Какой статистический метод применить? Опции

[Maria30]

Podska*ite, po*aluista, kakoi statisti4eskii metod lu4she primenit v moem slu4ae? Ja issleduju vosdeistvie dvuh veshestv na kletki pod*eludo4noi *elezy. Odin preparat(1) vysyvaet apoptosis(gibel kletok), drugoi preparat(2) podavljaet apoptosis. Polu4aetsja tablica
1opyt 2opyt 3opyt
control X X X
Preparat 1 koncentracija 1 X X X
Preparat 1 koncentracija 2 X X X
Preparat 1 koncentracija 3 X X X
Preparat 1 koncentracija 4
1opyt 2opyt 3opyt
Control + preparat2 X X X
Preparat 1 koncentracija 1+ preparat2 X X X
Preparat 1 koncentracija 2+ preparat2 X X X
Preparat 1 koncentracija 3+ preparat2 X X X
Preparat 1 koncentracija 4+ preparat2 X X X

Preparat 2 neizmennaja koncentracija
Neobhodimo sravnit control i preparat 1 koncentr.1........4, a zatem preparat 1konc.1 i preparat 1 konc.1+ prepar.2 i t.d.
Mo*no li v moem slu4ae primenit parnyi Student´s t-test?

Izvinjajus za latinskie bukvy:-(

[Maria30]

Thank you very much:-) The problem is that I have 4-5 experiments to work with. In the literature they describe that ANOVA is not god for small samples ( <12). "With very small samples, it may be impossible for the P value to ever be less than 0.05, no matter how the values differ" Should I stil try the ANOVA?

[плав]

Thank you very much:-) The problem is that I have 4-5 experiments to work with. In the literature they describe that ANOVA is not god for small samples ( <12). "With very small samples, it may be impossible for the P value to ever be less than 0.05, no matter how the values differ" Should I stil try the ANOVA?

Well, so far asyou are working with relatively homogenoeous groups it should not be the problem. The strength of ANOVA is that it estimates standard error based on all objects in the sample not just the two groups like in the pairwise comparison (t-test). Basically for two groups F is squared t, so that you will not lose power and t is the most powerful unbiased test (of course if we suspect underlying normal distribution). But again in the experimental setting the hypothesis of normality is much more plausible than in quasi-experimental works medicine relies mostly on. Basically experimental design in agriculture that started the whole frequentist statistics was dealing with plans with no repetitions at all and it is still very much so in clinical trials (Phase I especially - at least a few repetitions).
To sum up, I do believe that the ANOVA is most powerful analysis you can do (if there is quantitative outcome indicator). If you will calculate number of dead cells to the number of cells total then more general models like Poisson regression or GEE (general estimable equations) will be needed.