Type I and Type II SS for ANOVA, non-statistician

#1
Good Afternoon (non-statistician here),

I have to automate a GLM within an Excel sheet that will be analyzing some toxicity data, resulted from scoring cell aberrations. And, I'm using MiniTab to validate the estimates. MiniTab is using Type I (sequential) SS and Type II (adjusted) SS within their ANOVA GLM.

I can't find a place online that will give me a comprehensive guide to estimating the Type I and Type II SS.

I've also poured through "Applied Linear Statistical Models" and can't find out how to do these. The book only covers Type III SS, which is the classic unweighted means.

I also looked at "Four Types of Sums of Squares for ANOVA Effects". But, it was very vague in describing how to calculate the different SS Types.

Can someone point me in the right direction? Either a website or an example excel file?

Thank You,
BrainExploding
 

Dragan

Super Moderator
#3
Good Afternoon (non-statistician here),

I have to automate a GLM within an Excel sheet that will be analyzing some toxicity data, resulted from scoring cell aberrations. And, I'm using MiniTab to validate the estimates. MiniTab is using Type I (sequential) SS and Type II (adjusted) SS within their ANOVA GLM.

I can't find a place online that will give me a comprehensive guide to estimating the Type I and Type II SS.

I've also poured through "Applied Linear Statistical Models" and can't find out how to do these. The book only covers Type III SS, which is the classic unweighted means.

I also looked at "Four Types of Sums of Squares for ANOVA Effects". But, it was very vague in describing how to calculate the different SS Types.

Can someone point me in the right direction? Either a website or an example excel file?

Thank You,
BrainExploding
I'm not really clear either. But why would want to use, say, Type I sums of squares? I wouldn't even recommend it because it forces the sums of squares in the context of nonorthogonal designs to add up. The downside of achieving the additivity of the sums of squares comes at the cost of altering the hypotheses that are tested. In short, they do not express the difference between population means, making it difficult to give them a sensible meaning in experimental contexts.

You really should be using Type III sums of squares.
 

Miner

TS Contributor
#4
This article may answer the questions that you have regarding the different types of Sum of Squares.

To see how Minitab calculates the different types, go to General Linear Model > Options > Help > See Also > Methods and formulas > Sum of squares (SS) / Sequential sum of squares / Adjusted sum of squares.