context("Testing suberbPlot") test_that("PRELIMINARY TESTS (1/4)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') library(grid) library(gridExtra) # files are exported for validation with Mathematica's MeanPlot # write.table(ToothGrowth, file = "file0.dat", sep = "\t", col.names = FALSE) plt <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", statistic = "mean", plotStyle="bar" ) expect_equal( "ggplot" %in% class(plt), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("PRELIMINARY TESTS (2/4)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') library(grid) g0 <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", statistic = "mean", adjustments = list(purpose = "difference"), plotStyle="bar" ) g1 <- g0 + xlab("Dose") + ylab("Tooth Growth") + labs(title="adsf") + theme_light(base_size=20) + annotation_custom(grid.text("allo",x=.5,y=.5,gp=gpar(fontsize=20, col="grey"))) g2 <- g1 + theme(axis.text.x = element_text(size=30, colour="red") ) + coord_cartesian(ylim=c(5,45)) expect_equal( "ggplot" %in% class(g2), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("PRELIMINARY TESTS (3/4)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') res <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", showPlot=FALSE ) expect_output( str(res), "data.frame") # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("PRELIMINARY TESTS (4/4)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') p <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", statistic = "mean", plotStyle="line" ) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) ######################################### # SYSTEMATIC TESTS OF THE STATISTICS ######################################### test_that("test 1a: 3 groupes inpependants", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') dta1a <- GRD( SubjectsPerGroup = 20, BSFactors = "Group(3)", Population = list( mean=10, stddev = 5) ) # write.table(dta1a, file = "test1a.dat", sep = "\t", col.names = FALSE) p <- superbPlot(dta1a, BSFactor = "Group", variables = "DV", statistic = "mean", errorbar = "SE", plotStyle="line") expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 1b: factorielle a grps independants; 3 x 2", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') dta1b <- GRD( SubjectsPerGroup = 20, BSFactors = "Group(3): Sex(2)", Population = list( mean=10, stddev = 5)) # write.table(dta1b, file = "test1b.dat", sep = "\t", col.names = FALSE) p <- superbPlot(dta1b, BSFactor = c("Group","Sex"), variables = "DV", statistic = "mean", errorbar = "SE" ) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 2a: 1 facteur a 3 mesures repetees; (3)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'warnings') dta2a <- GRD( WSFactors = "Moment(3)", SubjectsPerGroup = 5, Population = list( mean=10, stddev = 5)) # write.table(dta2a, file = "test2a.dat", sep = "\t", col.names = FALSE) expect_message( p <- superbPlot(dta2a, WSFactor = "moment(3)", adjustments=list(decorrelation="CA"), errorbar = "CI", plotStyle="line", variables = c("DV.1","DV.2","DV.3") )) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 2b: 2 facteurs a mesures repetees; (3 x 2)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') dta2b <- GRD( WSFactors = "Moment(3): Dose(2)", SubjectsPerGroup = 5, Population = list( mean=10, stddev = 5, rho = .80)) # write.table(dta2b, file = "test2b.dat", sep = "\t", col.names = FALSE) p <- superbPlot(dta2b, WSFactor = c("moment(3)","Dose(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic="mean", errorbar = "CI", gamma = 0.90, plotStyle = "line", adjustments = list(purpose="difference", decorrelation="CM"), errorbarParams = list(position = position_dodge(width = .15)), pointParams = list(position = position_dodge(width = .15)), ) expect_equal( "ggplot" %in% class(p), TRUE) options("superb.feedback" = c("design","warnings") ) # restores to default }) test_that("test 3: scheme mixte; 3 x (3)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'warnings') dta3 <- GRD( BSFactors = "Group(3)", WSFactors = "Moment(3)", SubjectsPerGroup = 5, Population = list( mean=10, stddev = 5), Effects = list("Moment" = slope(5)) ) # write.table(dta3, file = "test3.dat", sep = "\t", col.names = FALSE) expect_message( p <- superbPlot(dta3, WSFactor = "Moment(3)", BSFactor = "Group", variables = c("DV.1","DV.2","DV.3"), statistic = "mean", errorbar = "SE", plotStyle="line", adjustments = list(purpose="single", decorrelation="CM") )) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 4a: scheme a trois facteurs, 2 etant between 3 x 3 x (3)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') dta4a <- GRD( BSFactors = "Group(3) : Dose(3)", WSFactors = "Moment(3)", SubjectsPerGroup = 4, Population = list( mean=10, stddev = 5), Effects = list("Moment" = slope(5)) ) # write.table(dta4a, file = "test4a.dat", sep = "\t", col.names = FALSE) p <- superbPlot(dta4a, BSFactor = c("Group","Dose"), WSFactor = "Moment(3)", variables = c("DV.1","DV.2","DV.3"), plotStyle = "line", statistic = "mean", errorbar = "SE", adjustments = list(purpose="difference", decorrelation="none"), factorOrder = c("Dose","Group","Moment"), showPlot = T) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 5a: scheme a quatre facteurs; 5 x 4 (3 x 2)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') dta5a <- GRD( SubjectsPerGroup = 20, BSFactors = "Group(5) : Dose(4)", WSFactors = "Moment(3):Hand(2)", Population = list( mean=10, stddev = 5, rho = .90), Effects = list("Moment" = slope(5), "Hand" = slope(10)) ) # write.table(dta5a, file = "test5a.dat", sep = "\t", col.names = FALSE) p <- superbPlot(dta5a, plotStyle="line", WSFactor = c("Moment(3)","Hand(2)"), BSFactor= c("Group","Dose"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "mean", errorbar = "CI", gamma = .9999, adjustments = list(purpose="difference", decorrelation="CM") ) expect_equal( "ggplot" %in% class(p), TRUE) options("superb.feedback" = c("design","warnings") ) # restores to default }) ######################################### # SYSTEMATIC TESTS OF THE OPTIONS ######################################### test_that("test 6: Some data", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') dta6 <- GRD( WSFactors = "Moment(3):Hand(2)", Effects = list("Moment" = slope(5), "Hand" = slope(3)), SubjectsPerGroup = 6, Population = list (mean = 20, stddev = 5, rho = 0.8) ) expect_output( str(dta6), "data.frame") # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6a: factorOrder", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') library(gridExtra) dta6 <- GRD( WSFactors = "Moment(3):Hand(2)", Effects = list("Moment" = slope(5), "Hand" = slope(3)), SubjectsPerGroup = 6, Population = list (mean = 20, stddev = 5, rho = 0.8) ) # factorOrder p1 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "mean", errorbar = "SE", factorOrder = c("Moment", "Hand") ) p2 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "mean", errorbar = "SE", factorOrder = c("Hand","Moment") ) p <- grid.arrange(p1,p2,ncol=2) expect_equal( "ggplot" %in% class(p1), TRUE) expect_equal( "ggplot" %in% class(p2), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6b: adjustments CA vs CM vs LM vs UA", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') library(gridExtra) dta6 <- GRD( WSFactors = "Moment(3):Hand(2)", Effects = list("Moment" = slope(5), "Hand" = slope(3)), SubjectsPerGroup = 6, Population = list (mean = 20, stddev = 5, rho = 0.8) ) # adjustments CA vs CM vs LM vs UA p1 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), adjustments = list(purpose="difference", decorrelation="CA") )+ coord_cartesian( ylim = c(8,30) ) + labs(title="CA") p2 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), adjustments = list(purpose="difference", decorrelation="CM") )+ coord_cartesian( ylim = c(8,30) ) + labs(title="CM") p3 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), adjustments = list(purpose="difference", decorrelation="LM") )+ coord_cartesian( ylim = c(8,30) ) + labs(title="LM") p4 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), adjustments = list(purpose="difference", decorrelation="UA") )+ coord_cartesian( ylim = c(8,30) ) + labs(title="UA") p <- grid.arrange(p1,p2,p3,p4,ncol=4) expect_equal( "ggplot" %in% class(p1), TRUE) expect_equal( "ggplot" %in% class(p2), TRUE) expect_equal( "ggplot" %in% class(p3), TRUE) expect_equal( "ggplot" %in% class(p4), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6c: statistics of central tendency mean, median and gmean", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') library(gridExtra) dta6 <- GRD( WSFactors = "Moment(3):Hand(2)", Effects = list("Moment" = slope(5), "Hand" = slope(3)), SubjectsPerGroup = 6, Population = list (mean = 20, stddev = 1, rho = 0.8) ) p1 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "mean", errorbar = "CI" ) + coord_cartesian( ylim = c(8,30) ) + labs(title="mean") p2 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "median", errorbar = "CI" ) + coord_cartesian( ylim = c(8,30) ) + labs(title="median") p3 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "gmean", errorbar = "CI" ) + coord_cartesian( ylim = c(8,30) ) + labs(title="geometric mean") p <- grid.arrange(p1,p2,p3,ncol=3) expect_equal( "ggplot" %in% class(p1), TRUE) expect_equal( "ggplot" %in% class(p2), TRUE) expect_equal( "ggplot" %in% class(p3), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6d: statistics of dispersion sd and MAD", { old <- options() on.exit(options(old)) options("superb.feedback" = "none") # to supress design confirmation; unneeded in tests library(gridExtra) dta6 <- GRD( WSFactors = "Moment(3):Hand(2)", Effects = list("Moment" = slope(5), "Hand" = slope(3)), SubjectsPerGroup = 6, Population = list (mean = 20, stddev = 5, rho = 0.8) ) # functions; SD should be asymmetrical; fisherskew should be about 0 # fisherkurtosis is less stable; MAD should be about 2.5; # gmean requires only positive data; p1 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "sd", errorbar = "CI" ) p2 <- superbPlot(dta6, WSFactor = c("Moment(3)","Hand(2)"), variables = c("DV.1.1","DV.2.1","DV.3.1","DV.1.2","DV.2.2","DV.3.2"), statistic = "MAD", errorbar = "CI" ) p <- grid.arrange(p1,p2,ncol=2) expect_equal( "ggplot" %in% class(p1), TRUE) expect_equal( "ggplot" %in% class(p2), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6e: adding ggplot graphics directives", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') # ggplot arguments p1 <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", statistic = "mean", errorbar = "CI", gamma = .999, adjustments = list(purpose = "difference") ) p2 <- p1 + # all these are added to the plot xlab("Dose per day") + ylab("Tooth Growth after study") + theme_light(base_size=14 ) expect_equal( "ggplot" %in% class(p2), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6f: adding ggplot arguments to the error bars, to the points", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') p <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", statistic = "mean", errorbar = "CI", gamma = .999, adjustments = list(purpose = "difference"), # see geom_errorbar for the possible arguments errorbarParams = list(width = .8, linewidth = 3, colour = "gray"), # see geom_point or geom_bar for possible arguments barParams = list(linetype = 3, colour = "black", linewidth = .5) ) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("test 6g: adding ggplot arguments to the error bars, to the points (bis)", { old <- options() on.exit(options(old)) options("superb.feedback" = 'none') p <- superbPlot(ToothGrowth, BSFactor = c("dose","supp"), variables = "len", statistic = "mean", errorbar = "CI", gamma = .999, adjustments = list(purpose = "difference"), plotStyle = "line", # see geom_errorbar for the possible arguments errorbarParams = list(width = .02, linewidth = 0.1, colour = "gray"), # see geom_point or geom_bar for possible arguments pointParams = list(colour = "gray", size = 10.5) ) expect_equal( "ggplot" %in% class(p), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) ######################################### # TESTS WITH ICC ######################################### test_that("Explorations for ICC", { old <- options() on.exit(options(old)) options("superb.feedback" = c('warnings')) library(gridExtra) dta99 <- GRD( WSFactors = "Moment(3)", SubjectsPerGroup = 15, Population = list( mean=20, stddev = 5), Effects = list("Moment" = slope(3) ) ) # add cluster information at the end dta99$myclus <- sort(rep(1:5, 3)) # create some intraclass correlation artificially dta99$DV.1 <- dta99$DV.1 + 10 * dta99$myclus dta99$DV.2 <- dta99$DV.2 + 10 * dta99$myclus dta99$DV.3 <- dta99$DV.3 + 10 * dta99$myclus # write.table(dta99, file = 'file9.dat', sep = "\t", col.names = FALSE) expect_message( noncluster <- superbPlot(dta99, WSFactor = "moment(3)", adjustments = list(decorrelation="CM"), errorbar = "CI", showPlot=TRUE, variables = c("DV.1","DV.2","DV.3") )+ labs(title="Without cluster information") + coord_cartesian( ylim = c(40,60) ) ) expect_message( yescluster <- superbPlot(dta99, WSFactor = "moment(3)", adjustments = list(decorrelation="CM", samplingDesign = "CRS"), clusterColumn = "myclus", errorbar = "CI", showPlot=TRUE, variables = c("DV.1","DV.2","DV.3") )+ labs(title="with cluster information") + coord_cartesian( ylim = c(40,60) ) ) p <- grid.arrange(noncluster, yescluster, ncol=2) expect_equal( "ggplot" %in% class(noncluster), TRUE) expect_equal( "ggplot" %in% class(yescluster), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) ######################################################## # latest tests based on dta3 verifying CM and popSize # ######################################################## test_that("Verifying CM and popSize ", { old <- options() on.exit(options(old)) options("superb.feedback" = c('warnings')) library(gridExtra) dta3 <- GRD( BSFactors = "Group(2)", WSFactors = "Moment(3)", SubjectsPerGroup = 5, Population = list (mean = 20, stddev = 5, rho = 0.8), Effects = list("Moment" = slope(5)) ) # write.table(dta3, file = "file3.dat", sep = "\t", col.names = FALSE) expect_message( p1 <- superbPlot(dta3, WSFactor = "Moment(3)", BSFactor = "Group", variables = c("DV.1","DV.2","DV.3"), statistic = "mean", errorbar = "SE", adjustments = list(purpose="single", decorrelation="CM", popSize = Inf ) ) + labs(title="Infinite populations") ) expect_message( p2 <- superbPlot(dta3, WSFactor = "Moment(3)", BSFactor = "Group", variables = c("DV.1","DV.2","DV.3"), statistic = "mean", errorbar = "SE", adjustments = list(purpose="single", decorrelation="CM", popSize = c(Inf,6) ) ) + labs(title="population of 6 in grp 2") ) p <- grid.arrange(p1,p2,ncol=2) expect_equal( "ggplot" %in% class(p1), TRUE) expect_equal( "ggplot" %in% class(p2), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) ######################################### # testing pre and post processing: OK! ######################################### test_that("Testing pre and post processing", { old <- options() on.exit(options(old)) options("superb.feedback" = c('warnings')) library(ggplot2) library(gridExtra) dta9 <- GRD( WSFactors = "Moment(3)", SubjectsPerGroup = 5, Population = list( mean=20, stddev = 5), Effects = list("Moment" = slope(3) ) ) # write.table(dta9, file = "file9.dat", sep = "\t", col.names = FALSE) expect_message( truecm <- superbPlot(dta9, WSFactor = "moment(3)", adjustments=list(decorrelation="CM"), errorbar = "CI", showPlot=T, variables = c("DV.1","DV.2","DV.3") )+ labs(title="With decorrelation = CM") ) altcm <- superbPlot(dta9, WSFactor = "moment(3)", adjustments=list(decorrelation="none"), preprocessfct = "subjectCenteringTransform", postprocessfct = c("biasCorrectionTransform"), errorbar = "CI", showPlot=T, variables = c("DV.1","DV.2","DV.3") )+ labs(title="with pre and post processing") pcm <- grid.arrange(truecm,altcm,ncol=2) expect_message( truelm <- superbPlot(dta9, WSFactor = "moment(3)", adjustments=list(decorrelation="LM"), errorbar = "CI", showPlot=T, variables = c("DV.1","DV.2","DV.3") )+ labs(title="with decorrelation = LM") ) altlm <- superbPlot(dta9, WSFactor = "moment(3)", adjustments=list(decorrelation="none"), preprocessfct = "subjectCenteringTransform", postprocessfct = c("biasCorrectionTransform","poolSDTransform"), errorbar = "CI", showPlot=T, variables = c("DV.1","DV.2","DV.3") )+ labs(title="with pre and post processing") plm <- grid.arrange(truelm,altlm,ncol=2) expect_message( truecmvslm <- superbPlot(dta9, WSFactor = "moment(3)", adjustments=list(decorrelation="LM"), errorbar = "CI", showPlot=T, variables = c("DV.1","DV.2","DV.3") )+ labs(title="with decorrelation = LM") ) expect_message( altcmvslm <- superbPlot(dta9, WSFactor = "moment(3)", adjustments=list(decorrelation="CM"), postprocessfct = c("poolSDTransform"), errorbar = "CI", showPlot=T, variables = c("DV.1","DV.2","DV.3") )+ labs(title="with decorrelation = CM and pooling") ) pcmvslm <- grid.arrange(truecmvslm,altcmvslm,ncol=2) expect_equal( "ggplot" %in% class(truecm), TRUE) expect_equal( "ggplot" %in% class(altcm), TRUE) expect_equal( "ggplot" %in% class(truelm), TRUE) expect_equal( "ggplot" %in% class(altlm), TRUE) expect_equal( "ggplot" %in% class(truecmvslm), TRUE) expect_equal( "ggplot" %in% class(altcmvslm), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) ######################################### # testing multiple formats! ######################################### test_that("Many tests with TMB1964r", { old <- options() on.exit(options(old)) options("superb.feedback" = c('none')) library(ggplot2) mee = TMB1964r[TMB1964r$Language == "English"|TMB1964r$Language == "French",] mp <- function(style, ...) { superbPlot(mee, WSFactor = "T(7)", BSFactor = c("Condition","Sex"), variables = c("T1","T2","T3","T4","T5","T6","T7"), adjustments = list(purpose="difference", decorrelation="CM"), plotStyle = style, ... ) } ###### BASIC PLOTS ###### plt1 <- mp("bar", errorbarParams = list(linewidth=0.75, position = position_dodge(.95) ), barParams = list(linewidth=0.5) ) + scale_colour_manual( name = "asdf", labels = c("Context 0", "Context 2", "Context 4", "Context 8"), values = c("blue", "black", "purple", "red")) + scale_fill_manual( name = "asdf", labels = c("Context 0", "Context 2", "Context 4", "Context 8"), values = c("blue", "black", "purple", "red")) + theme_bw(base_size = 16) + labs(x = "Exposure duration (ms)", y = "Mean of correct responses" )+ scale_x_discrete(labels=c("1" = "16.67", "2" = "33.33", "3"="50.00", "4" = "66.67", "5"="83.33", "6"="100.00", "7"="116.67")) plt2 <- mp("line", errorbarParams = list(linewidth=0.75, width = 0.2, position = position_dodge(.5) ), pointParams = list(size=2.5, position = position_dodge(.5)), lineParams = list(linewidth=0.25) ) plt3 <- mp("point", errorbarParams = list(position = position_dodge(.5) ), pointParams = list(size=2.5, position = position_dodge(.5)) ) ###### ADVANCED PLOTS ###### plt4 <- mp("pointjitter", errorbarParams = list(position = position_dodge(.5) ), pointParams = list(size=3.5, position = position_dodge(.5)), jitterParams = list(size = 0.5) ) plt5 <- mp("pointjitterviolin", errorbarParams = list(position = position_dodge(.5) ), pointParams = list(size=3.5, position = position_dodge(.5)), jitterParams = list(size = 0.5), violinParams = list(alpha =0.7) ) dta <- superb::GRD( WSFactors = "timepoints (100) : condition(2)", SubjectsPerGroup = 40, RenameDV = "activation", Effects = list("timepoints" = extent(5), "condition" = extent(3) ), Population=list(mean=50,stddev=10,rho=0.75) ) plt6 <- superbPlot(dta, WSFactors = c("timepoints(100)", "condition(2)"), variables = colnames(dta)[2:201], ## all the names of the dataframe except "id" adjustments = list( purpose = "single", decorrelation = "CM" ## or none for no decorrelation ), plotStyle="lineBand", # note the uppercase B pointParams = list(size= 1) ) expect_equal( "ggplot" %in% class(plt1), TRUE) expect_equal( "ggplot" %in% class(plt2), TRUE) expect_equal( "ggplot" %in% class(plt3), TRUE) expect_equal( "ggplot" %in% class(plt4), TRUE) expect_equal( "ggplot" %in% class(plt5), TRUE) expect_equal( "ggplot" %in% class(plt6), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("Heterogeneous variances", { old <- options() on.exit(options(old)) options("superb.feedback" = c('none')) dta <- GRD( BSFactors = "dif(3) : grp (2)", WSFactors="day(1,2)", SubjectsPerGroup = 3, Population=list( mean = 100, scores = "rnorm(1, mean = GM, sd = 100 * (grp-1) +0.1)" ) ) options("superb.feedback" = c('warnings')) expect_message( superbPlot(dta, BSFactor = c("dif","grp"), WSFactor = "day(2)", variables = c("DV.1","DV.2"), adjustment = list( purpose = "difference") ) ) options("superb.feedback" = c('none')) plt <- superbPlot(dta, BSFactor = c("dif","grp"), WSFactor = "day(2)", variables = c("DV.1","DV.2"), adjustment = list( purpose = "tryon") ) expect_equal( "ggplot" %in% class(plt), TRUE ) # restores default information options("superb.feedback" = c('design','warnings','summary')) }) test_that("Tryon vs. difference", { old <- options() on.exit(options(old)) options("superb.feedback" = c('warnings')) #lets have data with massively heterogeneous variances dta <-GRD( BSFactors="grp(3)", RenameDV = "score", Population=list( mean = 100, scores = "rnorm(1, mean = GM, sd = 10 * grp)" ), SubjectsPerGroup = 50, Effects = list("grp" = slope(15) ) ) # do the plot twice with transparant background expect_message(plt1 <- superbPlot(dta, BSFactor = "grp", plotStyle="line", variables = "score", errorbarParams = list(color="blue",position = position_nudge(-0.1) ), adjustments = list( purpose = "difference") ) + labs(title="(blue) Difference-adjusted 95% confidence intervals\n(red) Tryon 95% confidence intervals") + coord_cartesian( ylim = c(65,135) ) + theme(panel.background = element_rect(fill = "transparent"), plot.background = element_rect(fill = "transparent", color = "white")) ) expect_message(plt2 <- superbPlot(dta, BSFactor = c("grp"), plotStyle="line", variables = "score", errorbarParams = list(color="red",position = position_nudge(+0.1) ), adjustments = list( purpose = "tryon") ) + labs(title="(blue) Difference-adjusted 95% confidence intervals\n(red) Tryon 95% confidence intervals") + coord_cartesian( ylim = c(65,135) ) + theme(panel.background = element_rect(fill = "transparent"), plot.background = element_rect(fill = "transparent", color = "white")) ) # transform the plots in manipulable objets plt1g <- ggplotGrob(plt1) plt2g <- ggplotGrob(plt2) # put the two grob onto an empty ggplot (as the positions are the same, they will be overlayed) plt <- ggplot() + annotation_custom(grob=plt1g) + annotation_custom(grob=plt2g) expect_equal( "ggplot" %in% class(plt), TRUE) # restores default information options("superb.feedback" = c('design','warnings','summary')) })