The Monthly Average Hides the Day It Happened

Two months with the same mean can contain completely different weeks.

2026-01-09 · Backyard Station Bulletin

Two lines sit near each other in the station's summary sheet: July 2024 and July 2025, both logged at a noon average of 81.2°F. For a while that agreement looked like proof the log was settling into something dependable — the same number, arrived at independently, a year apart. It took pulling the daily entries back out from under the summary row, rather than trusting the row itself, to see that the two months had almost nothing else in common. One was a slow, even simmer that barely moved for four weeks. The other spent ten days flirting with an unseasonal cold front before finishing with the hottest stretch the log had seen all year. The average was not wrong. It just never mentioned any of that.

Two Julys, One Number

Laid out day by day, July 2024 looks jagged: a cool opening week, a colder second week that dipped low enough to feel like a mistake in the notebook, then two closing weeks that climbed hard enough to erase the earlier deficit and then some. July 2025 looks almost flat by comparison — four weeks that barely strayed from the low eighties, no week standing out from the others. Averaged over thirty-one days, the jagged month and the flat month land on the same number, because that is exactly what an average is built to do: it does not ask whether the days that produced it looked alike.

What an Average Is Built to Discard

A monthly mean is a sum divided by a count, and division has no memory of order. It cannot tell whether a hot day and a cold day sat next to each other in the same difficult week or landed a continent apart on the calendar with three calm weeks between them. Both arrangements contribute the same two numbers to the same sum. This is the same blind spot covered from a different angle in what three summers of noon readings actually show, where a single fixed sampling hour missed the shape of the day around it. A monthly average makes the equivalent trade across time instead of across the clock: it keeps the total and throws away the sequence.

Where the Two Months Actually Diverge

The clearest way to see the difference is week by week, set side by side against the same final number.

Two Julys can agree on paper and disagree in every way that mattered while they were happening.

The Trouble With Trusting the Summary Row

None of this would matter much if the monthly figure were only ever read as what it is: a single compressed number, useful for exactly one kind of question. The trouble starts when it gets asked a different kind of question — whether a given July was "calm," whether a rain-sensitive plant would have survived either month, whether a maintenance job scheduled for "an average week" would have hit a heat wave or a cold snap. An average July, taken as a single figure, answers none of those questions, because no single week of either actual July matched it. The flat month came closest, and even that month's second week ran nearly a degree above the yearly figure.

• A monthly mean can tell you whether one month ran warmer than another, on the whole.
• It cannot tell you whether that month was steady or whether it swung wide and cancelled itself out.
• It cannot tell you which week, if any, actually resembled the number reported for the month.
• Two means matching exactly is not evidence the underlying days matched — it can just as easily be evidence they didn't.

What the Daily Log Is For

This is the argument for keeping the daily entries at all, rather than letting the notebook collapse into a spreadsheet of monthly summaries once the entries are old enough to feel like clutter. A summary row is a fine index for browsing three years at a glance, but it is a poor substitute for the record underneath it, and the two Julys are the clearest proof this log has produced so far. The same flattening effect that hides a jagged week inside a flat monthly number is also patient enough to hide something slower: a related worry, covered in the drift nobody notices until the second winter, is that a series of individually unremarkable monthly averages can still be quietly drifting year over year, in a way no single month's number would ever flag on its own.

Keeping Both Numbers, Not Just One

None of this is an argument against averaging — a monthly mean genuinely is the right tool for some questions, and discarding it in favor of raw numbers for everything would just trade one blind spot for a different kind of overload. The practical habit, after three years of this log, has been to keep the mean for comparing months and to keep the daily line for everything else: was this month steady or was it two extremes that happened to cancel, and did that shape matter for whatever the reading was actually being used to decide. The mean answers the first question well. Only the daily record answers the second.