README.md

In eduardohet/resulteR: Test Results for Reproducible Documents

ResulteR: Test Results for Reproducible Documents

by Eduardo L. Hettwer Giehl

ResulteR implements functions to generate strings of text for the use in markdown documents. It aims to help you build top level reproducible documents.

Installing

install.packages("devtools")
library(devtools)
install_github("eduardohet/resulteR")
library(resulteR)

Trying it out

T-tests

res <- t.test(extra ~ group, data = sleep)
sleep.res <- tText(res)

# Cheatsheet to produce the following neatly formatted result: 
# Now you just need to add your result with an inline call to that text string. Here you could say that means 
# are similar in both sleep groups (`r sleep.res`).

Now you just need to add your result with an inline call to that text string. Here you could say that means are similar in both sleep groups (t17.8 = -1.861; P = 0.079).

Regression

utils::data(anorexia, package = "MASS")
anorex.1 <- lm(Postwt ~ Prewt, data = anorexia)
anorexia.pre <- lmText(anorex.1)

# Cheatsheet to produce the following neatly formatted result: 
# In this case, there was no association between weight in pre- and post-treatment when treatments are not taken 
# into account (`r anorexia.pre`).

In this case, there was no association between weight in pre- and post-treatment when treatments are not taken into account (R² = 0.11; F 1, 70 = 8.695; P = 0.004).

anorexia$Treat <- relevel(anorexia$Treat, ref="Cont")
anorex.2 <- lm(Postwt ~ Treat + Prewt, data = anorexia)
anorexia.full <- lmText(anorex.2)
anorexia.ft <- lmText(anorex.2, type="coefs", which.coef="TreatFT")
anorexia.CBT <- lmText(anorex.2, type="coefs", which.coef="TreatCBT")

# Cheatsheet to produce the following neatly formatted result:
# However, not always our models contain only one variable. Therefore we may need to word our results a little  
# bit differently. For instance, when we take treatment and pre-weight into account, we reach a significant result 
# (`r anorexia.full`) and function `lmText` takes care to appropriately change from regular to adjusted-R² 
# ($\\bar{R}^{2}$). By further looking into the results, we see that patients in family (`r anorexia.ft`) and  
# cognitive behavioural (`r anorexia.CBT`) treatments improved weight over patients in the control group.

However, not always our models contain only one variable. Therefore we may need to word our results a little bit differently. For instance, when we take treatment and pre-weight into account, we reach a significant result (R² = 0.246; F 3, 68 = 8.713; P < 0.001) and function lmText takes care to appropriately change from regular to adjusted-R² (notation should change from R to R bar squared but I was unable to reproduce the exact output in GitHub). By further looking into the results, we see that patients in family (b = 8.66; t = 3.949; P < 0.001) and cognitive behavioural (b = 4.097; t = 2.164; P = 0.034) treatments improved weight over patients in the control group.

Regression plots

In addition to test results inline in your texts, what about including the key results of your regression in your plot? Just ckeck out the figure below.

plot(Postwt ~ Prewt, data = anorexia)
abline(anorex.1)
plotlmText(anorex.1, h=0.15, v=0.8, pos=4)

Anova

anorex.1 <- aov(Postwt ~ Treat + Prewt, data = anorexia)
aov.treat <- aovText(anorex.1, which.coef="Treat")
aov.prewt <- aovText(anorex.1, which.coef="Prewt")

# Cheatsheet to produce the following neatly formatted result:
# Depending on the type of data in your analysis, you may have to report your results differently. For example, 
# the analysis of anorexia treatment has two independent variables: treatment (Treat) and weight of patient 
# before study period (Prewt). While this is actually an analysis called Ancova, some may want to represent its 
# results by F and P-values for the two variables instead of betas from regressions. Therefore, one may say that 
# there is an effect of both treatment (`r aov.treat`) and weight of patient before study period (`r aov.prewt`).

Depending on the type of data in your analysis, you may have to report your results differently. For example, the analysis of anorexia treatment has two independent variables: treatment (Treat) and weight of patient before study period (Prewt). While this is actually an analysis called Ancova, some may want to represent its results by F and P-values for the two variables instead of βs from regressions. Therefore, one may say that there is an effect of both treatment (F 2, 68 = 9.436; P < 0.001) and weight of patient before study period (F 1, 68 = 7.266; P = 0.009).

Correlations

res <- cor.test(~Postwt + Prewt, data=anorexia)
pearson.cor <- corText(res)
res <- cor.test(~Postwt + Prewt, data=anorexia, method="spearman")
spearman.rho <- corText(res, digits=c(3, 1, 3))
res <- cor.test(~Postwt + Prewt, data=anorexia, method="kendall")
kendall.tau <- corText(res)

# Cheatsheet to produce the following neatly formatted result:
# Another commonly used set of analysis is that of correlations. Let's have a look on the output of such 
# analysison the relationship of weight before and after treatment in the anorexia dataset. Below, we will 
# apply the three methods available in function `cor.test`:

# 1. Pearson's product-moment correlation: weight of patients correlated positively with their weights after the
# treatment (`r pearson.cor`)...

Another commonly used set of analysis is that of correlations. Let's have a look on the output of such analysis on the relationship of weight before and after treatment in the anorexia dataset. Below, we will apply the three methods available in function cor.test:

1. Pearson's product-moment correlation: weight of patients correlated positively with their weights after the treatment (r = 0.332; t70 = 2.949; P = 0.004).
2. Spearman's rank correlation ρ: weight of patients correlated positively with their weights after the treatment (ρ = 0.344; S = 40789.9; P = 0.003).
3. Kendall's rank correlation τ: weight of patients correlated positively with their weights after the treatment (τ = 0.246; z = 3.049; P = 0.002).

Descriptive statistics

# Simple mean and standard deviation of a numeric vector
mean.postwt <- meansdText(anorexia$Postwt)
prewt <- with(anorexia, split(Prewt, list(Treat)))
postwt <- with(anorexia, split(Postwt, list(Treat)))
meandiff.cont <- meansdText(postwt$Cont-prewt$Cont)
meandiff.ft <- meansdText(postwt$FT-prewt$FT)

# Cheatsheet to produce the following neatly formatted result:
# What about describing your data? Let's have a look on the anorexia dataset again. The mean and standard 
# deviation of weights after the treatment is `r mean.postwt` lb. Having a look on patients in the control 
# group, it seems most weight lost (mean difference postwt-prewt: `r meandiff.cont` lb), whereas patients 
# receiveing the family treatment improved their weight (`r meandiff.ft` lb).

What about describing your data? Let's have a look on the anorexia dataset again. The mean and standard deviation of weights after the treatment is 85.2 ± 8 lb. Having a look on patients in the control group, it seems most weight lost (mean difference postwt-prewt: -0.4 ± 8 lb), whereas patients receiveing the family treatment improved their weight (7.3 ± 7.2 lb).

Neat lists

x <- c("apples", "oranges", "grapes")
fruits <- tidyList(x)
numbers <- tidyList(c(2.3, 5.4, 3), sep="; ", dec=",", last="&")

The last function in the package is a small one to produce lists of any stuff you want. You may either list some fruits: apples, oranges, and grapes, or some numbers: 2,3; 5,4; & 3. You may format the output as you wish. A final hint: this may be useful to grab names of significant parameters from a model result.

eduardohet/resulteR documentation built on Aug. 9, 2021, 3:02 a.m.