From Question to Comparison to Chart

The executive question: what comparison would answer this?

Chart choice looks like a design problem and is really a question-framing problem. Managers rarely need the name of every visualization technique. They need to know what comparison a business question is asking for, because the comparison determines the visual form. Get the comparison right and the chart is nearly automatic. Get it wrong and no amount of polish rescues the page.

The move has three steps, and only the first two are hard:

1. Question. What decision is this chart for? Should we staff Suburban stores differently on weekday mornings?
2. Comparison. What has to be set against what to answer it? Weekday-morning traffic in Suburban stores against the same hours in every other region.
3. Chart. Which visual form makes that comparison legible? A sorted bar of morning traffic by region, or a small-multiple grid of hour-of-day curves.

Most bad charts skip step two. They jump from a vague question to a familiar template, and the template quietly picks a comparison nobody asked for. Figure 1 is the shortest useful version of step three: ten business questions and the chart form each one implies. It is not exhaustive — every analytics team has its own variants — but it is the minimum vocabulary a manager should expect a dashboard designer to have.

The pattern across all ten rows is that the question shape determines the chart shape. Continuous time becomes a line. A fixed sequence of stages becomes a funnel. A categorical ranking becomes a sorted bar. Two snapshots become a slope. There is no entry in Figure 1 whose primary job is decoration; each exists because some specific kind of business question is hard to answer without it. The next section, the chart atlas, expands this table into the full working vocabulary — every form with a live example, the question it serves, and the way it can be misused.

The comparison hides in the baseline

Naming the comparison sounds abstract until you try to build the chart, at which point it becomes concrete and unavoidable: compared with what? The answer to that question is the baseline, and it is the first real decision in any chart — before color, before size, before software.

The Progresso soup case makes this visible, and it will run through the rest of Part II. It is monthly scanner data across more than two thousand stores from June 2001 through December 2006, with Progresso, Campbell's, and private label all in the same category, and store-level census region and ZIP coordinates attached. It is seasonal, competitive, geographic, and — several chapters from now — the case that carries the pricing analysis in Part III.

The business pattern is countercyclical: Progresso's price rises when soup demand weakens. That sentence is easy to say after the fact. The question is how to make it visible without asking the reader to inspect every number. Figure 1's table says a metric-over-time question wants a line chart, but three different lines are candidates here — volume, share, and price — measured in three incompatible units. Indexing all three to a common January baseline puts them on one scale, and the pattern that was buried in the units becomes the shape of the chart.

The useful feature of Figure 2 is not that January is the right baseline. It is that the baseline is explicit, which means the comparison is explicit, which means a reader can argue with it. Each alternative baseline asks a different question of exactly the same data:

• A January baseline asks: how far does the year move away from the winter level?
• A prior-year baseline asks: how different is this year from last year?
• A competitor baseline asks: does Progresso move differently from Campbell's?
• A region baseline asks: which markets are behaving unusually?
• An active-store baseline asks: is this a demand pattern or a panel-coverage artifact?

None of those is more neutral than the others. Baseline choice is a managerial choice because it names the comparison, and naming the comparison is most of the argument. The same chart indexed to June would emphasize winter recovery rather than summer collapse. Indexed to Campbell's, it would become a chart about competitive pricing behavior. An analyst who picks a baseline without saying so has made the manager's argument for her, silently.

Indexes make comparisons legible and levels invisible

An index is the workhorse of comparison because it strips out the thing that usually blocks comparison: incompatible units and incompatible starting points. Twelve metro housing markets starting at wildly different price levels cannot be ranked by growth on a raw price chart — the expensive coastal markets sit at the top no matter what they did. Rebase every market to 100 at a common date and the fastest grower becomes the steepest line, which is what the question actually asked for.

That power is also the trap. An index answers how much did this move? and deliberately discards how much is this worth? A niche product and a flagship both rise from 100 to 140 even when one moves ten times the units. Whenever the decision depends on magnitude — where to spend, what to staff, which market to enter — the indexed view has to be paired with the levels it hid.

Figure 3 does exactly that for the soup case: it drops the index and returns price to dollars, so the seasonal contrast a pricing manager would actually act on stays concrete. Winter is defined here as October through February, the months when soup demand is generally strongest — a business season, not a calendar one.

Figure 3 is intentionally blunt, and intentionally incomplete. It does not prove that price caused lower volume. What it does is create the next question — why does Progresso price look highest when demand is weakest? — and a visualization chapter should stop exactly there. Whether that pattern is causal, strategic, promotional, inventory-driven, or simply confounded by seasonality is a question for Part III, which has the tools to answer it. Part II's job is to make the question unavoidable.

The deeper pathology behind all four is the same: building charts from the data the team has instead of the question the manager has. The discipline that fixes it costs one sentence. Before opening the tool, write down: this chart compares ___ with ___, so that a manager can decide ___. If that sentence cannot be completed, the chart is not ready to be built, and no amount of design will hide it.