The participants of this study were current Kyung Hee University medical students in their first, second, and third years. Of 288 total students, except those who responded falsely (four students) and did not respond, there were 174 male students and 114 females.

The questionnaire that professor Noe created in 1988 at the University of Minnesota is composed of 29 items. However, this study was conducted with the Korean version (Appendix 1). All the items were measured using a 5-point Likert-type scale ranging from 1 (strongly disagree) to 5 (strongly agree). The original questionnaire was used after receiving prior consent from Professor Noe by email. Each item corresponds to an individual function, but the overall functions are largely divided into psychosocial (15 items) and career development functions (14 items). This classification of the functions is cited from the classification of mentoring functions [6], by Kathy Kram at Boston University. According to the classification of the original questionnaire, the psychosocial functions include acceptance and confirmation, role model, counseling and friendship, whereas the career development functions include coaching, protection, exposure and visibility, sponsorship, and challenging assignments. The survey questionnaire was provided as supplementary material to this article.

The sample estimation process should be conducted robustly in order to enhance the quality of studies. Larger samples are superior to smaller samples in many ways. Researchers can minimize the probability of committing errors, increase the generalizability of their findings, and maximize the accuracy of their conclusion through employing larger samples. According to researchers, 10 cases for each item should be sufficient to perform an exploratory factor analysis [7]. This study has a sufficient number of subjects with a total of 288 valid subjects.

This study used the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s test of sphericity to measure the adequacy of samples in terms of the distribution of values for the execution of factor analysis. Exploratory factor analysis was used to examine whether the application of the adapted questionnaire to the Korean medical students can correspond to the same categorization of these functions. In the exploratory factor analysis, the principal factor method and the varimax method were used. This study conducted the confirmatory factor analysis to examine the model validity of the factors that are tied to the same group from the exploratory factor analysis. Also, since it is important to retain the reliability and validity in the confirmatory factor analysis, Cronbach α statistic was calculated and construct, convergent, and discriminant validity were conducted. The confirmatory factor analysis was performed with the use of chi-square, root mean square error of approximation (RMSEA) and comparative fit index to examine the fitness and correspondence of the model as a whole. Standard regression weights were used for the construct validity. The convergent validity was examined by using the construct reliability and variance extracted index. The discriminant validity was examined by using coefficient of determination and variance extracted index. This study used SPSS version 21.0 (IBM Corp., Armonk, USA) and Amos 20.0 (IBM Corp.), respectively, and the significance level was set for 0.05.

The participants of this study were composed of 60.4% male students and 39.6% of female students. The number of people in each year is as follows (Table 1).

In this study, the KMO test and Bartlett’s test of sphericity were used to measure the adequacy of the collected samples prior to the execution of exploratory for its suitability in the factor analysis. The outcome for the KMO test was 0.963, so the conformity was confirmed to perform the factor analysis. Also, as the stage to figure out common factors and to indicate the suitability of the factor analysis model, Bartlett’s test of sphericity generated 7,490.33 (0.000), showing the statistical significance. Since the variables are correlated with each other, this study can use the principal factor method. According to these grounds, the principal factor method, which uses varimax rotation, was performed as the factor extraction method. The table of the rotational component matrix is shown in the Table 2. The factor analysis was conducted with the principal component analysis using of varimax rotation to group common factors within 29 items. The result of the exploratory factor analysis within the 29 items of the original questionnaire was different from the result of this Korean version research form. The original questionnaire was classified into two parts: psychosocial and career development functions.

However, this new study showed different outcomes; “friendship functions” was added. As a result, the researchers of this study decided to group the 12 items except two items in “psychosocial functions,” because the items were originally in the same category. With regard to the factor 2, the 12 items except one item were overlapped with the original questionnaire, so we decided to call them “career development functions” as the original ones are named. The last three items were classified as “friendship functions” in the original questionnaire as sub-functions. Therefore, we emphasize “friendship functions” as new functions of this study. The total explanation power of the three factors was 67.819% and each Eigen values were higher than 1.0 (Table 2).

Previously, through the exploratory factor analysis, we determined that the two categories the original research form proposed were inappropriate in South Korea. Accordingly, the confirmatory factor analysis was performed to examine the suitability of the three factor structures of the mentoring program functions that were discovered in the exploratory factor analysis. The fit index obtained in the confirmatory factor analysis is shown in the Table 3. The result of the structure equation analysis showed chi-square statistics, the degree of freedom (df), and p-value. Generally, goodness of fit is interpreted as statistically significant if p-value is higher than 0.05 and p-value gets higher when the statistics of chi-square gets smaller. The probability of the statistics of chi-square also increases when the size of samples gets higher. In other words, since the p-value is influenced by the chi-square and the size of samples, it cannot be said that the goodness of fit is lower in a situation where the size of samples is high and the data with many variables [8]. Therefore, it is important to make a decision regarding the goodness of fit with other various indexes.

There is RMSEA that is not affected by the size of samples. RMSEA is statistics that equals chi-square adjusted by using df and size of sample. Mostly, if RMSEA is lower than 0.08, it demonstrates that the goodness of fit is considerably high. Especially, within a 90% confidence interval, it demonstrates higher goodness of fit [9] when the upper value is lower than 0.08. The confirmatory factor analysis in this research demonstrated a favorable RMSEA (0.101), showing in 0.095 to 0.106 range in the 90% confidence interval. Also the confirmatory factor analysis demonstrated a reasonable comparative fit index as 0.853. A standard coefficient estimation shown in Fig. 1 demonstrates statistical significant in the level of significance 0.05. Standard coefficient estimation of the sub-function of three factors indicated 0.80 to 0.83.

Cronbach α (Table 4) regarding the 29 items and the three factors were calculated to examine the reliability that measured the mentoring programs. As a result of the internal consistency of the scales, Cronbach α statistics for the factor 1, 2, and 3 were 0.955, 0.954, and 0.814, respectively. The overall Cronbach α statistic including all the 29 items is 0.971, demonstrated competent level of the reliability.

Construct validity was secured because standardized regression weights (Table 5) in all items were higher than 0.5. Construct reliability (Table 6) was confirmed because the internal consistency of the observed variables was higher than 0.7. Convergent validity was retained since all variance extracted indices was higher than 0.5. Regarding discriminant validity, a comparison of variance extracted indices (Table 6) and coefficients of determination (Table 7) shows that three of them are higher and three of them are lower, therefore it secured partial discriminant validity.