People agree to more personal sacrifices when the attack is larger.
People agree to more personal sacrifices when identity questions come before the crisis.
We expect the same pattern for support for government action.
The four groups
The interim master file contains all four combinations. We can therefore show attack size and identity order separately. The table gives group size, mean and 95% confidence interval for each outcome. Sacrifice ranges from 0 to 4. Government action ranges from 0 to 3.
Green means the sign matches our expectation. The dashed line marks zero. The numbers are the same equal-country differences as in the table. Colour does not show statistical significance.
The table pools records. The differences and the graph give each country the same weight. These can show different patterns when country sample sizes differ. Green means only that the sign matches our expectation. The graphs now include 95% confidence intervals. A predicted sign and an interval excluding zero are different checks.
Results by country
Each row shows one of the four groups within a country. The sacrifice score counts selected personal actions (0 to 4). Government action records the strongest response selected, from no action (0) to troops (3).
Each dot is a group average. Blue means identity questions came before the crisis. Orange means they came after it. The numbers show the averages. A rising line means more support under the larger attack. At the same attack size, a higher blue dot means more support when identity questions came first.
Read this graph in the same way. Higher scores mean support for a stronger government response. The colours identify question order; they do not mark expected or unexpected results.
The assignment table has all four combinations. This check alone does not prove that randomization worked in every respect. These are simple unweighted direction checks, with 95% confidence intervals and unadjusted two-sided tests. The final registered analysis will use the completed dataset.
How to read the confidence intervals
The bars show 95% confidence intervals. If an interval crosses zero, the estimate is not statistically distinguishable from zero at the unadjusted two-sided 5% level. A green estimate can still have an interval crossing zero. The colour records the predicted sign, not statistical significance.
The current export has no Wave 2 weights. Regression intervals use HC2 standard errors. Each panel model uses one change score per person. The graphs of proportions use Wilson intervals. The paired country-change graph uses the variation in each person’s change. The histogram shows counts, so it does not need a confidence interval.
All results remain preliminary. Many comparisons are shown, and their p-values are not corrected for multiple testing. These simple baseline models do not implement every part of the final preregistered analysis.
Treatment estimates and confidence intervals
The main effects keep the same equal-country comparisons as before. A country-specific OLS model with the attack-by-identity interaction gives exactly the same point estimates. It also lets us calculate uncertainty for those comparisons. The interaction measures how the attack difference changes when identity questions come before the crisis.
We repeat the same model without Poland. The remaining six countries receive equal weight. A significant result in one sample and an insignificant result in the other does not, by itself, show a difference. The exported formal contrast tests Poland against the average interaction in the other countries directly.
Does identity-item order change the placement effect?
This is a new exploratory analysis. Order 1 runs from municipality to world. Order 2 runs from world to municipality. The questionnaire randomizes this order, and both orders are recorded in both placement groups. Order shown after the crisis cannot prime an earlier answer. We distinguish the order effect among those who see identity first from the placement-by-order interaction.
The predicted means average equally over attack size and country. The three-way test asks whether placement-by-order moderation differs with attack size.
Response percentages and enlistment among troop supporters
All valid answers to each item form its denominator. The final row uses only people who selected troop deployment and gave a valid enlistment answer. These percentages describe answers; the difference is not a causal effect.
The vignette names Latvia, not a particular attacked town. We use distance from a representative point of each matched NUTS2 region to Latvian territory. This is a regional proxy, not a respondent’s exact location or distance to Riga. The existing GISCO name mapping is reused. Unmatched regions and small regional samples are reported in the diagnostics.
The model allows distance moderation to vary between Germany and Poland. Within each country, distance interacts with attack size, identity placement and their combination. Its intervals cluster at NUTS2 because distance is shared within a region. Separate country fits use t intervals with the number of country-specific clusters minus one degrees of freedom. Curves are shown only over each country’s observed 10th to 90th percentile distances. This is exploratory and observational.
The primary regression intervals use HC2. An additional CSV repeats uncertainty calculations using observed administrative regions as clusters. These regions are not documented survey sampling units. This sensitivity checks possible within-region dependence without replacing the primary models.
