A Study On Nutrition Knowledge, Attitude And Behavior Of University Students By Log-linear Model

Purpose Contingency tables are usually used in medical study. The method for association of variables in low-dimensional Contingency tables is Chi-square test, for multidimensional Contingency tables, although the methods of combination for reducing dimension, and hierarchical Chi-square test are usually used, there are some problems that the frequency of each cell will be getting smaller even zero if the number of layer in contingency table is getting bigger, and illogical combination. It is helpful to analyze the association and interaction effect between the variables in multidimensional contingency table with multivariate analysis. The good health of a student mainly depends on better and regular dietary behavior. It widely exist some bad behavior among many students in present, this behavior has influenced not only many students'state of health, but also state of psychology and spirit. The nutrition knowledge level and the dietary attitude also mainly influence the dietary behavior. A study on nutrition knowledge, attitude and behavior of university students, especially the association and interaction effect between them, will be helpful to find their principle and offer theoretic basis for the better dietary behavior. In present, the statistical method in study of nutrition knowledge, attitude and behavior of university students is mostly univariate analysis, to draw a conclusion by way of the professional knowledge. This kind of method is not well in analysis of the association and interaction effect between several variables, because it probably dissevers the organic relations between these variables. Especially for multidimensional contingency table, it will probably get an instable, uncertain even incorrect conclusion because of illogical combination for contingency table and over many tiny cell frequencies by over many layers. The main purpose of this study is to improve the traditional statistical method with log-linear model that belong to multivariate analysis method in the study of nutrition knowledge, attitude and behavior of university students, to study the association and interaction effect between the several variables in the multidimensional contingency tables, to compare with the traditional statistical methods for confirmation the same conclusion, and to point out the advantages and disadvantages for log-linear model. Material and methods Material: "The Field Survey for Nutrition Knowledge, Attitude and Practice (K-A-P) about University Students"A questionnaire survey, design by K-A-P model. 949 students were drawn from three compuses (Xinmin, Nanling, Qianwei) in Jilin University with a cluster sampling method, involved four specialities (medical, foreign language, chemistry, automobile), two grades (2001 and 2002). The data was inputted computer with Visual FoxPro 6.0 software. Methods: Univariate analysis: Chi-square test, Kruskal-Wallis Test,Mann-Whitney Test, Spearman rank correlation, Linear-by-Linear Association; Multivariate analysis: log-linear model Statistical software: SAS for Windows Release 8.2 and SPSS for Windows Release 10.0.1. Results The preparative analysis with univariate analysis method gained the almost same results as the conclusions from the earlier papers, these results may apply to confirmation and comparison in late analysis too. Using log-linear model to analyze the data aiming at different contingency tables, concluding following results: 1. Two-way contingency table The log-linear model is the same as Chi-square test. 2. Three-way contingency table The analytic procedure for a representative three-way contingency table from the data explained the reason of conflict results from combination and hierarchical analysis separately. It cannot simply use combination to analyze multidimensional contingency table. The results showed log-linear model is advantaged and efficiency method. 3. Four-way contingency table It is innovative to analyze the four-way contingency table from different research groups, nutrition knowledge, attitude and behavior with log-linearmodel. Fitting the best model for each table, studying the association and interaction effect from these four factors, finding and discovering those significant effected items, explaining and evaluating the conclusion with the conclusion from earlier papers, the professional knowledge and the results from preparative analysis, illustrating with sketch map. The results indicated, the association between different research groups (include specialty, gender, father's education level and mother's education level) and nutrition knowledge is significant, the same as nutrition knowledge and dietary attitude, dietary attitude and behavior (include sleeping status, self-conscious learning vigor and self-conscious mood status); the specialty and father's education level is significant association with dietary attitude separately, the association between gender and mother's education level is not significant; the specialty and gender is probably significant association with the behavior, the association between father's education level and mother's education level is not significant; the same as nutrition knowledge and behavior. Discussion 1. To analyze multidimensional contingency table with log-linear model should choose the variables whose levels are smaller. Log-linear model is not suiting over four dimensions contingency table basing entire efficiency. 2. In fitting model, the influence size of an effected item not only depends on its Wald's test probability, but also considers the size of changed range for the goodness-of-fit and Wald's test probability. This is helpful to...