another stats question please help

S

hi its satchi again
this time its about student's t-test.
If I do t-test for 2 groups, say A and B
and also do t-test for C and D.
Where do I look to see the greater effect? Is it the t-value if both are <.001.
Is it a bigger t value means there is a greater effect.

thanks in advance
satchi

J

the T value = effect size NOT p value!! this just says how significant effect is, not how big...

S

hi ju-ju
so if I have one t-value which is 1.956 and the other is 1.2 but both are p<.001
does this mean that the first pair has more effect
is this right

thanks satchi

A

hi Stachi, I think so, and if it's positive or negative value indicates whether the effect is positive or negative :)

S

hi algaequeen
THANKS!
satchi

J

sort of... but why not analyse the data together? because doing multiple independent tests is not good (increases likely hood of type1 error...) use an ANOVA? one or two way, depending on exp design... this is the only way to really be able to conclude anything across effects - make sense?

S

hi again! i did anova (one way) but; just I couldnt tell where the change (of effect) happened
satchi

C

For effect sizes you could look at something liek Cohen's d. There are loads of free internet programmes that can calculte this for you (plus I think spss can do it) G*Power is good, you put in your means, SDs and n for the two groups and it can tell you the effect size along with the power your study had. There are others on the internet that give you an effect size from the T statistic.

S

thanks :-)
satchi

J

you need to do a proper post hoc test to determine where significance is coming from - not just multiple t-tests!!!

A

Hi Satchi, it's difficult to say exactly what kind of test you should be doing withough knowing what kind of data you have or what you're trying to find out, but Ju-Ju is right, when you do ANOVA it basically tells you there is a effect, but it's the post-hoc tests that tell you where the effect is or how strong it is etc. you said you have 4 groups ...are you trying to compare them all with each other (as in AB with CD; AC with BD, AD with BC) or all separately or just as you have mentioned A with B and C with D? you need to decide exactly how you want to test them, if you are doing multiple comparisons, t-tests aren't strong enough for that. there is a book you should get called Choosing and Using Statistics by Calvin Dytham, it's the only book I can use as it's so clear about what tests mean and how to do them! even the Andy Field book that everyone loves is way too technical for me! and it's only about 1/10th the size of the andy field one!

Avatar for sneaks

I agree with the ANOVA suggestion. If you do persist in using multiple T-Tests make sure to use the boneferroni correction.

S

oh no. what is a posthoc test! I better read up. oh my goodness
I havent seen bonferroni anywhere for t-tests (what have I been doing?!)
I also didnt explain my data clearly; it is a two group design: for four different types of intervention.
I was going to compare 2 types of intervention (in pairs) for the two groups.

dumb satchi

Avatar for sneaks

======= Date Modified 04 Mar 2010 17:24:37 =======
Bonferroni correction is a technique to avoid type 1 error. YOu basically either divide the value you are using to denote significance (usually .05) by the number of tests you conduct (e.g. .05/5 because you are conducting 5 tests) OR you can just say you will use .01 or .001 rather than .05 so you won't accidently think something is significant when it shouldn't be.

Sounds like you would be better off doing an ANOVA with post hocs. ANOVA will show you there is a difference between groups. But it won't show you where those differences actually are - this is what post docs can do after you have run the initial ANOVA.

Not sure my brain will compute what your data is but you may need a MANOVA - if you have more than 1 Dependent variable

S

thanks sneaks :-)
satchi

14147