plot of chunk sacrifice-skipped-steps
PEPS-Y Wave 2 · Baltic crisis experiment
We asked 6,388 people across seven countries what they would personally do if Latvia were attacked. They could select any of four actions, from paying an additional monthly tax to enlisting in the armed forces.
The proposed sequence was pay an extra monthly tax → do civilian work → host refugees → enlist. Under this sequence, someone who chooses to host refugees would also choose the tax and civilian work. People may see the costs differently. For example, someone may be able to offer their time even if they cannot afford an extra tax.
A skipped step is an action left unselected below the highest-ranked action chosen. For example, choosing civilian work alone skips one step (the tax), and choosing enlistment alone skips three. Zero actions counts as zero skipped steps.
plot of chunk sacrifice-skipped-steps
Redder cells contain a larger share of the full unweighted sample. Grey cells represent combinations that cannot occur.
plot of chunk unnamed-chunk-10
Each person appears in one bar. Among people who selected at least one action, 39.1% chose actions without skipping a step.
Each bar shows an exact combination and gives its count and share of all 6,388 respondents. For example, Tax + civilian work means those two actions only. Choosing no action counts as level zero.
plot of chunk unnamed-chunk-11
2,771 people (43.4% of all respondents) chose a combination that skipped at least one step. The most common was civilian work alone. Each respondent appears in exactly one bar across the two charts.
plot of chunk unnamed-chunk-12
322 of the 641 people willing to enlist skipped the tax. Civilian work alone was the most common departure from the proposed order (1,236 people). Just 199 selected all four actions.
plot of chunk unnamed-chunk-13
Putting civilian work before the tax gave the best fit of all 24 possible orders. Even then, about one third of respondents skipped a step. Each bar uses the people who chose the first action as its denominator. The same person can appear in several bars.
Among 3,146 respondents who followed the proposed order and had complete background data, giving 25 more points in the Wave 1 Solidarity Game was associated with 14% higher odds of reaching a higher level of personal sacrifice in Wave 2. More educated and higher-income respondents, older respondents, and men also tended to reach higher levels.
plot of chunk unnamed-chunk-14
Each dot shows an estimated comparison after accounting for the other factors. For example, 2× means twice the odds of reaching a higher level. Dots to the right of 1× indicate higher odds. Dots to the left indicate lower odds. The horizontal lines show 95% confidence intervals, which indicate how uncertain each estimate is. The link with earlier giving, and the differences by gender and education, also appeared when we included people who skipped steps.
People shown the larger attack tended to select a higher-ranked action when everyone was included and each person was scored by their highest choice. The estimate was less clear when we counted the number of actions, or looked only at people who followed the order.
plot of chunk unnamed-chunk-15
For example, someone who selects enlistment alone scores 4 on highest-ranked action and 1 on number of actions. The full-sample comparisons use the randomly assigned attack descriptions. The comparison among people who followed the order uses a group defined by their answers after assignment. Treat the highest-action result as exploratory.
We compared people who were asked about their territorial identities before the crisis scenario with those asked afterward. We also checked whether the effect of a larger attack depended on when those questions were asked.
We found little evidence that question timing changed either what people chose or how they reacted to a larger attack. For the highest-ranked action, asking about identity first gave an odds ratio of 0.93× (95% CI 0.82 to 1.05). The larger-attack estimate was 1.16× (95% CI 0.97 to 1.37) when identity was asked first and 1.13× (95% CI 0.96 to 1.34) when it was asked afterward. The direct comparison between those attack effects was 1.02× (95% CI 0.80 to 1.31).
plot of chunk unnamed-chunk-17
The first row in each panel compares identity before versus after, averaging the two attack scenarios on the log-odds scale. The next two rows compare larger versus smaller attacks within each identity timing group. The last row directly compares the two attack effects. Its interval includes 1× in all three panels. The subset of people who followed the order is defined by answers given after assignment, so its estimates are descriptive.
In Wave 1, respondents reported how close they felt to the European Union and to their own country. The chart shows the weighted average number of actions people chose in Wave 2 for each EU attachment category.
plot of chunk sacrifice-by-eu-attachment
For each one-category increase in EU attachment, the estimated odds ratio was 1.99× (95% CI 1.83 to 2.17) for choosing more actions and 1.62× (95% CI 1.48 to 1.77) for choosing a higher-ranked action. Among people who followed the proposed order, the estimate was 2.64× (95% CI 2.32 to 3.00). These are associations. The models adjust for national attachment, country, experimental assignment, earlier giving, and the background characteristics used above.
The charts about response patterns show unweighted sample shares. The models use longitudinal weights and account for clustering by NUTS2 region. They adjust for country and randomized survey-block order. Age, gender and education come from Wave 1. Higher household income means a reported Wave 2 income bracket above the median bracket in the respondent’s country.
There were 3617 cumulative response patterns in the sample. The adjusted difference in the share who followed the order between the larger and smaller attack groups was -2.9 percentage points (95% CI -6.7 to 1.0). The people who followed the order can still differ between groups.
| Actions | Respondents | Sample share | Weighted share | Follows order |
|---|---|---|---|---|
| None | 1841 | 28.8% | 31.1% | Yes |
| Civilian work | 1236 | 19.3% | 18.8% | No |
| Additional tax | 617 | 9.7% | 8.9% | Yes |
| Host refugees | 555 | 8.7% | 9.1% | No |
| Additional tax, Civilian work | 481 | 7.5% | 7.4% | Yes |
| Additional tax, Civilian work, Host refugees | 479 | 7.5% | 7.0% | Yes |
| Civilian work, Host refugees | 361 | 5.7% | 5.2% | No |
| Additional tax, Civilian work, Host refugees, Enlist | 199 | 3.1% | 2.9% | Yes |
| Enlist | 188 | 2.9% | 2.8% | No |
| Additional tax, Host refugees | 177 | 2.8% | 2.6% | No |
| Additional tax, Civilian work, Enlist | 82 | 1.3% | 1.3% | No |
| Civilian work, Enlist | 74 | 1.2% | 1.2% | No |
| Civilian work, Host refugees, Enlist | 44 | 0.7% | 0.8% | No |
| Additional tax, Enlist | 23 | 0.4% | 0.4% | No |
| Host refugees, Enlist | 16 | 0.3% | 0.2% | No |
| Additional tax, Host refugees, Enlist | 15 | 0.2% | 0.2% | No |
The ordinal model estimates a common odds ratio across four thresholds. The threshold-specific estimates below show how the comparisons vary when each threshold is modeled separately.
| Model | Predictor | N | Odds ratio | 95% lower | 95% upper | p |
|---|---|---|---|---|---|---|
| Cumulative intensity | Large vs small attack | 3146 | 1.082 | 0.898 | 1.304 | 0.403662 |
| Cumulative intensity | 25 more Solidarity Game points | 3146 | 1.144 | 1.056 | 1.239 | 0.001147 |
| Cumulative intensity | Women vs men | 3146 | 0.466 | 0.378 | 0.575 | < 1e-04 |
| Cumulative intensity | 10 years older | 3146 | 1.158 | 1.090 | 1.230 | < 1e-04 |
| Cumulative intensity | Medium vs low education | 3146 | 1.504 | 1.143 | 1.979 | 0.003849 |
| Cumulative intensity | High vs low education | 3146 | 3.241 | 2.447 | 4.292 | < 1e-04 |
| Cumulative intensity | Higher vs lower household income | 3146 | 1.746 | 1.379 | 2.211 | < 1e-04 |
| Cumulative intensity, all qualifying | Large vs small attack | 3617 | 1.100 | 0.925 | 1.307 | 0.278614 |
| Highest action selected | Large vs small attack | 5593 | 1.114 | 0.970 | 1.279 | 0.126542 |
| Highest action selected | 25 more Solidarity Game points | 5593 | 1.124 | 1.061 | 1.191 | < 1e-04 |
| Highest action selected | Women vs men | 5593 | 0.606 | 0.523 | 0.703 | < 1e-04 |
| Highest action selected | 10 years older | 5593 | 1.037 | 0.996 | 1.081 | 0.080562 |
| Highest action selected | Medium vs low education | 5593 | 1.339 | 1.074 | 1.669 | 0.009835 |
| Highest action selected | High vs low education | 5593 | 2.062 | 1.634 | 2.602 | < 1e-04 |
| Highest action selected | Higher vs lower household income | 5593 | 1.134 | 0.980 | 1.311 | 0.090983 |
| Number of actions selected | Large vs small attack | 5593 | 1.029 | 0.891 | 1.187 | 0.697672 |
| Number of actions selected | 25 more Solidarity Game points | 5593 | 1.119 | 1.058 | 1.184 | 0.000121 |
| Number of actions selected | Women vs men | 5593 | 0.590 | 0.503 | 0.692 | < 1e-04 |
| Number of actions selected | 10 years older | 5593 | 1.066 | 1.021 | 1.113 | 0.004198 |
| Number of actions selected | Medium vs low education | 5593 | 1.395 | 1.123 | 1.734 | 0.002934 |
| Number of actions selected | High vs low education | 5593 | 2.636 | 2.098 | 3.311 | < 1e-04 |
| Number of actions selected | Higher vs lower household income | 5593 | 1.384 | 1.155 | 1.658 | 0.000530 |
| Highest action, all valid responses | Large vs small attack | 6388 | 1.146 | 1.017 | 1.290 | 0.025259 |
| Number of actions, all valid responses | Large vs small attack | 6388 | 1.059 | 0.936 | 1.198 | 0.362981 |
| Model | Predictor | N | Odds ratio | 95% lower | 95% upper | p |
|---|---|---|---|---|---|---|
| Highest-ranked action | Identity before vs. after | 6388 | 0.933 | 0.825 | 1.055 | 0.2647 |
| Highest-ranked action | Larger attack: identity asked after | 6388 | 1.131 | 0.957 | 1.337 | 0.1481 |
| Highest-ranked action | Larger attack: identity asked before | 6388 | 1.155 | 0.971 | 1.374 | 0.1023 |
| Highest-ranked action | Difference between the two attack effects | 6388 | 1.021 | 0.798 | 1.307 | 0.8664 |
| Number of actions | Identity before vs. after | 6388 | 0.952 | 0.838 | 1.081 | 0.4443 |
| Number of actions | Larger attack: identity asked after | 6388 | 1.062 | 0.896 | 1.259 | 0.4863 |
| Number of actions | Larger attack: identity asked before | 6388 | 1.053 | 0.868 | 1.277 | 0.5969 |
| Number of actions | Difference between the two attack effects | 6388 | 0.992 | 0.759 | 1.296 | 0.9508 |
| Only people who followed the order | Identity before vs. after | 3617 | 0.982 | 0.817 | 1.181 | 0.8455 |
| Only people who followed the order | Larger attack: identity asked after | 3617 | 1.230 | 0.969 | 1.560 | 0.0882 |
| Only people who followed the order | Larger attack: identity asked before | 3617 | 0.987 | 0.752 | 1.295 | 0.9245 |
| Only people who followed the order | Difference between the two attack effects | 3617 | 0.803 | 0.550 | 1.171 | 0.2517 |
| Model | Predictor | N | Odds ratio | 95% lower | 95% upper | p |
|---|---|---|---|---|---|---|
| Number of actions | EU attachment (one category higher) | 5593 | 1.993 | 1.828 | 2.172 | <1e-04 |
| Number of actions | National attachment (one category higher) | 5593 | 1.005 | 0.913 | 1.106 | 0.925 |
| Highest-ranked action | EU attachment (one category higher) | 5593 | 1.620 | 1.484 | 1.768 | <1e-04 |
| Highest-ranked action | National attachment (one category higher) | 5593 | 1.043 | 0.953 | 1.140 | 0.359 |
| Only people who followed the order | EU attachment (one category higher) | 3146 | 2.636 | 2.317 | 3.000 | <1e-04 |
| Only people who followed the order | National attachment (one category higher) | 3146 | 0.937 | 0.827 | 1.060 | 0.297 |
plot of chunk unnamed-chunk-21