Mean
When to use: Near-symmetric numeric samples with few outliers.
When to avoid: Strong skew, heavy tails, or Tukey-flagged clumps that pull the average.
Assumptions: Treats every point equally; one extreme value moves it. Excel AVERAGE.
Statistics catalog
Depth for the pane’s DNA cards. Hypothesis tests are not in v1. Ranked summaries are “best supported among candidates on this sample.”
Which-summary rank
- Categorical → mode.
- Binary → proportion (mean/SD of 0/1 are comparison only).
- Ordinal-hint → median + IQR.
- Continuous / integer: outlier rate ≥ 5% or |skew| ≥ 1 → median + IQR; mild skew or a few outliers → 10% trimmed mean + IQR; else mean + STDEV.S.
Excel conventions: STDEV.S, SKEW, KURT, PERCENTILE.INC (Hyndman type 7). Trim uses floor(n×0.1) each tail; skip if that is 0.
10% trimmed mean
When to use: Mild skew or a few outliers when you still want an average-like center.
When to avoid: Very small n (cannot trim) or ordinal / categorical data.
Assumptions: Drops 10% of each tail, then averages the rest. Not Excel TRIMMEAN’s even-rounding discard.
Median
When to use: Skewed samples, outliers, or ordinal scores.
When to avoid: Near-symmetric data where the mean is more efficient — still fine, just less usual.
Assumptions: The middle order statistic. Excel MEDIAN.
Proportion
When to use: Binary 0/1 columns (pass/fail, yes/no).
When to avoid: Do not report this as a “typical value” for a continuous column.
Mode
When to use: Categorical labels or a few integer codes.
When to avoid: Continuous measurements (almost every value is unique).
Sample SD
When to use: With a mean on near-symmetric data.
When to avoid: As the only spread on skewed or outlier-heavy samples.
Excel STDEV.S (n−1), not STDEV.P.
IQR
When to use: With a median, or whenever tails should not dominate spread.
When to avoid: Very small n (quartiles are jumpy).
Q3 − Q1 with PERCENTILE.INC. Tukey fences use 1.5×IQR.
Scaled MAD
When to use: A robust companion to SD; check how much outliers inflate SD.
1.4826 × median absolute deviation, so it matches SD on a normal sample.