# Visual significance tests (due to Andreas Buha).

# Look at the nrow by ncol array of scatterplots, visually order them by
# decreasing strength of relationship, then see the position (row, col) 
# of the true scatterplot by typing .Last.value.  If the true plot shows 
# the i'th strongest relationship, the p-value is 1/(nrow*ncol).


# Test whether real-valued x and y are related.

scram.test.xy <- function (x, y, nrow=4, ncol=5)
{
  row.true <- sample(1:nrow,1)
  col.true <- sample(1:ncol,1)

  par (mfrow=c(nrow,ncol), mar=c(2,2,1,1), mgp=c(1,0.6,0))

  for (row in 1:nrow)
  { for (col in 1:ncol)
    { if (row==row.true && col==col.true)
      { py <- y
      }
      else
      { py <- y[sample(1:length(y),length(y))]  # can be done as py <- sample(y)
      }
      plot (x, py, xlab="", ylab="", pch=20)
    }
  }

  invisible(c(row.true,col.true))
}


# Test whether real-valued (x,y) and categorical c are related.  Here, c 
# should be a vector of colours, or a vector of positive integers.  

scram.test.xyc <- function (x, y, c, nrow=4, ncol=5)
{
  row.true <- sample(1:nrow,1)
  col.true <- sample(1:ncol,1)

  par (mfrow=c(nrow,ncol), mar=c(2,2,1,1), mgp=c(1,0.6,0))

  for (row in 1:nrow)
  { for (col in 1:ncol)
    { if (row==row.true && col==col.true)
      { pc <- c
      }
      else
      { pc <- c[sample(1:length(c),length(c))]  # can be done as pc <- sample(c)
      }
      plot (x, y, col=pc, xlab="", ylab="", pch=20)
    }
  }

  invisible(c(row.true,col.true))
}