I think that the problem could be the significance. The p-value observed between significant categories is near to the p significant. Can you perform Student-Newman-Keuls method and can you compare the results with the Tukey test. SNK test is more consvervative than Tukey. How much groups are you contrsting?

Hi Julio,

Many thanks for your reply. I have 45 contrast pairs with 10 categories of samples.

Contrast Difference Standardizeddifference Criticalvalue Pr>Diff Significant

PLLGH vs PLLGHA 5.728 2.412 3.207 0.326 No

Category LS means Groups

PLLGH 23.025 A

PLLTHA 22.517 A B

PLLGHA 17.297 B

It says here that there is NO significant in contrast group, but from the category grouping, it IS significant? How so?

I don't understand. Which are the p values between PLLGH vs PLLTHA, PLLGHvsPLLGHA and PLLTHA vs PLLGHA

Hi Julio,

Sorry for the confusion. Please ignore. Attached is my question.

Appreciate your kind help. Thank you again!

Hi Tina,

you don´t have problems. The group AB is similar to ABA but is more similar to the others groups. The distribution of the groups is overlaping when I suposse that the values extemes from the groups AB and ABA is overlaping. Also, is the same the sample size in the groups? I think the most important is the constrast groupo to group.

I hope that is of helpful for you

Hi Julio,

Thanks for the reply. Sorry but I don't get what you mean. What does "AB similar to ABA but is more similar to other groups" mean? What does overlapping mean?

Not signficatives differences of AB with the other categories are observed. Therefore AB could be grouped into the A group or in the B group. However, the p-value observed among AB vs ABA is smaller than the observed between AB and the categories of the group A. If the size of sample by categories is different also can alter the results, because can alter the standar error which is used in the tukey. Can you perform another post-hoc test as SNK

Hi Julio,

Ok, I did SNK and it gave totally different grouping compared to Tukey (pls see SNK1). I was wondering how come you recommend SNK as it increases Type 1 error and does not account for confidence interval? Sample size AB = 5 samples, ABA = 10 samples.

Another example of inconsistency is found in Tukey2. Based on your explanation, the majority of GH is not significant with other sample group, so GH can be grouped into either A, B or C group? Sample size C = 5, GH = 63 samples.

Although SNK test slightly increase Type 1 error, it is more powerful than Tukey. In summary, both test are similar.

Aditionally, Tukey and SNK test assumes equal sample size and equal variances. Tamhane test is adequate for unequal sample sizes.

Also, consider that the homogeneous subgroups (A, B or C) are constructed from the significance observed between comparisons pair to pair. Categories which not present significatives differences ,could be into the same sub-group, whereas categories that present significant differences are found in sub-groups different s. Therefore, a categories can be in two or more sub-groups.