Comparisons between groups and subgroups

[#1]

I have a file With data from 260 individuals that look more or less in this format (but more parameters):

	Person ID	Gender	Age group	Protein intake	kcal
1	1	Male	children	33	107
2	2	Female	adults	52	220
3	3	Female	elderly	50	150
4	4	Male	adult	75	175

The data is not not normally Distributed and I have done distribution analysis. Furthermore I have compared the data between males and females and also between age Groups and calculated ps for pairwise

I know the data are not normally Distributed and non-Parametric tests have to de bone

I have compared the data for gender.

And I have also plotted and compared data for different age Groups (categories).

But then I would like to divide the data into age categories AND gender at the same time and have not figured out how to do this in JMP without splitting the data into one file for male and one for females

Gunnar

Gunnar 

I know the data are not normally Distributed and non-Parametric tests have to de bone

I have compared the data for gender.

And I have also plotted and compared data for different age Groups (categories).

But then I would like to divide the data into age categories AND gender at the same time and have not figured out how to do this in JMP without splitting the data into one file for male and one for females

You should be able to do the separate analyses you are mentioning "male and female adolescent, male and female elderly etc", by specifying those columns as "By" columns in the analysis and graphical platforms you are running. The example below adds "Age" and "Sex" to a simple Fit Y by X for "Height" vs. "Weight"

Gunnar

[#2]

First of all, why is the normal distribution important for the response in this study?

Analyzing the effects of variables is not efficient. If you want to continue this way, though, use the By analysis role. Cast gender and age categories into this role for your analysis.

You might consider the more efficient and informative analysis in which you use both variables in the linear predictor. You could then also include a term for the possible interaction effect where the effect of one factor (e.g., age group) differs depending the on the level of the other factor (e.g., gender). Use Analyze > Fit Model for this approach.

The resulting Fit Least Squares platform provides regression diagnostics based on the residuals. The model assumes that the errors are normally distributed (not the response itself) and independent.