Research on Education Effect for Chinese Medical Students' Learning Style Based on Kirkpatrick Model and Structural Equation Model

Background: The recognition of learning style was not optimistic in China. The evaluating indices of education effect were vague and adopted some ready-made indices such as teaching satisfaction or academic achievement. Those indices have their own basis and significance, but one mere index can not reflect education effect comprehensively. Methods: We used Kirkpatrick model to appraise Chinese medical education effect by selecting appropriate indices for the four levels and structural equation model (SEM) to explore the correlation among those indices after a survey on 17460 medical students. Results: 1) Kirkpatrick model: Level 1 students’ satisfaction scored 4.19±0.993; Level 2 learning motivation 1.95±0.568; Level 3 the Visual 0.227±0.131, Aural 0.300±0.138, Read/Write 0.273±0.138 and Kinesthetic 0.324±0.144 for sensory learning style, Extroversion 0.531±0.207, Sensing 0.597±0.186, Thinking 0.585±0.207, Judging 0.662±0.225 for personality type and Concrete Experience 32.03±6.394, Reflective Observation 33.41±6.486, Abstract Conceptualization 34.13±6.650, Active Experimentation 34.00±6.600 for empirical learning style; Level 4 learning attitude 2.68±0.781, learning expectation 2.14±0.953 and academic achievement 2.43±0.793. 2) SEM: the fitting indices were 0.687-0.581. Most of the paths were valid except for VARK, T and J parts (t < 2, P > 0.05). Conclusions: Kirkpatrick model with selected indices evaluated comprehensively the education effect (learning style was one important index); there were actually relations among those indices; based on the importance tested by SEM and students’ satisfaction obtained from the survey, we sorted the college cultivating

works into four types, among which type B (important but unsatisfactory work) urge most to be strengthened.

Background
Learning style is a learner's persistent individual learning method and tendency [] .
Studying learning style is both helpful for students to improve learning ability and efficiency and for teachers to promote teaching effectiveness and pertinence.
Among those different categorizations of learning style the five taxonomy system The quality of medical education is an eternal topic, while most papers studied it only from a certain aspect, such as students' satisfaction, academic achievement and learning motivation etc., those indices have their own basis, but one mere index was too weak to comprehensively evaluate the education effect. Besides, the relations among those indices were rarely reported. but it only worked well in the first two levels, the latter two levels, which revealed the training effect more profoundly, were seldom used for lack of practical indices.
In recent years, K model has begun to be applied in education, such as evaluating employment guidance training in colleges, principal training in primary and secondary schools [] and MBA courses, [] etc., while its application in evaluating education effect for Chinese medical students' learning style has not been reported. Regarding the cultivating of medical students as a training, we selected alternative evaluation indices for the four levels of K model on the basis of experts' repeated demonstration to comprehensively assess the education effect of Chinese medical students for their learning style: students' satisfaction with college cultivating works obtained by self-compiled satisfaction questionnaire (Level 1); students' mastery of professional knowledge and skills represented by their academic achievement of theoretical and clinical skills collected from the administrative department (Level 2); the change of students' learning behavior and attitude with learning style and motivation as the main body gained by surveying with questionnaires of VARK, MBTI, KLSI and learning motivation, learning expectation, as well as symposium and individual interview of teachers, educational managers and students (Level 3); and finally students' contribution to economical and social development demonstrated through their comprehensive quality which represented by participation in the second classroom, quality and quantity of publications, selfmanagement ability and lifelong learning consciousness (Level 4). Based on the information collected at those four levels, the education effect of Chinese medical students concerning learning style was comprehensively evaluated.
On the other hand, since satisfaction, learning motivation, learning style and comprehensive quality are all abstract concepts [] ; they have the character of latent variables with measuring difficulty. Considering SEM's technical characteristic of allowing independent and dependent variables to contain measurement errors, this study, after investigating 17460 students from 11 medical colleges covering different regions of China, constructed SEMs containing multiple latent variables to explore relations among the indices of four levels in K model, namely, the correlation among students' satisfaction with college cultivating works, their learning motivation, learning style and comprehensive quality (including academic achievement) and then the cultivating works tested by SEM with good fit and great contribution were selected and further sorted to provide a reference for medical colleges to improve the education.

Methods
Firstly, a comprehensive questionnaire was compiled, it contained three widelyused, authoritative learning style inventories (VARK, MBTI and KLSI), which have high reliability and validity, as well as scales to measure students' satisfaction, learning motivation, learning attitude and learning expectation, etc., specifically it included the following parts: Others: including students' learning expectation and learning attitude, etc., selfassessment was adopted to streamline the questionnaire, 1-4 points were recorded accordingly from naturalness to excellence in learning expectation and slothful to extremely hardworking in learning attitude.
Cronbach's alpha reliability coefficient was applied in this study and KMO factor analysis was adopted as the structural validity index. The reliability and validity of each scale were shown in Error of Approximation). If the model fitting is bad, the model needs to be revised.
The GFI is between 0 and 1, the closer to 1, the better the fitting is, it's generally required to above 0.80 or at least close to 0.60, which indicate the model fits well. [15] The preliminary SEM included four parts: The first was students' satisfaction with college cultivating works, recorded as f4. In addition to the overall satisfaction (recorded as OS), 16 items (recorded as SS1-16) were subdivided in order to make them more practical and operable. Each observed variable corresponded to a specific question in the questionnaire and was evaluated by Likert 5-level scoring, ranking 1 to 5 from "no satisfaction" to "absolute satisfaction". The second was the intensity of students' learning motivation (non-intellectual factors), recorded as f5. The third was learning style of three types, namely, sensory learning style (related to physiological element) recorded as f1 and measured by VARK with the standardized score (between 0 and 1, actual score divided by full score) of type V, A, R and K being the index; personality type (related to psychological element) recorded as f2 and measured by MBTI with standardized score of the first endpoints (E, S, T, J) in the four dimensions (E-I, S-N, T-F and J-P) (the standardized scores of the other endpoints was 1 minus the first ones) being the index; empirical learning style (related to psychological element) recorded as f3 and measured by KLSI with the scores of four stages in empirical learning cycle (actual score divided by full score) being the index. In order to explore more accurately the education effect for learning style, we took these three types of learning style as indices of Level 3 in K model. The fourth was comprehensive quality, recorded as f6, the observed indices included learning attitude, learning expectation and academic achievement, accordingly recorded as ATTI, EXPE and SCORE.
The theoretical hypothesis was that the better college cultivating works are (the higher the evaluation score of satisfaction), the better its adaptability to students' learning style, the higher students' learning motivation, which contribute to their comprehensive quality; meanwhile motivation and appropriate learning style could also promote comprehensive quality; besides, motivation may directly contribute to learning style.

Data collection
The data mainly came from the questionnaire and students' final examination scores of last term collected from the administration department according to the Student ID filled in on the questionnaire, scores below 70 (inclusive), 70-80, 80-90 and 90-100 were classified from bad to excellent, accordingly scored 1-4 points.

Results
The four levels of K model Level

The relations among the levels of K model
When VARK was taken as the representative of Level 3 in K model, SEM 1 (Figure 1) based on the theoretical hypothesis was constructed to explore relations between latent variables of cultivating works, learning motivation, sensory learning style and comprehensive quality, as well as between latent variables and observed variables.
The GFI of SEM 1 was 0.5742 and AGFI 0.5357 which were slightly low, the model didn't fit well with the sample data model, thus we revised it by removing the f1 and overall satisfaction part and constructed SEM 2 ( Figure 2), the GFI was 0.6870, the model reflected relations between variables well and was adopted as the final model after testing. It was showed that the measures proposed in this study of improving students' comprehensive quality by promoting students' learning motivation was effective and meaningful (being successfully-tested by SEM with high fit index), but sensory learning style was irrelevant to them.
By maximum likelihood estimation, fitting results of the measurement model were obtained and collated into  The factor loads in SEM were all statistically significant (t>2, p<0.05). It could be seen from the standardized regression coefficient that college cultivating works had an influence on students' learning motivation (0.12069), and motivation had certain negative influence on comprehensive quality (-0.29132), excessive motivation could lead to decline of comprehensive quality. Therefore, the influence of cultivating works on students' learning motivation and comprehensive quality was obvious.
However, none of the f1 part appeared in the model, which meant that sensory learning style was irrelevant to learning motivation, comprehensive quality and cultivating works.
By combining the standardized regression coefficients of measurement model and SEM, the path graph and standardized solution of SEM 2 for cultivating works, learning motivation, sensory learning style and comprehensive quality were drawn (Figure 2), each solid-line path represented a corresponding hypothesis and the path parameters were listed next to them.

"Importance -Satisfaction" analysis
It could be found that the satisfaction obtained by general statistical method (Chisquare test, variance analysis, etc.) was not consistent with the importance gained by SEM based on standardized regression coefficient. In order to clarify the importance and satisfaction of specific cultivating works, we sorted them according to their importance to the superior indices and students' satisfaction on them. When MBTI was taken as the parameter of Level 3, SEM 3 (Figure 3) for cultivating works, learning motivation, personality type and comprehensive quality was preliminarily built. The E-endpoint of the first dimension was used as contrast to identify and modify the model. The GFI was 0.5870 and AGFI 0.5462, which basically met the model adaptation standard, the hypothesized causal model was acceptable and basically satisfied with the sample data model. Table 4 shows the fitting results, the regression path hypothesis test showed that the path coefficient difference were all statistically significant (t>2, P<0.05) except f5→f2 between latent variables (f4→f2 just came to pass) and T→f2, J→f2 between latent variable and observed variable. As expected, these variables could define well corresponding latent variables.
After the removing of bad fitting part, the path graph and standardized solution of SEM 4 for medical students' cultivating works, learning motivation, personality type and comprehensive quality were drawn (Figure 4, with E as the reference). It shows that relation between cultivating works and students' personality type was slight, For personality type, the Extroversion (E) was the most important (0.70722), far more than other dimensions; Sensing (S 0.02493) only had a slight contribution, while Thinking (T) and Judging (J) had no relation with personality type.
When KLSI was taken as the parameter of Level 3, SEM 5 ( Figure 5) for cultivating works, learning motivation, empirical learning style and comprehensive quality was preliminarily built. CE was used as the contrast to identify and modify the model.
The GFI was 0.5809 and AGFI 0.5395, basically meeting the model adaptation standard, the hypothesized causal model was acceptable and satisfied with the sample data model. Table 5 shows the fitting results, the regression path hypothesis test showed that the path coefficient difference were all statistically significant (t>2, P<0.05) except f5→f3 between latent variables. As expected, these variables could define well corresponding latent variables.
After the removing of bad fitting part (f5→f3), the final path graph and standardized solution of SEM 6 for cultivating works, learning motivation, empirical learning style and comprehensive quality were drawn (Figure 6), it shows that cultivating works were related to empirical learning style, but learning motivation was not.

Selected indices evaluated comprehensively education effect and broke through K model's previous practical limitation
As mentioned above, there were previously many indices to evaluate education effect, each had its own basis and significance, but one mere index was unilateral.
The building of K model solved the dilemma and unified the indices of different

There were actually relations among the indices
We not only used K model to evaluate Chinese medical education effect, but also applied SEM to explore the correlation. It was showed that there were actually potential relations among the indices of four levels in K model.
When VARK being applied (Model 2), cultivating works had influence on students' learning motivation (the standardized path coefficient was 0.12069), and motivation had negative influence on comprehensive quality (-0.29132), but there was no relation between sensory learning style and cultivating works, comprehensive quality.
When MBTI was adopted, cultivating works had a promoting effect on motivation

The observed variables measured corresponding latent variables well
Confirmatory factor analysis model could judge whether the exogenous observed variables had significant measurement effect on corresponding latent factors, the explanatory ability or prediction degree (the extent latent variables affected observed variables) can be manifested in factor load [16]. Only when the factor load reached a significant standard, could it be indicated that the measured variable effectively reflect the latent variable it measured, the measurement had good validity [17]. To increase the practical operability of education management, we subdivided the cultivating works into 16 specific ones, the effective works been tested by SEM were put forward concretely. According to their importance (i.e. path coefficient in SEM) and implementing performance (satisfaction), these works were divided into four types: A, advantage approach, i.e. the "important and satisfactory" work, the resources input of which was suggested to be maintained to keep existing advantage, there was only one item falling into this category, i.e. the guidance and service on students' study and life in college; B, repairing approach, i.e. the "important but unsatisfactory" work, the resource investment of which should be strengthened and special attention should be paid in order to repair the existing "loopholes" and improve the overall satisfaction, 7 items fell into this category, naming curriculum resources, whether students' choice being respected in teaching management and reform, practical teaching, the campus culture, the analysis and guidance the college provided on graduates' employment and development, canteen and other food support, the condition of dormitory and other accommodation, these works need to be noticed urgently and improved in the first sequence; C, maintaining approach, namely "satisfactory but unimportant" work, the investment could be reduced as appropriate on the basis of maintaining satisfaction or be temporarily shelved, these items are: suitability of requirements on students' graduation, importance the leaders attached to teaching work, teaching level, seriousness of teaching, the opening of teaching facilities, the college ethos and learning atmosphere; D, opportunity approach, i.e. "neither important nor satisfactory" work, there was only one item, i.e. assessment methods, which must be reformed and improved in the second sequence.
Students' satisfaction evaluation could provide first-hand material for decisionmaking of education administration departments and colleges; it had long been one of main methods for efficient education management in developed countries [18].
There were still two problems need to be clarified: 1)satisfaction alone cannot judge whether the item should be raised in resource input, at least its importance should be taken into consideration; 2)satisfaction, an emotional index, couldn't remain unchanged for a long time, this required colleges to take satisfaction survey as a regular work in order to improve the pertinence of decision-making.

Conclusions
Different evaluating indicators were showed simultaneously by using K model to evaluate the education effect for Chinese medical students' learning style. These indicators were proved mutual-connected by SEM, cultivating works had influence on students' learning style, which is mainly manifested in personality type and empirical learning style, appropriate cultivating measures could also improve students' learning motivation and then improve their comprehensive quality, which affirmed the effect and significance of cultivating works in medical colleges. The effective works been tested by SEM were put forward concretely and sorted into four types according to their importance (path coefficient) and students' satisfaction; among which type B, those important but unsatisfactory works, urge most to be strengthened.
Despite the significance and innovations of this study as mentioned above, it has deficiencies which provide future research directions. Firstly, indices at Level 2 and 4 in K model need to be further excavated. Since the original intention of level 2 was to evaluate trainee's improvement in knowledge and skills, learning motivation could be an alternative index as it's one of the main factors affecting student's knowledge and ability development, but it's not the best one. Also students' academic achievement, learning attitude and expectation at level 4 could represent learning effectiveness, they were still not the best and comprehensive indices since the original intention was to assess students' contribution to economical and social development. Secondly, classifying cultivating works based on the importance and satisfaction is meaningful for colleges to manage education reasonably and orderly, but specific cultivating works must be constantly adjusted and improved with the development of the times.

Ethics approval and consent to participate
Th ethical committees of Naval Medical University and all other universities involved in this study have verified that all the students who voluntarily participated in this survey were well-informed in advance. They were assured that their responses had no effect on their grades and were completely confidential. We considered the return of a completed questionnaire to represent the provision of informed consent and other ethics approvals were not required given the study design.

Consent for publication
Not applicable

Availability of data and materials
All data generated or analyzed during this study are included in this published article.

Competing interests
The authors declare that they have no competing interests.

Funding
This study was supported by grants from National Natural Science Foundation of    Figure 1 The path graph of preliminary SEM 1 for cultivating works, learning motivation, sensory learn The path graph and standardized solution of the final SEM 2 for cultivating works, learning m The path graph of preliminary SEM 3 for cultivating works, learning motivation, personality t The path graph and standardized solution of the final SEM 4 for cultivating works, learning m The path graph of preliminary SEM 5 for cultivating works, learning motivation, empirical lea The path graph and standardized solution of the final SEM 6 for cultivating works, learning m