Trends, Noise and the Difference

A month is down. Is something wrong, or is that just what months do? Long records answer this and short ones cannot.

A concrete way to test the method is outlined in this working reference. Use a short pilot and keep only the fields that still make sense when comparing the record months later.

Establishing what normal variation looks like

Before asking whether a period is unusual, you need to know the ordinary spread.

A related source worth keeping beside this note is American Psychological Association, particularly for decisions involving stress awareness.

Take three years of monthly figures. The range they cover, excluding anything you can explain, is your noise.

A month inside that range is unremarkable however it feels. A month outside it is worth investigating.

This is the whole method and it requires no statistics beyond looking at a spread.

The comparison that works

Same period, year on year.

January against January, not January against December.

Adjacent-period comparison confuses season with change and is the source of most false findings in personal data.

What a trend looks like

Movement in one direction across several comparable periods, larger than the ordinary spread.

Three years of the same quarter declining is a trend. Two is a possibility. One is nothing.

And a trend should have a cause you can name. If you cannot, look for a change in your recording before concluding it is a change in your life — the drift note covers why.

The commonest false trend

A category growing steadily because its definition widened.

Indistinguishable from real growth in the numbers, detectable only by reading old entries.

Which is why the annual read includes reading entries rather than only totals.

Smoothing

A rolling average over three months makes trends visible and removes the month-to-month panic.

Use it for reading, not for storage. The underlying data should stay raw; smoothing is a lens, not a transformation.

And be aware that smoothing hides the extremes, which are sometimes the finding — a year with two catastrophic weeks and ten good ones averages to an ordinary year and was not one.

What to do with a confirmed trend

Name a cause before acting.

Check the recording first, then the life. Many apparent trends in long records are artefacts of a tool change, a category change or a shift in what you bothered to record.

And date the intervention if you make one, so that in two years you can see whether it did anything.

What not to conclude

That a bad quarter means the direction has changed.

That an improvement after an intervention was caused by it — you changed one thing and so did the world.

And that a flat record means nothing is happening. Stability is a finding, and for many people it is the one they most needed.

The one measure worth watching

Pick one, consistent with your question, and follow it across years.

A dashboard of twelve measures produces twelve false trends a year.

One measure, read annually, produces findings you act on.

What to check

Do you know your ordinary monthly spread?

When you last thought something was changing, did you compare year on year or against last month?

Could a trend you believe in be a definition change?

And is there one measure you follow, or a dashboard you glance at?

The point

Establish the ordinary spread before asking whether a period is unusual.

A month inside it is unremarkable however it felt, and a month outside it is worth investigating.

Worth adding

A three-month rolling average makes trends visible and hides the extremes, which are sometimes the finding.

Use smoothing for reading and never for storage.

Additionally

Check the recording before the life.

Many apparent trends are artefacts of a tool change, a category change, or a shift in what you bothered to record.

Finally

Pick one measure and follow it across years.

A dashboard of twelve produces twelve false trends a year; one measure read annually produces findings you act on.

In summary

Three years of the same quarter declining is a trend.

Two is a possibility. One is nothing, and a trend should have a cause you can name.