Skip to content

abusjahn/wrappedtools

Repository files navigation

wrappedtools

The goal of ‘wrappedtools’ is to make my (and possibly your) life a bit easier by a set of convenience functions for many common tasks like e.g. computation of mean and SD and pasting them with ±. Instead of
paste(round(mean(x),some_level), round(sd(x),some_level), sep=‘±’)
a simple meansd(x, roundDig = some_level) is enough.

Installation

You can install the released version of ‘wrappedtools’ from CRAN or the latest development version from github with:

devtools::install_github("abusjahn/wrappedtools")

Examples

This is a basic example which shows you how to solve a common problem, that is, describe and test differences in some measures between 2 samples, rounding descriptive statistics to a reasonable precision in the process:

# Standard functions to obtain median and quartiles:
median(mtcars$mpg)
#> [1] 19.2
quantile(mtcars$mpg,probs = c(.25,.75))
#>    25%    75% 
#> 15.425 22.800
# wrappedtools adds rounding and pasting:
median_quart(mtcars$mpg)
#> [1] "19 (15/23)"
# on a higher level, this logic leads to
compare2numvars(data = mtcars, dep_vars = c('wt','mpg', "disp"), 
                indep_var = 'am',
                gaussian = FALSE,
                round_desc = 3)
#> # A tibble: 3 × 5
#>   Variable desc_all         `am 0`           `am 1`           p    
#>   <fct>    <chr>            <chr>            <chr>            <chr>
#> 1 wt       3.32 (2.53/3.66) 3.52 (3.44/3.84) 2.32 (1.90/2.81) 0.001
#> 2 mpg      19.2 (15.3/22.8) 17.3 (14.8/19.2) 22.8 (20.6/30.4) 0.002
#> 3 disp     196 (121/337)    276 (177/360)    120 (79/160)     0.001

To explain the ‘wrapper’ part of the package name, here is another example, using the ks.test as test for a Normal distribution, where ksnormal simply wraps around the ks.test function:

somedata <- rnorm(100)
ks.test(x = somedata, 'pnorm', mean=mean(somedata), sd=sd(somedata))
#> 
#>  Asymptotic one-sample Kolmogorov-Smirnov test
#> 
#> data:  somedata
#> D = 0.057517, p-value = 0.8954
#> alternative hypothesis: two-sided

ksnormal(somedata)
#> [1] 0.8953558

Saving variable selections: Variables may fall into different groups: Some are following a Gaussian distribution, others are ordinal or factorial. There may be several grouping variables like treatment, gender… To refer to such variables, it is convenient to have their index and name stored. The name may be needed as character, complex variable names like “size [cm]” may need to be surrounded by backticks in some function calls but must not have those in others. Function ColSeeker finds columns in tibbles or dataframes, based on name pattern and/or class. This is comparable to the selection helpers in ‘tidyselect’, but does not select the content of matching variables, but names, positions, and count:

gaussvars <- ColSeeker(data = mtcars,
                       namepattern = c('wt','mpg'))
gaussvars
#> $index
#> [1] 1 6
#> 
#> $names
#> [1] "mpg" "wt" 
#> 
#> $bticked
#> [1] "`mpg`" "`wt`" 
#> 
#> $count
#> [1] 2

#Exclusion based on pattern
factorvars <- ColSeeker(mtcars,
                        namepattern = c('a','cy'),
                        exclude = c('t'))
factorvars$names #drat excluded
#> [1] "cyl"  "am"   "gear" "carb"

ColSeeker(mtcars,varclass = 'numeric')$names
#>  [1] "mpg"  "cyl"  "disp" "hp"   "drat" "wt"   "qsec" "vs"   "am"   "gear"
#> [11] "carb"

Workflow with ColSeeker and compare2numvars to describe and test a number of variables between 2 groups:

compare2numvars(data = mtcars,
                dep_vars=gaussvars$names,
                indep_var = 'am',
                gaussian = TRUE)
#> # A tibble: 2 × 5
#>   Variable desc_all  `am 0`    `am 1`    p    
#>   <fct>    <chr>     <chr>     <chr>     <chr>
#> 1 mpg      20 ± 6    17 ± 4    24 ± 6    0.001
#> 2 wt       3.2 ± 1.0 3.8 ± 0.8 2.4 ± 0.6 0.001

This should give you the general idea, I’ll try to expand this intro over time…