Kate's answer definitely will help, but perhaps the more difficult problem is separating heritable behavior traits from learned ones. This is at least the problem I am trying to navigate with male wild horses, who spend a considerable amount of time in their family group before dispersing and forming their own group. I'm working on a social network analysis approach at the moment, with the intent on incorporating genetic data, but still a ways off from nailing this down i a robust quantitative framework. This may not be a problem if you're working with a less social organism or even one with less parental care. I'm happy to discuss potential quantitative solutions if you have similar challenges.

I see, thank you a lot!!! This page is really helpful!!!

Jason, I work on spiders...no challenges like yours...Good Luck!

sounds like an animal model type analysis would indeed help you.

For software pack/age options to fit animal models MCMCglmm is free and R based, but make sure you understand the importance of priors etc or you can generate nonsense. I'd v strongly recommend reading all Hadfield's course notes if this is a new area of stats for you. If you want free and likelihood, rather than Bayesian you could try WOMBAT, or there is an R package called PEDIGREEMM that may be useful - at least for univariate animal models. If you have access to funds then ASReml or ASReml-R might be worth a look. It is pretty cheap if you are a student, pretty pricey if you aren't!

Jason - If you can specify your social network as a matrix of association between all individuals, then you can include this in a mixed model in exactly the same way as an animal model uses a matrix of genetic relatedness. See e.g. Stopher et al 2011 (I think) Evolution paper on red deer on Rum. this was spatial overlap but could equally be temporal measures of association. Fitting random effects associated with genetic and social "relatedness" matrices simultaneously is easy (practically) in mixed models using ASReml or MCMCglmm. Of course if all your genetic relatives hang out together in social groups the data may lack power to tease the two sources of behavioural similarity apart. Worth a shot though... especially if you get a lot of stallions dispersing to spread genes across bands etc.

Although I completely agree with the other people posting here, that the MCMCglmm and Baysian method are the most optimal ways of analyzing your data, I would suggest to use several different methods. You could start by using simple (mid)parent (mean)offspring regressions and variation between nests, and end with the more complicated Animal Model approaches. The reason for this is that there are several assumptions in all methods. It is good to realize that only by knowing your data, you will be able to recognize the limitations and specificities of your dataset.

You now have the assumption that there are no maternal effects, but are you able to be certain about this and are you able to separate out such a variance component. What doe the additive genetic component include in your specific dataset? Are there common environment effects or possibilities for parental effects via food or hormones?

So in general I have the opinion that if all methods give similar results you should use the best method like proposed above by Kate and Alistair. If that is not the case you should look into what is the cause for that.

Hops this helps,