1.2 Data visualization as a tool for discovery

While communication is often the first purpose that comes to mind, visualization is equally powerful for discovery. Before publication, plots and visual summaries help researchers explore datasets, detect errors, and uncover hidden relationships. In modern biomedical science, where genomics, imaging, and electrophysiology generate massive, multidimensional data streams, visualization is indispensable for making sense of complexity.

Humans are poor at spotting trends in raw numbers. Patterns that are invisible in tables become obvious when plotted. For instance, a long list of neuronal firing rates across trials tells little about behavior, but a raster plot can immediately reveal rhythmic bursts or stimulus-linked changes. Visualization turns data into insight.

The difference between exploratory and confirmatory visualizations

Exploratory visualization, used for hypothesis generation and pattern discovery, and confirmatory visualization, used to test specific hypotheses and communicate results, are fundamentally different modes of data visualization. Each has different goals and standards, and understanding this distinction is critical because the same visualization techniques may be appropriate in some contexts but not others. Not uncommonly, researchers use exploratory methods but present findings as if they were confirmatory (see HARKing). This can inflate false positives and causes major errors in reproducibility.

To best understand the distinction between exploratory and confirmatory visualizations when using data visualization for discovery, consider the function of each. 

Exploratory visualizations are your discovery mode and are used when you first encounter your data and help you develop a hypothesis. Exploratory visualizations help answer important questions, such as:

• Are there patterns and relationships between various factors in my data?
• Do I detect outliers and data quality issues?
• What is the distribution and how do my data variables potentially impact each other?
• Have I made appropriate statistical assumptions in my data analysis plan?

As an example, imagine you have conducted a learning and memory behavioral study where rats learn to navigate a maze.

You have behavioral data, including number of trials to learning, errors made, time spent exploring and successful navigation occurred.

You also have counts of activated neurons from 15 different brain regions.

You begin exploring your data by creating:

1. Bar charts showing average neuron counts in each region.
2. Scatterplots of neuron activation compared to behavioral performance.
3. Histograms of distribution of activation across animals.
4. Heatmaps showing regions most active in successful vs unsuccessful learners.

While looking over your visualization, you notice that animals with higher activation in the hippocampus learned the maze in fewer trials. This is an exploratory discovery because you didn’t predict this specific brain region and behavior connection before analyzing your data. You found the relationship among all of the brain regions using data visualization as a tool.

Confirmatory visualizations are used to test specific, pre-determined hypotheses. This means you specified what you would visualize before seeing the data and you defined the analysis parameters in advance. This type of visualization demonstrates whether a predicated effect occurred, like in response to “We hypothesized that…” or “To test whether…”.

Building on the exploratory visualization use example above, after you detected the hippocampal activation, learning speed relationship, you design a further study. Your established hypothesis states that rats with higher hippocampal neuron activation will require fewer trials to reach successful learning performance in the maze. You define your analysis plan, including a scatterplot with regression lines, to visualize your data. After running your experiment, you create the pre-specified visualizations and find that they demonstrate support for a hypothesis that existed independently of this new experiment and cohort’s data.

Know the role of your data visualization, even in discovery mode. Exploratory visualizations will lead you to new conclusions and hypotheses; confirmatory visualizations will help you verify answers!

Liability in visualization-based data discovery

In discovery mode, however, visualization’s strengths can also become liabilities. The human brain is wired to detect patterns, even where none exist (See our unit on confirmation bias for more). Consider an example from cognitive research: a scatterplot of age versus working memory performance across 200 participants shows no significant correlation (r = -0.12, p = 0.09). But after smoothing and binning, the same dataset appears to reveal a sharp decline with age. The smoothing algorithm has imposed a pattern that isn’t real. 

1.3 Common pitfalls and principles for rigor

Rigor in data visualization means designing figures that faithfully represent data and allow others to interpret them accurately. This section focuses on three major sources of error: misrepresentation, overload, and inaccessibility. Each can undermine the credibility of research if not carefully managed. Management starts at the inception of creating visualizations and requires intentional choices that link all the way back to experimental design.

Misrepresentation:
Misrepresentation can be intentional or accidental. It occurs when design choices (like chart type, scale, or selective inclusion) distort the data story. Misrepresentation can be mitigated through early, careful planning of data reporting during stages of experimental design.  In neuroscience, truncating axes, excluding baseline periods, or plotting only “responsive” data points can all create the illusion of stronger effects.

In a scenario of EEG amplitude measurement across different brain regions during rest vs. task conditions, a graph shows only the most active recording electrodes, omitting those that show no effect. This selection bias gives the impression that the treatment was universally effective, when in fact the full dataset would show large variability.

Professional presentation with selection bias:

Visual comparison reveals where patterns are present and absent:

Checklist for avoiding misrepresentation:

• Is your chart appropriate for the data type and statistical test?
• Can a viewer quickly determine what was being measured and compared?
• Are axes complete and clearly labeled?
• Are excluded or combined data points reported transparently?
• Is the language in titles and legends objective rather than persuasive?
• If a colleague unfamiliar with your experiment viewed the plot, would they interpret it the same way you do?

Rigor requires honesty in visual storytelling. An accurate visualization shows both what is interesting and what is true.

Overload and complexity:
Even accurate data can be obscured by clutter. Overloaded figures with too many variables, excessive color schemes, or confusing legends prevent comprehension.

For example, a researcher might attempt to show multiple experimental conditions, time points, and subgroups in a single chart. The result is an unreadable tangle of lines and colors, where trends are lost. Simplifying or separating figures can dramatically improve clarity.

Checklist for managing complexity:

• Who is your target audience, and what level of familiarity do they have?
• What is the single message this figure should convey?
• Can any element be removed without losing meaning?
• Are there too many categories or overlapping visual cues?
• Would a simpler or alternative visualization communicate the same idea more effectively?

Simplification is not an omission when done for comprehension. Omission of any data, for any reason, should be acknowledged in paired text.

Rigor is not about showing everything. Rather, it’s about showing the truth clearly.

Accessibility and transparency:
A rigorous figure must be accessible to diverse audiences. Accessibility has two dimensions:

1. Conceptual accessibility: Can non-experts understand it?
2. Visual accessibility: Can individuals with differing abilities (e.g., color vision deficiencies, low vision) interpret it?

Accessibility is driven by plot and design choices. Conceptual accessibility considerations include using jargon-free labels, providing context through annotations and labels, choosing appropriate chart types, including figure legends to explain symbols and scales, and even hierarchy of important information within a figure.

Visual accessibility includes using color-blind–friendly palettes, distinct shapes, and sufficient contrast ensures that meaning is conveyed through multiple cues, not color alone. For digital publications, providing alt-text or tabular data summaries helps screen reader users access the same information.

Checklist for accessibility:

• Are colors perceptually distinct and high-contrast?
• Are shapes or markers distinguishable without color?
• Could the figure be understood in grayscale?
• Does the figure include an alt-text description or accessible data table?
• Are figure titles and axis labels free of jargon?

Rigor and inclusivity are inseparable. A figure that some readers cannot interpret correctly is, by definition, not rigorous.

1.4 Our responsibilities when producing data visualizations

As the producer of a visualization, you are the first audience for your data and the final arbiter of how it is shared. The rigor of your work depends not only on statistical correctness but also on how transparently you communicate findings. Every choice from chart type to color palette affects interpretation.

When you create a figure, ask yourself:

• Does this visualization accurately reflect the evidence?
• Have I unintentionally exaggerated or minimized effects?
• Can another researcher reproduce this figure from my data and reach the same conclusion?

In other words, rigor begins the moment you decide to put data into image form. Visualization is not an afterthought, it is part of the scientific method.

You are both storyteller and gatekeeper. Each decision you make when designing a figure shapes how others will understand and build upon your data.

2.5 Linking visualization and statistical thinking

In rigorous research, visualization and statistical analysis are not separate steps. They are two sides of the same reasoning process. The way you plot data should emerge naturally from how you collected it and what question you are trying to answer.

Before creating a figure, take a moment to think through the logic of your dataset. What kind of variable are you working with: continuous, categorical, or a proportion? Are your samples independent, or are they paired measurements from the same individuals? Are your values distributed symmetrically or with noticeable skew and outliers? And finally, what question is your analysis really asking: a difference in averages, a difference in medians, or a consistent change within individuals?

Each of these considerations points toward a specific statistical test and a visualization that communicates its results accurately. When the analysis and the plot type match, the figure feels intuitive and credible; when they diverge, the viewer is left uncertain about what is being shown.

For example, when data are normally distributed and independent, a t-test and a simple bar graph with error bars convey the main result clearly. But when distributions are skewed or contain outliers, a boxplot with the raw data points overlaid gives a truer picture and supports a non-parametric test like the Mann–Whitney U. Paired data, by contrast, call for a visualization that highlights within-subject changes, such as a paired-line plot or a dot plot connecting pre- and post-values, paired with a Wilcoxon or paired t-test.

The table below summarizes these relationships:

When the elements of data structure, test, and visualization align, the result signals methodological rigor. A figure that communicates how data were analyzed allows others to evaluate and reproduce your reasoning.

2.6 Why this matters for neuroscientists

Spanning cells, circuits, and behavior, neuroscience data are notoriously noisy, variable, and hierarchical. These complexities amplify the consequences of poor visualization. A truncated scale or hidden pairing can make an insignificant neural response look groundbreaking or mask subtle but real changes across trials.

For example, representing neural firing rates with mean ± SE might imply precision that does not exist, especially when variability between cells is large. Displaying the same data as individual points or boxplots reveals the heterogeneity inherent in biological systems.

Transparent visualization practices not only strengthen your conclusions but also protect others from building on misleading findings. Rigor in plotting is, therefore, as essential as rigor in experimental design or statistical analysis.

A figure that tells the truth, even when the truth is messy, is an invaluable contribution to scientific rigor.