Time Series Chart

Why it matters

A time-series chart is the picture of one measured value changing through time: time runs along the horizontal axis, the value rises and falls on the vertical, and the points are joined into a line so the eye can read the change directly. Its whole job is to turn a column of dated numbers — which no one can hold in their head — into a single shape that answers, at a glance, what is happening over time and whether the latest point is unusual. It is the most-used analytical chart in the world for one reason: nothing else carries the trend, the cycles, the noise, and the surprises of a quantity over time in one honest image.

For example: a team watches a dashboard number — weekly signups — and reacts to each week as it lands. One week is down, and an emergency meeting is called. Plotted as a time series across two years, the same number tells a calmer, truer story: a steady upward trend, a repeating dip every December (seasonality, the holidays), and — visible only against that backdrop — one genuine anomaly in March where the line falls off a cliff and never fully recovers. The down week that triggered the meeting was just the December dip. The March cliff, which no single week had flagged, was the thing that actually mattered.

  • What it shows. How one measured value moves over time — its level, its long-run direction, its repeating cycles, and the specific points that break the pattern.
  • When to reach for it. Any quantity recorded repeatedly over time — a price, a metric, a count, a sensor reading — where the question is about change, trajectory, or whether the newest value is normal.
  • How to read it. Read the line left to right: the overall slope is the trend, repeating humps are seasonality, a sudden step is a level shift, and a lone spike or drop is an anomaly worth explaining.
  • What you’d miss without it. The shape of change — a slow drift, a turning point, a seasonal rhythm — that a single number or a snapshot comparison flattens into nothing.
  • Where it misleads. The same data can lie depending on how it is drawn: a y-axis that doesn’t start at zero exaggerates a wiggle into a crisis, and the wrong aggregation (monthly averages over daily data) can erase the very pattern you came to see.

How to read it

Picture two axes. Along the bottom runs time — minutes, days, months, years, left to right — and up the side runs the value being measured, in stated units. Each dated measurement is a point at its time and height, and the points are joined into a line. That line is the whole instrument: its job is to let your eye read change without doing arithmetic. Four things live in its shape. The trend is the long-run direction — is the line climbing, falling, or flat across the whole span? Seasonality is the repeating cycle — the same hump every December, the same dip every weekend, the daily rhythm of business hours. A level shift is a sudden step up or down to a new baseline that holds. And an anomaly is a lone spike or drop that breaks the pattern and usually has a specific cause.

When more than one line shares the axes, the chart becomes a comparison: two or three series rising together or pulling apart, read against each other. Reference lines and shaded bands mark thresholds, targets, or a forecast region; a small annotation pinned to an anomaly — lockdown begins at a vertical drop — turns the chart from a passive display into an explanation.

But the line only tells the truth if it is scaled honestly, and this is where the craft lives. The vertical axis is the first lever: anchored at zero, a 2% rise looks like the 2% rise it is; floated to start just below the data, that same rise fills the frame and reads as a crisis. The second lever is aggregation — the time interval each point represents. Daily data rolled up to monthly averages smooths away the weekend structure; minute-by-minute data stretched across years drowns the signal in noise. The honest time series names its interval, declares its units, starts its axis at zero unless the quantity is genuinely multiplicative, and lets the right grain reveal the pattern rather than hide it. Read well, it answers two questions at once: what is this value doing over time, and is this latest point unusual?

When to use it

The time-series chart belongs to the STATISTICAL family of visual outputs — the charts that show the behaviour of measured quantities — and within that family it is the one whose horizontal axis is always time. That is exactly what marks its boundaries, and knowing them is how you pick the right chart instead of the familiar one:

  • A Distribution Plot answers a different question entirely — the shape of one variable (where the values cluster, how they spread, whether they’re skewed), with no time involved. Reach for it when you care how the numbers are distributed, not how they move.
  • A Scatter Plot shows the relationship between two measured variables — does one rise as the other does? Reach for it when the question is correlation between quantities, not the path of a single quantity over time.
  • A Comparison Chart (the bar chart and its kin) compares values across categories — regions, products, teams — where the x-axis holds labels, not a clock. Reach for it when you’re ranking or contrasting groups, not tracking change.

Reach for a time series whenever a value is measured repeatedly over time and the question is about change, trajectory, cycles, or whether the newest reading is normal — prices, operational metrics, web traffic, sensor streams, case counts, vital signs. Skip it when there is no time dimension: use a distribution plot for the shape of one variable, a scatter plot for how two variables relate, and a comparison chart for differences between categories. When the analyst’s question is “what does this do over time,” the time series is almost always the right place to start; the more specialized temporal views (decomposing trend from season, residual plots) build on it rather than replace it.

How Ora builds it

Ora produces a time series from a semantic spec — a structured description of the time axis (its field and explicit grain: daily, weekly, monthly), one or more measured series, optional layered marks (a rolling-average smoother with a stated window, reference lines for thresholds or targets, event markers and shaded bands for anomalies and forecast regions), and the scaling rules (units, and whether the vertical axis anchors at zero or uses a logarithmic scale). That spec is rendered to a line chart (a matplotlib- or Vega-style line mark over a temporal x-channel), with an enforced caption naming source, period, sample size, and units — and a four-level text description (the period and variable, the level and trend, the seasonal structure, any annotated anomalies) plus keyboard navigation across the points in time order, because a line alone is not accessible.

The diagram is the visual face of Ora’s monitoring and trend-reporting work: when you ask “how has this moved over time — show me the trend, and is the latest point unusual,” the chart is how that analysis shows its answer. The underlying discipline — honest axes, the right aggregation, anomalies named rather than left as mute spikes — is what separates a serviceable line from a truthful one.

The form is the work of William Playfair, who invented the statistical time-series line chart and published it in his 1786 Commercial and Political Atlas — his line of England’s imports and exports across the 18th century is the canonical first example of the genre, drawn before anyone had agreed that a quantity over time should be a line. The contemporary rendering discipline — anchoring scales at zero for non-multiplicative quantities, treating the line’s ink as something to spend carefully — is owed in large part to Edward Tufte’s The Visual Display of Quantitative Information, whose work on data-ink and on the sparkline reframes the time series as the smallest honest unit of analytical display.

  • Distribution Plot — the STATISTICAL family member for the shape of one variable: where its values cluster and how they spread, with no time axis.
  • Scatter Plot — the family member for the relationship between two measured variables, answering correlation rather than change over time.
  • Heatmap — extends a time series to a second dimension, showing the same value across time and a category as a coloured 2D field.
  • Comparison Chart — the bar chart and its kin: differences across categories rather than movement through time.