```{r}
sum(df_data$Variable)
```[1] 16.64 Summarising data
Often we are interested in comparing treatment means. Taking the mean of variables in our data is way of summarising it, along with various other ways we can summarise our data in R.
If we want the summary of different treatments, we are actually wanting to group our data by these treatments and then calculate something from them.
4.1 Split-apply-combine concept
As we will see, one of the most useful tools is the group_by() function. This works on the ‘split-apply-combine’ principal where you split the data up into smaller tables, do something to those smaller tables, then put all the data back together again.
Let’s say we wanted to know the sum of the Variable column, we could do:
As we learnt in the previous chapter, we could store this in a new column using mutate():
```{r}
df_data %>% mutate(sum_total = sum(Variable))
``` Animal_ID Timepoint Treatment_Group Variable sum_total
1 11 1 Treatment A 1.0 16.6
2 12 1 Treatment A 1.2 16.6
3 21 1 Treatment B 1.0 16.6
4 22 1 Treatment B 1.2 16.6
5 11 2 Treatment A 2.0 16.6
6 12 2 Treatment A 2.5 16.6
7 21 2 Treatment B 0.5 16.6
8 22 2 Treatment B 0.4 16.6
9 11 3 Treatment A 3.0 16.6
10 12 3 Treatment A 3.5 16.6
11 21 3 Treatment B 0.2 16.6
12 22 3 Treatment B 0.1 16.6But let’s say we wanted to know the sum of the variable column for each Treatment (not total). If we ‘group_by’ Treatment_Group first, then behind the scenes the sum function will first be given a subset of the dataframe containing only the Treatment A data, then it will be given all of the Treatment B data. To do it manually we would have to do do:
```{r}
df_treatmentA <- df_data %>% filter(Treatment_Group == 'Treatment A')
sum(df_treatmentA$Variable)
```[1] 13.2Then we would then have to repeat this for all of the treatments, which is not ideal. So, one might consider using a function to help, like:
```{r}
# 1. get all unique names in Treatment_Group
treatments <- unique(df_data$Treatment_Group)
treatments
```[1] "Treatment A" "Treatment B"```{r}
#2. Write a function
f_groupsums <- function(x){
df_temp <- df_data %>% filter(Treatment_Group == x)
return(sum(df_temp$Variable))
}
#3. use apply to iterate through the treatments
lapply(treatments, f_groupsums)
```[[1]]
[1] 13.2
[[2]]
[1] 3.4This takes a lot of code,is probably a little scary if you’re not used to making functions, and returns a list of numbers that you then need to process further to get the treatment names for them.
4.2 The easy way
Instead, we use group_by with the summarise function instead. The summarise function is like mutate, except you use functions inside it that summarise the data, returning single values per group:
```{r}
df_data %>%
group_by(Treatment_Group) %>%
summarise(sum_groups = sum(Variable))
```# A tibble: 2 × 2
Treatment_Group sum_groups
<chr> <dbl>
1 Treatment A 13.2
2 Treatment B 3.4This gives the same numbers, only in a nice data frame.
It’s also possible to group by multiple columns:
```{r}
df_data %>%
group_by(Treatment_Group, Timepoint) %>%
summarise(sum_groups = sum(Variable))
````summarise()` has grouped output by 'Treatment_Group'. You can override using
the `.groups` argument.# A tibble: 6 × 3
# Groups: Treatment_Group [2]
Treatment_Group Timepoint sum_groups
<chr> <int> <dbl>
1 Treatment A 1 2.2
2 Treatment A 2 4.5
3 Treatment A 3 6.5
4 Treatment B 1 2.2
5 Treatment B 2 0.9
6 Treatment B 3 0.3You could actually still use mutate if you wanted to store this in a column of the original data frame (useful for some calculations, such as % within a group)
```{r}
df_data %>%
group_by(Treatment_Group) %>%
mutate(sum_groups = sum(Variable),
perc_of_group_sum = Variable/sum_groups*100) %>%
arrange(Treatment_Group) #just to make it nicer to read
```# A tibble: 12 × 6
# Groups: Treatment_Group [2]
Animal_ID Timepoint Treatment_Group Variable sum_groups perc_of_group_sum
<dbl> <int> <chr> <dbl> <dbl> <dbl>
1 11 1 Treatment A 1 13.2 7.58
2 12 1 Treatment A 1.2 13.2 9.09
3 11 2 Treatment A 2 13.2 15.2
4 12 2 Treatment A 2.5 13.2 18.9
5 11 3 Treatment A 3 13.2 22.7
6 12 3 Treatment A 3.5 13.2 26.5
7 21 1 Treatment B 1 3.4 29.4
8 22 1 Treatment B 1.2 3.4 35.3
9 21 2 Treatment B 0.5 3.4 14.7
10 22 2 Treatment B 0.4 3.4 11.8
11 21 3 Treatment B 0.2 3.4 5.88
12 22 3 Treatment B 0.1 3.4 2.94This same principal applies to all base functions (e.g. mean, max, min, etc) or you can write custom functions and use them with mutate and summarise:
```{r}
df_data %>%
group_by(Treatment_Group) %>%
summarise(mean_variable = mean(Variable, na.rm=TRUE)) #use na.rm=TRUE if you want the mean function to ignore NA values
```# A tibble: 2 × 2
Treatment_Group mean_variable
<chr> <dbl>
1 Treatment A 2.2
2 Treatment B 0.5674.3 Summarising multiple columns
As we mentioned previously, we can use tidyselect with across() to efficiently iterate through our columns.
Let’s say we want to find the mean and sd of all columns that contain numbers. We could write it out in long form:
```{r}
penguins %>%
group_by(species) %>%
summarise(bill_length_mean = mean(bill_length_mm, na.rm = TRUE),
bill_length_sd = sd(bill_length_mm, na.rm = TRUE),
bill_depth_mean = mean(bill_depth_mm, na.rm = TRUE),
bill_depth_sd = sd(bill_depth_mm, na.rm = TRUE)
#etc....
)
```# A tibble: 3 × 5
species bill_length_mean bill_length_sd bill_depth_mean bill_depth_sd
<fct> <dbl> <dbl> <dbl> <dbl>
1 Adelie 38.8 2.66 18.3 1.22
2 Chinstrap 48.8 3.34 18.4 1.14
3 Gentoo 47.5 3.08 15.0 0.981Or, we can use across():
```{r}
penguins %>%
group_by(species) %>%
summarise(
across(.cols = where(is.numeric),
.fns = list(mean = ~mean(.x, na.rm = TRUE),
sd = ~sd(.x, na.rm=TRUE)),
.names = '{.col}_{.fn}') #see ?across for details on how this works
)
```# A tibble: 3 × 11
species bill_length_mm_mean bill_length_mm_sd bill_depth_mm_mean
<fct> <dbl> <dbl> <dbl>
1 Adelie 38.8 2.66 18.3
2 Chinstrap 48.8 3.34 18.4
3 Gentoo 47.5 3.08 15.0
# ℹ 7 more variables: bill_depth_mm_sd <dbl>, flipper_length_mm_mean <dbl>,
# flipper_length_mm_sd <dbl>, body_mass_g_mean <dbl>, body_mass_g_sd <dbl>,
# year_mean <dbl>, year_sd <dbl>4.4 Task:
- Choose 1 variable from you data and calculate the mean and sd for each treatment (or another relevant group)
- Mutate 2 new columns to your summary table called
yminandymax, containing the values calculated frommean - SDandmean + SD, respectively. - Advanced: re-calculate your summary table but also include the number of samples per group.