Every published figure about women in tech rests on a prior decision about who counts as "in tech" — and that decision is rarely stable across time or source.
The most consequential choice in any women-in-tech statistic is the occupational boundary. "Tech" can mean only software engineers and developers; it can extend to IT operations, systems administration and network roles; it can stretch further to include digital project management, UX, data analysis, or any job classified under the Standard Industrial Classification codes for "information and communication." Each boundary produces a legitimately different number, and none of them is wrong — but they are not interchangeable.
This matters because women are not distributed evenly across those layers. Coding-only definitions tend to return lower female shares; broader service-and-support definitions return higher ones. A figure of 17 percent and a figure of 26 percent can both be honest counts of the same workforce at the same moment, drawn from different scope decisions on the same dataset.
A trend built on a moving boundary is not a trend; it is an artefact.
The problem compounds when tracking trends. If a survey tightens its occupational coding between waves — typically when a classification standard is revised, as happened across UK datasets following ONS updates to Standard Occupational Classification — the apparent trend can reverse or flatten for purely administrative reasons. A decline in the female share between two survey years may record a real change in who is being hired, or it may record that the revised code collapsed two previously separate job families into one, and the one that was dropped happened to employ more women.
Comparisons across countries introduce the same risk in international form. European Labour Force Survey harmonisation smooths some of this, but employer self-reported figures, professional-body surveys and government administrative data each draw the boundary differently, making cross-national league tables unreliable guides to policy outcomes.
- Standard Occupational ClassificationONS updates its SOC periodically; each revision can reclassify tech-adjacent roles, altering which workers fall inside or outside headline counts
- Standard Industrial ClassificationSIC codes used to identify "information and communication" employers; inclusion or exclusion of sectors changes the denominator in employer-level figures
- European Labour Force Survey harmonisationcross-national effort to align occupational coding across EU member states; improves comparability between countries but does not eliminate boundary differences between national datasets
The practical implication is straightforward: before asking what a trend shows, ask what the surveys count — specifically, whether the scope held still. A trend built on a moving boundary is not a trend; it is an artefact.