agridat (version 1.23)

besag.endive: Presence of footroot disease in an endive field

Description

Presence of footroot disease in an endive field

Format

A data frame with 2506 observations on the following 3 variables.

col

column

row

row

disease

plant is diseased, Y=yes,N=no

Details

In a field of endives, does each plant have footrot, or not? Data are binary on a lattice of 14 x 179 plants.

Modeled as an autologistic distribution.

We assume the endives are a single genotype.

Besag (1978) may have had data taken at 4 time points. This data was extracted from Friel and Pettitt. It is not clear what, if any, time point was used.

Friel does not give the dimensions. Besag is not available.

References

N Friel & A. N Pettitt (2004). Likelihood Estimation and Inference for the Autologistic Model. Journal of Computational and Graphical Statistics, 13:1, 232-246. https://doi.org/10.1198/1061860043029

Examples

if (FALSE) {
  
  library(agridat)
  data(besag.endive)
  dat <- besag.endive

  # Incidence map.  Figure 2 of Friel and Pettitt
  libs(desplot)
  grays <- colorRampPalette(c("#d9d9d9","#252525"))
  desplot(dat, disease~col*row,
          col.regions=grays(2),
          aspect = 0.5, # aspect unknown
          main="besag.endive - Disease incidence")
  
  
  # Besag (2000) "An Introduction to Markov Chain Monte Carlo" suggested
  # that the autologistic model is not a very good fit for this data.
  # We try it anyway.  No idea if this is correct or how to interpret...
  
  libs(ngspatial)
  A = adjacency.matrix(179,14)
  X = cbind(x=dat$col, y=dat$row)
  Z = as.numeric(dat$disease=="Y")
  m1 <- autologistic(Z ~ 0+X, A=A, control=list(confint="none"))
  
  summary(m1)
  ## Coefficients:
  ##      Estimate Lower Upper MCSE
  ## Xx  -0.007824    NA    NA   NA
  ## Xy  -0.144800    NA    NA   NA
  ## eta  0.806200    NA    NA   NA

  
  if(require("asreml", quietly=TRUE)) {
    libs(asreml,lucid)
    
    # Now try an AR1xAR1 model.
    dat2 <- transform(dat, xf=factor(col), yf=factor(row),
                      pres=as.numeric(disease=="Y"))
    
    m2 <- asreml(pres ~ 1, data=dat2,
                 resid = ~ar1(xf):ar1(yf))
    # The 0/1 response is arbitrary, but there is some suggestion
    # of auto-correlation in the x (.17) and y (.10) directions,
    # suggesting the pattern is more 'patchy' than just random noise,
    # but is it meaningful?
    
    lucid::vc(m2)
    ##       effect component std.error z.ratio bound 
    ##     xf:yf(R)   0.1301   0.003798    34       P   0
    ## xf:yf!xf!cor   0.1699   0.01942      8.7     U   0
    ## xf:yf!yf!cor   0.09842  0.02038      4.8     U   0
  }

}