Question to measurable claim

“Germany is losing competitiveness” sounds decisive. Analytically, it is incomplete. It does not identify an outcome, an economic unit, a comparison group, a period, or the quantity we want to estimate.

For this lesson, the broad claim becomes a bounded historical question:

From 2000–2019, did Germany’s manufacturing real value added per hour grow more slowly than the median of Austria, Czechia, France, Italy, the Netherlands, and Sweden; was its rebased unit-labour-cost movement less favourable from 2020 to 2021; and what additional evidence would be needed to explain the productivity gap?

This formulation is deliberately narrower than the headline. It does not establish Germany’s current 2022–2026 position. It turns an argument into a measurement plan that another analyst can inspect, reject, or reproduce.

The evidence package separates four source roles instead of blending unlike statistics into one score.

• World Development Indicators (DS001) provides whole-economy productivity, manufacturing growth, and export-intensity context.
• OECD Data Explorer (DS003) provides internationally comparable GDP-per-hour, manufacturing productivity growth, and manufacturing unit-labour-cost series.
• Eurostat (DS004) provides a second, European national-accounts view of manufacturing real value added, value added per hour, and nominal unit labour costs.
• EU KLEMS & INTANProd (DS011) defines the growth-accounting extension for TFP and intangible-capital contributions. Its public download and citation are verified, but numeric redistribution permission and the advertised release vintage are not. No EU KLEMS values are republished here pending written permission.

The common comparison window is 2000–2021. The structural reading ends in 2019; 2020–2021 is shown as a pandemic sensitivity rather than treated as an ordinary endpoint.

The worked solution uses OECD manufacturing real-GVA-per-hour growth as the primary outcome, repeats the same calculation with Eurostat, and keeps the two providers separate. WDI remains context. EU KLEMS remains a named but blocked mechanism branch.

The metric tree keeps four branches separate before any calculation begins:

1. Outcome: manufacturing real value added per hour.
2. Cost position: manufacturing unit labour cost, indexed to 2020.
3. External exposure: exports as a share of GDP and manufacturing real growth.
4. Mechanism: TFP and intangible-capital contributions, retrieval-only until EU KLEMS reuse rights are resolved.

The tree defines the analysis. Every leaf names its provider, unit, denominator, period, and limitation. No leaf is averaged into a synthetic “competitiveness score.” The annual chart below the method then shows what happens when one branch is actually calculated.

Begin with the decision, not the dataset. Then make every design choice executable.

1. State who could act differently after seeing the result.
2. Select one primary outcome and distinguish contextual indicators from mechanisms.
3. Fix the unit of observation: country-year for whole-economy context or country-industry-year for manufacturing evidence.
4. Choose benchmarks before examining the ranking.
5. Define the estimand. Here it is the mean of 20 annual gaps: Germany’s manufacturing productivity growth minus the median growth of the six declared peers in each year from 2000 through 2019.
6. Record units and denominators. GDP per worker, GDP per hour, a growth rate, an index, and an export share are not interchangeable.
7. Preserve provider vintages and flags. Similar-looking OECD and Eurostat series remain separate observations.
8. Compound Germany’s annual rates and the annual peer-median path separately. Do not compound an average rate or treat the constructed peer path as one country.
9. Report 2020–2021 as sensitivity, repeat the calculation with Eurostat, and state what the result does not establish.

For year t, the calculation is gap_t = Germany_growth_t - median(peer_growth_t). The structural estimand is the arithmetic mean of gap_t for 2000–2019. The worked solution below executes this exact sequence on the published panel and exposes every annual input.

The primary OECD result is 2.35% per year for Germany versus 2.65% for the annual peer-median path, an average gap of −0.29 percentage points per year. A separate Eurostat provider cross-check gives 2.35%, 2.64%, and the same rounded −0.29-point gap. Compounding the annual OECD rates gives 54.6% for Germany and 66.8% for the constructed peer-median path.

First reproduce the worked result; then use the workbench to change the design.

1. Download the permitted M01 CSV.
2. Run the DuckDB query shown in the worked solution without changing it.
3. Confirm the rounded output 2.35, 2.65, -0.29, 54.6, 66.8 and inspect all 22 annual comparisons.
4. Check that Eurostat independently produces the same rounded annual gap.
5. Only then change one choice in the workbench: the outcome, benchmark, estimand, metric, or caveat.
6. Explain why your changed result answers a different question.

The workbench validates structure, not ideology. It does not replace the calculation above. Multiple trees can pass when their choices are explicit and coherent, but each one must eventually be executed against data. A printable blank tree, full worked solution, annual table, and expected output remain available when scripts or local storage are unavailable.

A leader should not ask for “the competitiveness number.” They should ask which outcome is deteriorating, relative to whom, over what period, and through which mechanism.

The worked evidence supports a modest historical manufacturing-productivity gap: about 0.29 percentage points per year against the declared peer median in 2000–2019. It does not support the narrower claim that Germany’s 2020–2021 unit-labour-cost movement was less favourable; both rebased provider indexes sit below the peer median in 2021. It also does not establish the mechanism or Germany’s current 2022–2026 position.

What should an executive do differently after seeing this? Treat the historical productivity gap as a reason to commission a current, mechanism-level diagnosis—not as permission to assert that present competitiveness or labour-cost pressure has already been proved.

• The complete four-source window ends in 2021 and cannot establish the present-tense 2022–2026 claim.
• World Bank productivity is whole-economy evidence, while OECD, Eurostat, and EU KLEMS manufacturing measures use different classifications and construction methods.
• GDP per worker, GDP per hour, growth rates, and 2020-based indices answer different questions.
• The peer benchmark is a new median in each year. Its compounded path is a transparent constructed benchmark, not the experience of one country.
• Higher unit labour costs may reflect wage convergence, labour hoarding, shorter hours, quality upgrading, or temporary utilisation rather than structural decline.
• A 2020=100 ULC index compares movement from each country’s own base. It does not compare absolute labour-cost levels across countries.
• Exports as a share of GDP measure openness and domestic-demand composition, not export market share or domestic value added in exports.
• TFP is a residual that also absorbs utilisation, reallocation, assumptions, and measurement error.
• Pandemic disruption, the post-2021 energy shock, automotive transition, China exposure, demographics, exchange rates, offshoring, and sectoral reallocation remain rival explanations.

Change exactly one design choice and explain how the interpretation changes.

• Replace the peer median with an employment-weighted manufacturing benchmark.
• End the structural window in 2018 instead of 2019.
• Replace value added per hour with value added per worker and defend the denominator.
• Replace the annual peer median with the median of country-specific 2000–2019 average growth rates and explain why the estimand changes.
• Add Poland to the first three sources while documenting why it cannot enter the selected EU KLEMS extension.
• Design a separately sourced 2022–2026 extension without splicing it into the historical series.

Your extension is complete when the revised metric tree still names its decision, outcome, unit, comparison, estimand, sources, denominators, caveat, and executive implication.