17.1 Mean, Median & Mode

17.1 Mean, Median & Mode

Quick Answer: Mean = sum of values ÷ count (sensitive to extremes). Median = middle value of ordered data (robust to skew/outliers). Mode = most frequent value or class (best for discrete/categorical or multimodal data). On the CQA exam (BoK V.C.1 Apply), you must choose the right measure for audit data and interpret what it does—and does not—say about process performance. Practice free questions at /practice/cqa.

Why central tendency matters in audits

Auditors constantly meet summary statistics: “average scrap is 1.2%,” “mean CAPA age is 28 days,” “typical supplier OTD is 96%.” Those phrases can be accurate yet deceptive if the wrong center measure is used, the sample is stratified badly, or extremes dominate the mean. Domain V expects you to apply measures of central tendency—not just define them—so you can:

• Verify calculations in process records and management reviews
• Compare periods, lines, shifts, or suppliers fairly
• Challenge “green dashboard” averages that hide tails of nonconformance
• Support findings with objective numeric evidence, not anecdotes alone

Central tendency answers: Where is the bulk of the data located? Dispersion (next section) answers how spread out it is. Capability and variation concepts later in this chapter depend on both.

Mean (arithmetic average)

For a sample of n observations $x_1, x_2, \ldots, x_n$:

$\bar{x} = \dfrac{x_1 + x_2 + \cdots + x_n}{n}$

Population mean $\mu$ uses the same idea with all values in the population (rarely known in audit settings; auditors almost always work with samples).

Worked mini-calculation — mean

Five consecutive audit sampling results for defect counts per lot: 2, 1, 3, 2, 12.

• Sum = 2 + 1 + 3 + 2 + 12 = 20
• Mean = 20 / 5 = 4.0 defects per lot

Without the outlier 12, the mean of the first four values is 8 / 4 = 2.0. One extreme observation doubled the reported average. That sensitivity is the mean’s strength (every point counts) and its audit risk (one bad lot or data-entry error can dominate the story).

When auditors should prefer the mean

Audit caution: When management reports only a long-run mean, ask for the distribution shape, time window, and whether special-cause events were included. A mean that includes a one-time fire-sale of scrap or a known equipment failure can misrepresent the current process.

Median

Order the data from smallest to largest. The median is:

• Odd n: the middle value (position $(n+1)/2$)
• Even n: average of the two middle values (common continuous-data convention)

Worked mini-calculation — median

Same defect data ordered: 1, 2, 2, 3, 12

• n = 5 (odd) → median = middle value = 2

Compare to mean 4.0. For this skewed set, the median better describes “typical” lot behavior; the mean is pulled by the extreme 12.

Even n example — CAPA ages (days): 10, 14, 18, 45

• Two middle values: 14 and 18 → median = (14 + 18) / 2 = 16 days
• Mean = (10 + 14 + 18 + 45) / 4 = 21.75 days

If leadership claims “average CAPA age is about three weeks,” ask whether they mean mean or median—and whether open vs. closed CAPAs are mixed.

When auditors should prefer the median

• Right-skewed data: repair times, lead times, complaint ages, cost of poor quality per event
• Left-skewed or truncated data: yields near 100%, cleanliness scores near a ceiling
• Outlier-prone measures where extremes are real but rare
• Ordinal ratings sometimes summarized by median rather than mean of arbitrary codes

Exam framing: Choosing median over mean is not “anti-statistics”—it is matching the summary to the shape and purpose of the data so audit conclusions stay fair.

Mode

The mode is the value (or class interval) that occurs most frequently.

• Unimodal: one clear peak
• Bimodal / multimodal: two or more peaks (often a signal of mixed populations—two machines, two shifts, two suppliers)
• No unique mode: all values equally frequent, or continuous data with no repeated values (mode may be estimated from histogram bins)

Worked example — mode in audit data

Nonconformance codes in 20 records: A, B, A, C, A, B, A, D, A, B, A, C, A, B, A, A, B, C, A, A

• Code A appears 11 times (B appears 5, C appears 3, D appears 1) → mode = A

Even if “mean severity” is low, the mode tells auditors where frequency of failure modes concentrates—critical for Pareto analysis and sampling focus.

When auditors should prefer the mode

Comparing mean, median, and mode

Shape rules of thumb (for roughly unimodal data):

• Roughly symmetric: mean ≈ median ≈ mode
• Right-skewed (long high tail): mean > median (often > mode)
• Left-skewed: mean < median

Auditors use these relationships as a quick integrity check. If a report shows a mean far above the median for cycle time, expect a long right tail—and dig into chronic delays, not only the “average.”

Stratification before you average

A classic audit failure is computing one plant-wide mean when the data mix distinct processes.

Example: Line A cycle times mean 12 min; Line B mean 28 min. Combined mean might be 20 min—a value neither line runs. Correct practice: stratify by line, product, shift, supplier, or measurement method before summarizing center. Multimodal histograms and dual modes are red flags for mixed streams.

Audit application checklist (Apply-level)

1. Define the characteristic and unit (e.g., ppm, %, days, mm).
2. Confirm the dataset (period, inclusion/exclusion of rework lots, special runs).
3. Plot or scan for shape (histogram, ordered list, box plot if available).
4. Select measure(s): mean and/or median; mode for categories.
5. Compute and show sample size n.
6. Interpret in process language: typical performance vs. tail risk.
7. Challenge auditee statements that use “average” without specifying measure, n, or period.

Common exam traps

• Treating mode as “average” for continuous unique data
• Using mean of ranks or letter grades without care
• Ignoring that median of even n may not be an observed value
• Assuming mean always “best” because it uses all data—sometimes that is exactly why it misleads
• Reporting central tendency without any sense of spread (incomplete picture)

Link forward

Center alone does not prove process health. A mean on target with huge dispersion still produces defects. Next sections cover dispersion, qualitative vs. quantitative patterns, common vs. special cause, and capability/outliers—the full auditor toolkit for statistical evidence. Study the full guide at /study-guides/cqa.