Hi Ruhi,

to my understanding you have to use the Two-Way ANOVA since you have more than one factor which is affecting the result, isn´t it. I use mice for my experiments. When I do a measurement of a varience in eg inflammatory pain between 2 genotypes I allways do a 2-way anova with Bonferoni posttest. If I compare just a single varience (eg. the amount of Interleukin 1ß) in one genotype I do a 1-way ANOVA most times with Newmann-Keul´s posttest.

If you have different genotypes and eg treated vs. untreated or healthy vs sick both the status (sick or healthy) as well as the genotype might have an influence. Therefore the two-way ANOVA.

If the results are not significant in a two-way maybe you have to have higher n-numbers since the variences are higher with more factors.

Best wishes,

Claus

Hallo, your case is a very interesting issue. I think that the right apporach is analyzing healty and non-healty together. If I have well undestood, you have, let me say, X healty patients and Y non-healty patients (two level for a so-called hard-to-change factor: the healtheness), Furthermore, you are interested in studying the effect of "genotype" on some clinical and anthropometric variables. However you shoud specify better the characteristi of the factor "genotype". How many different genotypes you have to deal with? Are they equally represented within the two "healtmess" groups? Are some genotypes lacking in a group? Are you interested in evaluating the effect of the genotype within the effect of the first factor (healtness)? Fell free to contact me.

Pier Paolo

Please see the attached. It may help

Without understanding the specific data I'd suggest linear regression. The outcome(s) would be one of the continous variable (anthropometric?) and the predictors are the binary "treatment" variable (patient/control) and if you wish to include covariates, the genotype and clinical variables. This is called also ANCOVA - for using covariates (variables other than the "factor" variables defining the "groups"). You can have also interactions, e.g. genotype times "treatment".

Hi Ruhi,

I dont know the data but i think that there is an interction between genotypes and trial. Try the factorian ANOVA. If interaction term is exist than make traits such as genotype1+patient, genotype1+control, genotype2+patient and genotype2+control and use one way ANOVA.

I think to give a more informed answer we need more information. I'm assuming that the independant variable is "health status". Is that true? In that case what is the DV (dependant variable). If you don't know these terms - then perhaps state the research question. For example are you trying to predict health status from the other variables (including genotype)?

Dear Ruhi!

The effect of the statistical analysis will be maximized, given the distribution of the variable (parametric and non-parametric). Cross tables can help to "feel" the studied data, using quantitative interpretation of variability in quality (use a range of quantitative data). Nonparametric analogue of ANOVA (the Kruskal–Wallis one-way analysis of variance) to help in the search for patterns.

Ruhi, I'm with David Cottrell on this one. Could you spell out the variables and design in more detail? What are the independent and dependent variables? Are you asking whether genotype affects the variables being measured? If so, are there differences in the frequencies of genotypes between patients and controls? Usually, a "covariate" is a correlated or even "nuisance" variable that may obscure the relationship between the independent and dependent variables. Is that true here?

Yours, John

Thank You all for the suggestions..... John Ringo, I am using clinical parameters like BMI, WHR, Cholesterol, HDL, Triglyceride as covariates and genotype as Factor. I am

confused whether to apply ANOVA in pooled (Cases+Controls) or in grouped data. yes I am asking whether genotype affects the variables being measured?

Hi Ruhi Now we are getting somewhere. I appears that your trying to predict BMI, WHR, Cholesterol etc from genotype, hence these variables are your DV's. If your asking questions like "Do those with genotype A have a higher BMI than those with genotype B" then you could do a series of ANOVA's testing each question independently. That is you would have two between subjects factors: GROUP (TD2 and control) and GENOTYPE (I think you had two levels here as well). GROUP and GENOTYPE are thus your independent variables in a quasi-experimental design. This 2x2 analysis will also tell you if group and genotype interact (as pointed out by Hasan).

If you want to look at interactions between the DV's then things are a little more complex.

Whenever I do statistics consulting I always start with the research question. You really need to give us that information if things are more complex than what I have guessed at. Statistics should really be seen as a tool to help answer a clearly constructed research questions.

people do apply ANOVA in pooled samples(cases+controls) and also separately in controls. but in my sense applying ANOVA in pooled samples is better as u r seeing the affect of genotypes on the confounding factors (BMI, weight , WHR etc). Genotypes ll not differentiate between cases and controls.

Recently there was a paper from your department on ADIPOQ association with T2D, Follow that pattern.

Thank you Ankit. Paper you are talking was from my lab only. well i go through this problem and even talked to statisticians also and I think its better to apply ANOVA on cases and controls separately.