Allow me to paint a scene: you get a new dataset and need to explore it, so you don’t change a single number, but you make three different visualizations. If you showed those visualizations to three different people, they'd likely walk away with three slightly different impressions of what the data means. That is one of the things I find most fascinating about data visualization. We often talk about visualization as if it were simply the final step in a data-analysis pipeline, but that is not accurate at all! Visualization isn't just a picture of the data; it is an interpretation layer between the data and the person looking at it. That means when we choose a chart, an axis, a scale, a grouping, or even what to leave out, we are making decisions about the story the reader will see.

Anyone who works with data knows that the difficult part is rarely just getting a graph onto the screen. The difficult part is deciding which graph we show. Should we focus on the trend? The difference between groups? The variability? The outliers? Two visualizations can be completely accurate and still lead the viewer toward very different conclusions. One important question here is: Which chart should I use? But a better question is: What aspect of the data am I asking the reader to notice?

Suppose I want to represent the relationship between how long someone has been training and their strength. Using a simple line graph, you can see that as training time increases, strength increases. But the line makes the relationship look continuous and almost linear. Instead, a step-like representation emphasizes that strength tends to increase in stages. Using a logarithmic scale for strength can emphasize large improvements early on, known as "newbie gains." Neither scale is inherently more truthful.

We can go even further: a box plot could emphasize variability across training sessions, a bar chart could compare different training periods, and a scatter plot could show individual observations. The numbers haven’t changed; our interpretation has, and so has the message we're delivering.