A time series is the most natural form of data. Something happened, then something else happened. The horizontal axis is time. The vertical axis is quantity. The line connects one moment to the next. It is the oldest form of statistical graphic, and it remains the most useful.
It is also the most frequently ruined.
The Raw Signal
Daily website traffic is noisy. Daily revenue is noisy. (As I discuss in Choosing the Right Chart, the first question is always what comparison the reader needs to make.) Daily temperature is noisy. (For how to handle this noise in compact form, see Sparklines: The Quiet Revolution.) The noise is real. It reflects the actual variability of the phenomenon being measured. A time series chart that shows this noise honestly is showing the truth.
But truth is uncomfortable. Executives do not want to see jagged lines. They want trends. They want to know if things are going up or down. And so the analyst reaches for the smoothing function.
Moving averages. LOESS curves. Exponential smoothing. Savitzky-Golay filters. Each of these replaces the raw data with an approximation that suppresses short-term variation in favor of long-term direction. This is sometimes appropriate. It is never neutral.
Every smoothing function is an editorial decision. It says: this variation does not matter. The reader deserves to know what was removed.
What Smoothing Hides
A seven-day moving average of daily sales will erase the weekend effect. If the business is retail, the weekend effect is not noise. It is the most important signal in the data. A thirty-day moving average will erase monthly billing cycles, promotional spikes, and the effects of holidays. A ninety-day moving average will erase entire product launches.
The analyst who applies a moving average and shows only the smoothed line has made a decision about what matters. That decision is invisible to the reader. The reader sees a gentle upward curve and concludes that growth is steady. The raw data might tell a different story: growth driven entirely by two promotional periods, with flat or declining baseline performance.
This is not dishonesty in the conventional sense. The analyst may genuinely believe the trend is the story. But suppressing the evidence that might contradict the trend is a form of selection bias, performed visually.
The Dual Display
The solution is not to avoid smoothing. It is to show both. The raw data in light grey. The smoothed trend in a darker, more prominent line. The reader sees the trend and the evidence simultaneously. They can evaluate whether the smoothing is appropriate or whether it is hiding something important.
A seven-day moving average would report this week as ~2,500 daily sessions, steady. It would erase the fact that weekends account for less than a third of weekday traffic. For a business deciding when to run campaigns, that distinction is everything.
Aspect Ratio
William Cleveland demonstrated in 1988 that the aspect ratio of a time series chart directly affects the reader's ability to judge the rate of change. When the chart is too wide, slopes appear shallow. When too tall, slopes appear steep. The principle he established — banking to 45 degrees — recommends choosing an aspect ratio such that the average absolute slope of the line segments is approximately 45 degrees.
This is almost never done. Most time series charts are dictated by the width of the dashboard panel or the PowerPoint slide. The data is stretched or compressed to fit the container. The resulting slopes are artifacts of the layout, not of the data.
A chart that shows a dramatic decline might, at a different aspect ratio, reveal a gradual drift. A chart that shows stability might, at the proper aspect ratio, reveal oscillation. The aspect ratio is not a cosmetic choice. It is a perceptual variable that encodes information about rate of change.
Annotation Over Decoration
The most neglected element of time series visualization is the annotation. A line rises sharply in March. Why? The chart does not say. The reader must consult a separate document, ask a colleague, or guess.
The best time series charts annotate directly on the graphic. A vertical reference line at the point of a product launch. A shaded region during a promotional period. A text label at the moment a competitor entered the market. These annotations transform the chart from a picture of data into an explanation of data.
Edward Tufte's principle applies: the chart should be self-sufficient. A reader encountering it without any surrounding text should be able to understand what happened and, ideally, why.
Aggregation Windows
The choice of aggregation window — hourly, daily, weekly, monthly, quarterly — determines what the reader can see. Hourly data reveals server outages. Daily data reveals day-of-week patterns. Weekly data reveals seasonal trends. Monthly data reveals year-over-year growth. Quarterly data reveals almost nothing.
The analyst who shows only quarterly data has decided that short-term variation is irrelevant. This may be true for a board presentation. It is almost never true for an operational review. Yet the quarterly view dominates corporate reporting, because it produces the smoothest, most reassuring lines.
The antidote is small multiples: the same data at multiple aggregation windows, displayed side by side. The reader sees the forest and the trees. Nothing is hidden. Nothing is assumed on their behalf.