Author: little bit helper
Source: Dingdian help you
We have already talked about the non-parametric test for comparing two groups, similar to t-test and analysis of variance. When comparing more than two groups of data, we need to change the method.
The non-parametric KW test is compared to the Mann-Whitney test explained in the previous article. We can understand it as "analysis of variance of non-parametric tests."
The full name of the KW test is the Kruskal-Wallis test, which is an alternative to the analysis of variance for multiple independent samples when the normal distribution conditions are not met.
Case:
In order to understand the damage of DON (a certain toxin) to the joints, 15 rabbits were randomly divided into control group, low-dose group and high-dose group according to their body weight. They were injected with saline, 0.05mg/g and 0.10mg/g of DON, respectively. The toxin is processed experimentally.
After the expiration of the experiment, the level (μg/L) of tumor necrosis factor (TNF-α) in the joint washing fluid was measured, and the data obtained are shown in Table 10-6 below. Now compare the three groups of rabbit joint washing fluid TNF-α determination results are statistically different?
When doing medical statistics-related research, we often encounter the above-mentioned complex technical terms. We must have the ability to simplify the complex, and look carefully at the question. We write "Tumor Necrosis Factor (TNF-α)" as "Y" , In fact, is a simple one-way analysis of variance to compare whether the "Y" of the three groups is different.
Because the sample size here is only 15 (5 rabbits in each group), it is a typical small sample study. When the sample number is too small, it is difficult to reliably judge the normality of the data, and it is impossible to use one-way analysis of variance to test .
Because the non-parametric test does not require the normality of the data, when the sample size is small, the more robust Kruskal-Wallis test can be used for statistical inference.
Simply put, the basic idea of the Kruskal-Wallis test is to use the ranks of all observations instead of the original observations for one-way analysis of variance, and the test statistic is the H value:
The steps and results of SPSS operation are as follows:
As shown in the above table, P=0.004 <0.05, reject H 0 at the level of α=0.05 , and accept H 1. It can be considered that the difference in the TNF-α measurement results of the three groups of rabbit joint washing fluid is statistically significant.