tests/testthat/test-extract_svyfgtdec.R

In convey: Income Concentration Analysis with Complex Survey Samples

context("FGT decomposition output survey.design and svyrep.design")

skip_on_cran()

library(laeken)
library(survey)

data(api)
apistrat[ , sapply( apistrat, is.integer ) ] <- apply( apistrat[ , sapply( apistrat, is.integer ) ], 2, as.numeric )
dstrat1<-convey_prep(svydesign(id=~1,data=apistrat))
for ( this_thresh in c( "abs" , "relm" , "relq" ) ){
  for (this_g in 2:3 ) {
    test_that("svyfgtdec works on unweighted designs",{
      svyfgtdec(~api00, design=dstrat1, g=this_g, type_thresh= this_thresh, percent = 1, abs_thresh=600 , na.rm  = FALSE )
    })
  }
}

test_that("output svyfgtdec",{
  skip_on_cran()

  data(eusilc) ; names( eusilc ) <- tolower( names( eusilc ) )
  eusilc[ , sapply( eusilc, is.integer ) ] <- apply( eusilc[ , sapply( eusilc, is.integer ) ], 2, as.numeric )

  des_eusilc <- svydesign(ids = ~rb030, strata =~db040,  weights = ~rb050, data = eusilc)
  des_eusilc <- convey_prep(des_eusilc)
  des_eusilc_rep <-as.svrepdesign(des_eusilc, type= "bootstrap" , replicates = 20 )
  des_eusilc_rep <- convey_prep(des_eusilc_rep)

  # database-backed design
  library(RSQLite)
  library(DBI)
  dbfile <- tempfile()
  conn <- dbConnect( RSQLite::SQLite() , dbfile )
  dbWriteTable( conn , 'eusilc' , eusilc )

  dbd_eusilc <-
    svydesign(
      ids = ~rb030 ,
      strata = ~db040 ,
      weights = ~rb050 ,
      data="eusilc",
      dbname=dbfile,
      dbtype="SQLite"
    )

  dbd_eusilc <- convey_prep( dbd_eusilc )

  # create a hacky database-backed svrepdesign object
  # mirroring des_eusilc_rep
  dbd_eusilc_rep <-
    svrepdesign(
      weights = ~ rb050,
      repweights = des_eusilc_rep$repweights ,
      scale = des_eusilc_rep$scale ,
      rscales = des_eusilc_rep$rscales ,
      type = "bootstrap" ,
      data = "eusilc" ,
      dbtype="SQLite" ,
      dbname = dbfile ,
      combined.weights = FALSE
    )

  dbd_eusilc_rep <- convey_prep( dbd_eusilc_rep )


  for ( this_thresh in c( "abs" , "relm" , "relq" ) ){
    for (this_g in 2:3 ) {

      a1 <- svyfgtdec( ~eqincome, design=des_eusilc, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      a2 <- svyby( ~eqincome, by = ~rb090, design = des_eusilc, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)
      b1 <- svyfgtdec( ~eqincome, design=des_eusilc_rep, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      b2 <- svyby( ~eqincome, by = ~rb090, design = des_eusilc_rep, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)

      se_dif1 <- max(abs(SE(a1)-SE(b1)))
      se_diff2 <- max(abs(SE(a2)-SE(b2)))

      expect_is(coef(a1),"numeric")
      expect_is(coef(a2), "numeric")
      expect_is(coef(b1),"numeric")
      expect_is(coef(b2),"numeric")
      expect_equal(coef(a1), coef(b1))
      expect_equal(coef(a2), coef(b2))
      expect_is(SE(a1),"numeric")
      expect_is(SE(a2), "svyby" )
      expect_is(SE(b1),"numeric")
      expect_is(SE(b2),"svyby")
      expect_lte(confint(a1)[1,1], coef(a1)[1])
      expect_gte(confint(a1)[1,2], coef(a1)[1])
      expect_lte(confint(a1)[2,1], coef(a1)[2])
      expect_gte(confint(a1)[2,2], coef(a1)[2])
      expect_lte(confint(a1)[3,1], coef(a1)[3])
      expect_gte(confint(a1)[3,2], coef(a1)[3])
      expect_lte(confint(a1)[4,1], coef(a1)[4])
      expect_gte(confint(a1)[4,2], coef(a1)[4])
      expect_lte(confint(a1)[1,1], coef(a1)[1])
      expect_gte(confint(b1)[1,2], coef(b1)[1])
      expect_lte(confint(b1)[2,1], coef(b1)[2])
      expect_gte(confint(b1)[2,2], coef(b1)[2])
      expect_lte(confint(b1)[3,1], coef(b1)[3])
      expect_gte(confint(b1)[3,2], coef(b1)[3])
      expect_lte(confint(b1)[4,1], coef(b1)[4])
      expect_gte(confint(b1)[4,2], coef(b1)[4])
      expect_equal(sum(confint(a2)[,1]<= coef(a2)),length(coef(a2)))
      expect_equal(sum(confint(a2)[,2]>= coef(a2)),length(coef(a2)))
      expect_equal(sum(confint(b2)[,1]<= coef(b2)),length(coef(b2)))
      expect_equal(sum(confint(b2)[,2]>= coef(b2)),length(coef(b2)))

      c1 <- svyfgtdec( ~eqincome, design=dbd_eusilc, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      c2 <- svyby( ~eqincome, by = ~rb090, design = dbd_eusilc, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)

      # database svyfgtdec
      expect_equal(coef(a1), coef(c1))
      # expect_equal(rev(coef(a2)), coef(c2)) # inverted results
      expect_equal(SE(a1), SE(c1))
      # expect_equal(rev(SE(a2)), SE(c2)) # inverted results


      # compare subsetted objects to svyby objects
      sub_des <- svyfgtdec( ~eqincome, design=subset( des_eusilc, rb090 == "male"), g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      sby_des <- svyby( ~eqincome, by = ~rb090, design = des_eusilc, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)
      sub_rep <- svyfgtdec( ~eqincome, design=subset( des_eusilc_rep, rb090 == "male"), g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      sby_rep <- svyby( ~eqincome, by = ~rb090, design = des_eusilc_rep, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)

      # subsets equal svyby
      expect_equal(as.numeric(coef(sub_des)), as.numeric(coef(sby_des)[ grepl( "^male", names(coef(sby_des)) ) ]))
      expect_equal(as.numeric(coef(sub_rep)), as.numeric(coef(sby_rep)[ grepl( "^male", names(coef(sby_rep)) ) ]))
      expect_equal(as.numeric(SE(sub_des)), as.numeric(SE(sby_des)[1,]))
      expect_equal(as.numeric(SE(sub_rep)), as.numeric(SE(sby_rep)[1,]))

      # coefficients should match across svydesign & svrepdesign
      expect_equal(coef(sby_des), coef(sby_rep))

      # coefficients of variation should be within five percent
      cv_dif <- abs(cv(sby_des)-cv(sby_rep))
      expect_lte( max( unlist(cv_dif) ) , .05 )




      sub_dbd <- svyfgtdec( ~eqincome, design=subset( dbd_eusilc, rb090 == "male"), g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      sby_dbd <- svyby( ~eqincome, by = ~rb090, design = dbd_eusilc, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)
      sub_dbr <- svyfgtdec( ~eqincome, design=subset( dbd_eusilc_rep, rb090 == "male"), g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE )
      sby_dbr <- svyby( ~eqincome, by = ~rb090, design = dbd_eusilc_rep, FUN = svyfgtdec, g=this_g, type_thresh=this_thresh, percent = .6, abs_thresh=15000 , na.rm  = FALSE, deff = FALSE)

      # compare database-backed designs to non-database-backed designs
      # dbi subsets equal non-dbi subsets
      expect_equal(coef(sub_des), coef(sub_dbd))
      expect_equal(coef(sub_rep), coef(sub_dbr))
      expect_equal(SE(sub_des), SE(sub_dbd))
      expect_equal(SE(sub_rep), SE(sub_dbr))



      # compare database-backed subsetted objects to database-backed svyby objects
      # dbi subsets equal dbi svyby
      expect_equal(as.numeric(coef(sub_dbd)), as.numeric(coef(sby_dbd[2,])) ) # inverted results!
      expect_equal(as.numeric(coef(sub_dbr)), as.numeric(coef(sby_dbr[2,])) ) # inverted results!
      expect_equal(as.numeric(SE(sub_dbd)), as.numeric(SE(sby_dbd[2,])) ) # inverted results!
      expect_equal(as.numeric(SE(sub_dbr)), as.numeric(SE(sby_dbr[2,])) ) # inverted results!


    }
  }

  dbRemoveTable( conn , 'eusilc' )
  dbDisconnect( conn )

})