Tinystatr, an automated process of selecting statistical methods

https://github.com/yunzhu0304/tinystatr

🦖About tinystatr

The tinystatr is an R package for statistical analysis. It automatically assesses the normality and homogeneity of variance of input data, selects appropriate statistical methods based on the characteristics of the data, and returns the statistical analysis results.

🦒Introduction

Our statistical analysis process is based on the following two workflows: two groups, and more than two groups.

🍒For two groups:

• Normality test and homogeneity of variance test.

1. Both datasets are normally distributed and variance equal, will apply the t-test.
2. Both datasets are normally distributed but the variance is unequal, will apply the Welch’s t-test.
3. One of the datasets is non-normally distributed, and the sample size in a single group is more than 30, the choice of statistical method will depend on their variances. Equal to the Welch’s t-test, unequal to the Mann-Whitney test(Wilcoxon test).
4. One of the datasets is non-normally distributed but the sample size in a single group is less than 30 or both datasets are non-normally distributed, will apply the Wilcoxon test.

🍇For more than two groups:

• Normality test and homogeneity of variance test.

1. One of the datasets is non-normally distributed, will apply the Kruskal-Wallis test and Dunn’s test (Bonferroni test) as the post-hoc test.
2. All of the datasets are normally distributed, the choice of statistical method will depend on their variances. Equal to the ANOVA test (post-hoc: Tukey HSD), unequal to the Kruskal-Wallis test (post-hoc: Dunn’s test).

🏗️Installation

if (!require("devtools", quietly = TRUE))
    install.packages("devtools")
    
devtools::install_github("yunzhu0304/tinystatr")

📚Load the package

library(tinystatr)

📏Load data and explore the package

📘Load the data

data("ToothGrowth")
data("HairEyeColor")

📈Statistical analysis for two groups(stat2())

📌More than just grouping and value information

df <- ToothGrowth%>%
filter(dose %in% c("0.5","1"))

> df
    len supp dose
1   4.2   VC  0.5
2  11.5   VC  0.5
3   7.3   VC  0.5
4   5.8   VC  0.5
5   6.4   VC  0.5
6  10.0   VC  0.5

After executing the function(stat2()), the results will include assessments of normality, homogeneity of variance, and the selected statistical method.

# Dataframe with multiple columns, need to filter variable
> stat2(data = df,variable = "supp",id="VC",group = "dose",value = "len",
formula = len ~ dose)

Normally distributed  
Variance equal  
 t-test
# A tibble: 1 × 7
  group1 group2    n1    n2           p variable method
  <chr>  <chr>  <int> <int>       <dbl> <chr>    <chr> 
1 0.5    1         10    10 0.000000681 VC       t test

We also obtain a list named stat2result, which contains two data frames. One is stat, used to store the statistical result. The other is narmal, used to store the results of normality, mean and sd value.

> stat2result[["stat"]]
# A tibble: 1 × 7
  group1 group2    n1    n2           p variable method
  <chr>  <chr>  <int> <int>       <dbl> <chr>    <chr> 
1 0.5    1         10    10 0.000000681 VC       t test

> stat2result[["normal"]]
  group variable normal meanvalue       sd
1   0.5       VC   TRUE      7.98 2.746634
2   1.0       VC   TRUE     16.77 2.515309

📌Statistical analysis based solely on grouping and value data

If the datasets only contain information on groups and values, we will ignore the variable and id.

# Dataframe with only two columns (group,value)
data("HairEyeColor")
df <- as.data.frame(HairEyeColor)[,c(3,4)]

> stat2(data = df,group = "Sex",value = "Freq", formula = Freq ~ Sex) # Ignoring variable and id

Non-normally distributed  
 wilcoxon test
# A tibble: 1 × 7
  group1 group2    n1    n2     p variable method       
  <chr>  <chr>  <int> <int> <dbl> <chr>    <chr>        
1 Male   Female    16    16  0.88 id       Wilcoxon test

> stat2result[["stat"]]
# A tibble: 1 × 7
  group1 group2    n1    n2     p variable method       
  <chr>  <chr>  <int> <int> <dbl> <chr>    <chr>        
1 Male   Female    16    16  0.88 id       Wilcoxon test

> stat2result[["normal"]]
   group variable normal meanvalue       sd
1   Male       id  FALSE   17.4375 16.00820
2 Female       id  FALSE   19.5625 20.71382

📊Statistical analysis for more than two groups(stat3())

📍More than just grouping and value information

# Dataframe with multiple columns, need to filter variable
data("ToothGrowth")
df <- ToothGrowth

> df
    len supp dose
1   4.2   VC  0.5
2  11.5   VC  0.5
3   7.3   VC  0.5
...
11 16.5   VC  1.0
12 16.5   VC  1.0
13 15.2   VC  1.0
...
21 23.6   VC  2.0
22 18.5   VC  2.0
23 33.9   VC  2.0
...
31 15.2   OJ  0.5
32 21.5   OJ  0.5
33 17.6   OJ  0.5
...

After executing the function(stat3()), we will obtain the statistical result and a list named stat3result.

> stat3(data = df, group = "dose", value = "len", variable = "supp", id = "OJ", formula = len ~ dose)

All groups have 3 or more samples. 
Normally distributed  
Variance equal  
 Anova
  group2 group1        p.adj posthoc variable           p1 P1method p.adj.signif
1      1    0.5 1.584138e-05     hsd       OJ 8.887164e-08    ANOVA         ****
2      2    0.5 9.386773e-08     hsd       OJ 8.887164e-08    ANOVA         ****
3      2      1 1.309258e-01     hsd       OJ 8.887164e-08    ANOVA           ns

> stat3result[["stat"]]
  group2 group1        p.adj posthoc variable           p1 P1method p.adj.signif
1      1    0.5 1.584138e-05     hsd       OJ 8.887164e-08    ANOVA         ****
2      2    0.5 9.386773e-08     hsd       OJ 8.887164e-08    ANOVA         ****
3      2      1 1.309258e-01     hsd       OJ 8.887164e-08    ANOVA           ns

> stat3result[["normal"]]
  group variable normal meanvalue       sd
1   0.5       OJ   TRUE     13.23 4.459709
2   1.0       OJ   TRUE     22.70 3.910953
3   2.0       OJ   TRUE     26.06 2.655058

📍Statistical analysis based solely on grouping and value data

If the datasets only contain information on groups and values, we will ignore the variable and id.

# Dataframe with only two columns (group,value)
data("HairEyeColor")
df <- as.data.frame(HairEyeColor)[,c(2,4)]

> stat3(data = df,group = "Eye",value = "Freq", formula = Freq ~ Eye) # Ignoring variable and id

All groups have 3 or more samples. 
Non-normally distributed  
Variance unequal  
Kruskal-Wallis 
# A tibble: 6 × 8
  group1 group2 p.adj posthoc    variable     p1 P1method p.adj.signif
  <chr>  <chr>  <dbl> <chr>      <chr>     <dbl> <chr>    <chr>       
1 Brown  Blue   1     bonferroni id       0.0637 K_W      ns          
2 Brown  Hazel  0.671 bonferroni id       0.0637 K_W      ns          
3 Brown  Green  0.368 bonferroni id       0.0637 K_W      ns          
4 Blue   Hazel  0.347 bonferroni id       0.0637 K_W      ns          
5 Blue   Green  0.176 bonferroni id       0.0637 K_W      ns          
6 Hazel  Green  1     bonferroni id       0.0637 K_W      ns  

> stat3result[["stat"]]
# A tibble: 6 × 8
  group1 group2 p.adj posthoc    variable     p1 P1method p.adj.signif
  <chr>  <chr>  <dbl> <chr>      <chr>     <dbl> <chr>    <chr>       
1 Brown  Blue   1     bonferroni id       0.0637 K_W      ns          
2 Brown  Hazel  0.671 bonferroni id       0.0637 K_W      ns          
3 Brown  Green  0.368 bonferroni id       0.0637 K_W      ns          
4 Blue   Hazel  0.347 bonferroni id       0.0637 K_W      ns          
5 Blue   Green  0.176 bonferroni id       0.0637 K_W      ns          
6 Hazel  Green  1     bonferroni id       0.0637 K_W      ns      
    
> stat3result[["normal"]]
  group variable normal meanvalue        sd
1 Brown       id   TRUE    27.500 23.348295
2  Blue       id   TRUE    26.875 21.463840
3 Hazel       id  FALSE    11.625  9.694439
4 Green       id   TRUE     8.000  4.598136

📖References

1. Comparing Means in R
2. Learning Statistics with R
3. HOW CAN I DO POST-HOC PAIRWISE COMPARISONS IN R? | R FAQ