In geiser/rshinystatistics: Shiny Statistics

## Install and Load Packages
if (!'remotes' %in% rownames(installed.packages())) install.packages('remotes')
if (!"rshinystatistics" %in% rownames(installed.packages())) {
  remotes::install_github("geiser/rshinystatistics")
} else if (packageVersion("rshinystatistics") < "{{ rshinystatistics.version }}") {
  remotes::install_github("geiser/rshinystatistics")
}

wants <- c('ggplot2','ggpubr','rshinystatistics','utils')
has <- wants %in% rownames(installed.packages())
if (any(!has)) install.packages(wants[!has])

library(utils)
library(ggpubr)
library(ggplot2)
library(rshinystatistics)

library(knitr)
opts_chunk$set(echo = TRUE)
defaultW <- getOption("warn")
options(warn = -1)

Initial Variables and Data

• R-script file: ../code/{{ test }}.R
• Initial table file: ../data/initial-table.csv {{ paste0(paste0('* Data for ', dvs, ' ../data/table-for-',dvs,'.csv, collapse = '\n') }}
• Table without outliers and normal distribution of data: ../data/table-with-normal-distribution.csv
• Other data files: ../data/
• Files related to the presented results: ../results/

wid <- "{{ wid }}"
{{ ifelse(length(covar) > 0, paste0('covar <- "',covar,'"'), '') }}
between <- c({{ paste0(paste0('"', between, '"'), collapse = ',') }})
dvs <- c({{ paste0(paste0('"', dvs, '"'), collapse = ',') }})
names(dvs) <- dvs
{{
if (length(skewness) > 0) {
  paste0('skewness <- c(', paste0(lapply(names(skewness), FUN = function(i) {
        paste0('"',i, '"="', skewness[[i]],'"') }), collapse = ','),')')
} else {
  paste0('skewness <- c()')
}
}}

dat <- lapply(dvs, FUN = function(dv) {
  data <- read.csv(paste0("../data/table-for-",dv,".csv"))
  rownames(data) <- data[["{{ wid }}"]]
  return(data)
})
rdat <- dat
sdat <- dat

Descriptive statistics of initial data

{{
if (length(covar) > 0) {
  paste0(c('df <- dat; df[[covar]] <- dat[[1]]',
           '(df <- get.descriptives(df, dvs, between, include.global = T, symmetry.test = T))'), collapse = '\n')
} else {
  paste0(c('(df <- get.descriptives(dat, dvs, between, include.global = T, symmetry.test = T))'), collapse = '\n')
}
}}

kable(df, digits = 3)

{{
sformula <- paste0(paste0('`',between,'`'), collapse = '*')
box.plot.code  <- paste0(lapply(dvs, FUN = function(dv) {
  paste0("car::Boxplot(`",dv,"` ~ ",sformula,", data = dat[[\"",dv,"\"]], id = list(n = Inf))")
}), collapse = '\n')
box.plot.code
}}

Checking of Assumptions

Assumption: Symmetry and treatment of outliers

Applying transformation for skewness data when normality is not achieved

{{ code.skewness }}

Dealing with outliers (performing treatment of outliers)

{{ code.outliers }}

Assumption: Normality distribution of data

Removing data that affect normality (extreme values)

non.normal <- {{ code.non.normal }}
sdat <- removeFromDataTable(rdat, non.normal, wid)

Result of normality test in the residual model

{{
if (length(covar) > 0) {
  paste0('(df <- normality.test.by.residual(sdat, dvs, between, c(), covar, skewness = skewness))')
} else {
  paste0('(df <- normality.test.by.residual(sdat, dvs, between, skewness = skewness))')
}
}}

kable(df, digits = 3)

Result of normality test in each group

This is an optional validation and only valid for groups with number greater than 30 observations

(df <- get.descriptives(sdat, dvs, between, include.global = F, normality.test = T, skewness = skewness))

kable(df, digits = 3)

Observation:

As sample sizes increase, parametric tests remain valid even with the violation of normality [1]. According to the central limit theorem, the sampling distribution tends to be normal if the sample is large, more than (n > 30) observations. Therefore, we performed parametric tests with large samples as described as follows:

• In cases with the sample size greater than 100 (n > 100), we adopted a significance level of p < 0.01

• For samples with n > 50 observation, we adopted D'Agostino-Pearson test that offers better accuracy for larger samples [2].

• For samples' size between n > 100 and n <= 200, we ignored the normality test, and our decision of validating normality was based only in the interpretation of QQ-plots and histograms because the Shapiro-Wilk and D'Agostino-Pearson tests tend to be too sensitive with values greater than 200 observation [3].

• For samples with n > 200 observation, we ignore the normality assumption based on the central theorem limit.

{{ linearity.code }}

Assumption: Homogeneity of data distribution

{{
if (length(covar) > 0) {
  paste0('(df <- homogeneity.test(sdat, dvs, between, c(), covar, skewness = skewness))')
} else {
  paste0('(df <- homogeneity.test(sdat, dvs, between, skewness = skewness))')
}
}}

kable(df, digits = 3)

Saving the Data with Normal Distribution Used for Performing {{ title.test }}

ndat <- sdat[[1]]
for (dv in names(sdat)[-1]) ndat <- merge(ndat, sdat[[dv]])
write.csv(ndat, paste0("../data/table-with-normal-distribution.csv"))

Descriptive statistics of data with normal distribution

(df <- get.descriptives(sdat, dvs, between))

kable(df, digits = 3)

for (dv in dvs) {
  car::Boxplot(`dv` ~ {{ paste0(paste0('`',between,'`'), collapse = '*') }}, data = sdat[[dv]] %>% cbind(dv=sdat[[dv]][[dv]]), id = list(n = Inf))  
}

Computation of {{ title.test }} and Pairwise Comparison

{{ title.test }}

{{ code.hypothesis }}

{{ code.hypothesis.tbl }}

{{ code.pwc }}

{{ code.pwc.tbl }}

{{ code.emms }}

{{ code.emms.tbl }}

{{ code.plots }}

Textual Report

{{ hypothesis.text }}

{{ hypothesis.pwc.text }}

Tips and References

• Use the site https://www.tablesgenerator.com to convert the HTML tables into Latex format

• [2]: Miot, H. A. (2017). Assessing normality of data in clinical and experimental trials. J Vasc Bras, 16(2), 88-91.

• [3]: Bárány, Imre; Vu, Van (2007). "Central limit theorems for Gaussian polytopes". Annals of Probability. Institute of Mathematical Statistics. 35 (4): 1593–1621.

options(warn = defaultW)

geiser/rshinystatistics documentation built on Feb. 18, 2024, 6:07 p.m.