The Measurement Convention Is the Mechanism: Earnings Concepts, Self-Employment, and What Cross-National Comparison Can and Cannot Establish about Tourism Work

Comparative claims about tourism employment are routinely built by setting one country's tourism satellite account against another's. We show that this practice cannot in principle yield a comparable quantity, and we set out an alternative. The problem is not data availability but definitional: "tourism employment" spans business lines whose pay differs by a factor of five within a single country, while the earnings concepts embedded in national statistics differ in whether they count own-account workers and their mixed income at all. We formalise the second problem as measurement-convention dependence: the association between self-employment and measured earnings per person is a property of whichever earnings concept the statistician happens to be using, not of the labour market. We test this on a new harmonised panel covering 31 European countries, 17 years (2008-2024), and 19 NACE divisions - 9,738 sector-year observations - assembled from Eurostat national accounts (nama_10_a64, nama_10_a64_e), which unlike labour-cost surveys carry both mixed income and the employment split between employees and the self-employed. We compute the same estimand under three earnings concepts from one data source. Four results follow. (i) The mechanism is concept-dependent, and the dependence is large. Under the convention that books mixed income as labour income - the practice that generates the Chinese labour-share puzzle - the selfemployment gradient is β = −1.045 (SE 0.238, p = 0.0001). Under an SNA-consistent split at θ = 0.66 it is −1.216 (p