test_GEE_WRM.R
In spind: Spatial Methods and Indices

context('Test GEE and WRM outputs')

test_that('Coefficient estimates + SEs are as expected', {
  skip_on_cran()
  data(musdata)
  coords <- musdata [ ,4:5]

  formula <- as.formula(musculus ~ pollution + exposure)

  mgee <- GEE(formula, family = "poisson", data = musdata,
              coord = coords, corstr = "fixed", plot = FALSE, scale.fix = FALSE)

  mwrm <- WRM(musculus ~ pollution + exposure, family = "poisson",
              data = musdata, coord = coords, level = 1, plot = F)
  expect_equal(as.vector(mgee$b), c(-1.904747, 3.362160, -1.463483),
               tolerance = 1e-7)
  expect_equal(as.vector(mwrm$b), c(-1.935992, 3.184136, -1.228579),
               tolerance = 1e-7)
  expect_equal(as.vector(mgee$s.e.), c(1.3109113, 0.9141623, 0.8801014),
               tolerance = 1e-7)
  expect_equal(as.vector(mwrm$s.e.), c(1.917704, 1.225106, 1.506331),
               tolerance = 1e-7)

  expect_true(is.list(mgee))
  expect_is(mgee, 'GEE')
  expect_true(is.list(mwrm))
  expect_is(mwrm, 'WRM')

  expect_is(mgee$plot, 'ggplot')
  expect_is(mwrm$plot, 'ggplot')
})

test_that('predict.GEE and .WRM interact properly with parent functions',{
  skip_on_cran()
  data(musdata)
  coords <- musdata [ ,4:5]

  formula <- as.formula(musculus ~ pollution + exposure)

  mgee <- GEE(formula, family = "poisson", data = musdata,
              coord = coords, corstr = "fixed", plot = F, scale.fix = FALSE)


  mwrm <- WRM(musculus ~ pollution + exposure, family = "poisson",
              data = musdata, coord = coords, level = 1, plot = FALSE)
  expect_match(class(predict.GEE(mgee, newdata = musdata)), 'numeric')
  expect_match(class(predict.WRM(mwrm, newdata = musdata,
                                 sm = TRUE, newcoord = coords)), 'numeric')
  expect_match(class(predict.WRM(mwrm, newdata = musdata)), 'numeric')

  expect_equal(predict.GEE(mgee, newdata = musdata), mgee$fitted,
               tolerance = 1e-5)
  expect_equal(predict.WRM(mwrm, newdata = musdata), as.vector(mwrm$fitted),
               tolerance = 1e-5)

})

test_that('fails correctly', {
  skip_on_cran()
  data(musdata)
  coords <- musdata [ ,4:5]
  # rm(formula)

  expect_error(GEE(formula, family = "poisson", data = musdata,
               coord = coords, corstr = "fixed", plot = FALSE, scale.fix = FALSE),
               "formula: specified notation is missing")

  formula <- as.formula(musculus ~ pollution + exposure)

  expect_warning(GEE(formula, family = "poisson", data = musdata,
                  coord = coords, corstr = "fixed",
                  plot = TRUE, scale.fix = FALSE,
                  customize_plot = theme(axis.title = element_blank())),
                 regexp = '"customize_plot" and')


  coord <- cbind(rnorm(400), runif(400))

  expect_error(GEE(formula, family = "poisson", data = musdata,
                   coord = coord, plot = FALSE),
               "coordinates not integer")

  coord <- cbind(round(runif(300)), round(runif(300)))

  expect_error(GEE(musculus ~ pollution + exposure, family = "poisson",
               data = musdata, coord = coord, plot = F),
               "length of data does not match length of coordinates")


  coord <- cbind(rnorm(400), runif(400))

  expect_error(WRM(formula, family = "poisson", data = musdata,
                   coord = coord,plot = F),
               "coordinates not integer")

  coord <- cbind(round(runif(300)), round(runif(300)))

  expect_error(WRM(musculus ~ pollution + exposure, family = "poisson",
                   data = musdata, coord = coord, plot = F),
               "length of data does not match length of coordinates")

})


test_that("GEE's alternative corr+error structures work",{
  data(musdata)
  coords<- musdata[,4:5]

  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="exchangeable", cluster = 2,
            plot=F)

  expect_equal(as.vector(mgee$b), c(-1.539848, 3.783996, -2.34400),
               tolerance = 1e-7)
  expect_equal(as.vector(mgee$s.e.), c(.8850701, .8142585, .7519995),
               tolerance = 1e-7)


  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="quadratic", cluster = 2,
            plot=F)

  expect_equal(as.vector(mgee$b), c(-1.778806, 3.999191, -2.336889),
               tolerance = 5e-7)
  expect_equal(as.vector(mgee$s.e.), c(.8890819, .7705253, .7116026),
               tolerance = 1e-7)


  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="independence",
            plot=F)

  mglm<-glm(musculus ~ pollution + exposure, family = poisson(),
            data = musdata)

  expect_equal(as.vector(mgee$b), as.vector(mglm$coefficients),
               tolerance = 1e-7)
  expect_equal(as.vector(mgee$s.e.),
               as.vector(summary(mglm)$coefficients[ ,2]),
               tolerance = 1e-7)


  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="exchangeable", cluster = 3,
            plot=F)

  expect_equal(as.vector(mgee$b), c(-1.671703, 4.092487, -2.51481),
               tolerance = 1e-6)
  expect_equal(as.vector(mgee$s.e.), c(0.963654, .8203059, .6312397),
               tolerance = 1e-7)


  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="quadratic", cluster = 3,
            plot=F)

  expect_equal(as.vector(mgee$b), c(-2.715358, 4.492083, -1.850096),
               tolerance = 1e-7)
  expect_equal(as.vector(mgee$s.e.), c(.9045938, .8047187, .5717413),
               tolerance = 1e-7)


  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="exchangeable", cluster = 4,
            plot=F)

  expect_equal(as.vector(mgee$b), c(-1.685432, 3.237147, -1.620231),
               tolerance = 1e-6)
  expect_equal(as.vector(mgee$s.e.), c(.7889265, 1.1213838, .8657069),
               tolerance = 1e-7)

  mgee<-GEE(musculus ~ pollution + exposure, "poisson", musdata,
            coord=coords, corstr="quadratic", cluster = 4,
            plot=F)

  expect_equal(as.vector(mgee$b), c(-2.303986, 3.829125, -1.660454),
               tolerance = 1e-7)
  expect_equal(as.vector(mgee$s.e.), c(1.1420045, .9939361, .6020102),
               tolerance = 1e-7)



})