Forum: open-discussion

RE: Bayesian version of the Fisher-Lee mod [ Reply ] By: Claudio Agostinelli on 2009-06-05 06:11	[forum:1602]
Hi Harry, OK, I will wait for your code. Probably I will be able to include it in the package in the begin of fall. Please make sure about the copyright of the code. Claudio

RE: Bayesian version of the Fisher-Lee mod [ Reply ] By: Harry Southworth on 2009-06-04 18:30	[forum:1601]
I'll dig out what I have and decide what to do. I'm no longer doing any work with circular data, so I don't think I'll be very motivated to spend much time on this. So I might just send you the code. 2006 was probably the last time I looked at this. The circular-linear regression functions I mentioned in 2006 are probably the same ones I mentioned in this thread.

RE: Bayesian version of the Fisher-Lee mod [ Reply ]
By: Claudio Agostinelli on 2009-06-04 14:25

[forum:1600]

Hi Harry,
thanks for providing us such information. I will be very happy to include your code in the next version of the package. If you are interest please join us in the project and add the code yourself (this will be the faster way). In such a case you need to write two functions. The first that does the calculation for radians (or degree) with fixed system of coordinate, the second that does the user interface with arguments checking and with conversion to the appropriate units/coordinates. If instead you would like to send me the code, I will do the port, but in this second case, it is going to take much longer. Also, in a 2006 email you wrote me, you mention about functions for circular-linear regression. Do you have them?

Claudio

Bayesian version of the Fisher-Lee mod [ Reply ]
By: Harry Southworth on 2009-06-02 17:47

[forum:1593]

I once wrote some code to do inference for the Fisher-Lee regression model using a Metropolis algorithm. It was written for S+ and (I think) the Metropolis algorithm was written in C.

The model was a little different than described by Fisher & Lee because their formulation is sensitive to simple shifts of the predictors. My version fixes that by explicitly including an intercept in the linear predictor rather than mean-centring the predictors. It also means that the uncertainty involved in estimating the mean/intercept is properly accounted for in the model. It might be problematic in some circumstances, but I never had any trouble with it. It would be easy to convert back to the Fisher-Lee formulation.

Anyway, if you want the code, let me know and I'll try to dig it out. The code belongs to my employer, not me, so I'd have to get agreement to release it. I don't think it would be a problem, but might take me a little while.

Harry