Hi all, I'm new to this forum and could use a little help and you all seem like a friendly bunch :-). I'm just entering my 4th year in October and am now in my write up phase and have an issue with some data for one of my chapters - I don't think I can do stats on it!

I work on hybridisation in fish, and fertilised a salmon females eggs with sperm from both a hybrid male (salmon x trout) and a pure salmon male to see how many of the eggs the hybrids can win when competing against the pure males. The hybrids are pants and only 3  got any paternity at all, with a maximum of 10%.

So my answer is clear- hybrids don't appear to win in sperm competition, but my problem is the stats. I have proportion data which for a start which is always a nightmare and I have loads of zeros from the hybrids and loads of 1's from the pures. If the data were normally distributed I would just do an independent samples t-test, but even arcsine transforming doesn't produce normality due to the high number of 0's and 1's. I also have a small sample size (9) as the experiment went wrong half way through the molecular work :-(

Are stats possible? Is a simple mann whitney sufficient?! Or am I simply going to have to describe the data with graphs. Any suggestions would be so gratefully received and any solutions may prompt me to send you a pint by post!

Siany xx

Completely outside my realms of knowledge so I can't tell you for sure one way or the other. But I wonder whether you could use 'zero inflated' models here?
http://www.kent.ac.uk/ims/personal/msr/webfiles/zip/zip.html
http://www.ingentaconnect.com/content/routledg/cjas/2000/00000027/00000003/art00009
http://www.kent.ac.uk/IMS/personal/msr/webfiles/zip/ibc_fin.pdf

Might be barking up the wrong tree though!

I'm no stats expert, but your data sounds similar to my masters data where I had small sample sizes, data that was not normally distributed and lots of zeros, I used a Chi Square Goodness of Fit Test. Perhaps that's something to look into? But Mann Whitney U might be sufficient too I think, but again I'm no expert.

Do you have a stats advisor at your Uni to ask? Or can you Supervisor give any insight?

Good luck!

Have you got multiple observations for each thing (fish thing?) if so you could look at multilevel modelling.

Thanks for the replies guys,

Regrettably due to the cost of genotyping offspring each male-male competition was done only once. So multi level modelling wouldn’t be an option. I’ll look into the chi sq and the zero inflated models though.

Unfortunately my supervisor is a bit clueless on stats and is away for three weeks anyway! I’ve asked a couple of fellow PhD’s in my dept but no one knows and seems a bit unwilling to help, I don’t want to pester people.
Thanks again for the help! Xx