Katkat
Back

Visualize Trip Duration, Fees, and Survey Outcomes

AmazonBusiness Intelligence Engineer · Technical Screen
EasyBusiness Intelligence EngineerTechnical ScreenAnalytics & Experimentation

Last updated

This question tests the ability to design an analytical framework for identifying patterns in user satisfaction relative to trip costs and duration, while accounting for potential data quality issues like response bias.

Scenario

You are a Business Intelligence Engineer at a ride-sharing company. You are tasked with analyzing the relationship between trip duration, total fees, and a binary post-trip survey outcome (Satisfied vs. Dissatisfied). Additionally, you are interested in segmenting these metrics by 'Driver Experience Level' (New, Intermediate, Veteran).

Your goal is to provide a comprehensive visualization strategy that helps stakeholders understand what factors drive dissatisfaction and whether specific segments are disproportionately represented in the feedback loop.

Task

  1. Define the visualization strategy: Which chart types would you select to show the distribution of trip fees and durations, and how would you visualize the impact of the binary survey outcome on these continuous variables?
  2. Address Overplotting: With millions of rows of trip data, standard scatter plots become unreadable. Propose a technique to visualize these relationships without losing nuance.
  3. Identify Bias: If the 'Dissatisfied' population shows a significantly different distribution of trip durations compared to the 'Satisfied' population, how would you investigate whether the survey data is biased (e.g., are dissatisfied customers more likely to complete the survey)?

Requirements

  • Provide a specific recommendation for chart types for continuous vs. categorical relationships.
  • Explain how you would use faceting or color encoding to incorporate the 'Driver Experience Level' segment.
  • Propose a statistical or graphical approach to mitigate the risks of response bias in your analysis.

Hints

Hint 1

Consider using kernel density estimation (KDE) or box plots instead of raw scatter plots to handle the high volume of trip data.

Hint 2

Think about how faceting by 'Driver Experience Level' can help distinguish whether dissatisfaction is an artifact of specific driver segments.

Hint 3

To address response bias, consider comparing the distribution of trip durations for respondents versus the total population of all completed trips.

Follow-up questions

  • How would you modify your visualization if you discovered that 'Total Fees' has a significant right skew due to long-distance premium trips?
  • If we wanted to predict the probability of a 'Dissatisfied' outcome, what features would you engineer from the trip duration and fee data?
  • How would you communicate to non-technical stakeholders that a correlation between high fees and dissatisfaction does not necessarily imply causation?