Mean, median, and mode: which measure for task completion times?

WHAT THIS TESTS: Whether you recognize that raw task times are non-normal and heavily right-skewed, and whether you know that the arithmetic mean is distorted by a few slow outliers while the median stabilizes only with adequate sample size. The interviewer wants to see statistical literacy applied to UX metrics, not just textbook definitions.

A GOOD ANSWER COVERS: First, define the three measures in one sentence. The arithmetic mean is the sum divided by the count. The median is the middle value when times are sorted. The mode is the most frequent value, which is rarely useful for continuous task times because durations are spread across many unique values. Second, state that task completion times are almost always right-skewed due to a few users who struggle or get distracted. Third, explain the sample size rule from MeasuringU: use the geometric mean for small samples under twenty-five users because it dampens the effect of outliers, and use the median for larger samples because it becomes more stable as n grows. Fourth, mention that confidence intervals should be computed after log-transforming the data, since standard parametric methods assume normality. Fifth, clarify which task time you are averaging, because the reference distinguishes between average task completion time for successful users only, mean time to failure, and average time on task regardless of outcome.

COMMON WRONG ANSWERS: Defaulting to the arithmetic mean without mentioning skew or outliers. Claiming the median is always better without acknowledging it is less accurate for small sample sizes. Confusing average time on task with average task completion time and including failed attempts when the metric should reflect successful paths. Ignoring log transformation and proposing a plain standard deviation or standard error. Mentioning mode as a viable central tendency for continuous task times.

LIKELY FOLLOW-UPS: How would you report a confidence interval for task times. When is it appropriate to exclude failed tasks from the average. How do you compare task times between two designs. What is Keystroke Level Modeling and when would you use it. How do you handle users who were obviously distracted during an unmoderated test.

ONE CONCRETE EXAMPLE: Suppose you run a summative usability test with twenty users and collect successful task completion times. The arithmetic mean is four minutes because two users took twelve minutes after hitting a confusing label. The geometric mean is about two and a half minutes, and the median is two minutes. Because the sample is under twenty-five, you report the geometric mean and add a confidence interval derived from log-transformed data. If you had one hundred users, you would shift to the median and still log-transform for the interval. You would not report a mode because twenty continuous measurements rarely cluster on the exact same second.