http://spssx-discussion.165.s1.nabble.com/How-to-compare-more-than-2-percentages-tp5715878p5715885.html
It sounds to me as though Hianghao has not collected replicate data, but rather pooled the mortalities for the reps before calculating the percentages. Viz:
This is unfortunate, if true. What was the point of the replicates? Assuming it is indeed the case, the data reduce to a simple 2x2 frequency table.
Ian D. Martin, Ph.D.
> I think you need to study more about experimental design and testing,
> because here are multiple problems.
>
> 1) If you had some samples you compared with mortalities of
> 2% vs 51% with "no significant diff" -- then your *power* was
> apparently so small that you should not have bothered mentioning
> the analysis. However, "2%" implies an N of 50 or so. Unless the
> other group at 51% was tiny, you seem to have been doing a
> ridiculous amount of "correcting for testing", or you screwed up
> your testing.
>
>
> 2) If you bother to collect replications, you certainly should look
> at them to see that they are equivalent. Unless there is special
> gain from the simplification, you could use the separate numbers
> in an analysis. You have an N of 6+12=18 as proportions or
> logits, that might be entered into some sort of ANOVA... as one
> simple way of constructing the problem. Or, that seems to be
> 2*18= 36, if your second experiment with Predator was the same
> design.
>
>
> 3) The total number of tests to correct for is not the same as the
> total number of tests. You correct for the "interesting hypotheses"
> that exist as the purpose of the experiment.
>
> It hardly seems to be an interesting hypothesis - certainly, it
> is an hypothesis of a different kind - that the mortality is higher
> with a predator versus without. You want to confirm that this
> is true, as a simple, rational validation of the basic condition for
> considering the *difference* in surviving. So, you don't have
> as many "main tests" as you are counting.
>
> The test on Predator is uninteresting, except as confirmation
> that you have data to analyze further.
>
> The test on Species is (probably) relatively uninteresting.
> Whether they are the same or different seems dull, compared
> to the questions about relative survivorship under different
> conditions of Predator or Density. I think that the three
> actual, interesting hypotheses are the interactions of species
> with P, with D, and with PxD.
>
> Hope this helps.
>
> --
> Rich Ulrich
>
>
> > Date: Thu, 25 Oct 2012 20:21:40 -0700
> > From:
[hidden email]
> > Subject: How to compare more than 2 percentages
> > To:
[hidden email]
> >
> > I have 2 strains of mosquito: A & B. I reared both of them under low density
> > and high density conditions (3 reps for high density, 100 larvae per rep/
> > strain; 6 reps for low density, 50 larvae/rep/strain), and recorded total
> > percentage (not the percentage of each replicate) of them which reach
> > adulthood. At first i assigned "1" for each larva survived and 0 for dead,
> > then i conducted Mann Whitney test. Because i have 4 grps, I need to do 6
> > comparison and there'll be familywise error. Correction (Bonferroni/Sidak)
> > caused most of the analysis to accept null hypothesis, including another
> > experiment where i calculate the mortality with and without predator for
> > both strain A & B. FYI the mortality with predator was 51% and the one
> > without was 2 %, yet it became no significant diff btwn the mortality.
> >
> >
> > I also want to compare percentage mortality of 4 groups of treatment (strain
> > A of insect with predator, strain A without predator; strain B with
> > predator; strain B without predator). I found that percentage can be
> > compared using 2x2 contingency table. but i have 4 groups instead of 2. Or i
> > should conduct 6/4 chi square tests and divide the alpha with 6 or 4
> > (because i don't want to compare A+predator vs B-predator & A-predator vs
> > B+predator)?
> >
> > Something came into my thought: do i need to correct the mortality in
> > predator grp with mortality without predator grp?? How? Abott's correction?
> >
> > ...
command. To leave the list, send the command