Regarding Proc T test

[anilgvdbm]

Dear All,

Here i have an data of BMI for household men and women like variables are Gender, Age, Village, Year and BMI, BMI_catergories in this i have to test the significance of BMI with gender and BMI in Years and BMI with the BMI catergories like Normal, overweight and underweight so for theis i am using 2 sample t test is it correct?

I request you all to suggest for an Idea.

Here is the codes for your reference

proc ttest data=Nutr_old.antpfin_05;

where village ="village";

class sex;

var BMI;run;st

Results are like this

                                                                                                                            The TTEST Procedure

                                                                                                                               Variable: BMI (BMI)

         

1	86	19.2561	4.8538	0.5234	2.3671	33.0138
2	88	19.1482	4.6338	0.4940	10.9051	32.0001
Diff (1-2)		0.1079	4.7438	0.7193

          

1		19.2561	18.2154	20.2967	4.8538	4.2210	5.7114
2		19.1482	18.1663	20.1300	4.6338	4.0358	5.4415
Diff (1-2)	Pooled	0.1079	-1.3119	1.5277	4.7438	4.2911	5.3041
Diff (1-2)	Satterthwaite	0.1079	-1.3127	1.5285

       

Pooled	Equal	172	0.15	0.8809
Satterthwaite	Unequal	171.17	0.15	0.8810

       

Folded F	85	87	1.10	0.6675

[Ksharp]

If it was two level, you can use proc ttest for parameter test and proc npar1way(Wilconx test) for non-parameter test.

It it was three or more levels ,that means you need multi-comparison , check proc glm + means .

Xia Keshan

[Doc_Duke]

for comparing the BMI Categories by sex, you would need PROC FREQ, as that is a 2x3 table.

If you scored the BMI categories to get a rank, you could use NPAR1WAY, but it is much less sensitive to differences than the original scale.