I want to compare more than two independent sample in SPSS, so how can I define the sample size? is it the number of the units that I will compare or the total number of specimens in the compared units? and if the number of the compared units is inferior to 30, should I change the confidence level and the sample size in the exact of the test Kruskal Wallis or ANOVA or any test I will use? Thank you previously for the answers.
thank you Ramnath Takiar and Mercedes Orús-Lacort , it is kind from you to answer my question. My problem is that when I compare three groups whit Kruskal-Wallis test because the non normality of data, the results show non significant difference, but when I try to compare each two groups with Mann-Whitney test I find one case of significant difference between two groups.
Dear Nabti Ismahane,
I am sure, you know that t-test is used for small samples particularly when the sample size is less than 30. If your specimen is associated with some measurement then you can use t-test. This can be easily done on SPSS.
Enter the data in one column. In second column show the type or category of specimen. Now proceed as follows:
Analyze --> Compare Means --> Independent Sample T-test --> Specimen values (Test variable) --> Grouping variable (category of specimen) --> Range (1 and 2) ---> OK.
Note that if there are two categories of sample then the range is 1 and 2. If there are 5 specimen then the range would be 1 and 5.
The output gives Mean, SD and Standard error in one table. In second tables, the t- values are provided for sample variances being equal and not being equal. Note the significance and interpret the result.
Defining of sample size depends on the variance seen in the data. For this, use the data from some previous study or conduct a pilot study to decide the variance in the measurement concerned. A simple formula of 4S^2/L^2 can be used, where S^2 is the variance seen in the measurement concerned and L is the tolerable limit in the Error.
thank you Ramnath Takiar and Mercedes Orús-Lacort , it is kind from you to answer my question. My problem is that when I compare three groups whit Kruskal-Wallis test because the non normality of data, the results show non significant difference, but when I try to compare each two groups with Mann-Whitney test I find one case of significant difference between two groups.
Dear Nabti Ismahane
Welcome. What you have reported is possible in few cases. One school of thought says that if you have not found the significance by ANOVA, you should not look further for between sub-groups comparisons. Take it that there is no significance differences between the group. The result you have obtained also suggests that your data (may not be representative) is not supporting the hypothesis of significance between the group. You may also note that on obtaining the significance by ANOVA, you are permitted to go for sub-group comparisons say Duncan Multiple Range Test. This points out to the fact that all groups considered need not show significance among themselves.
However, I suggest that you can do the log transformation of the data and then carry out the usual analysis.
By default, we take the level of significance to be α =0.05. Sometimes, It is worth considering the level of significance at which your results are significant say α =0.10.
Try this two suggestions and may report again.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA