Validating a combined ZAP GAM

[Sophie Loca]

Hi folks,

I'm trying to run a two part ZAP GAM on r-inla, one bernoulli part (with the smoother) and a zero truncated poisson part (no smoother). I am having trouble combining the two parts, the code for this i am taking from the second volume of 'Beginners guide to Spatial, Temporal and Spatial-Temporal Ecological Data Analysis with R-INLA'. 

The covariate matrix was taken from the Bernoulli part of the model to avoid having components of different lengths (page 391, vl 2).

Covariates <- data.frame(

  Intercept   = rep(1, N),

  bath1  = basis_bath[,"bath1"],

  bath2  = basis_bath[,"bath2"],

  bath3  = basis_bath[,"bath3"],

  bath4  = basis_bath[,"bath4"],

  bath5  = basis_bath[,"bath5"],

  bath6  = basis_bath[,"bath6"],

  bath7  = basis_bath[,"bath7"],

  bath8  = basis_bath[,"bath8"],

  bath9  = basis_bath[,"bath9"],

  dco1  = basis_dco[,"dco1"],

  dco2  = basis_dco[,"dco2"],

  dco3  = basis_dco[,"dco3"],

  dco4  = basis_dco[,"dco4"],

  dco5  = basis_dco[,"dco5"],

  dco6  = basis_dco[,"dco6"],

  dco7  = basis_dco[,"dco7"],

  dco8  = basis_dco[,"dco8"],

  dco9  = basis_dco[,"dco9"],

  bath.std = MyStd(Skate$bath),

  dcoast.std = MyStd(Skate$dcoast),

  btemp.std = MyStd(Skate$btemp),

  gravel.std = MyStd(Skate$gravel_sq),

  current.std = MyStd(Skate$current),

  sal.std = MyStd(Skate$sal_sq),

  mud.std = MyStd(Skate$mud_sq)

)

So after running each model, validating and completing model selection, I have tried to run the following code:

# Bernoulli part

X                           <- as.matrix(Covariates)

Beta01                 <- I.01$summary.fixed[,"mean"]

Pi                          <- exp(X %*% Beta01) / (1 + exp(X %*% Beta01))

#Count part

betaTP                <- I.pos$summary.fixed[1:N,"mean"]

mu                       <- exp(X %*% betaTP)

mu.ZTruncPois <-  mu / (1 - exp(-mu))

The problem is that when i try to run the lines with the %*% in it, i get an error saying:

"Error in X %*% Beta01 : non-conformable arguments". The same happens with the line in the count part.

Can anyone give any pointers as to how i could fix this problem? I have a working understanding of the code but the underlying programming logic is still something i am new to. 

Thanks,

Soph

[high...@highstat.com]

Sophie,

That means that the number of columns in X does not match the number of regression parameters in Beta01 and betaTP. So I guess that your 'Covariates' thing is not good defined.

Just type:

dim(X)  and length(Beta01) and length(betaTP) and you will see what the problem is.

If you do X %*% b,  then the number of columns in X needs to match the number of elements in b. Highschool maths..:-).

Alain

[Sophie]

Hi Alain,

Thank you for the quick reply, i managed to figure this out yesterday evening. When i read the book i misunderstood it as meaning the covariate matrix would have to be the same length for both parts of the model in order for it to work - silly me. 

Thanks a million,

Sophie