Test If Initial Video Uploads Are Shorter Than Later Ones
Quick Overview
Evaluates hypothesis testing for whether users' first video uploads are shorter than later uploads. Strong answers define a clean data slice, paired user-level metrics, robust tests, and SQL-style implementation.
Test If Initial Video Uploads Are Shorter Than Later Ones
Company: LinkedIn
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: medium
Interview Round: Technical Screen
##### Scenario
A video-sharing platform wants to know whether users’ first uploaded videos tend to be shorter than their later uploads.
##### Question
Describe how you would test the hypothesis that users’ initial video posts are shorter. Define the data slice, metrics, statistical test, and provide high-level SQL or pseudo-code for the analysis.
##### Hints
Compare first-video length vs. subsequent average per user; control for outliers and posting date distribution.
Quick Answer: Evaluates hypothesis testing for whether users' first video uploads are shorter than later uploads. Strong answers define a clean data slice, paired user-level metrics, robust tests, and SQL-style implementation.
Hypothesis Test: Are Users' First Video Uploads Shorter?
You are given event-level data for a video-sharing product. Each record represents a published video with its duration and uploader. Determine whether a user's first uploaded video tends to be shorter than their subsequent uploads.
Assume a table like video_uploads(user_id, video_id, upload_ts, duration_sec, status, visibility, category, source).
Constraints & Assumptions
Define "first upload" by timestamp after cleaning and filtering.
Compare within users where possible to control for creator-level style.
Account for skewed video-duration distributions.
Exclude corrupt, private, test, deleted, or non-comparable uploads if appropriate.
Clarifying Questions to Ask Guidance
Should imports, ads, reposts, livestreams, or bulk migrations be excluded?
What duration range is valid for published videos?
Are we testing all users or only users with at least two uploads?
Should the analysis be segmented by category, source, creator type, or cohort?
Part 1 - Data Slice
Define inclusion and exclusion criteria.
What This Part Should Cover Guidance
Include published, visible, valid-duration videos.
Require users with at least one first upload and later uploads for paired analysis.
Handle ties in upload timestamp and duplicate records.
Consider cohort windows to avoid right-censoring newer users.
Part 2 - Metrics and Aggregation
Define the comparison metric.
What This Part Should Cover Guidance
Compare each user's first-video duration with their later-upload average or median.
Consider log-transformed duration because durations are often right-skewed.
Aggregate user-level paired differences rather than letting heavy uploaders dominate.
Report effect size and confidence interval.
Part 3 - Statistical Tests and Implementation
Specify tests, assumptions, and high-level SQL or pseudocode.
What This Part Should Cover Guidance
Use paired t-test on transformed or user-level differences if assumptions are reasonable.
Use Wilcoxon signed-rank, sign test, or bootstrap as robust alternatives.
Show window functions to rank uploads by timestamp and aggregate later durations per user.
Check robustness by segment and sensitivity to outliers.
Follow-up Questions Guidance
What if first uploads are shorter only for new creators in the last month?
How would you avoid survivorship bias from requiring later uploads?
How would you communicate a statistically significant but tiny duration difference?