Welcome to the HMX Data Science & Analytics takehome exam. Please submit answers in the form of a Jupyter notebook, in any scripting language of your choice. Please email zipped files to matt_bunch@hms.harvard.edu within five days of receiving the exam.

Scenario

Imagine our team is tasked to understand learner behavior within an online course. The broader team would like to better understand learner behavior and predictability of learner completion.

The data in data.csv is a sample of learner question submission. All data is tied to registered and enrolled users.

The objective is to understand learner behavior and to predict learner completion.

Note: Each row in data.csv represents a learners question submission.

Data

Column Name	Description	Desc Cont
course_id	the course id
max_grade	the maximum grade the learner could earn
created	the datetime of when the original record was created
grade	the learners earned points
learner_id	the learners unique identifier
modified	the datetime in which the record was modified
state	JSON formatted string containing:
- last_submission_time	of the attempts taken, the last time
- attempts	number of question attempts taken
- score
	- raw_possible	the points possible for a learner to earn
	- raw_earned	the points a learner earned
- student_answers
	- question_id	the question_id itself (the key) is in the form of a hash

Questions

1. What percent of each question did learners attempt twice?

2. Which questions were most frequently answered incorrect?

3. Create a visualization exploring the relationship between any of the content characteristics (such as learners, attempts, time etc...). This is an open-ended task. Briefly describe the visualization and the insight.

4. Create a simple model predicting the likelihood of a learner passing the course. (A passing grade is 80% or above)

5. What additional data would you add to improve the model? (Answer in 3-5 sentences)