Core 2: Conrad data and voting-preference checks

Published

October 15, 2026

Executive summary

Conrad’s efficiency measure gives the same broad, inconclusive pattern as EQI in Germany. Neither measure has a statistically clear association with willingness to contribute in the main matched sample. Perceived governance remains positively associated with willingness to contribute.
Controlling for voting preference does not explain away the main finding. In the six-country check, the EQI coefficient stays near zero. The perceived-governance coefficient changes from 0.381 to 0.407. The separately labelled French bloc check gives the same broad result.
The additional indirect-association and moderation checks remain inconclusive. Adding voting preference does not provide clear support for those proposed relationships.

Main limitation: Conrad’s scores describe German districts around 2014. They measure efficiency, whereas EQI 2024 measures broader governance quality. This is a useful additional check, not a test of equivalence between the measures.

Table notation: n.s. = p ≥ .05; * = p < .05; ** = p < .01. Markers use the unrounded, two-sided p-values from the corresponding analysis. Confidence intervals remain in brackets.

What this checks

This report adds two checks to Core 2.

Does the German local-government efficiency measure supplied by Conrad show a similar association to EQI?
Do the governance associations change after controlling for whether respondents would vote for a non-cabinet party, the prime minister’s electoral option, or another cabinet party?

The main German comparison uses 1,453 respondents. The exact voting check retains about 5,500 respondents across six countries.

All models use the existing Core 2 outcome, survey weights, and demographic controls. The outcome is willingness to contribute across municipal, national, and EU public provision.

Conrad’s data: Germany only

The supplied file accompanies Ziller and Goodman’s Local Government Efficiency and Anti-immigrant Violence. Its main lge measure is a bias-corrected efficiency score for German districts around 2014, using 2013 tax revenues as the input. Higher scores mean greater efficiency. The alternative lge_exp uses expenditure and covers fewer districts.

This is a different construct and an older period than EQI 2024. It is a Germany-only alternative-indicator check, not a replacement for the seven-country analysis. German EQI varies at the state (NUTS1) level; efficiency varies at the district level.

Districts are linked by official five-digit district codes using the Eurostat LAU 2024 correspondence.

Measure	Respondents	Districts
All German respondents	1600	360
Tax-input efficiency	1453	353
Expenditure-input efficiency	953	231

Main efficiency measure: tax input (`lge`)

Dependent variable in every column: respondents’ willingness to contribute to public provision. This combines their municipal, national, and EU contribution answers. The analysis uses German respondents only.

EQI and efficiency are predictors of willingness to contribute. We do not use one governance measure to predict the other. Perceived governance is a third predictor, taken from respondents’ answers in the same PEPSY survey wave as the outcome.

Each column asks a different version of the same question:

EQI alone: do respondents in states with higher EQI report greater willingness to contribute?
Efficiency alone: do respondents in districts with higher Conrad efficiency scores report greater willingness to contribute?
Both external measures: how is each measure associated with willingness when the other is included too?
Both plus perceptions: how are EQI, efficiency, and respondents’ own governance perceptions each associated with willingness when all three are included?

All columns also include the same demographic controls. These are regressions using individual respondents, rather than simple correlations between country averages. Within each table, all four models use the same respondents.

Example: the 0.023 in the first column means that a one-standard-deviation higher EQI is associated with a 0.023-point higher willingness-to-contribute score, after demographic adjustment. It does not describe a change in efficiency. Its interval includes zero, so this association is statistically unclear. In the last column, 0.458 is the association between perceived governance and willingness to contribute, accounting for both external measures and demographics.

Each cell gives the estimate and its 95% confidence interval. A dash means that predictor is not included.

Main measure: tax-input efficiency (1453 respondents; 14 states). Dependent variable: willingness to contribute
Predictor	EQI alone	Efficiency alone	Both external measures	Both plus perceptions
External governance quality (EQI)	0.023 ^n.s. [-0.042, 0.088]	–	0.030 ^n.s. [-0.051, 0.111]	-0.008 ^n.s. [-0.073, 0.056]
Conrad's local-government efficiency	–	-0.011 ^n.s. [-0.080, 0.058]	-0.022 ^n.s. [-0.107, 0.064]	-0.001 ^n.s. [-0.063, 0.061]
Perceived governance quality	–	–	–	0.458 ^** [0.401, 0.515]

Alternative efficiency measure: expenditure input (`lge_exp`)

Dependent variable in every column: willingness to contribute to public provision. The predictors and the four questions are the same as above. The only change is that Conrad’s efficiency predictor now uses spending as its input, rather than tax revenues. EQI and perceived governance remain predictors of willingness to contribute.

This version covers fewer districts, so the table uses a smaller German sample. It does not introduce a new dependent variable or a cross-country comparison.

Alternative measure: expenditure-input efficiency (953 respondents; 8 states). Dependent variable: willingness to contribute
Predictor	EQI alone	Efficiency alone	Both external measures	Both plus perceptions
External governance quality (EQI)	0.062 ^n.s. [-0.020, 0.143]	–	0.079 ^n.s. [-0.007, 0.165]	0.010 ^n.s. [-0.072, 0.093]
Conrad's local-government efficiency	–	-0.034 ^n.s. [-0.129, 0.060]	-0.057 ^n.s. [-0.154, 0.040]	0.002 ^n.s. [-0.075, 0.078]
Perceived governance quality	–	–	–	0.469 ^** [0.381, 0.558]

External measures are standardized separately within each matched German sample. Coefficients represent a one-standard-deviation increase. EQI and efficiency are always compared on the same respondents within each panel. The expenditure panel has a smaller sample, so its coefficients also reflect different coverage. Perceptions retain the original Core 2 standardization.

Confidence intervals are clustered by German state, with a t reference using the number of states minus one degrees of freedom, for all models, because EQI is measured at that level. There are few states, so these intervals should be read cautiously. District-level variation in efficiency does not turn EQI into a district-level measure.

Reading the figure

The outcome is the same in both panels: willingness to contribute to public provision. Tax input and expenditure input are two ways of constructing Conrad’s efficiency predictor. They are not dependent variables. The tax-input version assesses service provision relative to tax revenues. The expenditure-input version assesses it relative to spending. The second version is available for fewer districts. Each panel compares EQI with that version of efficiency on exactly the same respondents.

The figure shows the models that enter each external measure separately. EQI estimates are slightly positive; efficiency estimates are slightly negative. All four confidence intervals include zero. We therefore find no clear positive or negative association in these comparisons. Their opposite signs alone are not convincing evidence of a difference. Overlapping intervals do not establish equivalence, and no formal test of equality between these coefficients is reported.

Voting preference: exact-category check without France

The voting variable is a factor, with non-cabinet parties as the reference. It is never treated as a numerical scale. Other cabinet parties must hold full ministerial posts. Parliamentary support alone does not qualify. Thus Sweden Democrats, Spanish investiture supporters, and Croatian parliamentary support parties remain in the non-cabinet category. Italian Noi Moderati has junior executive posts rather than full ministers and is also coded non-cabinet under this definition. “Opposition” below therefore means non-cabinet, including external supporters.

The survey groups CDU/CSU and Poland’s KO into electoral options. The prime-minister category follows those recorded options and cannot distinguish their constituent parties. Unknown party, other party, refusal, blank vote, non-voting, and missing answers are excluded, not coded as opposition. Early and later questionnaires are recoded separately because several party codes changed.

France is excluded from the primary check because its main questionnaire combines governing parties as Ensemble Citoyens. A separate sensitivity below uses this bloc as an explicitly approximate prime-minister option.

Respondent counts by voting category (before model-specific exclusions)
Country	Opposition / non-cabinet	Prime minister option	Other cabinet party	Excluded / unclassified
Croatia	599	214	18	293
France	694	27	4	390
Germany	814	350	180	256
Italy	585	194	62	276
Poland	534	146	177	148
Spain	433	238	77	306
Sweden	658	150	69	174

Voting categories as percentages of all respondents within each country (unweighted)
Country	Opposition / non-cabinet	Prime minister option	Other cabinet party	Excluded / unclassified
Croatia	53.3%	19.0%	1.6%	26.1%
France	62.2%	2.4%	0.4%	35.0%
Germany	50.9%	21.9%	11.2%	16.0%
Italy	52.4%	17.4%	5.6%	24.7%
Poland	53.1%	14.5%	17.6%	14.7%
Spain	41.1%	22.6%	7.3%	29.0%
Sweden	62.6%	14.3%	6.6%	16.6%

Each row is a country. In the first table, each cell is the number of respondents. The second table shows percentages of all respondents in that country, including those excluded from the voting models. These are unweighted sample descriptions, so they show the composition of the analysis file rather than population voting intentions. The France row describes the approximate bloc coding used only in the separate sensitivity. The excluded column contains respondents without a classifiable named-party answer.

Does adding voting preference change the governance findings?

Dependent variable in every column: willingness to contribute to public provision.

Read across a predictor row to compare models. Read down within a model to see which predictors it includes. Each result cell shows before and after adding the voting categories. The numbers in brackets are 95% confidence intervals. A dash means the predictor is not included in that model.

Both versions use exactly the same respondents within each model. This matters because excluding respondents with unclassifiable voting answers already changes the sample. Comparing the two versions isolates what changes when voting preference is added. Country differences, weights, demographic controls, and predictor scales are handled in the same way in both versions.

The first table covers the main governance associations. The second covers interactions and change over time. An interaction asks whether an association becomes stronger or weaker as another characteristic changes. The joint model includes EQI and perceptions together. It does not by itself establish mediation. Welfare-state interaction intervals in this overview use regional clustering; the more appropriate country-level check appears below.

Main governance associations. Dependent variable: willingness to contribute
Predictor	EQI only	Perceptions only	Two perception components	EQI + perceptions
External governance quality (EQI)	Before: 0.004 ^n.s. [-0.083, 0.091] After: 0.005 ^n.s. [-0.081, 0.091]	–	–	Before: -0.014 ^n.s. [-0.095, 0.067] After: -0.014 ^n.s. [-0.093, 0.064]
Perceived governance quality	–	Before: 0.381 [0.334, 0.428] After: 0.407 [0.361, 0.452]	–	Before: 0.381 [0.334, 0.428] After: 0.407 [0.362, 0.452]
Public-service evaluations	–	–	Before: 0.208 [0.155, 0.261] After: 0.213 [0.160, 0.265]	–
Institutional confidence	–	–	Before: 0.207 [0.151, 0.263] After: 0.229 [0.173, 0.286]	–
Respondents	5496	5498	5498	5496

Interactions and governance change. Dependent variable: willingness to contribute
Predictor	Welfare-state interaction	Ideology interaction	Governance change
External governance quality (EQI)	Before: 0.003 ^n.s. [-0.085, 0.090] After: 0.003 ^n.s. [-0.084, 0.090]	–	–
EQI × welfare-state effort	Before: 0.032 ^n.s. [-0.102, 0.167] After: 0.038 ^n.s. [-0.095, 0.172]	–	–
Perceived governance quality	–	Before: 0.348 [0.314, 0.383] After: 0.355 [0.318, 0.393]	–
Right-wing ideology	–	Before: -0.253 [-0.290, -0.217] After: -0.242 [-0.277, -0.208]	–
Perceived governance × right-wing ideology	–	Before: -0.006 ^n.s. [-0.031, 0.019] After: -0.004 ^n.s. [-0.029, 0.021]	–
Baseline EQI (2021)	–	–	Before: 0.087 ^n.s. [-0.018, 0.192] After: 0.086 ^n.s. [-0.019, 0.191]
Change in EQI (2021–2024)	–	–	Before: -0.021 ^n.s. [-0.060, 0.018] After: -0.020 ^n.s. [-0.059, 0.018]
Respondents	5496	5498	4946

Optional: do voting groups themselves differ?

This is a different question from the main robustness check. The main check asks whether adding voting preference changes the governance coefficients. This small table instead shows the voting-category coefficients from the joint EQI-and-perceptions model.

Compared with respondents choosing non-cabinet parties, respondents choosing the prime minister’s electoral option score about 0.30 points lower on willingness to contribute, after the model’s other adjustments. The other-cabinet-party difference is smaller and statistically unclear. These are adjusted group differences, not effects of voting for a party. This table is supplementary and is not needed to read the main result.

Voting option (compared with non-cabinet parties)	Adjusted difference
Prime minister’s electoral option	-0.302 ^** [-0.407, -0.198]
Other cabinet party	-0.062 ^n.s. [-0.194, 0.069]

Indirect-association check

The proposed chain is better external governance → better governance perceptions → greater willingness to contribute. We ask whether this indirect association remains after accounting for voting preference. We separate average perceptions in a region from individual deviations around that average. This keeps the proposed regional link distinct from the association between individual perceptions and willingness to contribute.

The main table below shows the indirect association only. Rows identify the sample. Columns compare the model before and after adding voting preference to both parts of the chain. Every interval includes zero, so neither version gives clear evidence of the proposed indirect association.

Intervals use the paper’s existing approximation: 50,000 independent normal draws for the two path coefficients. This is not a resampling bootstrap, and it does not establish a causal mechanism.

Sample	Before voting adjustment	After voting adjustment
Six countries	0.020 ^n.s. [-0.022, 0.065]	0.021 ^n.s. [-0.023, 0.068]
Seven countries, French bloc	0.020 ^n.s. [-0.020, 0.062]	0.021 ^n.s. [-0.020, 0.065]

Optional: the separate links in the proposed chain

EQI → perceptions: whether people in better-governed regions report better perceptions.
Regional perceptions → willingness: the link between a region’s average perceptions and willingness to contribute, accounting for EQI and individual deviations.
Individual perceptions → willingness: the link between an individual’s perceptions and willingness, relative to their region’s average.
Remaining EQI association: the EQI association after those perception measures are included.

These rows describe different links. They should not be read as interchangeable estimates of one relationship.

Sample	Link	Before voting adjustment	After voting adjustment
Six countries	EQI → perceptions	0.048 ^n.s. [-0.053, 0.149]	0.047 ^n.s. [-0.052, 0.146]
Six countries	Regional perceptions → willingness	0.411 ^** [0.256, 0.566]	0.446 ^** [0.296, 0.596]
Six countries	Individual perceptions → willingness	0.380 ^** [0.333, 0.427]	0.405 ^** [0.360, 0.450]
Six countries	Remaining EQI association	-0.016 ^n.s. [-0.098, 0.066]	-0.017 ^n.s. [-0.097, 0.062]
Seven countries, French bloc	EQI → perceptions	0.047 ^n.s. [-0.048, 0.142]	0.048 ^n.s. [-0.044, 0.140]
Seven countries, French bloc	Regional perceptions → willingness	0.413 ^** [0.267, 0.559]	0.443 ^** [0.301, 0.585]
Seven countries, French bloc	Individual perceptions → willingness	0.385 ^** [0.342, 0.427]	0.407 ^** [0.366, 0.448]
Seven countries, French bloc	Remaining EQI association	-0.010 ^n.s. [-0.089, 0.070]	-0.011 ^n.s. [-0.088, 0.066]

Welfare-state interaction: country-level uncertainty

This reruns the paper’s country-level wild cluster bootstrap with Webb weights and 9,999 draws. The reported method column identifies any fallback. Six-country fallback intervals use five degrees of freedom; seven-country intervals use six. Few countries limit precision.

Sample	Method	Countries	b	95% CI	p
Six countries: before adjustment	Wild cluster bootstrap, Webb weights	6	0.032 n.s.	[-0.236, 0.136]	0.889
Six countries: vote adjusted	Wild cluster bootstrap, Webb weights	6	0.038 n.s.	[-0.229, 0.136]	0.864
Seven countries, bloc: before adjustment	Wild cluster bootstrap, Webb weights	7	0.029 n.s.	[-0.164, 0.122]	0.803
Seven countries, bloc: vote adjusted	Wild cluster bootstrap, Webb weights	7	0.032 n.s.	[-0.150, 0.123]	0.796

France: separate bloc sensitivity

This repeats the comparison with France included. Ensemble Citoyens is treated as the prime-minister bloc in the main questionnaire; Renaissance is used in the early questionnaire. Early MoDem and Horizons are other cabinet options. These are proxies, not exact identification of the prime minister’s party. Les Républicains is treated as non-cabinet because the party refused participation, despite individual appointments. This ambiguity is a reason to keep France out of the primary check.

Main governance associations. Dependent variable: willingness to contribute
Predictor	EQI only	Perceptions only	Two perception components	EQI + perceptions
External governance quality (EQI)	Before: 0.010 ^n.s. [-0.075, 0.096] After: 0.011 ^n.s. [-0.074, 0.095]	–	–	Before: -0.008 ^n.s. [-0.087, 0.071] After: -0.009 ^n.s. [-0.085, 0.068]
Perceived governance quality	–	Before: 0.386 [0.343, 0.428] After: 0.408 [0.368, 0.449]	–	Before: 0.386 [0.343, 0.429] After: 0.408 [0.368, 0.449]
Public-service evaluations	–	–	Before: 0.211 [0.164, 0.259] After: 0.215 [0.168, 0.262]	–
Institutional confidence	–	–	Before: 0.208 [0.158, 0.259] After: 0.229 [0.177, 0.280]	–
Respondents	6221	6223	6223	6221

Interactions and governance change. Dependent variable: willingness to contribute
Predictor	Welfare-state interaction	Ideology interaction	Governance change
External governance quality (EQI)	Before: 0.006 ^n.s. [-0.079, 0.091] After: 0.006 ^n.s. [-0.079, 0.090]	–	–
EQI × welfare-state effort	Before: 0.029 ^n.s. [-0.089, 0.147] After: 0.032 ^n.s. [-0.086, 0.150]	–	–
Perceived governance quality	–	Before: 0.343 [0.311, 0.375] After: 0.348 [0.313, 0.383]	–
Right-wing ideology	–	Before: -0.267 [-0.301, -0.233] After: -0.261 [-0.294, -0.228]	–
Perceived governance × right-wing ideology	–	Before: 0.006 ^n.s. [-0.018, 0.030] After: 0.007 ^n.s. [-0.016, 0.031]	–
Baseline EQI (2021)	–	–	Before: 0.090 ^n.s. [-0.012, 0.192] After: 0.089 ^n.s. [-0.012, 0.191]
Change in EQI (2021–2024)	–	–	Before: -0.022 ^n.s. [-0.060, 0.017] After: -0.021 ^n.s. [-0.059, 0.017]
Respondents	6221	6223	5671

Variable names: reference only

The main tables use readable labels. This lookup connects them to the analysis code. _z means standardized: a one-unit increase is one standard deviation. _imp identifies a predictor completed using the existing Core 2 procedure. Voting preference is not imputed. German external measures are standardized within each matched German sample; the remaining predictors keep the original Core 2 scales.

Label used in tables	Variable in the code
External governance quality (EQI)	eqi_germany_z
External governance quality (EQI)	regional_governance_quality_z
Conrad’s local-government efficiency	efficiency_z
Perceived governance quality	perceived_governance_index_imp_z
Public-service evaluations	service_quality_index_imp_z
Institutional confidence	institutional_trust_index_imp_z
EQI × welfare-state effort	regional_governance_quality_z:welfare_state_size_z
Right-wing ideology	ideology_right_imp_z
Perceived governance × right-wing ideology	perceived_governance_index_imp_z:ideology_right_imp_z
Baseline EQI (2021)	eqi_2021_z
Change in EQI (2021–2024)	regional_governance_change_z

Perceived governance combines public-service evaluations and institutional confidence. Public-service evaluations cover general satisfaction, six service domains, and fairness, efficiency, prompt help, and integrity. Institutional confidence covers parliament, courts, local and national government, the European Commission, and satisfaction with democracy. Welfare-state effort is the paper’s country-level welfare-state-size measure. Right-wing ideology is the respondent’s ideological position. The governance-change model controls for EQI in 2021 while examining change from 2021 to 2024.

Coding audit and sources

The following table lists every named cabinet-coded response option. The local coding CSV also records all non-cabinet and excluded response labels. Cabinet membership is defined for the survey period, not by a party’s later status.

country_code	instrument	party	vote_group
HR	pilot	DOMOVINSKI POKRET – DP	Other cabinet party
HR	pilot	HRVATSKA DEMOKRATSKA ZAJEDNICA – HDZ	Prime minister option
HR	main	HRVATSKA DEMOKRATSKA ZAJEDNICA – HDZ	Prime minister option
HR	main	DOMOVINSKI POKRET – DP	Other cabinet party
DE	pilot	Christlich Demokratische Union Deutschlands/ Christlich Soziale Union (CDU/CSU)	Prime minister option
DE	pilot	Sozialdemokratische Partei Deutschlands (SPD)	Other cabinet party
DE	main	Christlich Demokratische Union Deutschlands/ Christlich Soziale Union (CDU/CSU)	Prime minister option
DE	main	Sozialdemokratische Partei Deutschlands (SPD)	Other cabinet party
IT	pilot	Fratelli d’Italia	Prime minister option
IT	pilot	Lega	Other cabinet party
IT	pilot	Forza Italia	Other cabinet party
IT	main	Fratelli d’Italia	Prime minister option
IT	main	Forza Italia	Other cabinet party
IT	main	Lega	Other cabinet party
PL	pilot	Koalicja Obywatelska	Prime minister option
PL	pilot	Nowa Lewica	Other cabinet party
PL	pilot	Polska 2050	Other cabinet party
PL	pilot	Polskie Stronnictwo Ludowe	Other cabinet party
PL	main	Koalicja Obywatelska	Prime minister option
PL	main	Nowa Lewica	Other cabinet party
PL	main	Polskie Stronnictwo Ludowe	Other cabinet party
PL	main	Polska 2050	Other cabinet party
ES	pilot	PSOE	Prime minister option
ES	pilot	SUMAR	Other cabinet party
ES	main	PSOE	Prime minister option
ES	main	SUMAR	Other cabinet party
SE	pilot	Moderaterna	Prime minister option
SE	pilot	Kristdemokraterna	Other cabinet party
SE	pilot	Liberalerna	Other cabinet party
SE	main	Moderaterna	Prime minister option
SE	main	Kristdemokraterna	Other cabinet party
SE	main	Liberalerna	Other cabinet party
FR	pilot	MoDem	Other cabinet party
FR	pilot	Renaissance	Prime minister option
FR	pilot	Horizons	Other cabinet party
FR	main	Ensemble Citoyens	Prime minister option

Government sources used to check the coding: Germany’s CDU/CSU–SPD coalition, Sweden’s three cabinet parties, Spanish government press conference, March 2026, Polish coalition agreement, Italian ministers, Croatian government, and French cabinet.

Reproducibility

Open this QMD inside the existing project in RStudio and click Render. It reads the same survey and Core 2 preparation functions as the paper. The shared H3 helper accepts an optional additional control; the H4 helper uses the actual country count for its fallback degrees of freedom. Existing seven-country defaults are preserved. Aggregate coefficient tables and the explicit party coding are written to outputs/core2/additional_robustness/.

File	MD5
POLITSOLID_Hauptfeld_weighted_18052026.dta	fa84df48a4022b0da434cdd57c2a1d38
ZG_JOP_study1 (1).dta	03d546daf89c65bab4ac2d2f9d4bacaf
eurostat_ags_nuts3_2024.csv	9509f1f1494fb34f75bf7e58dcda88c7