Coursera Course completion EDA
This is the on-site interview portion of the interview where I will conduct EDA of the course data to examine what we can explore to improve course completion
library(tidyverse)## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.2 ──
## ✔ ggplot2 3.3.6 ✔ purrr 0.3.4
## ✔ tibble 3.1.7 ✔ dplyr 1.0.9
## ✔ tidyr 1.2.0 ✔ stringr 1.4.0
## ✔ readr 2.1.2 ✔ forcats 0.5.1
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()library(skimr)
library(knitr)data <- read_csv('~/Downloads/onsite.csv')## Rows: 85607 Columns: 39
## ── Column specification ────────────────────────────────────────────────────────
## Delimiter: ","
## chr (1): user_id
## dbl (38): registration_time, grade, passed, age, education, employment, gend...
##
## ℹ Use `spec()` to retrieve the full column specification for this data.
## ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.skim(data) %>% knitr::kable()| skim_type | skim_variable | n_missing | complete_rate | character.min | character.max | character.empty | character.n_unique | character.whitespace | numeric.mean | numeric.sd | numeric.p0 | numeric.p25 | numeric.p50 | numeric.p75 | numeric.p100 | numeric.hist |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| character | user_id | 0 | 1.0000000 | 40 | 40 | 0 | 85607 | 0 | NA | NA | NA | NA | NA | NA | NA | NA |
| numeric | registration_time | 0 | 1.0000000 | NA | NA | NA | NA | NA | -5.592963e+06 | 5.589680e+06 | -33403322 | -10228307.50 | -3543499.00 | -845833.500 | 1214678 | ▁▁▂▃▇ |
| numeric | grade | 0 | 1.0000000 | NA | NA | NA | NA | NA | 1.246811e+01 | 2.961924e+01 | 0 | 0.00 | 0.00 | 1.998 | 100 | ▇▁▁▁▁ |
| numeric | passed | 0 | 1.0000000 | NA | NA | NA | NA | NA | 9.598510e-02 | 2.945726e-01 | 0 | 0.00 | 0.00 | 0.000 | 1 | ▇▁▁▁▁ |
| numeric | age | 75630 | 0.1165442 | NA | NA | NA | NA | NA | 3.204801e+01 | 1.028923e+01 | 13 | 25.00 | 29.00 | 37.000 | 150 | ▇▂▁▁▁ |
| numeric | education | 75792 | 0.1146518 | NA | NA | NA | NA | NA | 7.138869e+00 | 1.860033e+00 | 0 | 7.00 | 7.00 | 8.000 | 10 | ▁▂▁▇▂ |
| numeric | employment | 76076 | 0.1113344 | NA | NA | NA | NA | NA | 1.584723e+00 | 2.359023e+00 | 0 | 0.00 | 0.00 | 3.000 | 8 | ▇▁▁▂▁ |
| numeric | gender | 75531 | 0.1177007 | NA | NA | NA | NA | NA | 8.691941e-01 | 3.462092e-01 | 0 | 1.00 | 1.00 | 1.000 | 2 | ▁▁▇▁▁ |
| numeric | reading | 75936 | 0.1129697 | NA | NA | NA | NA | NA | 3.506876e+00 | 5.664977e-01 | 0 | 3.00 | 4.00 | 4.000 | 4 | ▁▁▁▆▇ |
| numeric | speaking | 75946 | 0.1128529 | NA | NA | NA | NA | NA | 3.271193e+00 | 7.609432e-01 | 0 | 3.00 | 3.00 | 4.000 | 4 | ▁▁▂▇▇ |
| numeric | writing | 75923 | 0.1131216 | NA | NA | NA | NA | NA | 3.343556e+00 | 6.995069e-01 | 0 | 3.00 | 3.00 | 4.000 | 4 | ▁▁▂▇▇ |
| numeric | num_lectures_watched | 0 | 1.0000000 | NA | NA | NA | NA | NA | 8.953508e+00 | 1.856616e+01 | 0 | 0.00 | 0.00 | 11.000 | 602 | ▇▁▁▁▁ |
| numeric | num_unique_lectures_watched | 0 | 1.0000000 | NA | NA | NA | NA | NA | 5.823905e+00 | 9.632940e+00 | 0 | 0.00 | 0.00 | 8.000 | 27 | ▇▁▁▁▂ |
| numeric | num_lectures_streamed | 0 | 1.0000000 | NA | NA | NA | NA | NA | 5.968344e+00 | 1.233228e+01 | 0 | 0.00 | 0.00 | 4.000 | 168 | ▇▁▁▁▁ |
| numeric | num_unique_lectures_streamed | 0 | 1.0000000 | NA | NA | NA | NA | NA | 4.131590e+00 | 8.015672e+00 | 0 | 0.00 | 0.00 | 3.000 | 27 | ▇▁▁▁▁ |
| numeric | num_lectures_downloaded | 0 | 1.0000000 | NA | NA | NA | NA | NA | 2.985165e+00 | 1.336439e+01 | 0 | 0.00 | 0.00 | 0.000 | 602 | ▇▁▁▁▁ |
| numeric | num_unique_lectures_downloaded | 0 | 1.0000000 | NA | NA | NA | NA | NA | 2.179658e+00 | 6.744922e+00 | 0 | 0.00 | 0.00 | 0.000 | 27 | ▇▁▁▁▁ |
| numeric | num_quizzes_submitted | 0 | 1.0000000 | NA | NA | NA | NA | NA | 1.140982e+00 | 2.652004e+00 | 0 | 0.00 | 0.00 | 0.000 | 54 | ▇▁▁▁▁ |
| numeric | num_unique_quizzes_submitted | 0 | 1.0000000 | NA | NA | NA | NA | NA | 6.584742e-01 | 1.355048e+00 | 0 | 0.00 | 0.00 | 0.000 | 4 | ▇▁▁▁▁ |
| numeric | total_quiz_first_score | 66411 | 0.2242340 | NA | NA | NA | NA | NA | 1.253699e+01 | 5.667405e+00 | 0 | 7.75 | 14.00 | 17.750 | 20 | ▂▃▃▃▇ |
| numeric | total_quiz_last_score | 66411 | 0.2242340 | NA | NA | NA | NA | NA | 1.386138e+01 | 6.229622e+00 | 0 | 9.00 | 15.25 | 20.000 | 20 | ▁▂▂▂▇ |
| numeric | total_quiz_min_score | 66411 | 0.2242340 | NA | NA | NA | NA | NA | 1.236008e+01 | 5.644016e+00 | 0 | 7.50 | 13.75 | 17.500 | 20 | ▂▃▃▅▇ |
| numeric | total_quiz_max_score | 66411 | 0.2242340 | NA | NA | NA | NA | NA | 1.390936e+01 | 6.218716e+00 | 0 | 9.00 | 15.50 | 20.000 | 20 | ▁▂▂▂▇ |
| numeric | num_assignment_parts_submitted | 0 | 1.0000000 | NA | NA | NA | NA | NA | 9.082318e-01 | 2.833249e+00 | 0 | 0.00 | 0.00 | 0.000 | 102 | ▇▁▁▁▁ |
| numeric | num_unique_assignment_parts_submitted | 0 | 1.0000000 | NA | NA | NA | NA | NA | 6.742556e-01 | 1.931242e+00 | 0 | 0.00 | 0.00 | 0.000 | 7 | ▇▁▁▁▁ |
| numeric | total_assignment_parts_first_score | 74653 | 0.1279568 | NA | NA | NA | NA | NA | 9.969509e+01 | 4.826193e+01 | 0 | 60.00 | 100.00 | 150.000 | 150 | ▂▂▁▅▇ |
| numeric | total_assignment_parts_last_score | 74653 | 0.1279568 | NA | NA | NA | NA | NA | 1.186233e+02 | 4.545989e+01 | 0 | 100.00 | 150.00 | 150.000 | 150 | ▁▁▁▃▇ |
| numeric | total_assignment_parts_min_score | 74653 | 0.1279568 | NA | NA | NA | NA | NA | 9.950886e+01 | 4.827799e+01 | 0 | 60.00 | 100.00 | 150.000 | 150 | ▂▂▁▅▇ |
| numeric | total_assignment_parts_max_score | 74653 | 0.1279568 | NA | NA | NA | NA | NA | 1.187105e+02 | 4.545821e+01 | 0 | 100.00 | 150.00 | 150.000 | 150 | ▁▁▁▃▇ |
| numeric | num_threads | 0 | 1.0000000 | NA | NA | NA | NA | NA | 1.620190e-02 | 1.781373e-01 | 0 | 0.00 | 0.00 | 0.000 | 10 | ▇▁▁▁▁ |
| numeric | num_posts | 0 | 1.0000000 | NA | NA | NA | NA | NA | 1.363323e-01 | 1.334242e+00 | 0 | 0.00 | 0.00 | 0.000 | 131 | ▇▁▁▁▁ |
| numeric | num_up_voted_posts | 0 | 1.0000000 | NA | NA | NA | NA | NA | 6.382660e-02 | 7.203496e-01 | 0 | 0.00 | 0.00 | 0.000 | 76 | ▇▁▁▁▁ |
| numeric | num_down_voted_posts | 0 | 1.0000000 | NA | NA | NA | NA | NA | 1.133100e-03 | 3.819550e-02 | 0 | 0.00 | 0.00 | 0.000 | 3 | ▇▁▁▁▁ |
| numeric | num_votes_given | 0 | 1.0000000 | NA | NA | NA | NA | NA | 4.778231e-01 | 8.937769e+00 | -9 | 0.00 | 0.00 | 0.000 | 1413 | ▇▁▁▁▁ |
| numeric | num_votes_received | 0 | 1.0000000 | NA | NA | NA | NA | NA | 4.744355e-01 | 9.248649e+00 | -27 | 0.00 | 0.00 | 0.000 | 1752 | ▇▁▁▁▁ |
| numeric | num_up_votes_given | 0 | 1.0000000 | NA | NA | NA | NA | NA | 4.948778e-01 | 9.002486e+00 | 0 | 0.00 | 0.00 | 0.000 | 1417 | ▇▁▁▁▁ |
| numeric | num_up_votes_received | 0 | 1.0000000 | NA | NA | NA | NA | NA | 4.912098e-01 | 9.370208e+00 | 0 | 0.00 | 0.00 | 0.000 | 1755 | ▇▁▁▁▁ |
| numeric | num_down_votes_given | 0 | 1.0000000 | NA | NA | NA | NA | NA | -1.705470e-02 | 4.201418e-01 | -78 | 0.00 | 0.00 | 0.000 | 0 | ▁▁▁▁▇ |
| numeric | num_down_votes_received | 0 | 1.0000000 | NA | NA | NA | NA | NA | -1.677430e-02 | 4.226480e-01 | -75 | 0.00 | 0.00 | 0.000 | 0 | ▁▁▁▁▇ |
Overall completion rate: ~10%
mean(data$passed)## [1] 0.09598514Question1: Demographics: are there certain demographics that have higher course completion rate?
data %>%
filter(!is.na(employment)) %>%
ggplot(aes(x = factor(employment))) +
geom_bar() +
labs(x = 'Employment code', y = 'num users', title = 'User employment distribution')
data %>%
ggplot(aes(x =age)) +
geom_histogram() +
labs(x = 'Age', y = 'num users', title = 'User age distribution')## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.## Warning: Removed 75630 rows containing non-finite values (stat_bin).
data %>%
group_by(age) %>%
summarise(completion_rate = mean(passed)) %>%
ggplot(aes(x =age, y = completion_rate)) +
geom_point() +
geom_smooth() +
labs(x = 'Age', y = 'Completion rate', title = 'User age vs completion rate')## `geom_smooth()` using method = 'loess' and formula 'y ~ x'## Warning: Removed 1 rows containing non-finite values (stat_smooth).## Warning: Removed 1 rows containing missing values (geom_point).
Question2: Do users who watched more video tend to have higher completion rate?
data %>%
mutate(num_unique_lectures_watched_quantile = ntile(num_unique_lectures_watched, 4)) %>%
group_by(num_unique_lectures_watched_quantile) %>%
summarise(completion_rate = mean(passed)) %>%
ggplot(aes(x = num_unique_lectures_watched_quantile, y = completion_rate)) +
geom_col() +
labs(x = 'Num unique lectures watched quantile',
y = 'Completion rate',
title = 'Num unique lecture watched quantile vs completion rate')
data %>%
mutate(num_times_watched_per_lecture = round(num_lectures_watched/num_unique_lectures_watched,1)) %>%
group_by(num_times_watched_per_lecture) %>%
summarise(completion_rate = mean(passed)) %>%
ggplot(aes(x = num_times_watched_per_lecture, y = completion_rate)) +
geom_point() +
geom_smooth(method = 'lm') +
labs(x = 'Num watches per lecture', y = 'Completion rate', title = 'Num watches per lecture vs completion rate')## `geom_smooth()` using formula 'y ~ x'## Warning: Removed 1 rows containing non-finite values (stat_smooth).## Warning: Removed 1 rows containing missing values (geom_point).
Question3: Do users who submit more assignment tend to have higher completion rate?
data %>%
group_by(num_unique_assignment_parts_submitted) %>%
summarise(completion_rate = mean(passed)) %>%
ggplot(aes(x = factor(num_unique_assignment_parts_submitted), y = completion_rate)) +
geom_col() +
labs(x = 'Num unique assignment submissions',
y = 'Completion rate',
title = 'Num unique assignment submissions vs completion rate')
data %>%
filter(!is.na(num_unique_assignment_parts_submitted)) %>%
mutate(num_submission_per_assignment = round(num_assignment_parts_submitted/num_unique_assignment_parts_submitted,1)) %>%
group_by(num_submission_per_assignment) %>%
summarise(completion_rate = mean(passed)) %>%
ungroup() %>%
ggplot(aes(x = num_submission_per_assignment, y = completion_rate)) +
geom_point() +
geom_smooth(method = 'lm') +
theme(axis.text.x = element_text(angle = 90, vjust = 0.5, hjust=1)) +
labs(x = 'Num submissions per assignment',
y = 'Completion rate',
title = 'Num assignment submissions vs completion rate')## `geom_smooth()` using formula 'y ~ x'## Warning: Removed 1 rows containing non-finite values (stat_smooth).## Warning: Removed 1 rows containing missing values (geom_point).
Conclusion:
With our current data, demographic information is not as useful because we are missing 90% of the demographic information. –> Recommendation: We should try to collect more demographic information and verify the validity of those information before conducting analysis to arrive at meaningful result.
However, with the behavioral information, we are able to extract some interesting insights: those who watch more unique lectures tend to have higher completion rate, though frequency of watching each video does not improve completion rate –> Recommendation: We should explore the hypothesis that “Watching more unique lectures in the course could improve completion rate”. We can experiment with encouraging students to watch more lectures and verify the course lecture watch rate and completion rate.
Assignment submission also has position relationship with completion rate: those who have more unique submissions tend to have higher completion rate. We also see some signs of students who make more attempts in each assignment have higher completion rate; however, this data is quite noisy and should be further verified –> Recommendation: We should explore the hypothesis that “Completing more assignments in the course could improve completion rate”. We can experiment with encouraging students to attempt assignment and verify the course lecture watch rate and completion rate.