Process_only with or without the fixed effect

[Garyhsu1982]

Dear Brian

In your Cladocera Example, it says that "this will fit the “process error” model of MixSIR, which we MUST do when we only have one mix datapoint (or here, one mix datapoint per fixed effect)."

I have run the model for both the single data point and multiple data points with ID as the fixed effect, but it gives two different results. I have 20 sources in total For the same data point, I obtain

(1) process_only with the fixed effect p.ID 2.Berigische Land 0.083 0.088 0.002 0.004 0.020 0.051 0.117 0.270 0.327 p.ID 2.SM-Azuaga 0.229 0.122 0.036 0.051 0.138 0.215 0.304 0.448 0.494

(2) process_only (single point) p.global.Berigische Land 0.112 0.086 0.003 0.005 0.037 0.096 0.171 0.266 0.300 p.global.SM-Azuaga 0.113 0.088 0.003 0.005 0.036 0.096 0.175 0.274 0.301

I was wondering why there is a difference and which one is more correct or appropriate.

Best, Gary

[brianstock]

Hi Gary,

20 sources is a LOT... mixing models typically can't resolve more than 5-7, even if you have several tracers. Can you describe the nature of your data? What model options are you using? MixSIAR fits the source means and variances (and covariance if raw data are available), so the source positions could be changing between the two models. If you are fitting source data and only have 1 mix datapoint, the sources will be the main contributor to the likelihood.

Are you sure the models are both converged?

[Garyhsu1982]

I am dealing with lead isotope ratios of objects and my source data are potentially ore sources. The reason having some many sources is that lead isotope ratios of each ore source do not follow a normal distriubtion so I have to separate it into several normally distributed subgroups. Is normality a necessary condition for the mixing model? If it is not, I can treat each ore source as one source and reduce the numbers. In addition, I have more than 100 objects and want to know the contribution of each source. I can either run them together with ID in one run or do it individually for 100 runs. May I ask which way make more sense?

Many Thanks, Gary