Wave 2 interim analysis: changes in political solidarity
What we check
This report follows the preregistration on change. It checks which changes go with changes in solidarity.
We match Wave 2 records to Wave 1 by country and serial number. We use only Wave 1 main-field records (phase = 2). Germany has no matches. The survey provider still needs to confirm that the serial numbers identify the same people in both waves.
This export contains 2188 Wave 2 records and 1823 possible matches. All 365 German records lack serial numbers. The change models below use the matched records from the other six countries. They show associations, not experimental effects.
The match check finds 2 gender differences and 3 age gaps larger than one year. We keep these pairs in this overview. Their identities need review before the final panel analysis.
A positive change means more willingness to pay for public services that help others. The score averages three answers: support for people in the local area, the country and other EU countries. It ranges from 1 to 5.
We use a pair only when all three answers are available in both waves. Both wave averages in the table use the same people. “SD of change” shows how much the change scores vary.
Matching age and gender gives a useful check. It does not prove that a serial number identifies the same person in both waves.
We expect solidarity to increase with more left-wing views, greater financial security, more risk, more trust and better views of public services. We also expect an increase after an adverse event in the last six months.
Each model uses one predictor and accounts for country. We subtract the Wave 1 answer from the Wave 2 answer. For adverse events, we use the Wave 2 timing questions.
A positive estimate means that an increase in the predictor goes with a more positive change in solidarity. A negative estimate means the reverse. We expect positive signs for these predictors, except satisfaction with democracy, which has no separate sign prediction.
Each row shows how many matched pairs the model uses. We do not fill in missing answers. The updated figures and tables show uncertainty separately from sign agreement.
Code
change_vars <-c("economic_left", "cultural_left", "financial_security", "personal_risk","unemployment", "benefit_receipt", "social_trust", "institutional_trust","satdem", "pssatgen", "pssatloc", "service_quality", "service_delivery")for (v in change_vars) { changes[[paste0("delta_", v)]] <- changes[[paste0(v, "_w2")]] - changes[[paste0(v, "_w1")]]}results <-direction_models(changes, "delta_solidarity", c(paste0("delta_", change_vars), "recent_adverse")) %>%mutate(direction =case_when(predictor =="delta_satdem"~"No stated expectation", estimate >0~"Expected", TRUE~"Opposite or zero"))table_labels <-c(delta_economic_left ="Support for income equality and government intervention",delta_cultural_left ="Support for LGBT freedom and less traditional gender roles",delta_financial_security ="How long the household could cover living costs after losing most income",delta_personal_risk ="Expected risk of money shortage, unemployment, inability to work or care duties",delta_unemployment ="Being unemployed and actively looking for work",delta_benefit_receipt ="Number of unemployment, disability, housing or minimum-income benefits received",delta_social_trust ="Trust in other people",delta_institutional_trust ="Trust in political and public institutions",delta_satdem ="Satisfaction with how democracy works",delta_pssatgen ="Overall satisfaction with public services",delta_pssatloc ="Rating of local public services compared with the rest of the country",delta_service_quality ="Quality ratings for healthcare, schools, tax, transport and benefits services",delta_service_delivery ="Fairness, efficiency, prompt help and integrity of public services",recent_adverse ="An adverse life event in the last six months (yes versus no)")read_csv("../../outputs/preliminary_wave2/tables/solidarity_change_models.csv", show_col_types =FALSE) %>%filter(specification =="baseline") %>%transmute(`What changes between waves`=unname(table_labels[predictor]), `Pairs used`= n,Estimate = estimate, SE = std_error, `95% lower`= conf_low, `95% upper`= conf_high,`CI excludes zero`= excludes_zero) %>%kable(digits =3)
What changes between waves
Pairs used
Estimate
SE
95% lower
95% upper
CI excludes zero
Support for income equality and government intervention
1599
0.048
0.029
-0.010
0.106
FALSE
Support for LGBT freedom and less traditional gender roles
1716
-0.030
0.032
-0.092
0.032
FALSE
How long the household could cover living costs after losing most income
1613
-0.001
0.022
-0.043
0.042
FALSE
Expected risk of money shortage, unemployment, inability to work or care duties
1002
0.018
0.045
-0.070
0.107
FALSE
Being unemployed and actively looking for work
1811
0.002
0.096
-0.187
0.191
FALSE
Number of unemployment, disability, housing or minimum-income benefits received
1745
0.017
0.037
-0.055
0.089
FALSE
Trust in other people
1823
0.009
0.007
-0.005
0.023
FALSE
Trust in political and public institutions
1631
0.056
0.022
0.013
0.099
TRUE
Satisfaction with how democracy works
1760
0.040
0.013
0.015
0.065
TRUE
Overall satisfaction with public services
1777
0.044
0.021
0.004
0.085
TRUE
Rating of local public services compared with the rest of the country
1708
0.055
0.025
0.006
0.105
TRUE
Quality ratings for healthcare, schools, tax, transport and benefits services
1122
0.053
0.028
-0.001
0.107
FALSE
Fairness, efficiency, prompt help and integrity of public services
1823
0.068
0.032
0.005
0.130
TRUE
An adverse life event in the last six months (yes versus no)
1772
0.089
0.042
0.006
0.171
TRUE
The rows describe changes from Wave 1 to Wave 2. The last row is different: it compares people with and without an adverse event in the last six months.
For example, the estimate for economic views is 0.048. A one-point increase in the economic-left score goes with this many points of difference in solidarity change, after accounting for country. This is a score difference, not a percentage. The estimates describe associations, not cause and effect.
“Expected” means only that the estimate has the predicted sign. It does not mean that the hypothesis has been confirmed.
Size and direction of each link
Green means the sign matches our expectation. Orange means the opposite. Grey marks satisfaction with democracy, which has no separate stated sign prediction. Each estimate uses a one-unit change on that predictor’s scale. Different scales limit comparisons of bar lengths.
We reverse the social-trust scale so that higher values mean more trust. We also reverse the two attitude items that run in the opposite direction. An index needs all its answers in this overview.
We do not analyse income change here. First, we need to convert each country’s income brackets to amounts of money. A change in bracket number would not measure the registered income change.
Does recent service use make the link stronger?
H6 asks whether the link between service satisfaction and solidarity change is stronger among recent service users. H7 asks the same question for each service.
Each model includes the change in the service rating, recent use, their interaction and country. The interaction measures the difference in the rating-change coefficient between users and non-users. We expect a positive interaction.
A positive number means the rating-change coefficient is more positive among recent users. We expect a positive sign for all six comparisons. The graph shows the interaction column from the table.
A positive interaction agrees with our expectation. It means that the link between rating change and solidarity change is more positive among users. Small groups can make this estimate unstable.
What the numbers show
The next table counts how many estimates point in each direction. These are checks of signs, not tests of the predictions.
The country table shows how much change scores vary. It does not show that welfare-state differences explain that variation.
These models show links between changes. They do not show cause and effect. We leave the full models, income conversion, missing-data replacement and panel weights for the main analysis.
Data source and limits
We use only w2_data/POLITSOLID_W2_Interim_261007.dta, with 2188 records. Collection is still ongoing. We do not read or add either soft-launch file. We match this export to the Wave 1 main field by country and serial number.
The matched sample excludes Germany because its serial numbers are missing. The survey provider still needs to confirm that serial numbers identify the same people across waves. The models use available complete pairs, without imputation or weights. The results below include 95% confidence intervals and unadjusted two-sided tests.
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.
Age and gender robustness
Age and gender use Wave 1 values. These characteristics can predict different changes between people, even though the outcome is a within-person difference. The original model and a baseline using the adjusted model’s complete cases are both retained. This separates adjustment from a change in sample composition. Education is included in a further specification in the exported CSV.
The graph shows the interaction between recent use and the change in service evaluation. This is not an unconditional effect of using a service. The main use coefficient refers to respondents with no change in their service evaluation. All terms are in the CSV.
This extra graph rescales continuous predictor changes to one standard deviation. It does not replace the original estimates. Binary events and unemployment are excluded from this additional comparison.
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.
Baseline and adjusted models side by side
The comparison tables distinguish sample loss from adding controls. The figures below display the original baseline and the age/gender model together.
The table compares whether the primary HC2 interval and the observed-region clustered interval exclude zero. Changes here concern uncertainty, not a change in the estimated association.
Quality ratings for healthcare, schools, tax, transport and benefits services
FALSE
-0.001
0.107
0.002
0.104
TRUE
TRUE
An adverse life event in the last six months (yes versus no)
TRUE
0.006
0.171
-0.003
0.180
FALSE
TRUE
The dynamics registration specifies one-sided tests and 90% intervals for the final models. This update follows the requested 95% intervals and reports two-sided, unadjusted zero tests; it is not presented as the final registered decision rule.
The additional strict age check flags any age change outside 0 or 1 years. It finds four pairs, whereas the earlier absolute-gap-over-one-year check finds three. All flagged pairs remain in this overview.
