See How Outliers Change Mean and Median

Select a timestamp to watch that moment on YouTube.

1. 00:00

   A dashboard says the average value is 22.8, but four of its five observations are no larger than five. Is the dashboard wrong? It may simply be using the arithmetic mean, which includes the magnitude of every value. The median, the center of the sorted data, is four. Both are correct calculations, but a vague label such as average hides the difference. Today we will make a report that shows both measures and their signed gap. We will also reproduce a practical Python bug that appears when the input is a generator: one statistic consumes the iterator, leaving no values for the second statistic.

2. 00:41

   The fix is to materialize the data once before asking for two passes. We will test an odd-sized skewed group, an even-sized group, a generator, and an empty input. The value one hundred might be a rare valid event, a typing mistake, or a member of another group. The difference between mean and median tells us to inspect context; it does not authorize deleting data. You can run every sample in the lesson's browser console and then complete the checked function. The mean call receives all five generated values, adds them, and returns 22.8. A generator is a one-shot iterator; after those values have been yielded, it has

3. 01:23

   nothing left. median then sees empty input and raises StatisticsError. The try block catches that error so our video can display the symptom rather than stopping. This bug can hide when a function is tested only with lists, because lists can be traversed repeatedly. The public function should accept an iterable, and that includes a generator unless we deliberately restrict its contract. We could redesign a streaming statistical algorithm for very large datasets, but computing an exact median generally needs more information than a simple running sum. For this small lesson, copying the iterable into a list once is the clear solution.

4. 02:05

   We must also decide what happens when the original input is truly empty. Returning zero would invent a measurement; this task rejects emptiness with ValueError. Before implementing that repair, look closely at what the two statistics would mean on the actual data. With one hundred included, the mean is 22.8 and the median is four. When we temporarily remove it, both summaries are 3.5. This exercise is a comparison, not a recommendation to discard the last observation. The mean responds strongly because one hundred contributes directly to the sum and the denominator still counts five values.

5. 02:46

   The median of the five-item list is the third value after sorting. It remains four even though the largest value is much farther away. For the four-item list, median has no single center value, so it averages the two middle values, three and four, to obtain 3.5. This distinction is useful when explaining salaries, transaction sizes, response times, or any distribution with a long tail. It is also easy to misuse. A median can conceal the magnitude of a severe high-end event, while a mean can be dominated by that event. Showing both often tells a more honest story than choosing one silently.

6. 03:27

   Our function returns a signed mean-minus- median gap to make the difference visible. It is not an outlier detector; no statistical threshold has been defined, and a positive gap alone cannot tell us whether the high value is valid. The list conversion is the central repair. It consumes the generator exactly once and gives both statistics the same observations. The empty check provides the function's own useful failure message instead of asking the library to fail later. We compute mean and median, then return each alongside their signed difference. In this dataset the gap is positive 18.8 because the large upper value pulls the

7. 04:08

   mean above the middle position. A different distribution could produce a negative gap. Keeping the input as a list has a memory cost proportional to the number of observations, which is reasonable for a small report but should be reconsidered for millions of records. This function also leaves a caller's existing list unchanged: list(values) makes a new outer list and statistics.mean and median do not sort the input in place. Do not conflate that with deep copying nested objects; the function only reads numeric values. On the companion page, the starter code already materializes an iterable but

8. 04:49

   incorrectly calls the mean the median and hard-codes a zero gap. Replace those placeholders and handle the empty case explicitly. Then compare the result with the visible example before running the checker. We first confirm the skewed result and that the source list has not changed. The even-sized sample checks an important median rule: with four sorted values the two center numbers are three and four, so the median is 3.5. That case also produces a zero gap, which is a genuine result, not a default for every input. The generator assertion proves the function can consume a one-shot iterable once and still compute both statistics.

9. 05:32

   An empty list is deliberately rejected because no mean or median exists for it; we print a short message only after catching ValueError. If any implementation returns a made-up zero or silently ignores the generator problem, the check will fail. Consider a final exercise: add a second large value, perhaps 90, and predict what happens to both measures before running the code. The center position may move once enough large values are present. That helps explain why neither statistic is magically immune to a dataset's composition. For a real dashboard, include the sample size and units, inspect whether extreme

10. 06:13

    observations are legitimate, and label the measure you show. The task on the website checks the function's output and behavior; the quiz checks the interpretation. Completing both makes the lesson more than a memorized call to statistics.mean or statistics.median.