This package gives a number of functions to aid common data analysis processes and reporting statistical results in an RMarkdown file. Data analysis functions combine multiple base R functions used to describe simple bivariate relationships into a single, easy to use function. Reporting functions will return character strings to report p-values, confidence intervals, and hypothesis test and regression results. Strings will be LaTeX-formatted as necessary and will knit pretty in an RMarkdown document. The package also provides a wrapper for the CreateTableOne function in the tableone package to make the results knitable.

## Data analysis functions

Suppose we have the following data:

pred1 = sample(letters[1:3], size=50, replace=TRUE)
out1 = sample(letters[4:6], size=50, replace=TRUE)
out2 = rnorm(50)

We can investigate the relationship between pred1 and out1 using cat_compare():

cat_compare(x=pred1, y=out1)
## Warning in chisq.test(tab_no_miss): Chi-squared approximation may be incorrect

## \$counts
##      y
## x      d  e  f Sum
##   a    8  7  3  18
##   b    6  2  9  17
##   c    7  4  4  15
##   Sum 21 13 16  50
##
## \$chisq
##
##  Pearson's Chi-squared test
##
## data:  tab_no_miss
## X-squared = 6.5486, df = 4, p-value = 0.1618
##
##
## \$CramersV
## [1] 0.2559017
##
## \$plot

We can investigate the distribution of out2 across levels of pred1 using num_compare():

num_compare(y=out2, grp=pred1)
## \$summary_stats
##    n obs mis        mean     stdev         med         q1        q3
## a 18  18   0 0.006755781 1.0851542  0.04793630 -0.4201223 0.8188215
## b 17  17   0 0.001604250 0.8911016  0.07865256 -0.2591153 0.5775767
## c 15  15   0 0.198539217 1.0332958 -0.07142822 -0.2773362 0.6744763
##
## \$decomp
## Call:
##    aov(formula = y ~ grp, data = mydat)
##
## Terms:
##                      grp Residuals
## Sum of Squares   0.39657  47.67131
## Deg. of Freedom        2        47
##
## Residual standard error: 1.007116
## Estimated effects may be unbalanced
##
## \$eta_sq
## [1] 0.008250299
##
## \$plot

## inline and write functions

• inline_test()
• inline_reg()
• inline_coef()
• inline_anova()
• write_int()
• write_p()
• as_perc()

Using the data above, we can obtain some inferential results:

x = rnorm(50)
y = rnorm(50)
a = sample(letters[1:3], size=50, replace=TRUE)
b = sample(letters[1:3], size=50, replace=TRUE)

test1 = t.test(x)
test2 = chisq.test(table(a,b))
model1 = lm(y ~ x)
model2 = lm(y ~ a)

We can then report the results of the hypothesis test inline using inline_test(test1) and get the following: (t(49) = -0.7), (p = 0.49). Simiarly, inline_test(test2) will report the results of the chi-squared test: (^2(4) = 4.85), (p = 0.3). So far inline_test only works for (t) and chi-squared tests, but the goal is to add more functionality - requests gladly accepted.

The regression results can be reported with inline_reg(model1) and inline_coef(model1, 'x') to get (R^2 = 0.02), (F(1,48) = 0.81), (p = 0.37) and (b = -0.14), (t(48) = -0.9), (p = 0.37), respectively. In addition, inline_anova(model2) will report the ANOVA F statistic and relevant results: (F(2,47) = 2.81), (p = 0.07). So far inline_reg and inline_coef currently work for lm and glm objects; inline_anova only works for lm objects.

We can also report the confidence intervals using write_int() with a length-2 vector of interval endpoints. For example, write_int(c(3.04, 4.7)) and write_int(test1\$conf.int) yield (3.04, 4.70) and (-0.37, 0.18), respectively. If a 2-column matrix is provided to write_int(), the entries in each row will be formatted into an interval and a character vector will be returned.

P-values can be reported using write_p(). This function will take either a numeric value or a list-like object with an element named p.value. For example, write_p(0.00002) gives (p < 0.01) and write_p(test1) gives (p = 0.49).

Many R functions produce proportions, though analysts may want to report the output as a percentage. as_perc() will do this. For example, as_perc(0.01) will produce 1%.

See the help files of all functions described above for more details and options. For example, all test and regression reporting functions have wrappers ending in _p which report only the p-value of the input.

## KreateTableOne

The package also provides the function KreateTableOne, a wrapper for CreateTableOne from the tableone package which makes the results knitable. First use KreateTableOne in an R chunk with results='hide' (or ouside the RMarkdown document), then recall the saved data frame in a new chunk. For example:

table1 = KreateTableOne(x=mtcars, strata='am',
factorVars='vs')
colnames(table1)[1:2] = c('am = 0', 'am = 1')

Then

knitr::kable(table1[, 1:3], align='r')
am = 0 am = 1 p
n 19 13
mpg (mean (SD)) 17.15 (3.83) 24.39 (6.17) <0.001
cyl (mean (SD)) 6.95 (1.54) 5.08 (1.55) 0.002
disp (mean (SD)) 290.38 (110.17) 143.53 (87.20) <0.001
hp (mean (SD)) 160.26 (53.91) 126.85 (84.06) 0.180
drat (mean (SD)) 3.29 (0.39) 4.05 (0.36) <0.001
wt (mean (SD)) 3.77 (0.78) 2.41 (0.62) <0.001
qsec (mean (SD)) 18.18 (1.75) 17.36 (1.79) 0.206
vs = 1 (%) 7 (36.8) 7 (53.8) 0.556
am (mean (SD)) 0.00 (0.00) 1.00 (0.00) <0.001
gear (mean (SD)) 3.21 (0.42) 4.38 (0.51) <0.001
carb (mean (SD)) 2.74 (1.15) 2.92 (2.18) 0.754