statistics help

[hana-]

hi guys
im currently doing a phd in immunology but kinda helpless and hopless when it comes to stats. is there anyone who can chat to me about which statistics test to use to analyse some of my data?

it would be much appreciated
hana

 

[JohnM]

hana-

Please post your question(s) or describe the nature of your study to the Statistical Consulting forum, and we'll do our best to point you in the right direction.

Best regards,
John

 

[hana-]

well basically its a biological experiment

theres one treated group and one untreated

each group is split into 3 subsections

and the values for the three subsections are all percentages that add up to 100% per group.

i need to know if the difference i see between say the percentage for subsection 1 in the treated group is significant to the % for subsection 1 in the untreated group...


did i explain that ok?
ill give you my email or something if you need more info or can help...

thanks
hana

 

[JohnM]

If the total sample size for the treated group is the same as the sample size for the untreated group, then this is a good candidate for a chi-square test of independence (or goodness-of-fit, depending on your viewpoint).

You could lay out the data in a 2 x 3 table and determine if the distribution of items (i.e., subset %'s) in the treated group is significantly different from the distribution of items in the untreated group.

this link should give you a good idea of how to carry out this analysis:
http://davidmlane.com/hyperstat/chi_square.html

 

[hana-]

thankyou so much

im forever indebted!

 

[hana-]

http://davidmlane.com/hyperstat/chi_square.html

at that site, in step 2 (more than two categories) my table is like that but with two columns, one for treated and one for untreated.

but the thing is, my experiment was in triplicate so for each subsection i have 3 values.

i dont really know how to reconcile all that to to the chi square test

 

[JohnM]

If you can construct a 2 x 3 table like the attached image, then you can definitely do a chi-square test.

 

[hana-]

but what do i do about the fact that i have three replicates so actually have 3 times as many boxes as that?

 

[JohnM]

You could run 3 separate chi-square analyses, and if the analyses pretty much turn out to be the same, you could then combine the replicates together and run an "overall" analyses.

If you get different results for the different replicates, then you've got another variable in there affecting the results (replicate-to-replicate).....

 

[hana-]

someone suggested a 2 way ANOVA with bonferroni post tests to compare means of each subset
is that valid?

 

[JohnM]

I recommended chi-square because your original post said that the data was in the form of percentages....correct?

- if so, I would not recommend ANOVA, since percentages are not normally distributed (they are bounded by 0% and 100%)

....a lack of normality doesn't usually throw off ANOVA results enough to worry about, but in this case, where they add up to 100%, ANOVA results may be very misleading....

 

[hana-]

if i modiify my data back to the original numbers (which are radioactivity CPS) can i use 2 way anova? i feel like im getting more information outof it with the post tests comparing each mean...

 

[JohnM]

But when you say that the 3 subsets add up to 100%......

Maybe you could post a sample 2x3 table of what your data looks like - is it percentages in the sense of "x out of n" or is it a reading / measurement?

I'm a little confused as to what exactly you're measuring and how it's being quantified....