What Does a Reliable Historical Comparison Require?
From the earlier reference collection. Read the current software blog.
Historical comparisons become useful when their periods, definitions, and data coverage are aligned and their differences are made explicit.
Compare equivalent periods
A comparison between two historical values only makes sense if the time windows represent the same thing. Comparing the first week of one month with a full month from another will usually produce a misleading result. So can comparing a calendar month with a reporting period that begins midweek. State the start and end dates, and make clear whether the endpoints are included.
For example, to compare weekday service delays across two months, select the same weekdays and equivalent hours in each period. If one interval includes a holiday or a service disruption, either account for it explicitly or explain why it remains in the comparison. A chart title such as “April versus May” is not enough when the actual windows differ.
Keep definitions stable
A metric may change meaning even when its label stays the same. “Daily users” might count unique accounts in one period and sessions in another. A sensor summary might shift from an hourly mean to a daily maximum. Before comparing values, identify the formula, unit, and population represented in each period.
If a definition changed, mark the break in the series and describe its effect. For instance, a dashboard that begins counting an additional group of locations should not draw an uninterrupted trend line as if the earlier values covered the same network. A note can state the effective date and whether older records were recalculated. If they were not, readers should see that the comparison crosses a methodological boundary.
Check completeness and missingness
A period with fewer observations can look artificially low or stable. Report how much data is available in each interval, and describe material gaps. If one week has complete daily readings while another has several missing days, a total is not a fair comparison. A per-day average may help, but only if the missingness and calculation are made clear.
Coverage should include the locations and fields that contribute to the result. When a source is revised after initial publication, retain the refresh date or version where practical so readers can understand why a past value differs from an earlier view.
Make the comparison reproducible
A reliable comparison gives readers enough context to recreate or assess it: exact dates, metric definition, units, filters, and coverage. A brief worked example in the method note can show how records become the reported figure. The visual can then focus on the pattern, while the accompanying explanation establishes that both sides of the comparison use a consistent basis.