Overview

For this dataset, I’ll be using the one I found and it represents test score data. This data has one row per student and includes the following columns:

student_data <- read.csv("https://gist.githubusercontent.com/Kimmirikwa/b69d0ea134820ea52f8481991ffae93e/raw/4db7b1698035ee29885d10e1a59bd902716ae168/student_results.csv")
head(student_data)

Ideally, we flatten this table to be one row per student, test, and term number rather than having columns for term scores. Additionally, the sex and age column should be split into two columns.

Tidying the Data

One Row per Observation

To start, let’s pivot the data to make one row per observation.

student_data_pivoted <- pivot_longer(
  student_data, cols=6:8, names_to="term_number", values_to="test_score"
  )
head(student_data_pivoted)

The data looks properly pivoted!

Test and Term Number

Next, we’ll be reformatting term_number to remove the . and replace it with a space. We’ll also rename the column test.number to test_number for consistency.

student_data_pivoted$term_number <- gsub("\\.", " ", student_data_pivoted$term_number)

student_data_pivoted <- student_data_pivoted |> rename("test_number" = "test.number")

head(student_data_pivoted)

Perfect!

Sex and Age Split

Next, we’ll split the sex and age column as it’s currently formatted as [sex]_[age]. We’d want separate columns for this situation, one for sex and the other for age. We’d also want age to be numeric.

student_data_pivoted <- student_data_pivoted |> 
  separate_wider_delim(sex.and.age, delim="_", names = c("sex", "age"))

student_data_pivoted$age <- as.numeric(as.character(student_data_pivoted$age))

head(student_data_pivoted)

The data looks clean for analysis now!

Analysis

Class Breakdown

First, let’s look at the breakdown of this class by sex and age.

student_by_sex_age <- student_data_pivoted |> 
  group_by(sex) |>
  summarise(mean_age = mean(age),
            student_count = n_distinct(id),
            max_age = max(age),
            min_age = min(age),
            .groups = 'drop')

student_by_sex_age

We can see that there are more females in this class at 3 vs. 2 and the average age for females is a little higher than males at 13 vs. 11.5. The distribution of ages is also different as females go from 12 to 14 while males only range from 11 to 12. Now, how did each group do in terms of test scores?

Test Performance

We’ll look at overall average per term, then break it down by sex and age.

term_averages <- student_data_pivoted |> 
  group_by(term_number) |>
  summarise(mean_test_score = mean(test_score),
            median_test_score = median(test_score),
            min_test_score = min(test_score),
            max_test_score = max(test_score),
            .groups = 'drop')

term_averages

On average, students performed best in term 2, however the median test score in term 3 was higher. It seems there was more variance in term 3 as the range is from 63 to 94 while the other terms were a little closer in score.

Now, does sex affect how well the students did?

sex_averages <- student_data_pivoted |> 
  group_by(sex) |>
  summarise(mean_test_score = mean(test_score),
            median_test_score = median(test_score),
            min_test_score = min(test_score),
            max_test_score = max(test_score),
            .groups = 'drop')

sex_averages

On average, it looks like males perform better than females from these test scores. The females have the highest test score as well as the lowest one though, signalling that they have the most variance in the group. However, remembering that there are only 2 males and 3 females in this class, it’s hard to say whether this data is conclusive enough to extrapolate any meaning outside this one set of students.

Conclusion

Using tidyr, it was pretty simple to pivot and split columns based on delimiters which was what this data needed in order to get tidy. Upon doing that, using dplyr to help summarise is getting easier with each dataset I work with. The findings of this analysis aren’t super interesting due to the small sample size, however the code can be recycled if given a dataset of more students for more than one year’s worth of data in order to find overall averages and perhaps even trends to see if the course is seemingly getting harder or easier.