The American Career Mobility Study 2026: What the Evidence Can Support

A MyTopMatch analysis of O*NET 31.0 and BLS occupational data found that a transparent four-domain content model recovered O*NET related-occupation links with a receiver operating characteristic area under the curve of 0.924. The result validates occupational-content resemblance. It does not yet validate the probability, accessibility, or direction of a worker's career transition.

What the model measures

The pilot compares occupations on skills, knowledge, abilities, and work activities. The frozen specification assigns weights of 30%, 20%, 20%, and 30%, respectively. Importance and level values are scale-normalized, suppression-recommended and not-relevant cells are excluded, and mapped O*NET suboccupations are averaged to the canonical detailed occupational spine.

O*NET related-occupation links are excluded from the similarity score. They are used as an external validation target. This separation prevents the model from being graded on a relationship it already contains.

The combined profile performed best

The weighted model produced an AUC of 0.924, compared with 0.921 for knowledge alone, 0.859 for skills, 0.836 for work activities, and 0.814 for abilities. At ten suggested neighbors, the combined model recovered an average of 45.7% of available related-occupation links.

The result was also stable to reasonable weight changes. Across 200 perturbations of the domain weights, the middle 90% of AUC values ranged from 0.919 to 0.928. That narrow band supports the conclusion that the validation result is not dependent on one exact weighting choice.

Why transition rankings remain withheld

Occupational resemblance is only one part of a feasible move. Real transitions also depend on work history, education, training, licensing, geography, hiring practices, and whether comparable workers have actually made the move. The current research package does not contain the linked CPS or SIPP transition samples needed to train and evaluate that second stage.

MyTopMatch therefore withholds Occupational Mobility, Career Switchability, and Career Resilience rankings. Before those products can pass the publication gate, the transition model must be trained and tested on separate cohorts and evaluated for calibration, AUC, top-k recall, subgroup error, wage-direction accuracy, and rank stability.

What job seekers can use now

The similarity model can support a research shortlist. A worker can use content resemblance to identify occupations worth investigating, then verify current credentials, licenses, local demand, pay, and actual hiring pathways. The pilot pair file should not be interpreted as a recommendation or transition probability.

What this study cannot establish

• The probability that a particular worker will be hired into a target occupation.
• Whether a transition is legally or practically accessible in a specific location.
• Whether an adjacent occupation will raise an individual's pay.
• Whether the model performs equally well across demographic groups or career stages.
• A public ranking of the most switchable or resilient careers.