to.79. A comparison of internal consistency coefficients indicated that the results of the present study and Schraw, Bendixen, and Dunkle’s study (2002) were quite similar on this dimension, although neither result produced factors that were highly reliable.

** Table 3.5 Factor Structure of the Epistemic Beliefs Inventory (U.**

S. Sample N = 191) Factor 1: Simple Knowledge (Eigenvalue = 2.50; α =.69) Things are simpler than most professors would have you believe. (.77) The best ideas are often the most simple. (.70) Too many theories just complicate things. (.69) Instructors should focus on facts instead of theories. (.37) It bothers me when instructors don’t tell students the answers to complicated problems. (.30) Factor 2: Quick Learning (Eigenvalue = 2.42; α =.65) If you haven’t understood a chapter the first time through, going back over it won’t help. (.75) Working on a problem with no quick solution is a waste of time. (.73) If you don’t learn something quickly, you won’t ever learn it. (.47) Factor 3: Certain Knowledge (Eigenvalue = 1.91; α =.65) Absolute moral truth does not exist. (.73) Sometimes there are no right answers to life’s big problems. (.70) The moral rules I live by apply to everyone. (.46) What is true today will be true tomorrow. (.42) Factor 4: Omniscient Authority (Eigenvalue = 1.84; α =.66) People should always obey the law. (.80) When someone in authority tells me what to do, I usually do it. (.80) People who question authority are trouble makers. (.53) Factor 5: Innate Ability (Eigenvalue = 1.81; α =.53) Some people will never be smart no matter how hard they work. (.73) Really smart students don’t have to work as hard to do well in school. (.72) Smart people are born that way. (.41) Syllogisms Syllogisms (Appendix C) were used to provide a measure of cognitive reasoning.

The syllogisms test developed by Bendixen et al. (1998) includes 12 items in a twostatement logical form that provides premises to the test taker (e.g., “No mammal is a

asked to select a valid conclusion from among four alternatives (e.g., “Some quadrupeds are not reptiles.”). Premises differ from item to item on several dimensions, including whether they are positive or negative (e.g., “No mammal is a reptile”), include universals such as “all” or “always,” are abstract (e.g., “Some K’s are P’s”), and are empirically plausible (e.g., “Glass always bounces when it falls”).

To measure Korean students’ cognitive skill, a new Korean translation of Bendixen, Schraw, and Dunkle’s (1998) 12-item Syllogisms test was developed. In order to minimize possible linguistic and cultural discrepancies between the original scale and the new scale, the same procedures used with the EBI were applied.

Both the difficulty and discrimination indices (i.e., the point biserial and biserial correlations) via a computer program for classical item analysis (CIA) (Kim, 1999) were determined for twelve items. The computer program CIA provides classical item analyses. The results of the item analysis from the sixty undergraduates enrolled in a cyber-ethics class at Chuncheon National University of Education are listed in Table 3.6.

In the Table 3.6 below, PROP indicates the proportion of examinees who selected the correct response. RPBI is the point biserial correlation between the dichotomous item score and the total score, whereas RBIS is the biserial correlation between the same variables with the bivariate normal assumption (RBIS values in general are slightly higher than that of RPBI). It is suggested that items with discrimination indices (RPBI) below.25 should be rewritten or discarded (Payne, 1997). The difficulty index (PROP) of.625 might be considered optimal because there are four choices. The results reported in Table 3.5 showed that the discrimination levels for all items were acceptable, with the

Two internal consistency estimates of reliability were computed for the syllogisms test: coefficient alpha and a split-half coefficient expressed as a SpearmanBrown corrected correlation. The split-half coefficient was.60, indicating satisfactory reliability. However, a coefficient alpha of.452 indicated a relatively low degree of reliability. With regard to a relatively low degree of internal-consistency reliability, it should be recognized that internal-consistency reliability estimates are the results of an

among the items (Pedhazur & Schmelkin, 1991). By and large, alpha becomes increasingly larger, as the number of items is increased. Therefore, because the number of items in Bendixen, Schraw, and Dunkle’s 12-item Syllogisms is relatively small, the measure has a low estimate of internal-consistency reliability.

Demographic Questionnaire Demographic questions (Appendix D) include: (a) age, (b) gender, (c) academic major, (d) current GPA, (e) ethnic background, and (f) educational level.

For the investigation of the relationship among epistemological beliefs and moral reasoning between Korean and U.S. college students, a multiple regression analysis was employed. Especially, an all-possible multiple regression model was used to select the best regression model from among all possible regressions. Although popular, the stepwise procedures have been criticized for their reliance on multiple tests, for their inappropriate use of the F distribution, for their claim of identifying the best subset of explanatory/predictor variables, and for yielding nonreplicable models (Henderson & Denison, 1989; Huberty, 1989; Olejnik, Mills, & Keselman, 2000; Snyder, 1991;

Thompson, 1995; Wilkinson, 1979). As an alternative to the stepwise procedures, many methodologists have recommended that researchers examine all of the possible regression models that might be developed from the list of possible explanatory/predictor variables (Olejnik et al., 2000; Thompson, 1995). The two criteria used most frequently are the value of R2 achieved by the least squares fit and the Cp statistic. Mallow’s Cp statistic [Cp = RSSp/s2 – (n – 2p), where RSSp is the residual sum of squares from a model containing

residual mean square from the largest equation postulated containing all the Z’s and is presumed to be reliable unbiased estimate of the error variance σ2 (Draper & Smith, 1998; Montgomery & Peck, 1992). The expected value of Cp is p when model bias is 0, so one definition of the best model is the one in which the absolute value of the difference between Cp is p is smallest.

An examination of the individual variables entered in the multiple regression will reveal (a) whether epistemological beliefs are related to moral reasoning over and above the effects of other critical variables (i.e., age, education, gender, and basic reasoning skills) in each group, and (b) which of five epistemological beliefs (i.e., certain knowledge, innate knowledge, quick learning, simple knowledge, and omniscient authority) explains the greatest amount of sample variation in each group.

Valid data analyses require several important assumptions: (a) the observations should be independent (independence), (b) the observations on the dependent variables should follow a multivariate normal distribution in each group (multivariate normality), (c) homogeneity of the dependent variable variance across the independent variable score possibilities should be satisfied (variance homogeneity), (d) there should be a significant relationship between a set of dependent variables and a set of covariates (linearity), and (e) independent variables should not be perfectly correlated (collinearity) (Huberty & Petoskey, 1999; Pedhazur, 1997).

Before the application of the assumption check-up, diagnostic procedures to detect any observations that demonstrate real uniqueness in comparison with the remainder of the population (i.e., outliers) were used. For these purposes, the data were

which is a measure of the distance in multidimensional space of each observation from the mean center of the observations, was used (Stevens, 1996). As a result of these diagnostic tests, no observations seemed to demonstrate the characteristics of extreme outliers.

Then, the assumptions of multivariate normality and variance homogeneity were checked to see if there were any violations of the assumptions for multiple regression analysis. Multivariate normality was assessed by examining a normal probability plot.

The plot was virtually linear, indicating that the condition of normality was satisfied.

Also, homogeneity of Y-variable variance across the X-variable-score possibilities was assessed by examining a residual plot. The plot of residuals showed that the condition of variance homogeneity was met.

Collinearity was diagnosed by examining the variance inflation factor (VIF), as it indicates the inflation of the variance of b as a consequence of the correlation between the independent variables (Pedhazur, 1997). The variance inflation factor (VIF) is at a minimum (1.00) when the correlations between the independent variable in question with the remaining independent variables are zero. The test for the assumption of collinearity indicated that all the VIFs were close to the minimum with a range of 1.11 to 1.63, thus showing that the condition of collinearity was satisfied.

For the two continuous variables (moral reasoning and cognitive skill), measures used are simply the test scores. The two dichotomous variables (academic major and gender) were converted into sets of variables by dummy variable coding. In the case of sixteen academic majors as a categorical variable, a dummy variable transformation

most reasonable when the number of predictor variables is not too large (Olejnik et al., 2000), sixteen academic majors were grouped under two categories (sciences and nonsciences). Mathematics, science, and computer science education were categorized into sciences, while the remaining thirteen majors were categorized into non-sciences. For the two categorical variables with ordered categories (epistemological beliefs and educational level), this study used integer scaling to obtain variable measures. For example, for epistemological beliefs with five ordered categories, a “1” was assigned to the lowest category, …, and a “5” to the highest category.