ttest with 'by' options

[Alan Jeddi]

Hello, I have a chart of my variable "score" which is presented stratifying by sex and by age-group.
I would like to compare the score between men and women, for each age group separately. I will thus use 4 t-tests.
I am finding it hard to generate the code. So far I have used

ttest score, by(sex) if age_grp==1
ttest score, by(sex) if age_grp==2
ttest score, by(sex) if age_grp==3
ttest score, by(sex) if age_grp==4

but I keep getting the message that the option [if] if not allowed. I suppose I could do a Factorial ANOVA, but I am really trying to understand whether Score varies by sex, separately in each agegroup.

Age Group	N men (2299)	Score Men Mean(SD)	N women (4762)	Score Women Mean(SD)	P value*
<60	187	16.54(3.34)	324	15.86(2.95)
60-69	1028	16.91(3.57)	1426	16.78(3.37)
70-79	846	17.07(3.60)	885	16.89(3.39)
>80	238	17.48(3.91)	212	16.76(3.51)

[Nick Cox]

Look at the help for ttest. An if qualifier is not an option and should be specified before the comma. (There is an exception to this, but it would be a distraction to explain it.)

ttest score if age_grp==1, by(sex)

[Carlo Lazzaro]

Alan:
the -if- qualifier is misplaced in your code.
Try something along the following lines, instead:

ttest score if age_grp==1, by(sex) unequal *-unequal- option relaxes the often untenable assumption of equal variance*

PS: late to the party once more!

A different take would consider switching to regression, interacting -sex- with -age_grp-:

regress score i.sex##i.age_grp

[Marcos Almeida]

Also late. With 4 comparisons you may take in consideration adjustments for familywise error. in short, as pointed out before, linear regression or ANOVA seem to be the best approach.