Skip to Content
Back to Library

Interactive Laboratory

When you measure changes what you find

The same model, the same 32 countries — estimated on different periods. Drag the window and watch the coefficient change sign.

Selected window

Estimated on2014–2023
Coefficient0.1831
Std. error0.0679
Significant at the 5% level

Note the change of sign. Estimated on windows ending before 2015 the relationship is weak or negative; in windows dominated by post-crisis observations it turns positive and significant. Same model, same countries — only the period differs.

Only the 10-year window is published so far — these are the figures from Table 5.5 of the paper. Run scripts/generate-rolling-grid.py to add more window lengths; the slider enables itself automatically.

Why it matters

The literature on globalization and the welfare state usually settles the question with a single long-run estimate. Our rolling regressions show that number is unstable: the coefficient is weak or negative in windows ending before 2015, and turns positive and significant once the window is dominated by post-crisis observations. A Chow test rejects parameter stability at 2008 for all four indices.

Read the full paper →

Method

Panel data FE

Data

32 OECD countries, 1980–2023

Source

Table 5.5

Std. errors

Clustered by country