Seasonal patterns that look like trends

Many measurements follow the calendar. Heating use rises in cold months, retail sales often climb toward the end of the year, and outdoor activity peaks in warmer weather. When a series repeats the same shape every year, comparing one month with the month before can make ordinary seasonality look like a trend. A rise in December may simply be a normal December, and a fall in February may simply be a normal February.

The rule that prevents most of these mistakes is to compare each period with the same period in other years rather than with its neighbors. The rule works because it holds the time of year constant, so the difference left between two figures is more likely to reflect a real change. How to apply it depends on what the data look like and on how they have been prepared.

What counts as a seasonal pattern

A seasonal pattern repeats at a fixed interval, usually a year, although the interval can be shorter or longer. Daily data often show a weekly cycle, with more activity on some days of the week than others. Monthly data often show a yearly cycle. The key feature is regularity: the pattern recurs at the same point in each cycle, so it can be anticipated from the calendar alone.

Calendar effects are a related but separate source of variation. A month can contain four or five weekends, a holiday can fall on a different weekday from one year to the next, and some holidays move on the calendar altogether. Two months with the same underlying activity can contain different numbers of working days, which alone can produce a rise or a fall. These effects create apparent movements that have nothing to do with the underlying level.

Compare like with like

The most common way to remove a seasonal pattern by hand is to compare each period with the same period a year earlier. The change from one March to the March before it is a year-over-year comparison, and it holds the season constant. Because the seasonal shape is the same in both months, most of the difference reflects change in the underlying level rather than the time of year.

Year-over-year comparisons have a trade-off. They are insensitive to seasonality, but they respond slowly to a change that began partway through the year, because each month is measured against the same month a year ago. Comparing each quarter with the same quarter in the previous year has the same property on a coarser scale. Use the comparison that fits your question, and state which one you used.

Comparisons of neighboring months are not always wrong, but they must be read with the season in mind. A rise from a seasonally low month to a seasonally high one is expected and says little about direction. If you do compare neighboring months, ask whether the same change happened at the same time of year in earlier years.

Check whether a figure is seasonally adjusted

Many published series come in two versions: the raw figures and a seasonally adjusted version. Seasonal adjustment estimates the typical seasonal effect from past data and removes it, so the adjusted line shows the underlying movement more clearly. The label usually appears in the title, the table header or a note under the chart. When you compare a figure from one source with a figure from another, check that both are the same kind.

Adjusted figures are not perfect. The seasonal factors are estimates, and they can be revised as new observations arrive. The adjustment also cannot remove a pattern whose shape has changed, such as a season that started earlier than usual. A careful reader treats an adjusted series as the best available estimate of the underlying movement, and checks the revision notes before reading too much into the latest change.

Separate the calendar from the trend

When you have several years of monthly data, plot them on the same chart, with months along the horizontal axis and one line for each year. The lines often share a common shape. The question is whether the latest year's line sits above or below the earlier years' lines at the same point in the cycle. If it runs above all of them, that is evidence of a change in level. If it only rises and falls along the same path as before, the calendar explains the movement.

Be cautious with short histories. Two years of data show only one repetition of the cycle, so a shape that looks seasonal may be coincidence. Look for consistent behavior across at least three or four years where possible, and note any holiday or unusual weather that might explain a difference in a particular year. Finally, ask whether a change would still stand out if you compared each month with the same month in several earlier years rather than only the most recent one.

Seasonal patterns are not a nuisance to be removed from every analysis. They are real and often useful, because they tell you what to expect. The mistake is treating an expected rise as news. Once you know the calendar's contribution, what remains is the part of the change that deserves attention.