testing different models using the same group of longitudinal data

#1
Description of my research

Longitudinal growth data are fitted using parametric models giving growthcurves. Differentiating these growthcurves gives velocity curves. From every curve biological parameters can be determinated. Every parameteric model has function parameters.

The same longitudinal data are fitted using 5 different parametric models and 4 different age starting points. Resulting in 20 groups for every gender. Bad fits are excluded.

The questions I would like to find

  1. Are the parameters normally distributed?
  2. How is the goodness-of-fit?
  3. What's the effect on using different starting ages?
  4. What's the effect on using different parametric models?
  5. What's the influence the biological parameters have on eachother?

What I've come up with so far

(step 1 in the previous step is linked to step 1 in this step)

1.
- First I've calculated the mean, median, standard deviation, range, number of subjects, skewness, kurtosis and probability of Saphiro-Wilks for every function- and biological parameters in every group
- Histograms are obtained for every function- and biological parameters in every group

2.-pooled residual mean squares
- runs test
- autocorrelation test

3. -median or mean constant curves: if functionparameter has normal gaussian distribution mean is used, otherwise median. Plots are generated for every model containing all the groups with different starting ages
- next one way ANOVA is performed on the biological parameters

4. same as step 3 but here the age is fixed in every plot and the models vary.

5. Correlationmatrix and principal component analyses to find the relationship between the different biological parameters.


What I would like to know

Can anyone give some statistical insight in this. Are there other tests I can perform after the one way anova?
I'm a biology student and not a statistic, so chances are my questions can be a little dumb. I'm willing to learn however.

Thanks a lot
 
#2
I cannot emphasize strongly enough how wrong it is to fit multiple models to the same data. This makes utter nonsense of any claim of statistical significance. You must test models with other data (different years, different regions, etc.) or if necessary a small percentage of your full data set. You then decide on a model and apply it to the data of interest, or the entire set. One time only. You report those results. Or perhaps you are only interested in finding the best model, to be applied to future data? Then it would be OK to try multiple models on the full data set -- just don't attempt to reach any conclusions other than the best model to use in the future.