Subscribe Bookmark RSS Feed

How to do student's t pairwise comparison of two means in REML model?

Juno

Contributor

Joined:

Jan 4, 2017

Hello,

 

Please see attached picture. I have 3 fixed factors and 1 random blocking(Year*Location) effect. I ran the factorial analysis using REML. 

How do I do student's t pairwise comparison of any two means of the treatments in REML model? 

Also How do I obtain the MSE and MS Blocking from the given table?

 

Thank you.Picture1.png

1 ACCEPTED SOLUTION

Accepted Solutions
markbailey

Staff

Joined:

Jun 23, 2011

Solution

Click the red triangle at the top next to Response and select Estimates > Custom Test.

See Help > Books > Fitting Linear Models and search for "custom tests" for more information.

Learn it once, use it forever!
10 REPLIES
markbailey

Staff

Joined:

Jun 23, 2011

Solution

Click the red triangle at the top next to Response and select Estimates > Custom Test.

See Help > Books > Fitting Linear Models and search for "custom tests" for more information.

Learn it once, use it forever!
Juno

Contributor

Joined:

Jan 4, 2017

Thank you Mark!
Juno

Contributor

Joined:

Jan 4, 2017

Also where can I find the MSE and MSBlock for this desigb? 

markbailey

Staff

Joined:

Jun 23, 2011

It is right in front of you!

The MSE and MS Blocking are in the REML Variance Components Estimates table. The first row is the random block effect with its variance component (MS Block) and the second row is the residual variance component (MSE).

Learn it once, use it forever!
Juno

Contributor

Joined:

Jan 4, 2017

Oh I see!! So the Var Ratio is the F-test for the Blocking effect?

Does it mean my Blocking is ineffective if it is not significant? 

 

Thank you

markbailey

Staff

Joined:

Jun 23, 2011

No, the variance ratio is simply the variance component of a random effect to the variance component of the residual (error), such as blocking to residual. It is a relative measure, such as the variance due to changing blocks is three times the error variance.

There is no F ratio or hypothesis test. Instead you can use the equivalent confidence interval estimates. If the interval contains zero, then you do not reject the null hypothesis at the stated level of significance. If the interval does not contain zero, the you reject that the variance is zero.

Note: depending on your REML options, negative estimates are possible (they are really covariances) in order to produce unbiased estimates of the fixed effects and tests with the desired coverage.

Learn it once, use it forever!
Juno

Contributor

Joined:

Jan 4, 2017

So based on my analysis result, my confidence interval of Blocking contains zero, I do not reject the null hypothesis. Therefore I can conclude the variance is zero?

How do I interefere the effectiveness of my Blocking this way? 

Juno

Contributor

Joined:

Jan 4, 2017

And How do I know if there is any Block*Entry interactions?

markbailey

Staff

Joined:

Jun 23, 2011

We say that you blocked as a precautionary measure to minimize the impact of the potential external source of variation (not inherent random variation in the response, not a random effect due to changing a factor level) but it was unwarranted. The random effect of the blocking was not found to be significant in this experiment.

That's all.

Your second question is meaningless. A block is either a group-specific addition to the intercept (fixed block effect separate from interaction effects) or a random effect (sampled less frequently than residual). It makes no sense that a block is included in any interactions. If it did, then it means that the effect of the factors (the way things work) changes from block to block. Then it is a factor that you failed to identify or control.

Learn it once, use it forever!