Data Visualization

A chart turns values into visible properties: where a point sits, how long a bar is, or how dark a cell looks. It helps when a comparison, pattern, or exception is easier to see than to find in a table. Begin by writing the question in one sentence.

Name the task

“Which of our six warehouses shipped the most orders in March?” asks for a comparison across categories. “How did orders change each month?” asks for a trend. Other tasks ask about a whole and its parts, the spread of individual values, a relationship between measures, or a pattern across places. The question determines what the reader must compare.

Next, inspect the fields. A field is a column of information, such as warehouse or delivery time.

A numeric-looking identifier is still a label. Adding customer numbers does not answer a useful question. Conversely, “March” is a period in a time sequence, even when its name is stored as text.

Decide what one mark means

A mark is a visible item, such as a point or bar. Decide its meaning before building. One source row might represent an order, while one bar represents a warehouse's total orders. Summing quantity answers “How many items?” Counting order IDs answers “How many orders?” Averaging quantity answers “How large was a typical order?” These calculations can give different rankings from the same rows.

Write a small chart plan: the question, the fields, the calculation, and the meaning of one mark. For the warehouse question, it might be “Warehouse groups the rows; count orders in March; each bar is one warehouse.” This catches confusion before a polished chart makes it harder to notice.

Choose a useful first display

One category plus one measure often suggests bars. Time plus one measure suggests a line. One measure across individual records suggests a histogram or box plot. Two measures suggest a scatter plot. A place field suggests a map only when location helps answer the question.

For “How did monthly website visits change during 2025, and what happened around the March campaign?”, put month from left to right and visits vertically. Join the monthly values with a line and annotate March. The chart shows what happened around the campaign; testing whether the campaign caused the change needs an appropriate comparison or further analysis.

Match the display to the reader's job

Exploration helps you discover patterns. You may try several rough charts, filters, and groupings. Explanation helps an audience understand a finding. Select a focused display, title it clearly, and include the context the reader needs.

Sometimes the best display is a table. A stock manager checking exact quantities for 40 products needs reliable lookup. A KPI card, or scorecard, suits one headline value: “Orders this week: 1,284”, with a useful comparison such as “up 6% on last week”. A chart menu's first suggestion is a starting point; judge it by how quickly and correctly the reader can answer the question.

A bar chart shows categories through bar length. Put the categories on one axis and their values on the other. Horizontal bars leave room for long names; vertical bars, often called columns, suit a few short labels.

Make comparisons easy

Start the value axis at zero. A bar's length represents its amount, so cutting off the bottom changes the apparent ratio. Sort unordered categories by value when ranking is the task. Keep a natural order for satisfaction levels, age groups, or dates.

A support team received 1,000 tickets: email 420, chat 310, phone 180, and social media 90. Draw horizontal bars in that order and label their ends. Email represents 42% of tickets and more than twice the phone total. Length and aligned endpoints make those differences easy to see.

A dot plot places a point at each category's value. Readers compare position, so a clearly labeled dot-plot axis can begin above zero. It is useful when differences between close values matter and a full zero-based bar chart would make them hard to see.

Add a second category

Grouped bars are also called clustered or side-by-side bars. With two or three series, each pair or group remains easy to scan. With more, consider small charts or a display of the differences.

In a stacked bar, the total and bottom segment are easy to compare. Upper segments begin on shifting baselines, so comparing their thickness takes more effort. Choose grouped bars when precise comparison of every component is the main task.

Check that there really is a whole

A pie or donut divides one total into slices. It can communicate a rough share with a few parts, often about two to five. For shares of 42%, 27%, 18%, 9%, and 4%, a pie makes “no company has half” visible. A sorted bar chart makes close comparisons and ranking easier because people judge length more accurately than angle and area.

Survey answers can overlap. If people may select several services, percentages of 64%, 51%, and 33% total 148%. They each use all respondents as the denominator; they are not separate pieces of one whole. Show bars. For the composition of several distinct wholes, use 100% stacked bars and provide group totals if size matters.

Use specialized displays for their specific tasks

A treemap divides a whole into nested rectangles. It fits many categories arranged in a hierarchy, such as department and product, but areas support only rough comparisons. A waterfall connects an opening amount to a closing amount through increases and decreases. Opening stock, deliveries, sales, and closing stock form such a bridge. A waterfall's floating change bars have a different role from ordinary category bars: each shows a change from the preceding level.

A line chart connects values in time order. Put earlier dates on the left and later dates on the right. The shape reveals increases, falls, turning points, and repeated seasonal patterns. Columns can be better for a few separate period totals, such as four quarters.

Let distance mean elapsed time

Consider values of 100 in 2010, 120 in 2015, 140 in 2020, 150 in 2022, 160 in 2023, and 170 in 2024. The gaps are five, five, two, one, and one years. Equal spacing would make each gap look equally long and distort the apparent pace.

From 2010 to 2020, the increase is 40 over ten years, or 4 per year. From 2020 to 2024, it is 30 over four years, or 7.5 per year. A continuous date axis shows the faster later growth. If a display uses categories for irregular dates, make the uneven gaps explicit.

Choose a period that fits the question

The time grain is the size of each period: day, week, month, or year. Daily values can expose a short disruption but also contain noise. Monthly values reveal a broader trend. Annual totals can hide recurring busy and quiet seasons.

A 7-day rolling average uses each day and the six days before it. It smooths day-to-day variation while updating daily. Label the line as an average; it no longer represents the raw daily count. To examine seasonal change, compare a month with the same month a year earlier.

Treat incomplete periods visibly. If records stop on October 9, October contains nine days while September contains a whole month. Exclude the partial month, label it “October through October 9”, or compare equivalent portions of months. A full-height month axis label by itself hides the unequal coverage.

Read the vertical scale and the series

A line compares positions and slopes, so its vertical axis can begin above zero. Use a range that makes meaningful change readable without turning ordinary noise into dramatic swings. A line across unordered product names suggests a sequence that the categories do not have; use bars instead.

With about four or five series, direct labels at the line ends can work well. For many series, highlight the one that matters and mute the rest, or use small multiples: one chart per series with identical axes. Shared scales let readers compare levels as well as shapes.

Choose related time displays carefully

An area chart fills below a line to emphasize volume. Stacked areas show a changing total and its components, but the upper layers sit on moving baselines. For the trend of one component, use a separate line or place it at the bottom. A slope chart joins two time points for several categories, making the direction of each change visible.

Describe change with a calculation that matches the claim. Percent change uses the old value as the base:

The rise from 120 to 150 orders is 25%. Dividing by 150 would use the ending value and answer a different question.

An average can hide very different experiences. Deliveries that all take about 25 minutes and deliveries split between very short and very long times can have the same mean. A distribution shows how often values or ranges occur, revealing center, spread, shape, and unusual observations.

Build and read a histogram

A histogram divides a numeric scale into bins and counts observations in each. Use equal-width bins with touching bars. Unlike a category bar chart, the horizontal positions are numeric intervals, so their order is fixed.

For 50 deliveries, 10-minute bins have counts of 3, 14, 19, 9, 3, and 2, from 0–10 through 50–60 minutes. The tallest bar is 20–30 minutes. Most deliveries lie from 10 to 40 minutes, with a tail toward slower times. State the boundary rule: for example, include 20 in the interval starting at 20 and ending just before 30. Every observation belongs to one bin.

Try several sensible widths. Very wide bins conceal peaks; very narrow bins make the picture jagged. Unequal widths with count as height are misleading because wider bins have more room to collect observations.

A symmetric distribution has roughly matching sides. A right-skewed one has a long right tail; large values tend to pull its mean above its median. Left skew has a long left tail. Bimodal means two peaks, which can suggest mixed groups. An outlier lies unusually far from the main body of observations; investigate it before deciding what to do with it.

Read a box plot

A box spans the first quartile, Q1, to the third quartile, Q3. These mark the middle half of the data. A line inside shows the median. The interquartile range, or IQR, is Q3 minus Q1.

A five-number summary gives minimum, Q1, median, Q3, and maximum. A common box-plot convention draws whiskers only to the most extreme observed values within 1.5 IQR of the box. More distant values appear as separate points. In that version, the whiskers need not reach the minimum and maximum.

Work through the whiskers

Nine waiting times are 4, 6, 7, 8, 9, 11, 12, 14, and 30 minutes. The median is 9. Excluding that middle value, the lower half gives Q1 of 6.5 and the upper half gives Q3 of 13.

The value 30 exceeds the fence, so it appears as an outlier point; the upper whisker ends at 14, the largest observation inside the fence. The fence itself is a calculated boundary, not the whisker's endpoint. Quartile conventions differ: an inclusive method can give Q1 of 7 and Q3 of 12 for these same data. Check which convention the tool uses.

Choose shape or compact comparison

Side-by-side boxes compare many groups efficiently. Histograms and strip plots, which show one dot per observation, reveal detail that boxes hide. Two peaks can disappear inside an ordinary-looking box. For an audience new to box plots, explain the box and median in a short note. A bar chart of group averages cannot answer how much the individual values vary.

A scatter plot places each item using two numeric measures. One point might be a store, with floor area horizontally and monthly sales vertically. Read direction, form, strength, and unusual points before describing the pattern.

Read a relationship without overstating it

An upward pattern is positive: larger horizontal values tend to go with larger vertical values. A downward pattern is negative. Form may be straight or curved. Strength describes how closely points follow that form. Also inspect clusters and points far from the rest.

For six days, temperatures of 14, 17, 20, 23, 26, and 29 °C went with ice cream sales of 310, 380, 470, 520, 610, and 690 liters. The pattern is strongly positive and nearly straight; its correlation coefficient is about 0.998. A title such as “Sales were higher on warmer days” describes the observed association.

A chart alone cannot establish a cause. Summer conditions can raise both ice cream sales and swimming accidents without either causing the other. A fitted trend line summarizes a relationship within the observed range. Extending it to 40 °C would go beyond these observations and could give an unreliable prediction.

Handle more information carefully

Thousands of overlapping markers create overplotting. A dense patch may look no different from a moderately busy one. Reduce marker size, add transparency, or show a grid colored by counts.

A bubble chart adds a third quantity through point area. The area must be proportional to the quantity; proportional radii exaggerate differences. Bubble sizes support rough comparison, so keep the most important comparison on an axis.

A heat map colors a grid. Weekday as rows, hour as columns, and order count as color can reveal a Friday evening peak. Use a sequential palette for low-to-high counts. Add values in cells when exact lookup matters, because readers estimate color only roughly.

Make location earn its place

A filled map, or choropleth, shades regions. It is useful when geography matters, such as neighboring areas sharing a pattern. For ranking regions alone, sorted bars are often easier to read. A large region occupies more visual space even if few people live there.

Use a comparable rate for a filled map of outcomes across differently sized populations. Region A has 1,200 cases among 400,000 people; region B has 900 among 150,000.

A has more cases, but B has twice as many per 100,000 people. The denominator changes the interpretation. To show where the largest totals occur, proportional symbols can represent the counts without treating the region's filled area as the amount.

Geocoding turns place fields into map locations. “Paris” alone is ambiguous because the name occurs in France and Texas. Supply a country or state and inspect where the points land. A correct number in the wrong place still produces a wrong map.

Every chart maps data into something visible. Marks are its shapes: points, lines, bars, or areas. Channels are the properties that vary: position, length, angle, area, hue, lightness, saturation, shape, or size. This mapping is called visual encoding; reading the chart reverses it.

Identify what carries each field

In a scatter plot of stores, each point is a store. Its horizontal position could show floor area in square meters, its vertical position monthly sales, its hue the region, and its size the employee count. A reader must learn four mappings to understand the complete point. Labels and legends identify those mappings.

Adding a field to a chart is also adding a reading task. Position, color, shape, and size together can become hard to follow. Keep the main question visible and split the display if the extra channels compete with it.

Use the most accurate channel for the main comparison

Cleveland and McGill's 1984 work proposed a useful ordering for judging quantities. Some comparisons were tested directly, and later studies broadly supported the guidance. The general order runs from position on a common scale, through position on separate scales, then length, direction and angle, then area, then volume and curvature, and finally shading and saturation.

This is a guide to design, not a guarantee about every reader. Its practical lesson is direct: two aligned endpoints are easier to compare closely than two areas or shades. Bars and dots therefore work well for ranking close quantities. Pie angles and treemap areas are more suitable for broad impressions.

A stacked bar has mixed accuracy within the same picture. Its bottom segment starts at zero in each group, but an upper segment begins where the preceding segments end. To compare those upper components, the reader must mentally subtract the lower layers.

Match channels to field types

A triangle does not inherently mean more than a circle. Likewise, the colors of a rainbow have no dependable low-to-high order. A hue legend can name categories, but it makes a poor numeric scale unless the palette also provides a clear ordered progression.

Work backward from the comparison

If the question is “How did each region change from 2024 to 2025?”, place the two years together within each region. The reader can compare nearby bar endpoints. If the question is “Which region led in 2025?”, show the 2025 values along one shared axis and sort them.

For circles, people often underestimate area differences. A circle with twice the area may not look twice as large. Keep area for rough magnitude, and use position when a small difference drives a decision. Coloring every already-labeled bar a different hue adds decoding effort without adding a field; one neutral color may carry the same information more clearly.

A dark blue line among gray lines is noticed before a reader consciously searches for it. Such quickly noticed properties are called preattentive attributes. Color, size, position, orientation, shape, line width, and enclosure can all guide attention.

Emphasize with a clear purpose

Suppose eight product lines appear on one sales chart, and product C is declining. Draw C in dark, saturated blue and put its name beside the line. Use light gray for the seven context series. The reader can find C quickly while still judging its position among the others.

Contrast creates the emphasis. If all eight lines are bright, thick, and covered with arrows, the reader has eight competing starting points. Usually one or two emphasized elements are enough. A bold value, an annotation, or a carefully chosen position can also guide attention when color is unavailable.

Highlighting makes items easier to find; it does not perform every mental task for the reader. Blue 7s in a block of gray digits stand out, but counting a large number of them still takes attention.

Group elements so their relationships are visible

Gestalt principles describe how people organize what they see. These principles are useful because spacing and appearance can carry meaning even without an explicit label.

Connection can communicate a stronger relationship than simple closeness or matching color. A line joining product categories therefore suggests a progression even when none exists.

Reduce the distance between a label and its data

A line chart with four series and a distant legend forces repeated trips across the page: find a color in the legend, return to the data, and check the match. Write each name at its line's end in the same color. Proximity connects the name with the line; similarity reinforces the connection.

Place annotations beside the points they explain. Put the title and key message near the top left for an audience reading left to right. Leave enough space between unrelated elements to prevent accidental grouping.

Check the groups you created

Two unrelated categories with the same color can look related. One category shown in different colors across charts can look like different things. A shaded interval can suggest a meaningful period even if the shading is decorative. Inspect each gap, repeated appearance, border, and connection for the relationship it communicates.

Clutter is anything that takes attention without helping someone understand the data. A chart can have correct numbers and still be difficult to read because heavy lines, repeated labels, and decoration compete with its message.

Remove elements in a useful order

Edward Tufte's data-ink ratio describes the share of a graphic's ink that represents data. Removing ink that carries no useful information can make the data easier to see. He called unhelpful decoration chartjunk: background pictures, heavy patterns, shadows, and 3D effects are common examples.

Start with the frame and background. Remove the outer border if the axes already define the plotting area. Drop shadows, gradients, and 3D. Then lighten the gridlines or remove those that the reader does not need. A light grid can still support estimation, so judge it by its purpose.

Next, examine the legend. A chart with one series labeled “Orders” usually does not need a separate legend repeating that word. With several lines, direct labels can replace a distant color key. Keep a legend when it helps readers decode the chart.

Work through a default chart

Imagine a monthly orders chart for 2025. It has dark gridlines every 50 orders, the legend “Orders”, labels on every column, a value axis, and diagonal labels such as “January 2025”.

1. Put “2025” in the title and shorten month labels to “Jan” through “Dec”.
2. Keep month labels horizontal so readers can scan them without turning their heads.
3. Delete the one-series legend.
4. Decide whether exact values or the overall pattern is the main need.
5. For a few exact comparisons, keep the values on the columns and remove the redundant value axis. For a trend with many points, retain a light axis and label selected points.

Each change removes a particular reading burden. The year still appears once, the measure still has a name, and the values still have a way to be read.

Simplify the labels without changing their meaning

Use round tick steps, such as 0, 100, 200, and 300. Drop decimal places that have no useful meaning: “40%” is easier to scan than “40.00%” when the extra precision adds nothing. Use the same precision for comparable values. Preserve the unit and the zero baseline for ordinary bars.

Long category names often work better on horizontal bars. They can then remain horizontal and close to their data. Angled or vertical text saves room at the cost of slower reading.

Leave space for structure

Whitespace separates elements and makes groups visible. Align titles, subtitles, and notes to a consistent edge. Keep labels clear of marks and leave a useful gap between the plotting area and the source note. When charts appear together, align their plots so readers can compare them.

Decluttering has a limit: a necessary unit, target line, source, or scale carries information. Removing an axis from a chart with no data labels leaves values unreadable. The aim is to reduce effort while keeping the information needed to interpret the display.

Titles tell readers what to look for. Labels explain what the values mean. Notes explain the boundaries of the comparison. Write these parts as carefully as the chart itself.

Choose a title that fits the purpose

A descriptive title names the display: “Monthly orders, 2025”. A headline title, also called an action title, states a finding: “Orders fell 18% after the June price change”. Explanatory charts usually benefit from a headline, with a subtitle giving the measure, unit, period, and group.

Verify the number before putting it in a title. Orders fell from 1,250 in June to 1,025 in July:

The word “after” describes timing. Saying that the price change caused the fall would require more evidence than these two observations provide.

Make units and precision visible

Use axis titles such as “Delivery time (minutes)”, “Energy use (kWh)”, or “Revenue (thousands)”. Month tick labels may already explain a time axis, but an unlabeled numeric scale could mean several different quantities.

Format for reading. A stored value of 12,483.27 might appear as “12,483” or “12.5 thousand”, depending on the decision and required precision. Format a share of 0.45 as 45%. Keep decimal places consistent within a comparison. If readers may not know K and M, write “thousand” and “million” or define the abbreviations.

Prevent date and number confusion

Different locales use different punctuation for the same value.

The date “03/05/2026” is March 5 in the United States and May 3 in many other countries. Write “5 Mar 2026” or “2026-03-05” when readers may use different conventions. Set the report locale intentionally, and check exported labels as well as the interactive display.

Put explanations where they are needed

An annotation is a short note near the data it explains. “Price change, June 1” identifies an event. “Target: 95% on time” explains a reference line. “Warehouse closed for stock count” can identify an unusual observation when that event is known.

Keep annotations short enough to read beside the point. They should add a relevant fact, not repeat the plotted value without purpose. If several series need a legend, keep it close, use the same names throughout, and order its entries like the plotted series or stacked segments.

Identify the source and the coverage

A source note might say, “Source: order system, extracted 2026-01-05. Canceled orders excluded.” The source makes checking possible; the extraction date shows freshness; the exclusion defines what was counted. Put active filters in a visible subtitle so they remain clear in a screenshot or printed copy. A chart title about all orders becomes misleading if only online orders are included.

Before polishing a chart, identify the reader and the decision. An operations manager deciding where to assign staff needs a different view from an analyst investigating the source of delays. Detail is useful only when it supports that reader's task.

Give the chart a clear job

Ask who will read it, what they already know, and what they should understand or decide afterward. An explanatory chart works best with one main message. A fall in total sales and a shift in the product mix can require two displays because they call for different comparisons.

Sketch the arrangement before building. A rough bar chart or line chart is enough to test which values need to be near each other. If the main comparison is hard to show in a sketch, adding formatting will rarely solve it.

Adapt the detail to the medium

A tooltip can carry extra detail in an interactive display, but it disappears in print and may be hard to use on a phone. Keep the finding, units, and necessary context visible without hovering. Check black-and-white output if the report may be printed that way.

Make comparisons consistent

Keep the same color for the same category. Use the same number formats and definitions. For charts that readers compare side by side, use matching axis scales: a value of 500 and one of 5,000 must not both look equally tall because each chart independently fills its panel.

Small multiples repeat one chart for each category. Shared axes and a consistent layout let the reader compare levels and shapes. They can replace a tangled chart with many overlapping lines. If scales differ, a shape comparison can easily be mistaken for a comparison of absolute size.

Build different views of the same situation

Suppose monthly complaints are available for five stores. A regional director needs to notice the store whose complaints doubled: show a line chart with that store emphasized and a factual headline. Store managers need to compare their own patterns: small multiples with identical axes support that task. An analyst checking exact monthly counts may need a table.

The underlying observations can stay the same while the amount of detail, arrangement, and emphasis change. Avoid hiding information that the selected audience needs to make its decision.

Test the first impression

Show the chart briefly to a colleague. Ask what stood out first and what conclusion they drew. If they notice a bright context series instead of the intended finding, change the emphasis. If they misunderstand the population, clarify the subtitle or metric definition.

Before sharing, check the question, chart type, calculations, aggregation, scales, labels, units, clutter, accessibility, and overall impression. Test at the size and in the format readers will actually use.

Color can identify a group, show an ordered amount, or emphasize a finding. Decide which job it must do before selecting the palette. A color's hue is its family, such as blue or orange. Lightness runs from light to dark. Saturation runs from muted gray toward vivid color.

Match the palette to the data

A categorical palette might identify four sales channels. It should let readers tell the groups apart without suggesting that one channel is a larger quantity simply because its color is darker. About six to eight hues is a practical limit for many displays. With more categories, highlight selected ones or combine small categories as “Other” when that grouping suits the question.

Make low and high visible

For unemployment rates from 3% to 12%, a sequential scale can run from light blue to dark blue in five steps. On a light background, darker usually signals more. A carefully designed multihue palette, such as viridis, can also provide an ordered progression through changing lightness.

A rainbow is a poor quantity scale. Hue has no natural numeric order, and sudden lightness changes can create boundaries that the data do not contain. A viewer may see a dramatic break between yellow and blue where the underlying values change smoothly.

Use a center only when it means something

Population change from −5% to +5% has a meaningful center at 0%. A diverging scale could use orange for decline, a light neutral center for no change, and blue for growth. It lets readers see both direction and distance from the center.

A numeric midpoint chosen only because it is halfway between the minimum and maximum has no such meaning. For values that simply run from small to large, a sequential palette is usually clearer. Above-target and below-target results can use a diverging scale when the target is the meaningful dividing value.

Use emphasis consistently

One accent against muted context directs attention. A brand color can serve as that accent; surrounding every mark with equally vivid brand colors weakens the focus. Save useful fonts and palettes as a theme so related charts keep their visual rules.

Keep category meanings stable: if “online” is blue, use blue for it throughout the report. Color associations also vary across cultures. Red often means loss in Western financial charts, while red commonly marks rising stock prices in mainland China. Labels and a clear legend make the intended meaning visible instead of relying on an assumed association.

Check the finished palette in its actual setting. A sequence that looks distinct on a bright screen may become hard to read on a washed-out projector, and color distinctions still need accessible labels or other cues.

Accessible charts let people obtain the same information through different ways of seeing or reading. People may have color vision deficiency, low vision, or use a screen reader. Red–green color vision deficiency is the most common type; color vision deficiency affects about 1 in 12 men and is much less common in women.

Give meaning a second cue

Do not make color the only way to identify a series or status. Use direct labels, different marker shapes, solid and dashed lines, or patterns. Actual sales could use a solid line and the target a dashed line, with both names at their ends. Their identities then survive grayscale printing.

Blue and orange can be easier to distinguish than red and green. The Okabe–Ito palette includes orange #E69F00, sky blue #56B4E9, bluish green #009E73, and blue #0072B2. A palette designed for color vision differences still needs a check of the finished chart, including overlapping marks, background, and labels.

Check contrast for its actual role

Contrast compares the relative luminance of adjacent colors:

The scale runs from 1:1 for identical luminance to 21:1 for black against white. WCAG AA requires at least 4.5:1 for normal text and 3:1 for large text, defined as at least 18 point or 14 point bold. Essential graphic parts need 3:1 against adjacent colors unless equivalent visible information removes reliance on those parts.

On white, #767676 gray text reaches about 4.54:1 and passes for normal text. Light gray #AAAAAA reaches about 2.32:1, too low for essential text or a line relied on to read the data. Orange #E69F00 reaches about 2.25:1; add equivalent visible values or choose a darker line such as #D55E00, about 3.87:1. That darker color clears the graphic threshold but still falls below the normal-text threshold.

Write a useful text alternative

Alt text describes the chart for a screen reader. Include its type, subject, and main finding. For example: “Line chart of monthly orders in 2025. Orders rose from about 1,000 in January to 1,250 in June, fell 18% to 1,025 in July after the price change, and stayed near that level through December.” Describe the information; a list of colors alone does not convey the finding.

For a complex chart, add a longer description or an accessible table. Power BI offers “Show as a table”, and Tableau offers “View Data”. In Power BI, alt text is available in the visual's format settings under General.

Test the final reading experience

Keep labels horizontal and readable at final size. Avoid faint gray for important text. Check at 200% zoom and inspect whether labels overlap or disappear. Use grayscale, a color vision deficiency simulator, and a contrast checker. A chart needs both visible distinctions and a way to reach its actual values.

A chart can exaggerate a small change without changing a single data value. Inspect the scale, spacing, and shape before accepting its first impression.

Measure a truncated bar's distortion

Two branches average 48 and 52 orders per day. With an axis starting at 45, their drawn lengths are 3 and 7. The larger bar looks about 2.33 times as long, while the actual value is only about 1.08 times as large.

Tufte's lie factor compares relative change in the graphic with relative change in the data:

The data change is 4 divided by 48; the drawn-length change is 4 divided by 3. Their ratio is 16. A value close to 1 indicates faithful effect size; Tufte treated values above 1.05 or below 0.95 as substantial distortion. Fix the bars by starting at zero, or use a labeled dot plot when the small difference is the point.

Use line ranges with purpose

A line's vertical axis can begin above zero. A change from 36.8 °C to 38.6 °C is a meaningful rise of 1.8 °C; starting the axis at 0 °C would make it hard to notice. At the same time, an extremely tight range can make ordinary variation appear dramatic. Show a range appropriate to the question and label it visibly.

Aspect ratio matters too. Stretching a plot vertically steepens slopes; flattening it reduces their apparent steepness. Use a moderate shape and keep it consistent across charts that readers compare.

Inspect intervals and separate scales

On an ordinary linear axis, equal distances must represent equal differences. Ticks of 0, 10, 50, and 100 cannot be evenly spaced. A date axis must preserve elapsed time as well. Side-by-side plots need shared scales when their heights will be compared.

A dual-axis display gives two measures separate value scales. Changing those ranges can make the lines cross or look parallel without changing their values. Use aligned charts with a shared time axis, or compare relative growth by indexing each series to 100 at a meaningful start date.

Visits rising from 20,000 to 25,000 have an index of 125. Sales rising from 800 to 880 reach 110. The common index shows growth of 25% and 10%; it does not compare the original totals.

Recognize special scales

An inverted axis increases downward. That can suit a strong convention such as depth below a surface, but label it clearly because a falling line otherwise suggests a falling value.

On a logarithmic scale, equal distances mean equal ratios. Values of 10, 100, 1,000, and 10,000 can be equally spaced because each is ten times the previous one. This suits values spanning several orders of magnitude or steady percentage growth. State “log scale” in the axis title. A log scale has no zero baseline, so use points or lines instead of ordinary bars.

A picture can distort quantities through its geometry. When the quantity is encoded by area, compare the areas, not just the widths or radii. Flat charts also avoid perspective effects that make equal values look different.

Check how much the picture grows

Doubling both width and height makes a rectangle's area four times as large. If the data merely double, that enlarged icon shows a 300% area increase for a 100% data increase. To represent twice the amount through area, enlarge both dimensions by the square root of 2.

For bubbles showing 10 and 40, the larger value is four times the smaller. Setting radius proportional to value gives a radius ratio of 4 and an area ratio of 16. An honest area encoding needs a radius ratio of 2:

BI tools often scale by area, but inspect the setting. The label “Size” alone does not explain the geometry. For precise comparisons, position or length is easier to judge than area even when the area calculation is correct.

Remove perspective effects

A tilted 3D pie makes front slices appear larger than equal back slices. A 3D bar makes the relevant edge harder to align with a gridline. Use flat 2D forms when the data have two dimensions. Depth and shadows can look impressive while making values harder to read.

A pictogram can still be honest. If one person icon means 1,000 customers, repeat identical icons to show several thousand. Keep the unit icon's size fixed; stretching a single icon turns counting into an area comparison.

Verify the whole and the moving baseline

Pie slices and a 100% stacked bar must divide one meaningful whole. Separate survey questions and overlapping selections do not become parts simply because their values are percentages. Check the underlying categories and denominators before choosing the shape.

Displayed percentages of 99% or 101% can result from rounding parts of a valid whole. A short rounding note can explain that small discrepancy. Values totaling 148% from overlapping responses are a different issue: the categories do not partition the respondents, so a pie cannot represent them correctly.

In stacked areas, a layer's top edge includes the layers below it. It can rise even when that component stays constant. To read that component, inspect its thickness; for a clear trend comparison, use its own line or move it to the bottom. Return to a sorted bar or dot plot when decorative size, perspective, or moving baselines hide the comparison the reader needs.

Honest geometry cannot repair a misleading choice of data. Check the period, population, calculation, missing values, and comparison behind the marks.

Inspect the window and denominator

Cherry-picking means choosing a period or group because it supports a desired message. Starting a sales chart at an unusually bad month can make a recovery look like exceptional growth. Show the relevant period or explain why a shorter window answers the question.

Counts also depend on group size. A large city can have more incidents while having fewer per resident. For differently sized groups, compare rates such as cases per 100,000 people, orders per 1,000 visitors, or percentages. Name the denominator so readers know what is being compared.

Separate running totals from new activity

Monthly sign-ups of 120, 110, 95, and 80 produce cumulative totals of 120, 230, 325, and 405. The running-total line rises while monthly activity falls. With nonnegative additions, a cumulative total can only rise or stay level. Its upward direction alone says little about whether current performance is improving.

For “How many have joined so far?”, use the cumulative total. For “Are more people joining each month?”, show the monthly values. Label the series so readers know which question it answers.

Look inside a pooled percentage

Aggregation can hide or reverse group patterns, called Simpson's paradox. Consider two web page versions:

B has the higher conversion rate on each device. Yet most of B's visitors use mobile, the lower-converting device group, while most of A's use desktop. The overall rate combines group performance with group mix. Break the chart down by device before drawing a conclusion about the versions.

Show missingness and sample size

A missing month means the value is unknown. A zero means an observed absence. Draw a gap and add a note such as “No data: system outage” when records are missing. A line dropping to zero would make a claim the records do not support.

A percentage based on a small sample can swing sharply when one response changes. “80% satisfied” may mean only 4 of 5 people. Show the count or sample size alongside the percentage so the reader can judge its basis.

Restore the context needed for judgment

A current value may need a target, an earlier period, or a comparison group to mean anything useful. Check who produced the chart, its source, the exact measure and unit, its scales, included dates and groups, rates or counts, and the title's claim. Misleading choices can come from defaults, haste, or deliberate intent.

A business intelligence tool, or BI tool, connects to tables and turns fields into charts and reports. Tableau, Power BI, and Google Data Studio (formerly Looker Studio) share the same basic workflow: connect, inspect fields, choose calculations, build, format, filter, check, and share.

Power BI Desktop is free for Windows, with the Power BI service used for sharing. Tableau Public offers a free route for public work, and Data Studio is a no-cost browser-based reporting tool.

Load a table you can interpret

Begin with one header row, one row per record, and one column per variable. Merged cells, blank spacer rows, and embedded subtotals can cause incorrect groups or double-counted totals. Know whether one row is an order, order line, customer, or day.

After connecting a CSV, spreadsheet, database, or online source, inspect the data types. Numbers, text, dates, date-times, Boolean true/false values, and geographic roles have different uses. Text month names can sort alphabetically. Postal codes treated as numbers can lose leading zeros. Store identifiers as labels, even if they contain digits.

Separate groups from amounts

A dimension groups or labels records: region, category, or order date. A measure supplies a quantity to aggregate. Data Studio calls measures metrics. The dimensions in a chart determine the groups; the measure's aggregation determines each group's value.

Tableau's Data pane distinguishes dimensions and measures. Blue fields in a view are discrete and create headers; green fields are continuous and create axes. These colors describe discrete versus continuous, not simply text versus number. Power BI marks summable numeric columns with a Σ and authored DAX measures with a calculator icon. Data Studio shows each metric's aggregation.

Select the aggregation deliberately

The sum of unit prices 4, 5, and 6 is 15, which is not a typical price. Their average is 5. COUNT of an optional field skips empty values. To count records, use a row count such as Power BI's COUNTROWS, or count a field filled in every record. Use distinct customer IDs for unique customers, because one customer can appear on many rows.

Build a ratio from its totals

Store X has sales of 1,000 and profit of 100; store Y has sales of 9,000 and profit of 1,800. Their margins are 10% and 20%, but averaging those percentages gives the two stores equal weight.

Create the calculation from aggregated profit and sales. Tableau uses SUM([Profit]) / SUM([Sales]); Data Studio uses SUM(Profit) / SUM(Sales). In Power BI, a measure can use DIVIDE(SUM(Orders[Profit]), SUM(Orders[Sales])).

Keep customer IDs, years, and postal codes out of sums. Use them as dimensions, or select “Don't summarize” in Power BI where appropriate. Before charting, compare row counts and known totals with the source so type errors, duplicates, or unintended filters surface early.

Once fields and calculations are correct, place them in the chart's slots. Each tool uses different names, but the underlying choices are the same: grouping field, numeric value, chart type, and optional second grouping.

Build total quantity by category

Tableau: Drag Category to Rows and Quantity to Columns. Check that Quantity uses SUM. Choose bars if needed, sort descending, and drag Quantity to Label on the Marks card.

Power BI: Choose a clustered bar chart. Put Category in the Y-axis field well and Quantity in the X-axis well. Check the aggregation, sort by Quantity descending through the visual's More options menu, and turn on data labels in Format visual.

Data Studio: Add a bar chart. In Setup, set Category as Dimension and Quantity as Metric with SUM. Sort by Quantity descending. In Style, show data labels.

In all three, check that the value axis starts at zero. Replace a default title such as “Sum of Quantity by Category” with wording that tells the reader the period and the finding.

Build a monthly line chart

In Tableau, place Order Date on Columns and Sales on Rows. A date can initially appear as discrete YEAR; select continuous Month for a timeline at the needed grain. In Power BI, select a line visual, place the date on X-axis and Sales on Y-axis, and inspect whether the date hierarchy is using year, quarter, month, or day. In Data Studio, choose a time series with the date dimension at Year Month and Sales as metric.

A month label without its year can combine January from different years. Confirm that the chart represents separate year-month periods when the question follows a continuous history.

Add a second group only when useful

Tableau's Marks card has Color, Size, Label, Detail, and Tooltip. Power BI offers field wells such as Legend. Data Studio offers a Breakdown dimension in Setup. These can separate a line by region or split bars into series.

Each new dimension can create more marks. Two categories do not simply add two labels: they form combinations. Check whether the resulting groups still support the comparison and remain readable.

Format and provide context

Use a real title, unit labels, sensible decimal places, and purposeful colors. A theme applies consistent fonts and colors. A reference line can show a target or average; Tableau provides reference lines through Analytics, while Power BI and Data Studio expose them in the selected visual's settings.

A filter limits the records included. Interactive filters appear as Tableau filter cards, Power BI slicers, or Data Studio controls. State the active period and restrictions in a visible subtitle such as “2025, online orders only”.

Tooltips can add exact values and shares, but essential context must remain visible in exported copies. Check any automatic suggestion, including Show Me or an AI proposal, against the question, field types, calculations, and scales before keeping it.

Before sharing, compare the finished chart with independent source totals. A plausible picture is not enough: the same fields can produce a convincing chart with the wrong aggregation, filters, or row count.

Trace a discrepancy to its cause

Suppose a 2025 chart shows 14,960 orders, while a SQL count of the orders table gives 12,480. The chart's source joins orders to order lines. An order with several lines therefore appears several times. Counting distinct order IDs restores the correct 12,480 orders.

Check what one row means after loading or joining data. SUM of line quantity can be correct on that table while a plain count of order IDs overstates the number of orders. Choose the calculation for the thing being counted.

Inspect visual, page, and report filters. Confirm the date range, time grain, measure aggregation, and category labels. A spelling error such as “Nroth” can create a separate group. Handling of case and spaces varies by tool; Power BI ignores capitalization and trailing spaces but not leading spaces. Clean inconsistent labels at the source instead of depending on accidental merging.

Give missing values an explicit treatment

A null can mean unknown, unavailable, or not recorded. Tableau may show a null indicator, while Power BI may display “(Blank)”. Investigate the meaning before excluding it. Fix the source where possible, use an “Unknown” category when appropriate, or disclose a justified exclusion.

A missing time period should remain visibly missing. Do not turn an unknown value into a zero just to make the chart complete. Recheck totals after any exclusion, because removing unknown categories can change percentages and rankings.

Document what viewers are receiving

Include the period and active filters, source, refresh date, and metric definition. For example, “Active customer: ordered at least once in the last 90 days” gives a precise meaning to a familiar phrase. Check that titles and subtitles stay accurate after filtering.

Set alt text, readable fonts, accessible color and secondary cues. If a report has several visuals, give them a logical reading order. For an exported image or PDF, inspect the actual export: it may lose tooltips or contain text too small to read.

Match the sharing method to access needs

An image or PDF suits a document or slide. An interactive version lets readers filter and inspect the data through Tableau Cloud, Tableau Server, the Power BI service, or a shared Data Studio report.

Check access and underlying-data permissions. Never publish personal or confidential data to Tableau Public, where work can be seen on the internet. Use the authorized audience and sharing method for the information involved.

A refresh can introduce a new category, change a total, or add an incomplete current period. Repeat the data and visual checks after refresh so the next version remains readable and accurate.