Source: Levelica Canonical: https://www.levelica.com/en/guides/data-visualization # Data Visualization ## Summary Turning data into visual form that reads clearly, choosing what fits the purpose and blocking misreadings. ## Description Data visualization loads numbers onto visual channels like position, length, color, and shape so people can read them. You decide what to show, pick a visual encoding that suits the data type, strip out the misreadings that axes and colors create, and check that the same information reaches readers who can't see the picture. The weight sits on designing and delivering a chart that will be read, not on reading and interpreting one. --- ## Level Overview | Level | Proficiency Stage | Hashtags | | ----- | ----------------------------------- | ------------------------------------------------------------------------------ | | Lv.1 | Chart Decoding | #ChartAnatomy #AxisReading #ChartTypeIdentification #VisualChannelSpotting | | Lv.2 | Accurate Chart Making | #ChartTypeSelection #AxisLabeling #ScaleIntegrity #ChartDeclutter | | Lv.3 | Purpose-Driven Design | #EncodingChoice #MisleadingAxis #TextAlternative #ChartAnnotation | | Lv.4 | Audience-Adapted Design | #SmallMultiples #ColorVisionSafety #InteractionDesign #AudienceAdaptation | | Lv.5 | Standards and Review | #ChartStandards #VisualizationReview #DesignCritique #DataVizMentoring | | Lv.6 | Organizational Visualization System | #VisualizationGovernance #EncodingGuideline #PerceptualEvidence #VizCurriculum | | Lv.7 | Encoding Language Creation | #EncodingLexicon #NewChartForm #VisualizationCanon #DataVizPedagogy | > Proficiency stage names come from the core behavior of each level. Data visualization has no internationally recognized grading system, so no grade code is used as a stage name. The order of the stages follows the progression laid out by the conceptual framework in the PNAS visualization literacy study and by the multi-level task typology of Brehmer & Munzner. > **Integrative Framework:** The seven levels of this guide are not a copy of one grading table. They divide where four independent axes meet. The perceptual experiments of Cleveland & McGill(1984) and the conceptual framework in the PNAS visualization literacy study carry the **encoding decision axis** (how refined your judgment is about which visual element a data type rides on). The multi-level task typology of Brehmer & Munzner carries the **task layer axis** (how much of why a chart is needed, how it gets built, and what goes in and stays out you settle yourself). The VLAT visualization literacy assessment and the W3C accessibility rules for complex images carry the **reception check axis** (how your way of confirming that a result was read moves from your own eyes to reader responses and text alternatives). The practice of people who publish charts in the field carries the **misreading responsibility axis** (how far you carry the consequences of a misread chart, from your own single chart to the team, the organization, and the reading habits of a whole field). This guide covers the design and delivery of visual encoding. It leaves out analyzing data to reach conclusions and critically evaluating other people's data claims. > **Cross-Validation:** The level boundaries were set by checking where separate sources point to the same spot. > > - **Lv.2 to Lv.3 boundary**: VLAT is a tool its authors confine to the consumption dimension, so its list of reading tasks runs out of range here. At the same spot, the two layers of text alternative the W3C complex image rules require move responsibility for delivery over to the maker, and the how layer of Brehmer & Munzner is first demanded here too. Cleveland & McGill(1984) confirmed by experiment that accuracy differs across elementary perceptual tasks, which points at the same boundary: change the encoding and the accuracy people read at changes with it, so what rides where becomes a matter of judgment from here on. You move from drawing by convention to setting a purpose and choosing an encoding. > - **Lv.4 to Lv.5 boundary**: The why layer of Brehmer & Munzner only becomes necessary once you judge charts someone else made. The W3C rules change character too, from checking a single chart to setting rules for publishing, and the way the PNAS study designed definitions, exercises, and assessments as one package shows that demanding an ability from others requires tasks you can grade. Responsibility passes from making your own charts well to judging the charts of others. > - **Lv.6 to Lv.7 boundary**: Above this, no source names the next step. The PNAS conceptual framework and the typology of Brehmer & Munzner both stop at organizing forms already in use, so Lv.7 is defined not as a slot in the table but as the contribution that extends it. --- ## Level 1 #ChartAnatomy #AxisReading #ChartTypeIdentification #VisualChannelSpotting ### Summary Taking a chart apart to see which value rides on which visual element. ### Description You've just stepped into data visualization. On the encoding decision axis you're still checking mappings someone else chose, so you focus on telling bar, line, and scatter charts apart and on reading what axes and legends tell you. You break a single chart down into which values ride on position and which ride on color or size. Designing a chart yourself belongs to the next level. ### Checklist - [ ] Can tell bar, line, scatter, and pie charts apart on sight - [ ] Can point to the axes, axis labels, tick marks, legend, and title in a chart - [ ] Can say which of position, length, color, or size each value in a chart rides on - [ ] Can read the direction and size of change between two points on a line chart within 30 seconds - [ ] Can move data from a table into one basic chart - [ ] Can pick which of two charts of the same data makes values easier to compare and say why - [ ] Can show your first chart to someone else and write down what they read from it > If you've checked off most of this list, you're ready to enter Lv.2 Accurate Chart Making, where you pick the chart type the data calls for yourself. Lee, Kim & Kwon, who built the VLAT visualization literacy test, split reading into separate items: find a value, read a trend, compare extremes. Throw those same tasks at your own chart one at a time and the reading checklist flips straight into a making checklist. --- ## Level 2 #ChartTypeSelection #AxisLabeling #ScaleIntegrity #ChartDeclutter ### Summary Choosing a chart that fits the data type and making it readable. ### Description You start making charts yourself. You separate categorical from continuous data, pick the chart type each one calls for, and add a title, axis labels, units, and a source so the chart still makes sense on its own. On the encoding decision axis you still follow common convention. On the reception check axis you go no further than your own eyes. ### Checklist - [ ] Can separate categorical from continuous data and pick the chart type that fits each - [ ] Can build a chart carrying a title, axis labels, units, and the data source - [ ] Can start a bar chart baseline at zero and explain why it belongs there - [ ] Can change the sort order so the ranking of values shows at a glance - [ ] Can cut the number of series in one chart so overlapping lines stop competing - [ ] Can print a finished chart in black and white and check that the series stay distinguishable - [ ] Can draw the same data as two chart types and write one sentence on why you kept the one you did > If you've checked off most of this list, you're ready to enter Lv.3 Purpose-Driven Design, where you fix the reader's judgment first and choose the encoding around it. Cleveland & McGill(1984) confirmed by experiment that reading accuracy differs across elementary perceptual tasks. Draw the same data three ways, have someone read each one, and write down how closely they recover the values. Choosing stops being taste and becomes judgment. --- ## Level 3 #EncodingChoice #MisleadingAxis #TextAlternative #ChartAnnotation ### Summary Setting a purpose, picking the encoding, and finishing a chart alone. ### Description This is where dependence turns into autonomy. You write down what the reader has to judge, then load the comparison that judgment needs onto position or length. You strip out anything that distorts values: truncated axes, dual axes, areas inflated by volume. The reception check axis shifts here. You attach a short identifying description and a long one covering values, trends, and relationships, so information reaches readers who can't see the picture. ### Checklist - [ ] Can write in one sentence the judgment a reader should make from a chart and pick the encoding that serves it - [ ] Can find and strip out 3 or more value-distorting elements such as truncated axes, dual axes, and 3D effects - [ ] Can attach both a short identifying alt text and a long description covering values, trends, and relationships - [ ] Can keep color on the series you want to stress and mute the rest to set the order the eye travels in - [ ] Can annotate a chart with the conclusion sentence so readers don't miss the interpretation - [ ] Can show a draft to 2 or more people and revise it around the differences in what they read - [ ] Can find and record cases where the same data looks different once you change the aggregation unit or time span - [ ] Can check whether sentences alone carry the same conclusion and write down why the chart is needed > If you've checked off most of this list, you're ready to enter Lv.4 Audience-Adapted Design, where you rebuild the same data for each audience. The W3C accessibility guidance requires complex images to carry a short identifying description and a long one together. That habit of preparing one piece of information in two layers runs straight into designing two versions per audience. Write the long description first and hand the chart only as much as that text can drop. --- ## Level 4 #SmallMultiples #ColorVisionSafety #InteractionDesign #AudienceAdaptation ### Summary Reshaping layout by audience and medium to design complex visuals. ### Description Rule application gives way to situational adaptation. On the task layer axis you define why the chart is needed at all, then split the same data into one version for decision makers and one for practitioners. You lay several categories out against a shared axis, or add interaction so readers change the conditions themselves. You encode shape and position on top of color so nothing rests on color alone, and you tune density to the medium. ### Checklist - [ ] Can split one dataset into two versions, one for decision makers and one for practitioners, at different levels of detail - [ ] Can build small multiples that lay several categories out against a shared axis - [ ] Can add redundant encoding through shape or position so readers with color vision deficiency still tell series apart - [ ] Can add interactions such as filter, zoom, and detail on demand and write down the question each one answers - [ ] Can tune the density and type size of one chart for print, presentation screens, and mobile screens - [ ] Can judge whether a comparison is about time, distribution, part to whole, or relationship and pick the matching form - [ ] Can collect cases where your charts were misread and sort the causes into encoding, labeling, and context - [ ] Can record the times you fixed the conclusion instead of the chart when the data didn't support it > If you've checked off most of this list, you're ready to enter Lv.5 Standards and Review, where you judge other people's charts. Brehmer & Munzner's multi-level task typology splits a visualization into why it was made, how, and what goes in and stays out. Ask in that order, but treat encoding as the accuracy question Cleveland & McGill(1984) settled: put two options in front of one reader and record the gap. Taste complaints drop and fixable spots surface layer by layer. --- ## Level 5 #ChartStandards #VisualizationReview #DesignCritique #DataVizMentoring ### Summary Setting team chart standards and reviewing the visuals others make. ### Description Your scope widens from your own output to the standard the team works to. The misreading responsibility axis splits here: you carry the accuracy of every chart the team ships, not just your own. You write color, axis, and label rules into a document, harden them into templates, and review other people's visuals against encoding accuracy, misreading risk, and accessibility. You also rule on the cases where no chart is the better answer. ### Checklist - [ ] Can write a team chart standard covering color, type, axis, and label rules and get 5 or more people using it - [ ] Can review someone else's visual against three criteria: encoding accuracy, misreading risk, and accessibility - [ ] Can build chart templates for recurring reports and record how much production time they cut - [ ] Can set a pre-publication check that lists text alternatives and color contrast as items to confirm - [ ] Can teach 3 or more people how to justify encoding choices and have each submit a before and after chart - [ ] Can find 2 or more recurring bad charting habits in the organization and announce them with replacements - [ ] Can have 5 or more readers read the same chart and pinpoint where their answers diverge - [ ] Can judge which visualization requests need no chart and send them back as a table or a sentence > If you've checked off most of this list, you're ready to enter Lv.6 Organizational Visualization System, where you set guidelines that cross departments. The PNAS study designed the definition of visualization literacy, its exercises, and its assessments as one package. A standard you can teach comes with tasks you can grade. Rewrite each clause of the team standard as a single task that passes or fails, and the document grows from a rulebook into a course. --- ## Level 6 #VisualizationGovernance #EncodingGuideline #PerceptualEvidence #VizCurriculum ### Summary Designing the organization's visualization system and joining field debate. ### Description Your reach extends past the team into the organization and the field. You pull the forms each department used differently into one guideline, and attach visual perception research to every encoding rule so taste arguments turn into questions you can settle. You design a curriculum, measure ability before and after it, and put your methods into industry discussion. You also publish the exceptions where your own rules break down. ### Checklist - [ ] Can write a visualization guideline that 3 or more departments follow and check compliance every quarter - [ ] Can ground encoding rules in visual perception research and leave that evidence in the document - [ ] Can design a visualization course, run it for 20 or more people, and measure ability before and after - [ ] Can present visualization methods at industry events or write for professional outlets - [ ] Can define visualization quality metrics for the organization and report 12 or more months of trend - [ ] Can test a new chart form against an existing one for reader comprehension and decide whether to adopt it - [ ] Can find the data types your guideline fails on and publish them with exception rules > If you've checked off most of this list, you're ready to enter Lv.7 Encoding Language Creation, where you build forms no typology lists yet. Cleveland & McGill(1984) showed that you can settle which form wins by reading accuracy rather than by argument. When you put out a new form, draw the same data in the existing form beside it, measure accuracy, and publish the procedure as well. That leaves evidence other people can reproduce. --- ## Level 7 #EncodingLexicon #NewChartForm #VisualizationCanon #DataVizPedagogy ### Summary Founding a new encoding language that changes how later readers see data. ### Description You leave a mark on the language of data visualization itself. For data structures existing typologies can't hold, you found a new form of expression or a new encoding rule, and it becomes the default for other organizations and for the generation after you. No grade or credential names the next step, so the only evidence of arrival is how long the language you left keeps getting referenced. ### Checklist - [ ] Can confirm that a chart form or encoding rule you proposed is in use at 3 or more outside organizations - [ ] Can confirm 10 or more citations of your visualization work by other makers - [ ] Can show that your contribution has been referenced in both research and practice for 5 or more years - [ ] Can confirm that people you taught are setting visualization standards in their own organizations - [ ] Can create and validate a new form of expression for data structures existing typologies can't hold - [ ] Can publish what your own language hides and where it stops so later makers can build alternatives > If you've checked off most of this list, you stand where the default changes: what people reach for when they look at data. The reading ability VLAT measures and the delivery requirements W3C sets both assume forms that already exist. Adding to those forms is confirmed by no grade, only by what the generation after you takes for granted. --- ## References - [Data visualization literacy: Definitions, conceptual frameworks, exercises, and assessments](https://www.pnas.org/doi/10.1073/pnas.1807180116) — Proceedings of the National Academy of Sciences (PNAS) - [VLAT: Development of a Visualization Literacy Assessment Test](https://www.bckwon.com/pdf/vlat.pdf) — Lee, Kim & Kwon (IEEE TVCG) - [A Multi-Level Typology of Abstract Visualization Tasks](https://www.cs.ubc.ca/labs/imager/tr/2013/MultiLevelTaskTypology/brehmer_infovis13-corrections.pdf) — Brehmer & Munzner (IEEE InfoVis) - [Web Accessibility Tutorials: Complex Images](https://www.w3.org/WAI/tutorials/images/complex/) — W3C Web Accessibility Initiative (WAI) - [Graphical Perception: Theory, Experimentation, and Application](https://doi.org/10.1080/01621459.1984.10478080) — Cleveland & McGill (Journal of the American Statistical Association)