Non-linear comparisons

[SergeantMike]

I attribute the kinks to inconsistencies in the data collection. I have other data that is being collected that is much better controlled as far as the collection. This was the first year and we were working out the kinks.

However, that said the r-squared for each line using this data was all over O. 95 so I'm still pretty confident.

What I am trying "prove" is to create a pretty solid argument that using this approach is more robust and useful than using transformed data and a linear model based ANOVA. This would add a level of confidence in the differences we see on the graph. There will be concrete differences between the lines.

TBH My advisor is a pretty heavy advocate of the ANOVA method because it is what he was taught. I feel this is a better description of the data. I don't know what it means in nature when the stats are transformed using arcsine. The actual emergence on day 4 is the arcsine of what I counted?

 

[katxt]

What I am trying "prove" is to create a pretty solid argument that using this approach is more robust and useful than using transformed data and a linear model based ANOVA. This would add a level of confidence in the differences we see on the graph. There will be concrete differences between the lines.

I feel that you may be setting yourself a very difficult if not impossible task. All we have done so far is go through the mechanics of how to find a SE in an unusual situation, and use it to get some p values. There is no discussion of robustness, nor of goodness of fit, nor of usefulness. It is a brave (read risky) matter to go against a supervisor, and in this case I think you would need a much deeper and more solid foundation in statistical theory to prove your point, if indeed it is true. (There are other simpler models you could consider as well.)
As for the arcsine transformation and the anova, this isn't really an anova situation at all as far as I can see. Anova compares the means of different groups and sometimes the data that gives those means needs to be transformed so that the mathematical anova calculations will be valid.
On the other hand, insects are not my field. Your supervisor is a better source of advice than a random guy of the internet. If you are really concerned, consider buying an hour of a statisticians time where you can discuss the whole thing face to face.
Good luck, kat

 

[katxt]

Looking at your graphs, if you want to know which are different, another approach could be the Two-sample Kolmogorov–Smirnov test. See https://en.wikipedia.org/wiki/Kolmogorov–Smirnov_test

 

[SergeantMike]

Looking at your graphs, if you want to know which are different, another approach could be the Two-sample Kolmogorov–Smirnov test. See https://en.wikipedia.org/wiki/Kolmogorov–Smirnov_test

Ok, I will look into it, thank you! More and more learning.


In the mean time I was wondering what your experience is with Euclidean distance calculations. I have a program that I am developing that calculates these for a Bioinformatics program and my results in Excel and in the program seem to differ from the commercially available packages. I was hoping you could check my work.

 

[katxt]

Sure. Zip up the Excel and send it through and I'll see if understand.
The K-S test won't specifically test the t50s but will test for differences between the emergence timings as a whole. It is extremely simple, you already have all the data you need, and does not need the Gompertz model or the arcsine transform.

 

[SergeantMike]

Thank you again! Each sheet has two versions of the data, just transposed. I was using the second form with the sample number on the left and the allele number across the top.

 

[katxt]

And you want to know the Euclidian distances between the samples? Probably best to gmail me at katxt345