Hi

Just looking for a bit ofadvice, I recently sat my viva for transfer from Mphil to PhD, prior to theviva both my supervisors said i had enough data and a strong document, almostworthy of PhD as it stood.

I am working with insects andmeasuring there response to chemical cues originating from food products (theinsects are assumed to be visual predators and no chemo response has previouslybeen documented in the species) However there response is not normallydistributed, so i have used non parametric statistics to compare response timesto the chemical compared to control data (without the chemical) in a Y tubetype test.

I was told in the viva (bytwo bio-chemists) that i could not use non normal data and to either transformit to make it normal and then apply "real" stats to the results orrepeat the last two years worth of experiments so i could use a form ofPrinciple component analysis.

I have used various tests andthe data is still not normal, what is the predudice against non parametricstatistics? I understand in other sciences parametric data is very important,but in animals who have individual traits and behaviours i cant see how it ispossible!



I am obviously reluctant to repeat everything and theseconcerns were not raised in the last viva (Msc to Mphil)

Has anyone else had similar issues with non normal data?

Be grateful of any advice, should i just repeat it all , orre submit and have a strong argument as to my reasons behind using non parametricdata.

Many thanks

Jay

Interesting, I think given that most people are trained to focus on parametric data there's a bias and tendency to utilise parametric stuff despite violation of normality assumptions. Generally, amongst most academics there seems to be a real gap in knowledge regarding non para and I think that's probably filtered during the supervision process ie that people are not comfortable with working with non para analysis given knowledge gap.

I've used non parametric stuff throughout my upgrade (mphil to PhD upgrade), my submitted PhD, following my viva and now creating the final PhD draft following my viva. I integrated reasons for selecting non parametric tests and clearly stated why my quantitative data was not suitable for parametric techniques. My examiners were fine with it and have not recommended that I re-analysis everything.

Guess it all depends on supervisors, examiners and datasets!

Thanks for the reply, That's really interesting,
and nice to know.  I think i may stick with my original analysis and resubmit
with better explanations of why my data is suited to non-parametric stats. Do
you know, off hand,  of any recent papers that have used non-parametric data in
biological sciences? If you could let me know that would be great!  Thanks
Jay.

Not of the top of my head, but my PhD is within social sciences so I'm not familiar with biological science literature at all.

A really quick look at some biological articles reveals some usage of non parametric testing though......

https://www.hepato-gastroenterology.org/wp-content/uploads/2012/01/LIVER_11181.pdf
http://www.pregnancyhypertension.org/article/S2210-7789%2812%2900160-2/abstract
http://onlinelibrary.wiley.com/doi/10.1111/j.1600-051X.2011.01825.x/abstract

Perhaps that's another reason for their suspected fixation on parametric stuff? ie considering biological science seems to place considerable emphasis on scientific approach, positivism, deductivism etc etc rather than alternatives to situations or disciplines which are more willing to integrate alternatives? - although appreciate I may be barking up the wrong tree here :$

My examiners and my supervisors all work within the same field and they know that sometimes complying with normality assumptions with some datasets is just not possible or practical, so they have some understanding and appreciation of non-parametric tests. So maybe that's another reason why their more flexible with their approach? I've used MW tests, KW tests, Spearman Rho as well as chi sq and fishers extended tests and general tests for normality (ie K-S tests).

Hi Jay,

I also think that their comments are due to a gap of knowledge than anything else. I don't think you can "correct" the data- my opinion is that this is "cherry picking". If you have time watch statistics in Berkeley on i tunes (http://itunes.apple.com/us/itunes-u/statistics-2-001-fall-2009/id354822870). There is a very nice example of researchers deleting one point as an "outlier" while it was actually showing that the engine of the rocket will explode if outdoor temperature was too low. Which of course exploded.
In any case, in real life situations you will always get a skewed distribution... All my data are skewed and I plan to use non-parametric tests. Because I am still struggling with my statistics, why don't you ask for support from the smart nerdy guys.

Yes, it is a good idea to see how other people in your field analysed the data.

This sounds very similar to something I faced from the other side - I did a statistics phd with a biological problem and it was very hard to convince the biologists why I was doing certain statistical methods. It sounds like these people have a very fixed idea of how the problem should be solved and don't really understand your approach, a lack in their knowledge and not yours.

Maybe you just need to explain and reference it better why your approach is valid for your problem - what about doing an appendix on the general use of nonparametric testing to support your argument?

Also have you tried doing transformation to normal and seeing how the results compare for a parametric problem? Or would that be too much work to do?