Data Visuals
Good data visualization can make complex ideas easier to understand, reveal patterns at a glance, and bring research findings to life. In this section, I share a selection of data plots I’ve created as part of my work. Each visual is accompanied by a short explanation and a link to the code on my GitHub for those interested in how it was made.
Classification of Occupations: ISCO-08
Understanding how occupations are structured isn’t always intuitive.
Occupations are grouped and classified into hierarchical systems like the International Standard Classification of Occupations (ISCO-08), which organizes all jobs from broad categories to highly specific roles. This structure helps researchers, policymakers, and international organizations speak a common language — but it’s complex.
Things get even more challenging when we start layering in additional data — such as the potential impact of generative AI on different types of work. Visual tools like the one above can help bring order and clarity to these multilayered systems.
​
This radial plot illustrates the full hierarchy of ISCO-08, from broad 1-digit occupational groups (e.g. Managers, Professionals) down to the 4-digit level that identifies specific job titles. Each branch is colour-coded by major group.
​
Admittedly, it’s a lot to take in — but for data nerds, this radial plot might just be the ultimate tattoo inspiration.
Deep Hierarchy: Tree Plot of ISCO-08 with tasks
Understanding how occupations break down into specific tasks is a real challenge — especially when working with a classification system as detailed as ISCO-08. There are 436 occupations, each linked to one or more of 3,267 tasks, forming a deep hierarchy that’s hard to visualize in traditional formats.
​
This tree plot illustrates that full structure — from broad 1-digit groups, through intermediate levels, down to 4-digit occupations and their associated tasks. Each branch shows how tasks are nested within occupational groups, allowing you to explore both vertical and horizontal relationships in the classification.
​
At the most detailed level, tasks are color-coded by their exposure to generative AI, making it possible to see which areas of work may be most affected. The result is a powerful way to combine occupational classification with AI impact data — and to explore it all in one interactive visual.
Will Generative AI affect jobs of the poor?
Among existing occupations in 16 Latin American countries, how can we tell whether exposure to generative AI is linked to income — and whether it matters if someone is self-employed or a wage employee? This figure visualizes that relationship.
Each dot represents a country-specific occupation, with its position showing how far its median income is from the national median wage. The top panel shows employees, and the bottom panel shows self-employed workers. Dot size reflects employment share, and colored dots highlight occupations with significant GenAI exposure — those using a computer and falling into the categories of automation, augmentation, or uncertainty.
​
We see that GenAI-exposed jobs for employees tend to cluster around or above the median income, particularly in professional and managerial roles. More broadly, the plot shows that most of the exposed occupations fall into what could be described as middle- and upper-middle income jobs — with very few low-income occupations showing significant exposure.
In other words, the first-order effects of GenAI technologies are likely to be felt most by people in higher-income, higher-skilled jobs, while many lower-income occupations — at least for now — may remain outside the immediate reach of this technological transition.
Which of the 2,541 Occupations Are Exposed to Generative AI?
How can we visualize over 2,500 occupations in Poland — many of which overlap — in a way that still reveals meaningful patterns?
​
This plot maps each ISCO-08 4-digit occupation by its average GenAI exposure score (x-axis) and the variation in exposure across its tasks (y-axis). To address the issue of overlapping data points, a small amount of jitter was added to improve readability.
​
Each shape and color represents a different level of GenAI exposure — from not exposed (grey) to four increasing exposure gradients (from blue to red). The green trend line highlights a clear pattern: variation in exposure tends to be higher in the middle of the distribution and lower at both extremes.
​
This visualization provides insight into not only which occupations are more exposed to GenAI, but also how consistent that exposure is across the tasks within each occupation.
