## 13. Logit Regression on Support for Temporary Levy, Robustness Checks

## Source details

**Canonical URL:** [13. Logit Regression on Support for Temporary Levy, Robustness Checks](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf.json)

---

### Objectives and context
- Survey objective: elicit people’s views on taxation (temporary levy, temporary levy labels, progressive taxation, revenue-neutral reforms) and spending priorities in the context of the COVID-19 pandemic; relate attitudes to personal pandemic experiences, moral perspectives, and demographics.
- Sample: "2,519" US resident adults surveyed October 9–23, 2020 via Dynata using standard census quotas.

### Survey design and sample characteristics
- Randomized label variation: one of six terms randomly assigned (Tax; Solidarity Tax; COVID-19 Recovery Tax; Contribution; Solidarity Contribution; COVID-19 Recovery Contribution), roughly "420" questionnaires per label ("419" for “Tax”).
- Geographic matching: county-level urban influence codes (US Department of Agriculture) and COVID-19 cases/deaths (USAFACTS) via ZIP-code bridging files; urbanity code matched for all but "38" observations.
- Key demographics and descriptive statistics (exact values preserved):
  - Median age: "43" years; Mean Age: "44.6"; Standard deviation: "16.6"; Min: "18.0"; Lower quartile: "32.0"; Median: "43.0"; Upper quartile: "56.0"; Max: "99.0".
  - Number of children mean: "1.1"; Standard deviation: "1.3"; Min: "0.0"; Lower quartile: "0.0"; Median: "1.0"; Upper quartile: "2.0"; Max: "9.0".
  - Number of adults in household mean: "2.1"; Standard deviation: "1.3"; Min: "1.0"; Lower quartile: "1.0"; Median: "2.0"; Upper quartile: "2.0"; Max: "19.0".
  - COVID-19 cases in percent of county population, 10/8/20 mean: "2.3"; Standard deviation: "1.0"; Min: "0.1"; Lower quartile: "1.5"; Median: "2.2"; Upper quartile: "2.9"; Max: "12.3".
  - COVID-19 deaths in percent of county population, 10/8/20 mean: "0.1"; Standard deviation: "0.1"; Min: "0.0"; Lower quartile: "0.0"; Median: "0.1"; Upper quartile: "0.1"; Max: "0.4".
- Urbanity (sample vs US aggregates):
  - Large metropolitan: "59 percent" (sample) vs "55 percent" (US).
  - Small metropolitan: "29 percent" vs "30 percent".
  - Micropolitan: "7 percent" vs "8.8 percent".
  - Noncore: "5 percent" vs "6.2 percent".
- Education (sample vs US aggregates):
  - High school completed: "93 percent" vs "88 percent".
  - College completed: "57 percent" vs "42 percent".
  - Graduate school: "21 percent" vs "12 percent".
- Household income (sample vs US aggregates):
  - Up to $20,000: "23 percent" vs "13 percent".
  - Up to $50,000: "51 percent" vs "37 percent".
  - Above $150,000: "9 percent" vs "19 percent".
  - Above $100,000: "20 percent" vs "34 percent".
- Data quality note: some respondents reported age "99" (survey maximum); authors flag potential misreporting concentrated on age.

### Key findings on taxation and spending preferences (exact percentages preserved)
- Support for temporary levy (Question 1, pooled):
  - "62 percent" support a temporary levy to cover pandemic costs.
  - Label “COVID-19 Recovery Contribution” approval: "67 percent" (only term significantly different; p-value "2 percent").
- Preferred tax base for temporary levy (Question 2, conditional on accepting a temporary levy):
  - Majority prefer a progressive tax base over a proportional tax on everyone.
  - Respondents favored taxes on incomes or profits that are high regardless of underlying causes rather than pandemic-specific gains.
- Support for permanent progressive tax increase (Question 3):
  - "68 percent" support some permanent increase in taxes raised only on those who are well off.
  - Support by use: highest for social spending; around "25 percent" support for climate change mitigation; about one third support for infrastructure, education, local environment, and debt reduction (exact shares in source figures).
- Support for revenue-neutral reform (Question 4):
  - "72 percent" support raising taxes on the rich while reducing taxes for others so total revenue is unchanged.
  - Preference: raise taxes on the top decile more often than on the top percentile; among "963" people who supported a tax on the top decile, "420" did not select any other options.
- General tax sentiment (Question 5):
  - "56 percent" consider taxes to be too high; "33 percent" about right; "11 percent" too low.
  - Majority believe the rich do not pay their fair share; "40 percent" worry about the national debt.
- Spending priorities (Question 6):
  - On average favor more spending on health, education, and infrastructure.
  - Largest shares favoring cuts: border protection, the military, and the police.
- Perceived groups suffering most (Question 7):
  - Almost half identified poor people as those who suffered the most; second most frequent choice was old people.
- Personal impacts of COVID-19 (Question 8):
  - "45 percent" reported no direct impact on their household.
  - Employment loss reported by over "20 percent".
  - Illness reported by "9 percent".

### Regression analysis — methods and selected results
- Methods:
  - Binary responses analyzed via logit regressions (probit yielded extremely similar results); linear probability models also similar. Robust standard errors reported. Significance: "*** p<0.01, ** p<0.05, * p<0.1."
  - Observations: "2,519" in most models; "2,481" in model including urban indicator.
- Main marginal effects and selected coefficient signs (Table 2 / Table 8):
  - Education (1–4): marginal effect of completing one more level raises support by between "2 and 3 percentage points"; coefficient: "0.15*** (0.06)" in (1).
  - Has children: marginal effect between "6 and 7 percent"; coefficient: "0.33*** (0.09)" in (1).
  - Household size (equivalized): increases support by "2 percent" at the margin; coefficient: "0.08*** (0.03)" in (1).
  - Older half: associated with marginal reduction in support by "4-7 percentage points"; coefficient: "-0.31*** (0.09)" in (1).
  - Black/African American: marginal effect "6-8 percentage points"; coefficient: "0.35** (0.14)" in (1).
  - Ethnicity withheld: marginal effect of "-49 to -51 percentage points"; coefficient: "-2.21*** (0.35)" in (1).
  - Income effects:
    - Income surpassing "$20,000": marginal effect "7-8 percentage points"; coefficient: "0.34*** (0.12)" in (1).
    - Income > $50k: coefficient "-0.16 (0.11)" in (1).
    - Income > $150k: support falls significantly; coefficient: "-0.37** (0.16)" in (1).
  - Suffered (household employment loss or critical illness): raises support by "15 percentage points"; coefficient: "0.68*** (0.09)" in (3).
  - Older half × suffered interaction: not significant; coefficient: "-0.02 (0.18)" in (4).
- Table 2 model observations: "2,519" in (1), (3), (4); "2,481" in (2).

### Moral perspectives, PCA, and interaction with suffering
- Moral-perspective definitions (based on spending-preference responses):
  - Universalist: supports at least one of higher health, education or climate spending; does not ask for cuts in any of these three; wants to reduce at least one of police, military, or border protection.
  - Communitarian: supports at least one of higher police, military, or border protection spending; does not ask for cuts in any of these three; wants to reduce at least one of health, education, or climate spending.
  - Moderate: anyone not meeting other two labels.
- Resulting shares:
  - Universalists: "almost 50 percent".
  - Communitarians: "21 percent".
- PCA (Table 11) key outputs:
  - Number of obs = "2,519"; Component eigenvalues: Comp1 "2.40073" (Proportion "0.3001"), Comp2 "1.64689" (Proportion "0.2059"), Comp3 "0.8646" (Proportion "0.1081"); cumulative up to Comp8 "1.0000".
  - Selected loadings (Comp1): Health Cuts "0.3620"; Education Cuts "0.3844"; Climate change cuts "0.4709"; Police cuts "-0.2648"; Military cuts "-0.2856"; Border protection cuts "-0.3411"; Infrastructure cuts "0.1717".
- Main empirical findings using moral perspectives (Annex II, Table 9 and related):
  - Universalists: marginal effect of "26 percentage points" higher likelihood to support a temporary levy.
  - Interaction with household suffering:
    - Universalists: additional marginal effect "7 percentage points".
    - Communitarians: those with household suffering increase support by "22 percentage points".
    - Moderates: support rises by "14 percentage points".
  - Change among those who could change their mind (ceiling adjustment):
    - "26.3 percent" of unaffected universalists do not support a temporary levy; "63.8 percent" of unaffected communitarians do not support it.
    - Estimated shifts: "22 percent" of universalists otherwise opposed would support if exposed; "42 percent" of communitarians otherwise opposed would support if exposed; p-value for difference "8.8 percent".
- Robustness with PCA and alternative definitions:
  - Using PCA score: likelihood of supporting a temporary levy declines with communitarian (PCA) preferences; interaction with suffered reduces opposition (Communitarian (PCA) and suffered: "0.12**").

### Robustness checks on exposure, labeling, tax base, and uses
- Exposure measurement robustness:
  - Replacing subjective suffering with objective county incidence of cases or deaths preserves key results.
  - Universalists more likely to claim economic suffering or to have a critically ill acquaintance; not more likely to report illness in own household controlling for county incidence.
- Labeling (Table 4):
  - Terms generally do not differ significantly controlling for covariates.
  - Exception: “COVID-19 Recovery Contribution” marginal effect "5 percentage points" (p-value "3.5 percent").
- Tax base differentiation (Table 5):
  - Universalists show higher support for every progressive reform.
  - Communitarians show greater increases in support following COVID-19 exposure for many bases.
  - Exception: support for higher taxes on multinational enterprises (MNEs) does not rise following COVID-19 (except among moderates).
- Use-of-funds differentiation (Table 6) and revenue-neutral reforms (Table 7):
  - Universalists more supportive across spending uses; communitarians increase support more after suffering.
  - Specific heterogeneities:
    - Women less likely to support taxes restricted to COVID-19-related high income vs. general high incomes; less likely to support progressive reforms funding social spending or infrastructure; less likely to support redistributive reforms raising taxes only on the top percentile or decile.
    - People above median age more likely to favor tax increases on profits or multinationals.
    - People with children less likely to support revenue increases for fighting climate change.
  - Universalists significantly more likely to support revenue-neutral reforms raising taxes on Any, Top 1%, Top 10%, Top third, Top half, MNEs; communitarians who suffered increase support across many reform types.

### Data quality, puzzling responses, and sample notes
- County-level ratio note: "2.3 percent in the sample (average of county-level ratios over all respondents for which such data could be matched based on ZIP codes)."
- Inconsistencies and puzzling responses:
  - About "10 percent" of households saw their income rise during the pandemic.
  - Out of "1,138" respondents claiming no direct household effect: "25" reported employment loss, "24" reported critical COVID-19 illness in their household, "13" reported both.
  - Of "36" respondents who claimed no direct impact while facing employment loss or critical illness: "9" also picked option that everyone had suffered equally while singling out at least one category as having suffered most.
  - None of these puzzling respondents claimed age "99".
  - Excluding "55" respondents claiming to be "99" or providing contradictory answers did not affect results significantly.
- Sample caveats: potential biases (higher education, lower reported household incomes); authors focus on correlations and interactions rather than simple majority-rule conclusions.

### Conclusion and policy implications
- Main conclusion: COVID-19 exposure (household job loss, illness, or knowing someone affected) is associated with higher support for progressive taxation.
- Distributional nuance:
  - Universalists more likely to support temporary levies overall.
  - Communitarians who experienced suffering display larger increases in support, including when measured as percent of those who could change their minds.
- Policy stakes and recommendations:
  - Individuals harmed by the pandemic currently demand more redistributive policies.
  - If unmet, affected individuals may grow disillusioned and lose trust in government; unmet demand may morph into social unrest or embrace of simplistic policy solutions.
  - Policymakers should deliver on health and on policies that foster a more equal distribution of incomes and access to government services.
  - Policymakers should devote extra attention to political economy factors and gauge support for policy choices from various population segments.
- Authors’ calls for further research:
  - Strengthen causal identification.
  - Replicate in other countries to test hypothesis that those bearing the brunt of the pandemic become more favorable to progressive taxation.
  - Investigate persistence of attitude changes and potential entrenchment of oppositional views.
  - Consider simultaneous factors changing attitudes, including evidence that epidemics weaken trust in government.

*Source: wpiea2021024-print-pdf - https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf*

### References________________________________________________________________________________________________41

### References

### Figures
- 1. Description of Sample _________________________________________________________________________________ 13
- 2. Responses to Question 1______________________________________________________________________________ 14
- 3. Responses to Question 2______________________________________________________________________________ 14
- 4. Responses to Question 3______________________________________________________________________________ 15
- 5. Responses to Question 4______________________________________________________________________________ 15
- 6. Responses to Question 5______________________________________________________________________________ 16
- 7. Responses to Question 6______________________________________________________________________________ 17
- 8. Responses to Question 7______________________________________________________________________________ 17
- 9. Responses to Question 8______________________________________________________________________________ 18

### Tables
- 1. Descriptive Sample Statistics _________________________________________________________________________ 12
- 2. Logit Regression on Support of any Temporary Levy ______________________________________________ 21
- 3. Support for Temporary Levy with Estimated Moral Perspectives __________________________________ 25
- 4. Logit Regression on Support of a Temporary Levy Dependent on Labelling_____________________ 26
- 5. Logit Regression on Supported Tax Base, Assuming Introduction ________________________________ 28
- 6. Logit Regression: Support for Permanent Progressive Tax Hike, Depending on Use ____________ 29 
- 7. Logit Regression: Support for Revenue-Neutral Reform, Depending on Who Would Pay  
  More ____________________________________________________________________________________________________ 30
- 8. Breakdown of Variables Used in Table 2_____________________________________________________________ 36
- 9. Logit Regression with Breakdown of Spending Preferences _______________________________________ 37
- 10. Logit Regressions to Assess Plausibility of Spending-Based Moral Perspectives _______________ 38
- 11. Principal Component Analysis of Spending Preferences__________________________________________ 39
- 12. Logit Regression of Suffering on Moral Perspectives _____________________________________________ 39

*Source: wpiea2021024-print-pdf - References________________________________________________________________________________________________41*

### 13. Logit Regression on Support for Temporary Levy, Robustness Checks _________________________ 40

### 13. Logit Regression on Support for Temporary Levy, Robustness Checks

### Objectives and context
- Survey objective: elicit people’s views on taxation (temporary levy, temporary levy labels, progressive taxation, revenue-neutral reforms) and spending priorities in the context of the COVID-19 pandemic; relate attitudes to personal pandemic experiences, moral perspectives, and demographics.
- Sample: 2,519 US resident adults surveyed October 9–23, 2020 via Dynata using standard census quotas.

### Survey design and data sources
- Randomized label variation for the temporary post-pandemic levy: one of six terms randomly assigned (Tax; Solidarity Tax; COVID-19 Recovery Tax; Contribution; Solidarity Contribution; COVID-19 Recovery Contribution), roughly 420 questionnaires per label (419 for “Tax”).
- Geographic and exposure variables: matched county-level urban influence codes (US Department of Agriculture) and COVID-19 cases/deaths (USAFACTS) via ZIP-code bridging files; urbanity code matched for all but 38 observations.
- Survey structure: demographic screener → introductory paragraph → main questions on temporary levy and progressive tax reforms → spending preferences (elicits moral perspectives) → personal COVID-19 impacts → background (education, ethnicity, household size).

### Sample characteristics and descriptive statistics
- Demographics (selected exact statistics preserved):
  - Median age: 43 years (sample median noted as 43); mean Age: 44.6; Standard deviation: 16.6; Min: 18.0; Lower quartile: 32.0; Median: 43.0; Upper quartile: 56.0; Max: 99.0.
  - Number of children mean: 1.1; Standard deviation: 1.3; Min: 0.0; Lower quartile: 0.0; Median: 1.0; Upper quartile: 2.0; Max: 9.0.
  - Number of adults in household mean: 2.1; Standard deviation: 1.3; Min: 1.0; Lower quartile: 1.0; Median: 2.0; Upper quartile: 2.0; Max: 19.0.
  - Number of children in household mean: 0.6; Standard deviation: 1.2; Min: 0.0; Lower quartile: 0.0; Median: 0.0; Upper quartile: 1.0; Max: 19.0.
  - COVID-19 cases in percent of county population, 10/8/20 mean: 2.3; Standard deviation: 1.0; Min: 0.1; Lower quartile: 1.5; Median: 2.2; Upper quartile: 2.9; Max: 12.3.
  - COVID-19 deaths in percent of county population, 10/8/20 mean: 0.1; Standard deviation: 0.1; Min: 0.0; Lower quartile: 0.0; Median: 0.1; Upper quartile: 0.1; Max: 0.4.
- Urbanity (sample vs US aggregates):
  - 59 percent live in large metropolitan areas (sample) vs 55 percent (US).
  - 29 percent live in small metropolitan areas (sample) vs 30 percent (US).
  - 7 percent live in micropolitan areas (sample) vs 8.8 percent (US).
  - 5 percent live in noncore areas (sample) vs 6.2 percent (US).
- Education (sample vs US aggregates):
  - 93 percent have completed high school (sample) vs 88 percent (US).
  - 57 percent have completed college (sample) vs 42 percent (US).
  - 21 percent graduate school (sample) vs 12 percent (US).
- Household income (sample vs US aggregates):
  - 23 percent report household income up to $20,000 (sample) vs 13 percent (US).
  - 51 percent report incomes up to $50,000 (sample) vs 37 percent (US).
  - 9 percent report incomes above $150,000 (sample) vs 19 percent (US).
  - 20 percent report incomes above $100,000 (sample) vs 34 percent (US).
- Data quality note: some respondents reported age 99 (survey maximum); authors flag potential misreporting concentrated on age rather than other variables.

### Key findings on taxation and spending preferences (exact percentages preserved)
- Support for temporary levy (Question 1, pooled across labels):
  - 62 percent of respondents support a temporary levy to cover pandemic costs, with little variation across label terms.
  - The label “COVID-19 Recovery Contribution” had the highest approval rating at 67 percent and is the only term significantly different from the average (p-value of 2 percent).
- Preferred tax base for temporary levy (Question 2, conditional on accepting a temporary levy):
  - The great majority preferred a progressive tax base over a proportional tax on everyone.
  - Respondents more often supported taxes on incomes or profits that are high regardless of underlying causes, rather than pandemic-specific gains.
- Support for permanent progressive tax increase (Question 3):
  - 68 percent of respondents support some permanent increase in taxes raised only on those who are well off.
  - Support varies by intended revenue use: highest for social spending; around 25 percent support for climate change mitigation; about one third support for infrastructure, education, local environment, and debt reduction. (Exact spends-by-use shares are reported in source figures.)
- Support for revenue-neutral reform raising taxes on the rich while reducing taxes for others (Question 4):
  - 72 percent of respondents support such a tax law change.
  - Respondents favored raising taxes on the top decile more often than on the top percentile; out of 963 people who supported a tax on the top decile, 420 did not select any other options.
- General tax sentiment and beliefs (Question 5):
  - 56 percent considered taxes to be too high; 33 percent considered them about right; 11 percent considered them too low.
  - Majority believe the rich do not pay their fair share; 40 percent worry about the national debt; minorities believe taxes are squandered or that benefits discourage work (exact shares shown in the source figure).
- Spending priorities and elicited moral perspectives (Question 6):
  - Respondents on average favor more spending on health, education, and infrastructure.
  - Areas with largest shares favoring cuts: border protection, the military, and the police.
- Perceived groups that suffered most (Question 7):
  - Almost half of respondents identified poor people as those who suffered the most.
  - The second most frequent choice was old people.
- Personal impacts of COVID-19 (Question 8):
  - 45 percent reported no direct impact on their household.
  - Employment loss reported by over 20 percent of respondents.
  - Illness reported by 9 percent of respondents.

### Behavioral and heterogeneity insights (preserved qualitative findings)
- Moral perspectives:
  - Attitudes toward taxation, and progressive taxation in particular, are more favorable among individuals classified as “universalist” versus “communitarian” based on expenditure preferences.
  - Universalist/communitarian distinctions were elicited via spending-priority questions (health, education, police, military, border protection, climate change, local environment, debt reduction, infrastructure).
- Personal pandemic experience:
  - Respondents who experienced serious illness or job loss due to COVID-19, or know someone who has, favor progressive taxation more than others; this impact is stronger among respondents with communitarian perspectives.
- Labeling and framing:
  - Support for a temporary levy is relatively insensitive to variation in labels, but there is a marginal statistically significant preference for the term “COVID-19 Recovery Contribution.”
- Interpretation and caveats:
  - High aggregate support for tax increases appears conditional on progressivity (targeting the well off), which coexists with a plurality perceiving taxes as “too high.”
  - Some inconsistent response combinations (e.g., selecting “everyone suffered about equally” plus specific groups) may reflect misunderstanding or nuanced views; survey enforces at least one category cut in spending question to impose a minimal budget constraint.
  - Authors note potential sample biases (higher education, lower reported household incomes) and typical survey participation caveats; they focus on correlations and interactions rather than simple majority rule conclusions.

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf*

### 2.3 percent in the  sample (average of

### wpiea2021024-print-pdf - 2.3 percent in the  sample (average of

### Data quality, puzzling responses, and sample notes
- County-level ratios matched by ZIP codes show "2.3 percent in the  sample (average of county-level ratios over all respondents for which such data could be matched based on ZIP codes)."
- Possible explanations for discrepancies between reported illness and matched data:
  - Respondents who experienced symptoms but did not undertake a test may still report the illness as COVID-19.
  - A positive household-level response may reflect one case in the household and respondents may have interpreted "household" broadly.
- About "10 percent of households saw their income rise during the pandemic."
- Out of "1,138 respondents claiming that their household had not been directly affected by the pandemic":
  - "25 reported employment loss,"
  - "24 reported critical COVID-19 illness in their household,"
  - "13 reported both."
- Of the "36 respondents who claimed no direct impact of the crisis, while facing employment loss or critical illness":
  - "9 also picked the option that everyone one had suffered equally from the pandemic, while singling out at least one category as having suffered most (but only 2 of them picked all options under Question 7)."
- None of the respondents providing these puzzling answers claimed to be "99 years of age."
- The authors repeated analyses excluding the "55 respondents claiming to be 99 years of age or providing contradictory answers to Question 8," finding this exclusion "did not affect the results significantly."

### V. REGRESSION ANALYSIS — methods and general notes
- Binary responses analyzed via logit regressions (probit yielded extremely similar results; Akaike information criterion did not point strongly to one over the other).
- Linear probability models also yield very similar results.
- Results reported consistently using logit regressions; robustness checks and alternative specifications referenced in Annex II.

### A. General Support for a Temporary Levy — main findings
- Support for a temporary levy rises with the level of education:
  - Education measured categorically from "no high school (1) to graduate degree (4)."
  - "The marginal effect of completing one more level of education is to raise support by between 2 and 3 percentage points, depending on specification."
  - Completing high school has the greatest single impact; additional schooling raises support only slightly more.
- Family and household composition effects:
  - Respondents who "have children" show a marginal effect "between 6 and 7 percent."
  - "Equivalized additional household member raising support by 2 percent at the margin." (Each minor is counted as half an adult.)
  - The "has children" measure captures links to the following generation; household size also serves as an income indicator.
- Age effects:
  - Being "above the median age" is associated with a marginal reduction in support by "4-7 percentage points, depending on specification."
  - Finer age quartile breakdowns reported in Annex II (Table 8, regression (1)); dummy for pensionable age was insignificant.
- Ethnicity effects:
  - "Black/African American" respondents favor a levy more strongly, with a marginal effect of "6-8 percentage points, depending on specification."
  - Respondents who "choose not to report their ethnicity have much lower levels of support, with a marginal effect of -49 to -51 percentage points."
  - Further ethnicity breakdowns in Annex II (Table 8, regression (4)) show other ethnicities not statistically different.
- Income effects (nonlinear):
  - Support rises significantly when income surpasses "$20,000 per year—with marginal effect of 7-8 percentage points."
  - Further income increases have less impact until the highest incomes ("above $150,000"), where support falls significantly.
  - Intermediate income step details in Annex II (Table 8, regression (3)).
- Variables found to have no consistent impact:
  - Respondent’s gender.
  - Living in an urban ZIP code (Urban Influence Code definitions tested: metropolitan (1 or 2), broader/micropolitan (3, 5, 8), and large metropolitan (1) — none yielded significant results). The urban indicator reduces sample size due to ZIP code merging and is excluded in other regressions.
- Direct pandemic impact:
  - A dummy for households "directly affected by the crisis, through loss of employment or critical illness with COVID-19" raises support for temporary levies with a marginal effect of "15 percentage points."
  - Annex II (Table 8, regression (5)) shows illness had greater impact than employment loss, and there is a significant increase in support even for just knowing someone who lost employment or fell ill.
- Interaction of age and suffering:
  - Regression (4) includes interaction between "being above median age and having directly suffered from the crisis."
  - Although support among the older half is lower, "the increase in support after having suffered is almost the same in both age groups."

### Table 2 — Logit regression summary (selected coefficients and notes)
- Dependent variable: 1 if respondents support the introduction of a temporary levy.
- All explanatory variables are dummies, except household size (equivalized by giving half weight to children) and education (1–4).
- The variable "suffered" indicates household directly affected by COVID-19 via employment loss or critical illness.
- Selected coefficient signs and significance (standard errors in parentheses):
  - Female: 0.01 (0.09) in (1); not significant.
  - Older half: -0.31*** (0.09) in (1); -0.19 (0.12) in (4).
  - Has children: 0.33*** (0.09) in (1).
  - Household size: 0.08*** (0.03) in (1).
  - Education: 0.15*** (0.06) in (1).
  - Income > $20k: 0.34*** (0.12) in (1).
  - Income > $50k: -0.16 (0.11) in (1).
  - Income > $150k: -0.37** (0.16) in (1).
  - Black/African American: 0.35** (0.14) in (1).
  - Ethnicity withheld: -2.21*** (0.35) in (1).
  - Urban: 0.07 (0.13) shown in (2); not significant.
  - Suffered: 0.68*** (0.09) in (3).
  - Older half and suffered interaction: -0.02 (0.18) in (4); not significant.
- Observations: "2,519" in models (1), (3), and (4); "2,481" in model (2).
- Robust standard errors in parentheses; significance indicators: "*** p<0.01, ** p<0.05, * p<0.1."
- Constants range from -0.24 to -0.48** across specifications.

### B. Analysis of Estimated Moral Perspectives — definitions and distribution
- Moral perspective labels assigned based on spending priorities (respondents had to identify at least one area for budget cuts):
  - "Universalist:"
    - (i) supported at least one of higher health, education or climate spending, and
    - (ii) did not ask for cuts in any of these three areas, and
    - (iii) wanted to reduce at least one of police, military, or border protection spending.
  - "Communitarian:"
    - (i) supported at least one of higher police, military, or border protection spending, and
    - (ii) did not ask for cuts in any of these three areas, and
    - (iii) wanted to reduce at least one of health, education, or climate spending.
  - "Moderate:" anyone not meeting the conditions for the other two labels.
- Resulting shares:
  - Universalists: "almost 50 percent"
  - Communitarians: "21 percent"
- Detailed spending preference results are reported elsewhere (Annex II referenced).

*Source: Authors’ calculation.*

### Annex II (Table 9).

### Annex II (Table 9)

### Measurement of moral perspectives
- Spending-preference assignments were regressed on value statements of Question 5 (Annex III: Table 10) to reflect underlying moral perspectives.
- Principal component analysis (Annex III: Table 11) findings:
  - The first principal component aligns preferences for spending cuts in health, education, local environment, and climate change.
  - Infrastructure spending cuts load with the same sign but a lower coefficient.
  - Preference for cuts in police, military, and border protection have the opposite sign and similar loadings.
  - The first principal component’s signs align with priors about communitarians and universalists and is used as an alternative measure of communitarian preferences.
- The authors note they did not include local environmental spending preferences in the primary categorization because Enke et al. (2020) suggest environmental spending focused on the local area is less clearly aligned with universalist views.

### Main empirical findings on support for a temporary levy
- Using estimated moral perspectives from spending preferences:
  - Universalists are significantly more likely to support a temporary levy, with a marginal effect of 26 percentage points.
  - Interaction with household suffering (employment loss or illness) increases support:
    - Universalists: additional marginal effect of 7 percentage points.
    - Communitarians: although less likely to support levies overall (significantly so in one specification), those with household suffering increase support by a marginal effect of 22 percentage points.
    - Moderates: support rises with a marginal effect of 14 percentage points.
- To address ceiling effects among universalists, the authors estimate change among those who could change their mind:
  - Assumption: share of universalists and communitarians who pre-pandemic were not in favor of progressive taxation equals shares among respondents who did not personally experience illness or job loss (26.3 percent of unaffected universalists; 63.8 percent of unaffected communitarians do not support a temporary levy).
  - Using regression (3) LPM results and dividing increases in the affected group (5.8 percentage points among universalists and 26.7 percent among communitarians) by those who could potentially change:
    - 22 percent of universalists otherwise opposed to temporary levies support them if exposed to COVID-19.
    - 42 percent of communitarians otherwise opposed support them if exposed to COVID-19.
  - The p-value for the difference in ratios is 8.8 percent.
- Regression (4) and (5): replacing categorizations by the first principal component (PCA) score yields consistent results:
  - Likelihood of supporting a temporary levy declines with communitarian (PCA) preferences.
  - The interaction with household suffering reduces opposition to a temporary levy (Communitarian (PCA) and suffered: 0.12**, reported in Table 3).

### Robustness checks on exposure and measurement
- Regressions of household exposure indicators on moral perspectives (Annex III: Table 12) show:
  - Universalists were more likely to claim economic suffering or to have a critically ill acquaintance.
  - Universalists were no more likely to report illness in their own household controlling for county incidence of COVID-19 cases and deaths.
- Replacing subjective suffering with objective county incidence of cases or deaths (Annex III: Table 13, regressions (1) and (2)) preserves the key result:
  - Universalists are more likely to support temporary levies, but more communitarians change their view if affected by the pandemic.
- Alternative moral-perspective definition excluding healthcare spending (to address potential endogeneity from COVID-19 shaping health preferences) yields similar shares:
  - Universalists drop slightly to 40 percent.
  - Communitarians drop slightly to 18 percent.
  - Estimation results remain robust (Annex III: Table 13, regression (3)).

### Attitudes among respondents skeptical of taxation
- Respondents agreeing “most tax revenues are squandered anyway”:
  - Less likely to favor a temporary levy (marginal effect of -7 percentage points; Table 3, regression (6): "Taxes are squandered" coefficient -0.30**).
  - If their household is affected by the crisis, support rises by an additional marginal effect of 9 percentage points (i.e., 21 percentage points in total), overcompensating for previous opposition.
  - Suggests heterogeneity: some respondents simultaneously believe taxes are squandered and support a temporary progressive levy.

### Differences across labeling, tax base, and use of funds
- Labeling (Table 4):
  - Controlling for covariates, terms used to describe the temporary levy do not differ significantly in their effects on support.
  - Exception: the term “COVID-19 Recovery Contribution” fares significantly better (p-value of 3.5 percent; marginal effect of 5 percentage points).
  - Grouping terms by elements like “tax,” “COVID-19,” or “solidarity” did not yield significant differences.
- Differentiation by tax base (Table 5):
  - Patterns mirror temporary levy findings: universalists show higher support for every progressive reform; communitarians show greater increases in support following COVID-19 exposure for many bases.
  - Exception: support for higher taxes on multinational enterprises (MNEs) does not rise following COVID-19 (except among moderates).
  - Support for a tax on everyone rises among communitarians following exposure.
- Differentiation by use of funds (Table 6):
  - Results generally consistent with main findings: universalists more supportive of progressive tax hikes across spending uses; communitarians increase support more after suffering.
  - Specific differences noted:
    - Women are significantly less likely to support taxes restricted to COVID-19-related high income vs. general high incomes, less likely to support progressive reforms funding social spending or infrastructure, and less likely to support redistributive reforms raising taxes only on the top percentile or decile.
    - People above median age are more likely to favor tax increases on profits or multinationals.
    - People with children are less likely to support revenue increases for fighting climate change.
- Revenue-neutral reforms (Table 7):
  - Universalists are significantly more likely to support revenue-neutral reforms that raise taxes on specified groups (Any, Top 1%, Top 10%, Top third, Top half, MNEs).
  - Communitarians who suffered increase support across many of these reform types.

### Summary sample and statistical notes
- Observations in reported regressions: 2,519.
- Robust standard errors reported in parentheses; significance levels: *** p<0.01, ** p<0.05, * p<0.1.
- LPM indicates linear probability model used as an alternative to logistic regression in regression (3).
- Variables "Universalists," "Communitarians," and "Moderates" are identified based on spending preferences as explained in the text. PCA indicates use of principal component analysis of spending preferences. “Taxes are squandered” is a dummy indicating agreement with that statement.

### Conclusion (excerpted)
- COVID-19 exposure (household job loss, illness, or knowing someone affected) is associated with higher support for progressive taxation.
- Universalists are more likely to support temporary levies overall, but communitarians who experienced suffering display larger increases in support, including when measured as percent of those who could change their minds.
- The paper notes three caveats for further research, including uncertainty about the persistence of these effects over time.

*Source: Annex II (Table 9), wpiea2021024-print-pdf - Annex II (Table 9).*

### conclusion could have strengthened the effects. Second, further work will be necessary in other

### wpiea2021024-print-pdf - conclusion could have strengthened the effects. Second, further work will be necessary in other

### Conclusion: interpretation and policy stakes
- Individuals harmed by the pandemic currently demand more redistributive policies.
- If the state fails to meet that demand, affected individuals may grow disillusioned and lose trust in government; unmet demand may morph into social unrest or embrace of simplistic policy solutions.
- Policymakers face high stakes: deliver on the health front and on policies that foster a more equal distribution of incomes and access to government services.
- Policymakers should devote extra attention to political economy factors and gauge support for policy choices from various population segments.

### Survey design and questionnaire (Annex I)
- Screener variables collected:
  - Gender: Male / Female / Other/nonbinary
  - Age (open)
  - Pretax household income categories: Below $20,000; $20,001-$40,000; $40,001-$50,000; $50,000-$75,000; $75,001-$100,000; $100,000-$150,000; more than $150,000
  - Zip code
- Introductory framing: government spending up and tax revenues down due to COVID-19; respondents randomly assigned one of six CELL WORDING terms for two initial questions: (i) tax; (ii) solidarity tax; (iii) COVID-19 recovery tax; (iv) contribution; (v) solidarity contribution; (vi) COVID-19 recovery contribution.
- Key survey questions included:
  - Support for a temporary [CELL WORDING] charged on those who are well off to cover pandemic costs (Yes/No).
  - If introduced, which groups should be charged? (high incomes; high incomes as a result of pandemic; high wealth; companies with high profits; companies with high profits as a result of the pandemic; everyone proportional to income; None).
  - Support for permanent increase in taxes raised only on those who are well off (multiple motives listed).
  - Support for tax law change raising taxes on the rich while reducing for others so total revenue is unchanged (richest 1%, richest 10%, richest third, richer half, multinational corporations, No).
  - Attitudes toward overall tax levels and statements about the rich, solidarity, waste, national debt, and work incentives.
  - Spending preference grid across Health care, Education, Cleaner environment in your area, Fighting climate change, Police, Military, Border Protection, Infrastructure (More / No change / Less).
  - Questions on who suffered most during pandemic and direct household impacts (employment loss, illness, income rise, no direct impact).
  - Demographics: children count, household composition, education level, ethnicity.

### Main regression findings and robustness (Annex II, III)
- Sample size / observations:
  - Observations reported: 2,519 (Tables 8, 10, 11); 2,481 for some regressions (Tables 12, 13).
- Selected coefficients from Table 8 (breakdown of variables used in Table 2):
  - Female: 0.01 (standard error 0.09) across columns.
  - Older half: -0.31*** (0.09) in columns (1)-(3); -0.22** (0.09) in column (4).
  - Has children: 0.33*** (0.09) across columns; in column (5) 0.24*** (0.09).
  - Household size: 0.08*** (0.03) across columns; in column (5) 0.07** (0.03).
  - Education: 0.15*** (0.06) in columns reporting it; in column (5) 0.10* (0.06).
  - Income > $20k: 0.35*** (0.12) in column (1); 0.34*** (0.12) in (2); 0.35*** (0.12) in (4); 0.32** (0.13) in (5).
  - Income > $50k: -0.16 (0.11) in columns shown; in column (5) -0.20* (0.11).
  - Income > $150k: -0.37** (0.16) in column (1); -0.36** (0.16) in (2); -0.36** (0.16) in (4); -0.28* (0.16) in (5).
  - Black/African American: 0.35** (0.14) in columns (1)-(3); 0.37*** (0.14) in (4); 0.30** (0.14) in (5).
  - Prefer not to answer (ethnicity withheld): -2.20*** (0.35) in many columns; column (5) -1.93*** (0.36).
- Selected coefficients from Table 9 (Logit regression with breakdown of spending preferences):
  - Older half: -0.33*** (0.10).
  - Has children: 0.35*** (0.10).
  - Household size: 0.08*** (0.03).
  - Education: 0.14** (0.06).
  - Income > $20k: 0.32** (0.14).
  - Income > $150k: -0.51*** (0.17).
  - Health More: 1.23*** (0.12).
  - Education More: 0.27** (0.12).
  - Local environment More: 0.29** (0.13).
  - Climate change More: 0.33** (0.13).
  - Police Less: 0.30** (0.13).
  - Military Less: -0.28** (0.13).
  - Border protection More: -0.28** (0.13).
- Principal Component Analysis of spending preferences (Table 11):
  - Number of obs = 2,519; Number of comp. = 3; Trace = 8; Rho = 0.6140.
  - Component eigenvalues and proportions:
    - Comp1 eigenvalue 2.40073; Proportion 0.3001; Cumulative 0.3001.
    - Comp2 eigenvalue 1.64689; Proportion 0.2059; Cumulative 0.5060.
    - Comp3 eigenvalue 0.8646; Proportion 0.1081; Cumulative 0.6140.
    - Comp4 eigenvalue 0.801472; Proportion 0.1002; Cumulative 0.7142.
    - Comp5 eigenvalue 0.690115; Proportion 0.0863; Cumulative 0.8005.
    - Comp6 eigenvalue 0.563269; Proportion 0.0704; Cumulative 0.8709.
    - Comp7 eigenvalue 0.534905; Proportion 0.0669; Cumulative 0.9377.
    - Comp8 eigenvalue 0.498018; Proportion 0.0623; Cumulative 1.0000.
  - Selected loadings (Comp1, Comp2, Comp3) and unexplained:
    - Health Cuts: Comp1 0.3620; Comp2 0.3444; Comp3 -0.2829; Unexplained .421.
    - Education Cuts: 0.3844; 0.3726; -0.1582; Unexplained .395.
    - Local environment cuts: 0.4495; 0.0991; -0.0367; Unexplained .4976.
    - Climate change cuts: 0.4709; 0.0323; -0.0851; Unexplained .4597.
    - Police cuts: -0.2648; 0.4809; -0.2259; Unexplained .4068.
    - Military cuts: -0.2856; 0.4918; -0.1984; Unexplained .3719.
    - Border protection cuts: -0.3411; 0.3575; 0.1406; Unexplained .4931.
    - Infrastructure cuts: 0.1717; 0.3617; 0.8810; Unexplained .04268.
  - Note: Variables coded 1 for more spending, 2 for unchanged, 3 for cuts.
- Moral perspectives and suffering (Table 12):
  - Universalist is positively associated with suffering indicators:
    - Universalist coefficient 0.29*** (0.09) for Suffered; 0.28*** (0.09) for Suffered Empl. Loss, HH; 0.24** (0.11) for Empl. Loss; 0.38*** (0.11) for Acqu. Illness, HH.
    - Universalist and HH Income Rise coefficient -0.99*** (0.15).
  - Communitarian coefficients generally smaller and less consistently significant; example Communitarian 0.06 (0.11) in column (1).
  - Case incidence coefficient 0.17*** (0.04) in column (1); other case/death incidence coefficients shown with larger standard errors.
- Robustness and interaction results (Table 13, Logit on support for temporary levy):
  - Three robustness specifications shown: (1) "Suffered" defined by Covid-19 cases as share of county population; (2) "Suffered" defined by Covid-19 deaths as share of county population; (3) Moral perspectives defined without use of health spending category.
  - Common covariate coefficients across columns:
    - Female ~ 0.04 (0.09 / 0.09 / 0.09).
    - Older half: -0.27*** (0.09) in (1) and (2); -0.22** (0.09) in (3).
    - Has children: 0.40*** (0.09) in (1); 0.41*** (0.09) in (2); 0.37*** (0.09) in (3).
    - Household size: 0.09*** (0.03) in (1) and (2); 0.08*** (0.03) in (3).
    - Income > $20k: 0.33** (0.13) in (1) and (2); 0.35*** (0.13) in (3).
    - Income > $150k: -0.39** (0.16) in (1) and (2); -0.39** (0.16) in (3).
    - Ethnicity withheld: -2.38*** (0.37) in (1); -2.37*** (0.37) in (2); -2.23*** (0.34) in (3).
  - Moral perspective coefficients and interactions:
    - Universalist: 1.56*** (0.25) in (1); 1.29*** (0.16) in (2); 1.21*** (0.14) in (3).
    - Communitarian: -0.28 (0.28) in (1); -0.46** (0.19) in (2); -0.56*** (0.16) in (3).
    - Universalist and suffered: 0.02 (0.07) in (1); 0.55 (1.38) in (2); 0.23 (0.16) in (3).
    - Communitarian and suffered: 0.22*** (0.09) in (1); 6.23*** (1.90) in (2); 1.13*** (0.20) in (3).
    - Moderate and suffered: 0.14** (0.07) in (1); 0.48 (1.21) in (2); 0.38*** (0.13) in (3).
  - Constants: -0.90*** (0.25) in (1); -0.63*** (0.22) in (2); -0.66*** (0.20) in (3).

### Implications for research and further work
- Authors note limitations and call for:
  - Strengthened causal identification (conclusion could have strengthened the effects).
  - Further work in other countries to explore hypothesis that individuals directly bearing brunt of pandemic become more favorable to progressive taxation.
  - Investigation of whether views against progressive taxation become more entrenched among those who continue to hold them.
  - Consideration of simultaneous factors changing attitudes, including evidence that epidemics weaken trust in government (Aksoy, Eichengreen, and Saka 2020).

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021024-print-pdf.pdf_