A team lead once showed me a slide where revenue looked like it had doubled. The bars marched up the screen, tall and confident, and the room nodded along. Then I looked at the axis. It started at 4.8 million and topped out at 5.1 million. The real change across the year was about six percent. The chart had not lied with a single number, it had lied with a picture, and every person in that meeting believed the picture before they read the labels.

Choosing a chart is not decoration you sprinkle on after the analysis is done. It is the last and most public step of the work you started in Part 12, where you learned to keep correlations honest. A good chart makes the true story obvious in about a second. A bad one, even an honest mistake, aims a whole meeting at the wrong conclusion. This part is about picking the shape that fits your data, and steering clear of the charts that mislead, including the ones people reach for out of habit.

Start with the question, not the chart

The mistake beginners make is opening the chart menu first and picking whatever looks impressive. The order should run the other way. Decide what question you are answering and what you want the reader to see, and the right chart usually narrows down to one or two choices on its own. A chart is a sentence made of shapes, and like a sentence it should say one thing clearly rather than five things at once.

Two ingredients decide the shape. The first is the kind of data in play: whether each column is a category, like country or product, or a number, like revenue or age, and how many of each you are showing at once. The second is the comparison you want to draw. Are you ranking categories against each other, tracing one measure through time, breaking a total into its parts, or asking whether two numbers move together? Answer those, and the chart almost picks itself.

Four questions and the charts that answer them

Most everyday analysis comes down to four questions, and each has a natural chart. If you are comparing categories, reach for a bar chart and sort it so the reader can rank at a glance. If you are showing how something changes over time, use a line chart, because the eye reads the slope as movement. If you are breaking one total into parts, a pie can work when there are only a few slices, though a bar is often clearer. If you are asking whether two numbers relate, use a scatter plot, the same tool you met in Part 12. The table below is the quick reference I keep in my head.

Five questions, five charts. Match the question first and the trap in the last column is the thing to double check.

The same logic fits into a short decision walk. Run your question through it from the top, and you land on a chart without ever opening the full gallery of options your tool offers. The flow below is the exact path I take when I am not sure which chart a question deserves.

flowchart TD
  Q[What is your question?] --> C{Comparing categories?}
  C -->|Yes| BAR[Sorted bar chart]
  C -->|No| T{Change over time?}
  T -->|Yes| LINE[Line chart]
  T -->|No| P{Parts of one whole?}
  P -->|Yes, few parts| PIE[Pie or stacked bar]
  P -->|No| R{Relationship of two numbers?}
  R -->|Yes| SCAT[Scatter plot]
  R -->|No| DIST[Histogram for one number]

What bar, line, pie and scatter are each for

A bar chart is the workhorse, and it is the right answer more often than beginners expect. It compares categories by length, and length is the one visual measure people read accurately without effort. Sort the bars from largest to smallest unless the categories have their own order, such as months, and use horizontal bars when the labels are long. If you find yourself squinting to compare slices of a pie, a bar chart was almost always the better call.

A line chart is built for time. Points joined by a line invite the eye to read the slope between them as change, which is exactly what you want when the horizontal axis is dates. That same slope reading is why you should never join points that are not in a natural order; a line across unordered categories invents a trend that does not exist. Keep the number of lines small, because four or five lines in different colours turn into a plate of spaghetti that nobody can trace. The chart below is a line doing its proper job.

A pie chart shows parts of a whole, and it does one thing well: it tells a reader that a single slice is roughly a half, a third, or a quarter of the total. Past three or four slices it falls apart, because people cannot compare angles nearly as well as they compare lengths. A scatter plot, meanwhile, plots two numbers against each other, one per axis, and is the tool for spotting relationships, clusters and outliers, with the warning from Part 12 still standing: a pattern in the cloud is a lead, never a proof of cause.

The truncated axis, and why bars must start at zero

Here is the trick behind that opening slide, and it is worth understanding in your bones because it is everywhere. A bar chart works by encoding a value as the length of the bar, so the reader compares lengths to compare values. The moment the axis starts somewhere above zero, the lengths stop matching the numbers, and small differences balloon into cliffs. This is called a truncated axis, and it is the most common way a chart misleads while every number on it stays true.

Take four quarters of active users, in thousands, that barely move. The numbers themselves are calm. What changes everything is where you start the axis, as the table and the two charts below make plain.

The whole year moves inside a four percent band. Any excitement in the chart has to come from somewhere other than these numbers.

The rule that falls out of this is simple: bar charts start at zero, always, because the bar length is the message. Line charts are the exception, since a line encodes the position of points rather than a length, so a trimmed axis is acceptable when you are studying small movements, as with a stock price. Even then, label the axis clearly so the reader knows what they are looking at. The real danger is not the honest analyst zooming in, it is the spreadsheet tool that trims the axis for you by default and hopes you do not notice.

When a pie chart stops working

Pies get a bad reputation, and mostly they earn it, but the fault is usually how they are used rather than the shape itself. A pie asks the reader to compare angles and areas, and people are poor at both. Two slices of 18 and 21 percent look identical in a pie; as bars, the difference is obvious. So the pie holds up only when there are two or three slices and the point is a rough part to whole split, and it collapses the moment you crowd in eight categories and a legend the reader has to keep glancing back at.

Two habits make pies worse than they need to be. The first is stuffing in too many slices, which turns the chart into a colour wheel nobody can rank. The second is the three dimensional pie, where the software tilts the disc for looks. That tilt is not harmless decoration, because the slices near the front edge gain apparent area and the ones at the back shrink, so two slices of the same value no longer look the same. If a tool offers you a 3D pie, treat it as a warning label.

Dual axes and other honest looking traps

The most seductive trap is the dual axis chart, where two lines share one plot but each reads against its own vertical scale, one on the left and one on the right. Because you can slide each scale independently, you can make almost any two lines appear to track each other, which manufactures a correlation out of pure axis arithmetic. A viewer sees two lines rising and falling in step and reads a relationship that the data never contained. When you need to compare two measures in different units, two charts stacked one above the other, sharing the same time axis, are almost always the honest choice.

A few smaller traps round out the list. Inconsistent colours across a report make the reader relearn the code on every slide, so fix a colour to a category and keep it. A rainbow of categories with no order forces a hunt through the legend, when sorting and direct labels would do the work. And leaving off the axis labels or the units entirely, which happens more than you would think, quietly shifts the burden of guessing onto the reader. None of these are lies exactly. They are friction, and friction is where honest charts start to mislead by accident.

Match the chart to the question, then check the axis

If you carry two habits out of this part, carry these. First, pick the chart from the question, not the gallery: bar for comparison, line for time, scatter for relationships, and pie only for a few parts of a whole. Second, before you publish anything, look hard at the axis you did not choose on purpose. Most misleading charts are not built by liars, they are built by defaults, and the axis your tool trimmed for you is the first place to look. A chart that answers the right question on an honest axis is worth more than the prettiest picture in the deck.

You now have the judgement to choose a chart on purpose and to catch the ones that mislead. Part 14 puts these choices to work on a real canvas, building your first dashboard in Power BI or Looker Studio, where the charts you just learned to pick get arranged into something a stakeholder reads every morning. Keep the Data Analyst guide open as your map, and revisit Part 12 whenever a scatter tempts you to read cause into a cloud.

This week, take one chart from a recent report and put it on trial. Name the question it answers, decide whether its shape is the right one, and check where its axis starts. If either is off, rebuild it. Do that a handful of times and choosing the right chart stops being a decision and becomes a reflex.

References

• Essential Chart Types for Data Visualization, Atlassian
• Choosing a Chart Type, UC Berkeley Library
• Five ways to improve your chart axes, Nicola Rennie

About The Author

Dr. Pranay Jha

Dr. Pranay Jha is a Cloud and AI Consultant with 18+ years of experience in hybrid cloud, virtualization, and enterprise infrastructure transformation. He specializes in VMware technologies, multi-cloud strategy, and Generative AI solutions. He holds a PhD in Computer Applications with research focused on Cloud and AI, has published multiple research papers, and has been a VMware vExpert since 2016 and a VMUG Community Leader.

See author's posts