Three Ways to Describe a Week of Weather

Opening Scene

A forecaster reviewing a week of daily rainfall totals doesn’t reach for just one number. She’ll note the average rainfall across the seven days, but she’ll also check the middle value if she sorts the days low to high, because one freak downpour can drag the average up without a single other day actually feeling wet. And she’ll note how much the days varied from each other — a week where every day got roughly the same light drizzle tells a very different story than a week with six bone-dry days and one flood, even if both weeks average out to the same total.

Three numbers, three different jobs. None of them alone tells the whole story, and a forecaster who reported only one would be quietly hiding the other two.

In Plain English

The mean is the arithmetic average — sum every value and divide by how many there are. The median is the middle value once everything is sorted, immune to being dragged around by a single extreme reading. Variance (and its square root, standard deviation) measures how spread out the values are around their center — a low variance means the readings cluster tightly, a high variance means they’re scattered widely. The mean and median both try to answer “what’s typical,” while variance answers a different question entirely: “how much do individual readings actually differ from that typical value.” A dataset isn’t fully described until you’ve reported both a center and a spread.

The Old Way

Before this three-way distinction is made explicit, a few habits tend to fill the gap, each one incomplete in its own way:

Reporting only the mean — the classic mistake, since a single very large or very small reading can pull an average far from where most of the data actually sits, the way one record-breaking storm can make an otherwise dry week look average.
Assuming mean and median will always roughly agree — true for a symmetric distribution, false for a skewed one, and skewed distributions are far more common in the real world than most people assume, from income to rainfall to response times.
Ignoring spread entirely — comparing two weeks purely by their average rainfall while missing that one was a week of steady drizzle and the other was six dry days and a flood, a difference that matters enormously for anyone planning around it.

Each shortcut isn’t wrong so much as partial — it answers one question about the data while silently leaving another one unasked.

What’s Changing (and Why AI Is the Reason)

AI-assisted exploratory tooling now reports mean, median, and variance together by default, rather than requiring an analyst to specifically request all three. What used to take deliberate statistical discipline to remember is increasingly surfaced automatically the moment a dataset is loaded.
Automated anomaly detection increasingly relies on variance directly, flagging a reading as unusual precisely when it falls far outside the typical spread rather than merely far from the average. This makes variance a working, load-bearing concept in production systems, not just a classroom definition.
As dashboards increasingly summarize data into single AI-generated headline numbers, the risk of quietly collapsing three different questions — center, midpoint, and spread — into one misleading figure grows, unless someone insists the full picture stays visible.

The Metaphor, Fully Extended

Weather Element	Statistics Concept
The average rainfall total across all seven days	The mean — the arithmetic average of all readings
The middle day’s rainfall once the week is sorted low to high	The median — the midpoint value, resistant to extreme readings
How much each day’s rainfall actually differed from the others	Variance (and standard deviation) — the spread of readings around their center
One freak downpour dragging the week’s average upward	A skewed mean, distorted by an extreme value the median would resist
A week of steady drizzle versus six dry days and a flood, same total rainfall	Two distributions with an identical mean but very different variance

For Beginners: What to Actually Do

Whenever you compute a mean, compute the median alongside it, and treat any large gap between the two as a signal the data is skewed by extreme values.
Never report a “typical value” without also reporting a measure of spread — a mean or median means little without variance or standard deviation beside it.
Practice constructing your own counterexample: two small datasets with an identical mean but obviously different spreads, until the distinction feels concrete rather than abstract.
Ask, whenever a dashboard shows you a single summary number, which of these three jobs it’s actually doing, and whether the other two are quietly missing.

For Practitioners and Leaders: The Deeper Layer

Standardize on reporting center and spread together across your team’s dashboards and reports, rather than leaving mean-only summaries as the default.
Use variance-based thresholds for automated anomaly detection rather than fixed absolute thresholds, since what counts as “unusual” should scale with how much a given metric naturally varies.
Watch for skewed distributions — costs, latencies, response times — where the mean and median diverge meaningfully, and make sure leadership reporting reflects the median where it’s the more honest “typical” figure.
Push back on any AI-generated summary that reduces a dataset to a single headline number without preserving access to the underlying spread.

Quick Recap

The mean and median both describe what’s typical, but the median resists being dragged around by extreme values in a way the mean cannot.
Variance describes something entirely different — not what’s typical, but how much individual readings actually differ from that typical value.
Reporting only the mean, assuming mean and median always agree, or ignoring spread entirely are all common, incomplete habits that hide real information.
A dataset isn’t genuinely described until both a center and a spread have been reported side by side.

Where This Fits in the Series

This article gives you the core vocabulary for summarizing any distribution encountered in Article 2. Article 4 turns from describing what already happened to expressing genuine uncertainty about what hasn’t happened yet — the discipline of probability.