Country effects explorer
Every country contrast in the paper, and the cell means behind them. All numbers are aggregates over fifty repetitions per country–model cell; the underlying responses are not on this site.
1. Country effects relative to the United States
Country fixed effects from the pre-registered regression (equation 1 and its Q2 and Q3 analogues): model fixed effects, standard errors clustered by model, wild cluster bootstrap with Webb weights and 9,999 replications. The portfolio question is the primary outcome (Version B, age 40). The twenty contrasts of each question form one Holm family.
2. The same contrast, one model at a time
The regression pools thirteen models. This panel shows the raw difference between a model's mean recommendation in each country and its own U.S. mean, with no pooling, so you can see whether the country ordering is shared across models or driven by a few. The dots are differences of cell means (fifty responses each), not regression estimates, and carry no confidence interval.
3. Leave one model out
The primary regression re-estimated without one model at a time (Q1, Version B, age 40). The joint wild-bootstrap p-value tests that all twenty country contrasts are zero.
4. Language, country name and local context, model by model
Cell means for one country under the three prompt versions (Q1, age 40). Version B asks in the local language with generic instruments. Version B-cross asks the same question in English with the country named, so B minus B-cross is the effect of the language. Version C asks in the local language and names local funds, bonds and pension vehicles, so C minus B is the effect of local context. The six English-prompting countries have no B-cross cell.
5. Age
Cell means for one country at the three prompted ages (Q1, Version B). The 100-minus-age rule would put the equity share at 70, 60 and 45 percent.
6. The stated-investor arm
Cell means for one country when the English Version B question adds a sentence about the investor (Q1, age 40): risk tolerance crossed with pension generosity, and a one-year or twenty-year horizon in place of the five-year baseline. The baseline column is the Version B cell in the local language.