## Annex I. Data Source, Methodology — wpiea2023064-print-pdf

## Source details

**Canonical URL:** [Annex I. Data Source, Methodology — wpiea2023064-print-pdf](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023064-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023064-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023064-print-pdf.pdf.json)

---

### Key statistics and scope
- General government wage bill spending: ranges from 7.4 percent of GDP in developing countries to 10.2 percent in advanced countries.
- Meta-database coverage: 86 countries (26 advanced, 37 emerging, and 23 developing countries).
- Annual time-series dataset: 43 countries (14 EMDEs, and 29 AEs) covering 1995 to 2020.
- Quarterly time-series dataset: 32 countries (7 EMDEs and 25 AEs) covering 1990: Q1-2022: Q2.

### Theoretical framework (determinants of public wage premium or deficit)
- Drivers of a public wage premium:
  - Political economy: ability of public sector workers to compete over public budgets.
  - Institutional factors: pay regulation, unionisation, collective bargaining coverage; degree of unionisation tends to be higher in the public sector.
  - Political cycle: unions and politically sensitive services can create upward wage pressure, especially around elections; effects more marked in low-income and emerging countries.
  - Inelastic public-sector labor demand: unions can bargain higher wages without large employment reductions.
  - Policy objectives: wage setting for gender equity, poverty reduction, income equality.
- Drivers of a public wage deficit:
  - Compensating differentials: job security, longer holidays, generous pension schemes.
  - Public-sector monopsony power in certain professions or groups.
  - Macro objectives: public wages used to constrain overall wage inflation, support fiscal consolidation, and enhance competitiveness.
- Cross-country and temporal variation expected from differences in wage-setting mechanisms, union importance, public sector size, equal-pay prioritization, and trade openness.

### Empirical estimation approach
- Mincerian semi-logarithmic wage regression:
  - Y = α + β·X + γ·pub + u
    - Y is the log individual wage.
    - X is a vector of individual characteristics (education, work experience proxied by age, rural-urban or city indicator, gender).
    - pub is a binary variable equal to 1 if an individual works in the public sector.
    - γ estimates the average public-private wage differential (positive = premium; negative = deficit).
  - Wage variable typically based on reported gross labour earnings.
- Oaxaca–Blinder decomposition:
  - Y_pub = α_pub + β_pub·X_pub + u_pub
  - Y_priv = α_priv + β_priv·X_priv + u_priv
  - Mean wage differences decomposed into:
    - Endowment effect: β̂_pub · (X̄_pub − X̄_priv)
    - Wage premium (unexplained) component: [(α̂_pub − α̂_priv) + (β̂_pub − β̂_priv)·X̄_priv]
- Additional considerations:
  - Controls for hours worked or use of average hourly wages.
  - Allow premium heterogeneity by gender, skill, income via interaction terms or separate regressions.
  - Inclusion of squared terms or logs to capture non-linear education/experience effects.

### Meta-analysis and database construction: inclusion criteria
- Time window: studies from 1991 onwards.
- Methodology: include only regression-based estimates; exclude mean wage ratios without controls.
- Occupation control: exclude studies that include occupation without correcting for occupation selection endogeneity.
- Public sector definition: focus on general government (excluding state-owned enterprises); compare to wage-employed private sector (excluding self-employed).
- Age sample: workers between legal minimum working age and retirement age.
- Comparability: emphasize multi-country comparable studies; supplement with single-country studies following similar strategies.
- Notes:
  - Inclusion of state-owned enterprise employees does not significantly change average premium results.
  - When ISCED-based skill groupings are unavailable, ISCED-3 substituted with ISCED-2 if necessary.

### Skill- and gender-specific premium construction
- Gender-specific premiums drawn from same sample as average premium.
- Skill-specific classification:
  - Prefer ISCED-based:
    - Low-skilled: ISCED-1 and ISCED-0.
    - High-skilled: ISCED-3.
  - If ISCED unavailable, use ISCO occupational groups:
    - Low-skilled: ISCO-9 (elementary workers).
    - High-skilled: ISCO-1, 2, 3 (managers, professionals, technicians).
  - Accept study-specific low/high skill classifications when necessary.

### Time-series datasets and analytical uses
- Two country-level administrative time-series datasets:
  - Annual dataset: 43 countries (14 EMDEs, and 29 AEs) covering 1995–2020.
  - Quarterly dataset: 32 countries (7 EMDEs and 25 AEs) covering 1990: Q1–2022: Q2.
- Analytical uses:
  - Analyse evolution of relative public and private sector wages.
  - Examine effects of economic cycles and electoral cycles on relative wages.
  - Estimate effects of public wage shocks on private wages and inflation.
  - Examine heterogeneity across labor market characteristics, regulations, and macroeconomic conditions.

---

### Meta-database and data coverage (Annex Table 1 summary)

### Database composition
- Total premium estimates: 208 spanning 86 countries.
- By income group:
  - Advanced: 26.
  - Emerging: 37.
  - Low-income and developing countries (LIDCs): 23.
- Time periods in Table 1: 1991-2000 and 2001-2014.
- Coverage notes:
  - Around two-thirds of countries appear at least twice; 29 countries appear only once.
  - Regional coverage includes LAC, SSA, CEE, AP, CIS, MENA.
  - High coverage of AEs (26 out of possible 35) reflects greater survey availability.
  - Number of premium estimates higher in 2001-2014, likely due to increasing data availability.

---

### Descriptive analysis — average, gender, and skill-specific premiums

### Cross-country average public-private wage premium
- Cross-country average over last two decades: approximately 10 percent.
- By income group (Mean; Most recent year mean; # countries):
  - All: Mean 10.1; Most recent year mean 11.2; # countries 86.
  - AEs: Mean 5.4; Most recent year mean 5.2; # countries 26.
  - EMs: Mean 11.7; Most recent year mean 13.3; # countries 37.
  - LIDCs: Mean 12.8; Most recent year mean 14.8; # countries 23.
- Figure-note average premiums for most recent year: AEs 5.2 percent, EMs 13.3 percent, LIDCs 14.8 percent.
- Time-trend evidence:
  - Comparing 1990s to 2000-14: 16 out of 20 countries experienced an increase in the premium.
  - Mean change 1990s to 2000-14: 7.3 percentage points.
  - Median change 1990s to 2000-14: 7.8 percentage points.
  - Comparing 2001-07 to 2008-14: mean (median) change is -0.4 (-1.1) percentage points.
  - Panel trends: most AEs (14 out of 20 AE countries) saw a decline in the premium between 2001-07 and 2008-14; most LIDCs saw increases; EMs mixed with regional heterogeneity.

### Gender-specific public wage premium
- Overall: premium smaller for men than for women.
- By group — mean estimates (All years / Most recent year / # countries):
  - All: Men 6.3; Women 8.6; Diff -2.4. Most recent year: Men 6.9; Women 7.9; Diff -1.0; # countries 60.
  - AEs: Men 4.9; Women 6.5; Diff -1.6. Most recent year: Men 4.5; Women 5.0; Diff -0.4; # countries 22.
  - EMs: Men 4.6; Women 6.3; Diff -1.7. Most recent year: Men 7.2; Women 6.5; Diff 0.7; # countries 26.
  - LIDCs: Men 12.6; Women 17.6; Diff -5.0. Most recent year: Men 10.5; Women 16.5; Diff -5.9; # countries 12.
- Robustness: results unchanged using medians.
- Interpretation:
  - LIDC findings consistent with greater gender pay equality in the public sector and partial reflection of private-sector gender pay discrimination.
  - Temporal patterns: female premium larger in most LIDCs; in most AEs no significant male/female difference in recent years; AE female premium evident during 2001-07 but disappears in 2008-14; no clear pattern in EMs.

### Skill-specific public wage premium
- Average premium significantly higher for low-skilled than for high-skilled workers.
- By group — mean estimates (All years / Most recent year / # countries):
  - All: High 1.6; Low 7.6; Diff -6.1. Most recent year: High 0.7; Low 10.3; Diff -9.6; # countries 60.
  - AEs: High 2.7; Low 7.7; Diff -4.9. Most recent year: High 0.1; Low 11.0; Diff -10.8; # countries 21.
  - EMs: High 1.5; Low 6.0; Diff -4.5. Most recent year: High 1.7; Low 7.2; Diff -5.5; # countries 27.
  - LIDCs: High -0.3; Low 11.1; Diff -11.4. Most recent year: High -0.7; Low 16.3; Diff -16.9; # countries 12.
- Interpretation:
  - High-skilled premium range: -0.3 to 2.7 percent.
  - Low-skilled premium range: 6.0 to 11.1 percent.
  - Premium gap much larger in LIDCs.
  - Difference between low- and high-skill premiums increased over time, driven largely by rising low-skill public wage premium.
  - Evidence consistent with public-sector monopsony power over highly educated workers with fewer private opportunities, especially in developing countries.
- Business-cycle sensitivity:
  - In AEs, pre-crisis (2001-07) high-skilled workers tended to have higher public premium relative to low-skilled.
  - Post-crisis (2008-14), high-skilled premium in AEs declined substantially while low-skilled premium remained unchanged or increased in some countries.

---

### Database description (compensation per employee datasets)

### Annual dataset
- Data source: United Nations Statistics Division Database.
- Coverage: 43 countries (14 EMDEs and 29 AEs).
- Period: 1995 to 2020.
- Measure: nominal compensation per worker = total compensation of employees / total employment (SNA total compensation of employees).

### Quarterly dataset
- Data source: OECD quarterly database of national accounts.
- Coverage: 32 countries (7 EMDEs and 25 AEs) over 1990: Q1–2022: Q2.
- Sector definitions:
  - Public sector: ISIC Rev. 4 sections O, P and Q (public administration and defense, health and education services).
  - Private sector: ISIC Rev. 4 activity classification; preferred private benchmark is private manufacturing to control for skill-level differences.

### Advantages of SNA total compensation data
- Captures wages and salaries, bonuses, gratuities, income in kind, allowances, retroactive wage payments, among others.
- Reported on a gross basis before employee deductions.
- Provides measure of true cost of labor.
- SNA framework ensures consistent compilation across countries and over time.

### Cross-country pay ratio findings (1995–2020)
- Average public-private pay ratio: 1.12 (implying a pay differential of 12 percent).
- Medians:
  - EMDEs median: 1.34.
  - AEs median: 1.06.
- Temporal patterns:
  - 1997–2007: ratio increased on average among EMDEs due to rapid public pay increases.
  - 2008–2010: ratio increased in both AEs and EMDEs because private sector compensation fell by around 4 percent while public sector compensation declined by less than 1 percent.
  - 2011–2016: private compensation recovery led to gradual decrease in public-to-private pay ratios across income groups.
  - More recently: ratios continued to decline due to larger increases in private wages following recovery from the COVID-19 pandemic and subsequent downturn.

---

### Research question, empirical specification, and main annual results

### Regression specification (annual analysis)
- y_it = α + β EC_it + λ PC_it + γ z_it + δ T_it + f_i + ε_it  (Equation (1))
  - Dependent variable: log of the ratio of average public pay to average private manufacturing pay.
  - EC_it: recession/economic slack binary (equal to 1 in periods of economic slack or recession; defined using the output gap in percent of real potential output).
  - PC_it: election-year binary (equal to 1 in election years).
  - z_it: conditioning variables including ratio of public to private employment, inflation expectations, and real output per worker.
  - Inflation expectations: inflation forecast in the IMF Spring issue of the World Economic Outlook for the previous year.
  - Country-fixed effects f_i, linear and quadratic time trends included.
  - Standard errors clustered at the country level, robust to heteroscedasticity and autocorrelation, based on a non-parametric block bootstrap procedure.

### Main empirical results on economic and political cyclicality
- Recessions:
  - Pay ratio behaves counter-cyclically: increasing by about an extra 3 percent during recessions compared to non-recession years.
  - Robust to alternative recession definitions and transition functions of the state of the economy.
- Elections:
  - Public-to-private pay ratio varies over political cycle because average government compensation rises ahead of elections.
  - Effect concentrated in EMDEs: pay ratio in EMDEs increases by around 2 percent more, on average, during election years relative to non-election years.
  - Interpretation: stronger institutions in higher-income countries mitigate effects; public wage policy can be used to influence voting behavior in some countries.

### Macroeconomic considerations of public wage shocks
- Public wage policy often pro-cyclical; procyclicality reflects wage levels more than public employment and can exceed that of total public spending.
- Transmission channels:
  - Potential to trigger wage-price spirals via automatic indexation, COLA clauses, or large public-sector union agreements.
  - Fiscal-theory perspective: increases in public wages that raise public debt value can be inflationary if markets expect inadequate surpluses to repay debt.
- Structural determinants: labor market institutions and regulations shape private-sector and price responses to public wage increases.

---

### Quarterly analysis: identification, data, and impulse-response findings

### Empirical strategy and identification
- Local Projections (Jordà 2005) to estimate impulse responses over horizons h = 0, …, H.
- Baseline specification (2a): dependent variable y includes CPI, unemployment rate, private compensation per employee, nominal effective exchange rate, and central bank policy rate. PW is log average public-sector compensation per employee. z contains controls including public/private employment ratio.
- Country-fixed effects and year-fixed effects included. Variables in log first differences (except interest rate, first-differenced only). Four lags of each variable included.
- Standard errors clustered at country level; 90 percent confidence intervals reported.
- Identification of public wage shocks: recursive ordering with domestic retail energy prices ordered first; assumption that changes in public compensation per employee do not respond to macro realizations within same quarter.

### Data
- Quarterly data 1990: Q1–2022: Q2.
- Sample: 32 countries (25 advanced, 7 emerging), ~3250 observations.
- Employee compensation from the quarterly SNA database; other variables from IMF IFS.

### Key impulse-response findings — public wage shocks to private wages
- Public sector size:
  - Larger public employment countries: a 1 percent increase in average public wages → around 0.45 percent increase in average private wages (peak), persistent up to 15 quarters.
  - Smaller public sector countries: effect around 0.07 percent and very short-lived.
- Openness to trade:
  - A 1 percent increase in average private wages → around 0.27 percent increase in public sector wages in more open countries (persistent).
  - Less open countries: effect around 0.10 percent, significant for two quarters.
- Unionization, bargaining coverage, centralization:
  - Higher union density and bargaining coverage and greater centralization: 1 percent public wage shock increases average private wages by up to around 0.31 percent and 0.52 percent, respectively.
  - Lower unionization and less centralized bargaining: effects around 0.13 percent and 25 percent, respectively, and less persistent (note: 25 percent figure appears as reported in the source).
  - Similar patterns when measuring union density in the public sector.
- Inflation regime:
  - High-inflation environment threshold: 90th percentile of inflation across countries (roughly 4.5 percent).
  - A 1 percent increase in public wages → around 0.59 increase in private wages (peak) in high-inflation environments — roughly 3 times larger than in low-inflation environments.
- State of the economy (economic slack):
  - Effects larger and more persistent during periods of lower economic slack (higher bargaining power when labor demand is strong).

### Key impulse-response findings — public wage shocks to inflation
- General pattern: CPI responses follow private wage responses in magnitude and persistence.
- Public sector size:
  - Larger public sector: 1 percent public wage shock raises consumer price level by around 0.3 percent — nearly 5 times bigger than effect in small-public-sector countries (around 0.06 percent).
- Bargaining coverage and centralization:
  - Higher bargaining coverage: 1 percent public wage shock increases consumer price level by up to around 0.10 percent.
  - Greater centralization: similar shock increases consumer prices by up to around 0.22 percent.
  - Lower unionization/coverage and less centralized bargaining: effects up to around 0.03 percent and 0.06 percent respectively, less persistent, mostly insignificant.
- Inflation regime:
  - In high-inflation environments a 1 percent increase in public wages leads to around 0.44 increase in consumer prices (peak) — roughly 3 times larger than peak response in low-inflation environments (around 0.14 percent).
- Economic slack:
  - During periods of lower economic slack, consumer prices rise by about 0.2 percent in response to a public wage shock and remain positive and significant over long horizons.
  - No evidence of consumer price effects during higher economic slack periods.
- Interpretation: public wages can contribute to higher and more persistent inflationary pressures via spillovers to private wages; magnitude and persistence depend on unionization, bargaining coverage, centralization, economic slack, and inflation regime.

---

### Conclusion — summary statistics and broader findings
- Micro-econometric dataset: on average, public wages are around 10 percent higher than private wages for workers with similar socio-economic characteristics.
  - Premium by development level: 5.4 percent in AEs, 11.7 percent in EMs, and 12.8 percent in LIDCs.
  - Premium higher for women than men; driven primarily by a premium for lower-skilled workers.
- Time-series evidence (1990: Q1-2022: Q2):
  - Public wage premium present especially in EMDEs.
  - In high-income countries: premium relatively stable until 2008 financial crisis, then increased sharply (reflecting a sharp decline in private wages), and then gradually decreased as private wages recovered post-crisis.
  - Premium varies counter-cyclically (public wages decrease by less than private wages during bad times).
  - Premium higher during election years, but only in low-income countries sample.
- Overall: public wages tend to drive wage-setting where the public sector is relatively large; public wage shocks raise private wages and the price level, with heterogeneity driven by unionization, bargaining coverage, centralization, economic slack, and inflation regime.

*Source: Annex I. Data Source, Methodology — wpiea2023064-print-pdf*

### Annex I. Data Source, Methodology ......................................................................................

### Annex I. Data Source, Methodology

### Key statistics and scope
- General government wage bill spending ranges from 7.4 percent of GDP in developing countries to 10.2 percent in advanced countries.
- Meta-database of micro-econometric public wage premium estimates covers 86 countries: 26 advanced, 37 emerging, and 23 developing countries.
- Annual time-series dataset: 43 countries (14 EMDEs, and 29 AEs) covering 1995 to 2020.
- Quarterly time-series dataset: 32 countries (7 EMDEs and 25 AEs) covering 1990: Q1-2022: Q2.

### Theoretical framework (determinants of public wage premium or deficit)
- Public wage premium drivers:
  - Political economy: public sector wages depend on the ability of public sector workers to compete over public budgets (Gunderson, 1989; 1979; Mueller, 1998).
  - Institutional factors: pay regulation, unionisation, and collective bargaining coverage; evidence that degree of unionisation tends to be higher in the public sector (Lucifora and Meurs, 2006; Perez and Sanchez, 2010; Giordano et. al., 2011; Dickson et. al., 2014).
  - Political cycle: public sector unions and politically sensitive service delivery can lead to upward wage pressure, especially around elections (Rogoff 1990; Akhmedov and Zhuravskaya 2004); empirical impact on total wage spending more marked in low-income and emerging countries (Shi and Svensson 2006; Drazen and Eslava 2010; Cahuc and Carcillo 2012; Eckardt and Mills 2014; Gaspar et. al., 2017).
  - Inelastic public sector labor demand: unions can bargain for higher wages without large employment reductions (Ashenfelter and Ehrenberg, 1975; Gunderson, 1979).
  - Policy objectives: wage setting that promotes gender equity, poverty reduction, and income equality (Alesina et. al., 2000; Chatterji et. al., 2007).

- Public wage deficit drivers:
  - Compensating differentials: job security, longer holidays, generous pension schemes (Bellante and Link, 1981; Moore and Raisian, 1991).
  - Public-sector monopsony power in certain professions or groups (Mueller, 1998; Campos et. al., 2017).
  - Macro objectives: public wages used to constrain overall wage inflation, support fiscal consolidation, and enhance competitiveness (Gregory, 1990).

- Cross-country and temporal variation expected from:
  - Differences in wage setting mechanisms, union importance, relative public sector size, prioritization of equal-pay policies, and trade openness (Rattsø and Stokke, 2019).

### Empirical estimation approach
- Standard Mincerian semi-logarithmic wage regression used to estimate average public-private wage differential:
  - Y = α + β·X + γ·pub + u
    - Y is the log individual wage.
    - X is a vector of individual characteristics (education, work experience proxied by age, rural-urban or city indicator, gender).
    - pub is a binary variable equal to 1 if an individual works in the public sector.
    - γ estimates the average public-private wage differential (positive = premium; negative = deficit).
  - Wage variable typically based on reported gross labour earnings.

- Oaxaca–Blinder decomposition approach:
  - Separate wage regressions for public and private sector:
    - Y_pub = α_pub + β_pub·X_pub + u_pub
    - Y_priv = α_priv + β_priv·X_priv + u_priv
  - Mean wage differences decomposed into:
    - Endowment effect: β̂_pub · (X̄_pub − X̄_priv)
    - Wage premium (unexplained) component: [(α̂_pub − α̂_priv) + (β̂_pub − β̂_priv)·X̄_priv]

- Additional empirical considerations:
  - Controls for hours worked or use of average hourly wages to account for job-condition differences.
  - Allowing premium to differ by gender, skill level, or income group via interaction terms or separate regressions.
  - Studies sometimes include squared terms or logs to capture non-linear effects of education and experience.

### Meta-analysis and database construction: inclusion criteria
- Time window: studies from 1991 onwards.
- Methodology: include only studies that estimate public wage premium using regression methods; exclude studies that report mean wage ratios without controlling for individual characteristics.
- Occupation control: exclude studies that include occupation as a control without appropriately correcting for occupation selection endogeneity.
- Public sector definition: focus on general government (excluding state-owned enterprises); compare to wage-employed workers in the private sector (excluding self-employed workers).
- Age sample: workers between legal minimum working age and retirement age.
- Comparability: emphasize studies covering large sets of countries using comparable data, variables, and estimation techniques; supplement with single-country studies that follow similar estimation strategies.

- Notes on inclusions and substitutions:
  - Empirical evidence indicates inclusion of state-owned enterprise employees does not significantly change average wage premium results.
  - When ISCED-based skill groupings are not available, substitute ISCED-3 (at least secondary education) with ISCED-2 (at least lower secondary education) if necessary.

### Skill- and gender-specific premium construction
- Gender-specific premiums: drawn from the same sample of studies used for the average premium.
- Skill-specific premiums: classification approach
  - Prefer ISCED-based classifications where available:
    - Low-skilled: up to primary education (ISCED-1) and no formal education (ISCED-0).
    - High-skilled: at least secondary education (ISCED-3).
  - When ISCED not available, use ISCO occupational groups:
    - Low-skilled: elementary workers (ISCO-9).
    - High-skilled: managers, professionals, technicians (ISCO-1, 2, 3).
  - Accept study-specific classifications of low- versus high-skilled workers when necessary.

### Time-series datasets and analysis objectives
- Two country-level administrative time-series datasets constructed:
  - Annual dataset: 43 countries (14 EMDEs, and 29 AEs) covering 1995–2020.
  - Quarterly dataset: 32 countries (7 EMDEs and 25 AEs) covering 1990: Q1–2022: Q2.
- Analytical uses:
  - Analyse evolution of relative average public and private sector wages over time.
  - Examine how relative wages are affected by economic cycles and electoral cycles.
  - Estimate effects of public wage shocks on private wages and inflation.
  - Examine heterogeneity of these relationships across labor market characteristics, regulations, and macroeconomic conditions.

*Source: Annex I. Data Source, Methodology — wpiea2023064-print-pdf*

### Annex Table 1 summarises the data sources, methodologies, and country coverage. Table 1 (Panel A)

### wpiea2023064-print-pdf - Annex Table 1 summarises the data sources, methodologies, and country coverage. Table 1 (Panel A)

### Data coverage and meta-database
- Database comprises 208 premium estimates and spans 86 countries.
- By income group:
  - 26 are advanced.
  - 37 are emerging.
  - 23 are low-income and developing countries (LIDCs).
- Time periods covered in Table 1: 1991-2000 and 2001-2014.
- Around two-thirds of countries appear at least twice; 29 countries appear only once.
- Regional coverage includes Latin America & the Caribbean (LAC), Sub-Saharan Africa (SSA), Central & Eastern Europe (CEE), Developing Asia & the Pacific (AP), Commonwealth of Independent States (CIS), and Middle East & Northern Africa (MENA).
- Relatively high coverage of advanced economies (AEs; 26 out of a possible 35) likely reflects greater availability of labor force and household surveys.
- The number of premium estimates is higher over the 2001-2014 period, most likely reflecting increasing data availability.

### Descriptive analysis — average public-private wage premium
- Cross-country average public wage premium over the last two decades: approximately 10 percent.
- By income group (mean and most recent year mean):
  - All: Mean 10.1; Most recent year mean 11.2; # countries 86.
  - AEs: Mean 5.4; Most recent year mean 5.2; # countries 26.
  - EMs: Mean 11.7; Most recent year mean 13.3; # countries 37.
  - LIDCs: Mean 12.8; Most recent year mean 14.8; # countries 23.
- Figure note average premiums for most recent year: AEs 5.2 percent, EMs 13.3 percent, LIDCs 14.8 percent.
- Time-trend evidence:
  - Comparing 1990s to 2000-14 period, 16 out of 20 countries experienced an increase in the premium.
  - Mean change over 1990s to 2000-14: 7.3 percentage points.
  - Median change over 1990s to 2000-14: 7.8 percentage points.
  - Comparing 2001-07 to 2008-14, mean (median) change is -0.4 (-1.1) percentage points.
  - Panel-level trends: most AEs (14 out of 20 AE countries) experienced a decline in the premium between 2001-07 and 2008-14, while most LIDCs saw increases; EMs show mixed results with regional heterogeneity (Eastern Europe declining, Baltic countries increasing).

### Gender-specific public wage premium
- Overall finding: premium for working in the public sector is smaller for men than for women.
- By group — mean estimates (All years / Most recent year / # countries):
  - All: Men 6.3; Women 8.6; Diff -2.4. Most recent year: Men 6.9; Women 7.9; Diff -1.0; # countries 60.
  - AEs: Men 4.9; Women 6.5; Diff -1.6. Most recent year: Men 4.5; Women 5.0; Diff -0.4; # countries 22.
  - EMs: Men 4.6; Women 6.3; Diff -1.7. Most recent year: Men 7.2; Women 6.5; Diff 0.7; # countries 26.
  - LIDCs: Men 12.6; Women 17.6; Diff -5.0. Most recent year: Men 10.5; Women 16.5; Diff -5.9; # countries 12.
- Robustness: results unchanged when using medians.
- Interpretation: LIDC findings are consistent with greater gender pay equality in the public sector and likely partly reflect gender pay discrimination in the private sector.
- Temporal patterns:
  - In most LIDCs, female public employees enjoy a larger wage premium than male counterparts.
  - In most AEs, no significant difference between female and male premiums in the public sector for recent years.
  - In AEs, a higher premium for female workers was evident during 2001-07 but appears to disappear in 2008-14.
  - No clear consistent pattern among EMs.

### Skill-specific public wage premium
- Average public wage premium is significantly higher for low-skilled workers than for high-skilled workers.
- By group — mean estimates (All years / Most recent year / # countries):
  - All: High 1.6; Low 7.6; Diff -6.1. Most recent year: High 0.7; Low 10.3; Diff -9.6; # countries 60.
  - AEs: High 2.7; Low 7.7; Diff -4.9. Most recent year: High 0.1; Low 11.0; Diff -10.8; # countries 21.
  - EMs: High 1.5; Low 6.0; Diff -4.5. Most recent year: High 1.7; Low 7.2; Diff -5.5; # countries 27.
  - LIDCs: High -0.3; Low 11.1; Diff -11.4. Most recent year: High -0.7; Low 16.3; Diff -16.9; # countries 12.
- Interpretation:
  - Average public wage premium for high-skilled workers ranges between -0.3 to 2.7 percent in the data.
  - Average public wage premium for low-skilled workers ranges from 6.0 to 11.1 percent.
  - The premium gap between low- and high-skilled workers is much larger in LIDCs.
  - The difference between low-skill and high-skill premiums has increased over time, driven largely by a rising low-skill public wage premium.
  - Evidence consistent with public sector monopsony power over highly educated workers who may have fewer attractive private-sector opportunities, especially in developing countries.
- Business-cycle sensitivity:
  - For AEs, pre-crisis (2001-07) high-skilled workers tended to enjoy a higher public wage premium relative to low-skilled workers.
  - Post-crisis (2008-14), the public premium for high-skilled workers in AEs declined substantially while the low-skilled premium remained unchanged or increased in some countries.

### Database description (compensation per employee datasets)
- Two datasets compiled on average compensation per employee for public and private sectors using System of National Accounts (SNA) data.
- Nominal compensation per worker computed as ratio of total compensation of employees to total employment.
- Annual dataset:
  - Data source: United Nations Statistics Division Database.
  - Coverage: 43 countries (14 EMDEs and 29 AEs).
  - Period: 1995 to 2020.

*Source: IMF staff calculations.*

### 2. Quarterly dataset: we obtain data from the OECD quarterly database of national accounts. The data

### wpiea2023064-print-pdf - 2. Quarterly dataset: we obtain data from the OECD quarterly database of national accounts. The data

### Quarterly dataset, coverage, and sector definition
- Data source: OECD quarterly database of national accounts.
- Coverage: 32 countries (7 EMDEs and 25 AEs) over the period from 1990: Q1-2022: Q2.
- Public sector definition: activity classification given in sections O, P and Q of ISIC Rev. 4 (includes public administration and defense, health and education services).
- Private sector definition: activity classification under ISIC Rev. 4; preferred private benchmark for this analysis is the private manufacturing sector (used to control for differences in average skill levels across sectors and countries).

### Advantages of using SNA total compensation data
- The SNA category "total compensation of employees" captures:
  - wages and salaries, bonuses, gratuities, income in kind, allowances, retroactive wage payments, among others.
  - reported on a gross basis, prior to deductions for employees' contributions to income tax, employment insurance, and pension funds.
  - provides a measure of the true cost of labor.
- SNA framework ensures consistent compilation and presentation across countries and over time, allowing matching of employment levels and total compensation by the same classification of economic activities.

### Cross-country pay ratio findings (1995–2020)
- Average public-private pay ratio: 1.12 (implying a pay differential of 12 percent).
- Medians:
  - EMDEs median: 1.34.
  - AEs median: 1.06.
- Temporal patterns:
  - Between 1997 and 2007, ratio increased on average among EMDEs due to rapid and large increase in public pay.
  - Between 2008 and 2010 (onset of global recession and financial crisis): ratio increased in both AEs and EMDEs, largely because average compensation in the private sector decreased by around 4 percent while average compensation in the public sector declined by less than 1 percent.
  - Recovery in private compensation between 2011 and 2016 led to a gradual decrease in public-to-private pay ratios across both income groups.
  - More recently, ratios continued to decline due to larger increases in private wages following recovery from the COVID-19 pandemic and subsequent economic downturn.

### Research question and empirical specification
- Research aim: estimate the relationship between economic and political factors and the public-private pay ratio using annual data on average compensation per employee in private and public sectors over 1995 to 2020.
- Remaining variables sourced from IMF’s International Finance Statistics (IFS) and World Economic Outlook databases.
- Regression specification:
  - y_it = α + β EC_it + λ PC_it + γ z_it + δ T_it + f_i + ε_it  (Equation (1))
  - Dependent variable: log of the ratio of average public pay to average private manufacturing pay.
  - EC_it: recession/economic slack binary (equal to 1 in periods of economic slack or recession; defined using the output gap in percent of real potential output).
  - PC_it: election-year binary (equal to 1 in election years).
  - z_it: conditioning variables including ratio of public to private employment, inflation expectations, and real output per worker.
  - Inflationary expectations: inflation forecast in the IMF Spring issue of the World Economic Outlook for the previous year.
  - Country-fixed effects f_i, and linear and quadratic time trends included.
  - Standard errors: clustered at the country level, robust to heteroscedasticity and autocorrelation, based on a non-parametric block bootstrap procedure.

### Main empirical results on economic and political cyclicality
- Recessions:
  - The pay ratio behaves counter-cyclically, increasing by about an extra 3 percent during recessions compared to non-recession years.
  - Robustness: result holds using transition function of the state of the economy and when defining recession dummy as periods of negative growth rather than negative output gaps.
- Elections:
  - Public-to-private pay ratio varies over the political cycle mainly because average compensation per worker in the government sector rises ahead of elections.
  - Effect concentrated in EMDEs: the pay ratio in EMDEs increases by around 2 percent more, on average, during election years relative to non-election years.
  - Interpretation: effects mitigated in higher-income countries by stronger institutions and governance; suggests public sector wage policy can be used to influence voting behavior in some countries.

### Macroeconomic considerations of public wage shocks
- Literature context:
  - Public wage policy often found to be pro-cyclical; procyclicality reflects public wage levels more than public employment and can be larger than that of total public spending.
  - Political cycles also affect fiscal policy; evidence on wages and electoral cycles is mixed but points to more pronounced effects in developing and transition economies.
- Transmission channels from public wage shocks to the macroeconomy:
  - Potential to influence the likelihood and severity of an economy-wide wage-price spiral (e.g., via automatic wage indexation, COLA clauses, or large public-sector union agreements).
  - Fiscal-theory perspective: increases in public wages that raise the value of public debt can be inflationary if markets expect inadequate surpluses to repay debt or require higher returns to hold such debt.
- Structural determinants of transmission:
  - Labor market institutions and regulations (e.g., wage bargaining structures, degree of government influence over wage setting) shape how public wage increases affect private-sector wages and prices.

*IMF Working Paper content provided in the source PDF.*

### 2. Prevailing macroeconomic conditions. The likelihood of an economy entering a wage-price spiral

### 2. Prevailing macroeconomic conditions. The likelihood of an economy entering a wage-price spiral

### Overview
- The likelihood of a wage-price spiral depends in part on macroeconomic conditions: workers’ bargaining power is typically greater when labor demand is strong and labor markets are tight; firms have more pricing power when aggregate demand is strong.
- Spillovers from wages to prices are expected to be larger when inflation and its expectations are high and when unemployment is low.
- The credibility of monetary policy matters: a more credible commitment to a low and stable inflation rate anchors long-run inflation expectations and weakens feedback from wages to prices.

### Empirical strategy
- Method: impulse responses estimated using Local Projections (Jordà 2005) over horizons h = 0, …, H.
- Baseline specification (2a): dependent variable y contains CPI, unemployment rate, private compensation per employee, nominal effective exchange rate, and central bank policy rate. PW is log average public-sector compensation per employee. Vector z contains controls including the ratio of employment in the public and private sectors.
- Country-fixed effects and year-fixed effects included. Variables transformed into log first differences (except interest rate, first-differenced only). Four lags of each variable included.
- Standard errors clustered at the country level, robust to heteroscedasticity and autocorrelation; 90 percent confidence intervals reported.
- Identification of public wage shocks: recursive identification where domestic retail energy prices are ordered first; assumption that changes in public compensation per employee do not respond to macroeconomic realizations within the same quarter (predetermined within-quarter).

### Data
- Quarterly data over the period from 1990: Q1-2022: Q2.
- Sample: 32 countries (25 advanced countries, and 7 emerging countries), for a total of around 3250 observations.
- Employee compensation variables: quarterly database described earlier in the chapter. Other variables from IMF’s International Finance Statistics (IFS).

### Key empirical findings — public wage shocks and private wages
- Effect size depends on public sector size:
  - In countries with a larger level of public employment: a 1 percent increase in average public wages leads to around 0.45 percent increase in average private sector wages (peak response), persistent up to 15 quarters.
  - In countries with a relatively small public sector: effect is around 0.07 percent and very short-lived.
- Openness to trade:
  - A 1 percent increase in average private wages leads to around 0.27 percent increase in public sector wages in countries more open to trade (persistent).
  - In less open countries: effect around 0.10 percent and significant for only two quarters.
- Unionization, bargaining coverage, and centralization:
  - In countries with higher union density and bargaining coverage and greater centralization, a 1 percent public wage shock increases average private wages by up to around 0.31 percent and 0.52 percent, respectively (Figures 11-12, LHS).
  - In countries with lower unionization and less centralized bargaining the effects are around 0.13 percent and 25 percent, respectively, and less persistent.
  - Note: similar pattern when measuring union density in the public sector.
- Inflation regime:
  - Using the 90th percentile of inflation across countries as a threshold (roughly 4.5 percent inflation) to define high-inflation environment.
  - A 1 percent increase in public wages leads to around 0.59 increase in private wages (peak response) in high-inflation environments — roughly 3 times larger than in a low-inflation environment.
- State of the economy (economic slack):
  - Effects of public wage increases on private wages are larger and more persistent during periods of lower economic slack (measured using civilian employment shortfall via transition function), consistent with higher bargaining power when labor demand is strong.

### Key empirical findings — public wage shocks and inflation
- General pattern: responses of the consumer price level follow the response of private wages to public wage shocks in both magnitude and persistence.
- Public sector size:
  - In countries with a larger public sector: a 1 percent public wage shock raises the consumer price level by around 0.3 percent — nearly 5 times bigger than the effect for countries with a relatively small public sector (around 0.06 percent). Effects are persistent in the larger-public-sector group, transitory in the smaller-public-sector group.
- Bargaining coverage and centralization:
  - Higher bargaining coverage: a 1 percent public wage shock increases the consumer price level by up to around 0.10 percent.
  - Greater centralization of wage bargaining: similar shock increases consumer prices by up to around 0.22 percent.
  - Lower unionization/coverage and less centralized bargaining: effects up to around 0.03 percent and 0.06 percent respectively, less persistent, and mostly insignificant.
- Inflation regime:
  - A 1 percent increase in public wages leads to around 0.44 increase in consumer prices (peak response) in high-inflation environments — roughly 3 times larger than the peak response in low-inflation environments (around 0.14 percent).
- Economic slack:
  - During periods of lower economic slack, consumer prices rise by about 0.2 percent in response to a public wage shock and remain positive and significant over long horizons.
  - No evidence of an effect of public wage shocks on consumer prices during periods of higher economic slack.
- Interpretation: public wages can contribute to higher and more persistent inflationary pressures via spillovers to private wages; the magnitude and persistence depend on labor market characteristics and prevailing macroeconomic conditions.

### Conclusion — summary statistics and broader findings
- Micro-econometric dataset: on average, public wages are around 10 percent higher than private wages for workers with similar socio-economic characteristics.
  - Premium by development level: 5.4 percent in AEs, 11.7 percent in EMs, and 12.8 percent in LIDCs.
  - Premium higher for women compared to men; driven primarily by a premium for lower-skilled workers.
- Time series (1990: Q1-2022: Q2) evidence:
  - Public wage premium present especially in EMDEs.
  - In high-income countries: premium relatively stable until the 2008 financial crisis, then increased sharply (reflecting a sharp decline in private wages), and then gradually decreased as private wages recovered post-crisis.
  - Premium varies counter-cyclically (public wages decrease by less than private wages during bad times).
  - Premium higher during election years, but only in the sample of low-income countries.
- Overall: public wages tend to drive wage-setting in economies where the public sector is relatively large; public wage shocks raise private wages and the price level, with heterogeneity driven by unionization, bargaining coverage, centralization, economic slack, and inflation regime.

*IMF Working Paper — "Public-Private Wage Differentials and Interactions Across Countries and Time" (chapter: 2. Prevailing macroeconomic conditions. The likelihood of an economy entering a wage-price spiral)*

### Annex I. Data Source, Methodology

### Annex I. Data Source, Methodology

### Note
- 1/PSM = Propensity Score Matching (results not used in our analysis).

### Regression method, main data sources, and country coverage
- IMF
  - Public dummy: Yes
  - Decomposition: -
  - PSM: 1/
  - Main data sources: National labour force and household surveys
  - Country coverage: Bosnia, Botswana, Burkina Faso, Cameroon, Costa Rica, Egypt, Ghana, Gambia, Honduras, Hungary, Jamaica, Kenya, Lithuania, Moldova, Mozambique, Philippines, Rwanda, Serbia, El Salvador, Zambia, Tunisia
- ECB
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: EU-SILC
  - Country coverage: Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Germany, Denmark, Spain, Estonia, Finland, France, Greece, Hungary, Ireland, Italy, Lithuania, Luxembourg, Latvia, Malta, Netherlands, Poland, Portugal, Romania, Slovakia, Slovenia, United Kingdom
- Panizza (2001), Panizza and Qiang (2005), and Mizala et al (2011)
  - Public dummy: Yes
  - Decomposition: Yes
  - PSM: (not indicated)
  - Main data sources: ECLAC household surveys
  - Country coverage: Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, El Salvador, Uruguay, Venezuela
- Aminu (2011)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: General household surveys
  - Country coverage: Nigeria
- Hyder and Reilly (2005), and Aslam and Kingdon (2009)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Living Standards Measurement Survey, and Labour Force Survey
  - Country coverage: Pakistan
- Glinskaya and Lokshin (2007) and Azam and Prakash (2010)
  - Public dummy: Yes
  - Decomposition: Yes
  - PSM: (not indicated)
  - Main data sources: National Sample Survey
  - Country coverage: India
- Birch (2006), Gibson (2007), Siminski (2013)
  - Public dummy: Yes
  - Decomposition: Yes
  - PSM: Yes
  - Main data sources: Household, Income and Labour Dynamics Survey, and International Social Survey Program Work Orientations
  - Country coverage: Australia, New Zealand
- Filmer and Lindauer (2001)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Labour Force Survey, and Household Expenditure Survey
  - Country coverage: Indonesia
- Finan et al (2015)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: National Household Survey
  - Country coverage: Albania, Armenia, Egypt, Georgia, India, Indonesia, Iraq, Korea, Laos, Pakistan, Sri Lanka, Tajikistan, Timor Leste, Uganda, USA, Vietnam
- Casero and Seshan (2006)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: National Household Survey
  - Country coverage: Djibouti
- KIPPRA (2013)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Economic Survey
  - Country coverage: Kenya
- Lucifora and Meurs (2006)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Labour Force Survey, Survey of Household Income and Wealth, and Enquete Emploi
  - Country coverage: France, United Kingdom, Italy
- Melly (2005)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Socio-Economic Panel
  - Country coverage: Germany
- Mueller (1998), and Lammam et al (2015)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Labour Market Activity Survey
  - Country coverage: Canada
- Molato (2005)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Labour Force Survey, and Family Income and Expenditure Survey
  - Country coverage: Philippines
- Nielsen and Rosholm (2001)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Priority Survey, and Living Conditions Monitoring Survey
  - Country coverage: Zambia
- Ognjenovic (2011)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Living Standards Measurement Survey
  - Country coverage: Serbia
- Rubil (2013)
  - Public dummy: Yes
  - Decomposition: -
  - PSM: (not indicated)
  - Main data sources: Labour Force Survey
  - Country coverage: Croatia

### Methodological notes (as presented)
- The Annex lists regression methods used in the literature and links each method to:
  - Whether a public dummy is used (Yes/No).
  - Whether a decomposition is performed (Yes/No).
  - Whether Propensity Score Matching (PSM) is used (1/PSM or Yes where indicated).
  - The main household or labour survey data sources employed.
  - The specific country coverage for each study or data source grouping.

*Public-Private Wage Differentials and Interactions Across Countries and Time, Working Paper No. WP/2023/064*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023064-print-pdf.pdf_