labterm not found error

[dougwyu]

Hi,

I'm interested in trying to correct for spatial autocorrelation, and I have come up with this syntax, but i get a labterm not found error.

I don't get this error if i use lv.formula = x*y instead of formula = x*y, but that doesn't seem correct to me

(x and y are geographic coordinates; Site = individual lakes, and we have multiple samples per lake)

Thanks in advance

gllvm(y = community_tab, X = covariates, formula = ~ x*y, num.lv.c = 2, family = binomial(), studyDesign = Site, row.eff = ~(1|Site), control.start=list(n.init = 1, jitter.var = 0)))

Error in gllvm(y = community_tab, X = covariates, formula = ~x * y, num.lv.c = 2, : object 'labterm' not found

[BertvanderVeen]

Thanks for pointing out this issue, Douglas. The model you are trying to fit should throw an error, but for some reason it does not, I will look into what causes that. Either how, num.lv.c can only be used in combination with lv.formula, if formula is also given. The lv.formula interface is for predictors -inside- the ordination, and the formula interface for predictors -outside- of the ordination. So, the model should be one of the two following:

gllvm(y = community_tab, X = covariates, lv.formula = ~ x*y, num.lv.c = 2, family = binomial(), studyDesign = Site, row.eff = ~(1|Site), control.start=list(n.init = 1, jitter.var = 0))) or gllvm(y = community_tab, X = covariates, formula = ~ x*y, num.lv = 2, family = binomial(), studyDesign = Site, row.eff = ~(1|Site), control.start=list(n.init = 1, jitter.var = 0))) I hope that gets you going for the moment, but let me know if it does not.

[dougwyu]

Thanks Bert, Just to be clear, gllvm(y = community_tab, X = covariates, formula = ~ x*y, num.lv.c = 2, family = binomial(), studyDesign = Site, row.eff = ~(1|Site), control.start=list(n.init = 1, jitter.var = 0))) throws the labterm error, as you say it should.

But clearly, i have misunderstood something conceptual. I wanted to use xy as a fixed effect, and use the latent variables to model the residuals (i.e. after using xy to remove spatial autocorrelation). (I'm just exploring)

I haven't understood what you mean by inside and outside the ordination. i'll keep on this. thanks,

[BertvanderVeen]

Yes, that's what I am saying. In that case, you are after the second model above, i.e.: gllvm(y = community_tab, X = covariates, formula = ~ x*y, num.lv = 2, family = binomial(), studyDesign = Site, row.eff = ~(1|Site), control.start=list(n.init = 1, jitter.var = 0)))

with num.lv instead of num.lv.c. The difference is whether (num.lv.c) the latent variables are modelled as a function of the predictors, or (num.lv) do not include information that can be accounted for by the predictors. It sounds like you are after the latter.