hello, I am helping someone with her statistical analyses for her biology master-dissertation , basically, i would like to know if my approach is correct. I will first  explain the setup of the experiment: the study involves the response of 3 different types of mutants (:OE,WT or Ty-2) for a specific gene in the lenght of the primary root of a plant species. The measurements of the root lenghts were done on 2 different days and on 3 different concentrations of Nitrate. Because we only need to compare the root lenghts of the mutant types within a specific day for a specific concentration, there is no need to search for differences between concentration or between days. On e.g. day 9, (NO3-) 1mM, the average root lenghts of the three mutants look like this: I thus thought i could do pairwise comparisons of the root lenghts of the mutants with independant t-tests (the root lenght of one mutant seedline is not affected by the measurements of those of the other mutant). Compare OE with WT, OE with Thy-2 and Wt with Thy-2 and putting the p-values on top of the bars with letters crowned with with asterixes . My first thought on this was to run an Anova (proc glm?) on the 3 mutants, but when upon reading articles, i find that they all only use pairwise comparisons using  independant t-tests so i guess i can do this? When using an Anova i would do a post-hoc and end up with pairwise comparisons using  t-tests anyway, is this a just assumption or should i use an anova despite of the articles doing otherwise? For the t-test there is thus 1 independant categorical variable (mutant type) and 1 dependant continuous variable (root lenght) for each comparison to make. is the following testing then correct?: for e.g. day 9, Nitrate concentration 1milliMolair(mM) proc ttest data=thy2dag91mMthy2vsWT; class expression; var count; run; class because of the independant t-test, var lenght for testing equality of variances and the output gives me a qqplot where i can check for normality deviations (or should i do a proc univariate for testing normality) or mustn't I oblige to this assumtion if i find some of them are not normally distributed? given the data: data thy2dag91mMthy2vsWT; (the sample sizes for the two mutants are different in size due to some seeds not germinating) input expression $ count; cards; thy-2 2.580 thy-2 3.296 thy-2 4.179 thy-2 4.298 thy-2 4.459 thy-2 2.986 thy-2 3.413 thy-2 4.419 thy-2 4.368 thy-2 3.849 thy-2 3.962 thy-2 3.718 thy-2 4.106 thy-2 4.456 thy-2 4.483 thy-2 4.345 thy-2 4.248 thy-2 4.311 thy-2 3.336 thy-2 3.720 thy-2 3.874 thy-2 3.030 thy-2 3.888 thy-2 4.492 thy-2 4.332 thy-2 4.231 thy-2 2.752 thy-2 4.567 thy-2 4.379 thy-2 4.807 thy-2 4.843 thy-2 3.791 thy-2 4.049 thy-2 2.950 thy-2 4.231 thy-2 4.141 thy-2 4.746 thy-2 4.463 thy-2 4.555 WT    3.885 WT    3.968 WT    3.259 WT    3.670 WT    3.572 WT    3.309 WT    3.803 WT    4.074 WT    3.991 WT    4.057 WT    3.786 WT    3.855 WT    3.635 WT    3.415 WT    3.462 WT    3.835 WT    3.976 WT    3.946 WT    3.567 WT    3.344 WT    3.630 WT    1.992 WT    3.605 WT    3.670 WT    3.837 WT    3.702 WT    3.784 WT    3.766 WT    1.884 WT    3.999 WT    3.937 WT    4.144 WT    3.537 WT    3.704 WT    3.657 WT    4.256 run; kind regards, Bas ... View more

Hi, This actually is a follow up of a previous question i asked (proc glimmix, insect counts), but i shall give the information again: I am busy on a comparative study on different sowing types for field margins in their effect on abundancies of insects I have 4 different types of sowing type: G(rass),I(ndiginous),T(übinger) and C(lover), The experimental setup is as such that I have planted on each location these four different sowing types, so at every location there are four field margins, each containing one of the sowing types, but never two same sowing types on the same location. I then have 3 locations where all four sowing types are represented and 1 location where only I and T are present. So I have an unbalanced design and a very small sample size. This actually is data obtained from repeated measurements, I sampled the field margins four times in the season. I have already (with the help of this forum) done a repeated measurements analysis on the whole abundancies (all insects) and on one functional group of them , using a negative binomial generalized mixed linear model. So here I could fit a model count=treatment date treatment*date with random intercept for location and a repeated measurements cobvariance structure. but for another  functional group of intrest in the insect-counts I cannot obtain a converging model, even when adjusting the pconv, maxiter... there are a lot of zero counts for some locations/treatments/dates and i guess that is why it does not converge... Now comes the problem I would like to solve: Because I cannot obtain a model with the repeated measurements design, I thought I could look at it for the whole season then… So now I counted up the number of bumble bees of this functional group at each location at each treatment. Instead of having now for location x, treatment G for example 4 measurements (one on each date), I now have 1 measurement (the 4 measurements of the dates counted up). My first question is: is this reasonable to do so? Although it is a pilot study, I wouldn’t like to report p-values which are not very accurate because of huge confidence intervals and low power because I by doing this reduced the number of freedom degrees drastically… Second question: Is the following proposed justified to do in this experimental setup, given that the first question is positively answered? The counted up data-input  is as follows, below that the code I would want to use, but which I am not sure of… data jaarkortetong; input loc $ treatment $ count; cards; mb    g     2 md    g     33 pv    g     2 mb    i     1 md    i     44 pv    i     17 gb    i     38 mb    k     2 md    k     5 pv    k     1 mb    t     23 md    t     27 pv    t     27 gb    t     48 run; ods graphics on; proc glimmix data=jaarkortetong initglm method=mspl; nloptions maxiter=2000 tech=congra;; class treatment loc; model count=treatment/dist=negbin link=log solution ddfm=sat; random intercept/ subject=loc; lsmeans treatment/cl ilink diff plot=meanplot (ilink); run; ods graphics off; best regards, Bas ... View more