4 groups of mice: WT-Placebo, MUTANT-Placebo, WT-Treatment, MUTANT-Treatment

[Riccardo Urbanet]

Hello to everyone, 

new rules in research world and big trouble for the ones who don't have any biostatistical guy available in the lab.

I need to test two different mice phenotype, either receiving Placebo or a drug treatement.

I'd like to express the variation as % of the basal level (placebo treated wild type mice) and I'd like to express the SD as % of this value. 

I mean, if the parameter I'm testing in the plasma is 100 nM, I'd like to assume (following the bibliography) that, a variation in the mean over 100% is the minium I expected as significant and that the SD intra-group is less than 10% of the basal value.

How do I get this done ? 

Any help ? Idea ?

Piface seems a wonderful application, but how does I translate my wishes in statistical sense ?

Ric

[Lenth, Russell V]

Often, when people want to focus on relative differences, I suggest using a logarithmic scale, because it translates ratios to differences: log(x/y) = log(x) - log(y) -- or more to the point here, considering a 10% difference,

    log((y + .1y) / y) = log(1.1y) - log(y) = log(1.1) + log(y) - log(y) = log(1.1) ~= .1

on the natural log scale. That is, small percentage differences translate to that same fraction as a difference of natural logs. Moreover, the kinds of data for such scenarios are often skewed, and the log transformation makes things look more normal. You can of course explore that in your pilot data -- and you have some, right? (If not, you're just making up numbers.)

So my suggestion is to obtain the error SD of a suitable model fitted to the logged (base e) pilot data, and use that as the SD in the sample-size dialog, along with a target difference equal to the fractional difference of interest (e.g., 10% us .1, 20% use .2).

Russ

Russell V. Lenth  -  Professor Emeritus
Department of Statistics and Actuarial Science   
The University of Iowa  -  Iowa City, IA 52242  USA   
Voice (319)335-0712 (Dept. office)  -  FAX (319)335-3017

--
You received this message because you are subscribed to the Google Groups "PiFace discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to piface-discuss...@googlegroups.com.
To post to this group, send email to piface-d...@googlegroups.com.
Visit this group at http://groups.google.com/group/piface-discussion.
To view this discussion on the web visit https://groups.google.com/d/msgid/piface-discussion/2dd302e7-356c-4ef9-bf2f-d69d993166c0%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Riccardo Urbanet]

Thanks a lot Russ, could I ask you some more help, if you have time, I'd like to have some more explanation.

I have to say, that's is the first time I use "power statistics", and I need to use them cause It's important for ethical purpose to not use too many animals in research, I'd rather say, It's important to use the smallest amount of them.

So, looking back to real data, let's say, from a pilot study, I got that the over expression of my target gene (in mutant mice) causes en increase of my target plasma parameter by 300%. In this experiment, the SD is within around 30% of the mean in each group.

The, pilot considered only the, "placebo" groups, so only two groups. 

Now, what is my next move ? What do I have to do, for answer the question: How many mice do I need to demonstrate, that the treatment is able to reverse the effect observed in mutant mice ?

Is there any simple way to deal with this ?

Which test do I have to perform ? with which software ?

thank you

Riccardo 

[Lenth, Russell V]

The quick answer is that I don't know, without a fair amount of other information. This looks like important work, using expensive lab procedures. I think you should find a statistical consultant nearby who can talk to you and hash these things out. You want to get this done right, and it's not simple. It's worth the money - or rewarding co-authorship to a consultant. Most university statistics departments have people that can help.

Russ

Sent from my iPad