## wpiea2020068-print-pdf

## Source details

**Canonical URL:** [wpiea2020068-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020068-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020068-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020068-print-pdf.pdf.json)

---

### I. Introduction — context, motivation, and research contribution
- African policymakers committed to meet the Sustainable Development Goals (SDGs) by 2030.
- Corruption identified as a key obstacle to mobilizing domestic revenues to meet the SDGs (African Union Commission, 2019).
- Prior evidence cited:
  - IMF (2019a): among emerging and low-income countries, a country in the top 25 percent in terms of control of corruption collects 2¾ and 4 percent of GDP more in revenues, respectively, than a country in the lowest 25 percent.
  - Estimated additional annual spending needs of $2.6 trillion in low-income (and emerging markets) to reach the SDGs by 2030 (IMF, 2019b).
- Digitalization: benefits and risks
  - Potential benefits: reduce bureaucracy, improve efficiency, reduce human interactions and opportunities for corruption (Gupta et al., 2017; IMF 2018); digital tools could increase indirect tax collection at the border by up to 2 percent of GDP per year (IMF 2018); promote transparency, accountability, citizen participation (IMF, 2019a; IMF, 2018).
  - Risks: manipulation of digital records by officials with IT skills; cyberattacks; ambiguous net impact, context-dependent.
- Research focus and contribution
  - First paper to focus on digitalization’s effect on perception of corruption and trust in tax officials using individual-level Afrobarometer data and multiple digitalization indicators.
  - Uses multiple digitalization indicators: World Bank DAI and components; UN e-Government, e-Participation and Online services indices; World Bank open budget and PFM e-services indices.
  - Exploits exogenous variation from deployment of submarine cables at the subnational level to identify causal effects of internet use on corruption perception.
  - Assesses non-linearity: impact of intentional Internet shutdowns and government success in promoting ICT.

### II. Data, variables, and empirical strategy
- Data sources and sample
  - Primary data: sixth round of Afrobarometer surveys covering 36 African countries; surveys conducted in 2014/2015.
  - Sample used in main analysis: more than 23,000 individuals from 26 African countries; sample size (Obs) for descriptive statistics: 23,635.
  - Afrobarometer features: nationally representative individuals aged more than 18 years; multilevel random selection.
- Main dependent variables (Afrobarometer)
  - Perception of corruption of tax officials: Q53F — coded 0 = “None”; 1 = “Some of them”; 2 = “Most of them”; 3 = “All of them”.
  - Trust in tax officials: Q52D — coded 0 = “Not at all”; 1 = “Just a little”; 2 = “Somewhat”; 3 = “A lot”.
- Digitalization indicators
  - World Bank Digital Adoption Index (DAI) and components: aggregate DAI = simple average of three sub-indices (business, people, government).
    - DAI, Business = simple average of percentage of business with websites; number of secure servers; download speed; 3G coverage.
    - DAI, People = simple average of mobile access at home; internet access at home.
    - DAI, Government = simple average of core administrative systems; online public services; digital identification.
  - UN indices: E-Government Index; E-Participation Index; Online Service Index.
  - World Bank indices: Open Budget Index; PFM e-services Index.
  - Additional World Bank World Development Indicators used in robustness checks: use of digital payment methods; use of mobile phone to pay bills or access a financial institution account; use of internet for online services.
- Controls and covariates
  - Individual-level: age, education (0 to 9 where 0 = no formal education and 9 = some post-graduate education), employment, gender, urban (1) or not (0), wealth, perceptions on tax evasion, handling of public services, perceptions of democracy, unfair ethnic treatment, satisfaction with politicians, media channels used.
  - Country-level: World Bank Doing Business environment; oil deposit dummy (1 if respondent lives in area with petroleum site, 0 otherwise).
- Empirical strategy
  - Multilevel ordered probit model with two levels (individual, country).
  - Baseline equation: Corruption_ic = π Digitalization_c + β X_ic + μ_c + ε_ic, where π is main parameter of interest.
  - Identification: exogenous deployment of submarine cables at second administrative unit used to instrument internet use.

### III. Stylized facts and descriptive statistics
- Perception and trust (figures summary)
  - Perception of corruption of tax officials: around 9 out of 10 persons surveyed respond that tax officials are corrupt in some way.
  - Trust in tax department: only 1 out of 5 individuals surveyed have full trust in tax officials.
  - Strong negative association between digital adoption and corruption perception; positive association between digital adoption and trust.
- Selected sample descriptive statistics (Obs = 23,635)
  - Corruption of tax officials — Mean: 1.5; Std. Dev.: 0.9; Min: 0.0; Max: 3.0
  - Trust in tax officials — Mean: 1.5; Std. Dev.: 1.1; Min: 0.0; Max: 3.0
  - Aggregate DAI — Mean: 0.3; Std. Dev.: 0.1; Min: 0.2; Max: 0.6
  - DAI Business — Mean: 0.4; Std. Dev.: 0.1; Min: 0.2; Max: 0.7
  - DAI People — Mean: 0.2; Std. Dev.: 0.1; Min: 0.0; Max: 0.5
  - DAI Government — Mean: 0.4; Std. Dev.: 0.1; Min: 0.2; Max: 0.8
  - E-government index — Mean: 0.3; Std. Dev.: 0.1; Min: 0.1; Max: 0.5
  - Open budget — Mean: 0.4; Std. Dev.: 0.2; Min: 0.0; Max: 0.7
  - PFM e-services — Mean: 0.6; Std. Dev.: 0.1; Min: 0.3; Max: 0.7
  - Oil deposits — Mean: 0.0; Std. Dev.: 0.2; Min: 0.0; Max: 1.0
  - GDP per capita, log — Mean: 8.1; Std. Dev.: 0.9; Min: 6.7; Max: 9.8

### IV. Baseline empirical findings
- Main results (Table 1 and narrative)
  - Digitalization coefficient negative and strongly significant at the 1 percent level across specifications — greater digital adoption correlated with lower perception of corruption in tax officials.
  - Marginal effect example (aggregate DAI, column (1)):
    - Increase in index of digital adoption from first quartile to third quartile associated with decline in probability that respondents assess that all tax officials are corrupt from 11.2 to 7.5 percentage points.
    - Simultaneously, probability that respondents assess that any tax officials are corrupt increases from 11.7 to 16.7 percentage points.
  - Disaggregated DAI: DAI Business, DAI People, DAI Government coefficients negative and strongly significant at 1 percent; DAI Business coefficient larger than sub-indices.
  - E-Government and E-Participation: coefficients negative and significant at 1 percent.
  - Online Services, Open Budget, PFM e-services: negative correlation with corruption perception.
- Selected control variable effects (examples from column (1))
  - Handling of improving living standards: -0.0954***.
  - Reducing crime: -0.0814***.
  - Infrastructure: -0.0399***.
  - Unfair treatment of own ethnic group: 0.0842***.
  - Difficulty of finding which taxes to pay: 0.0702***.
  - Satisfaction with politicians: -0.2034***.
  - Media (newspapers/social media use): 0.0762***.
  - Oil deposits: 0.2682**.
  - GDP per capita, log: -0.0993* (mixed significance across columns).
- Sample and clustering details (examples)
  - Observations: 21,866 in some columns; up to 23,752 in others.
  - Number of regions: 301 to 334 depending on specification.
  - Number of countries: ranges from 26 to 32.
  - Region, ethnicity, and religion fixed effects included; standard errors reported; significance: *** p<0.01, ** p<0.05, * p<0.1.

### V. Robustness checks and alternative estimators
- Digital payments (Table 2)
  - Digitalization coefficients negative and significant:
    - Made or received digital payments (% age 15+) — -0.5143*** (SE: 0.175).
    - Made digital payments (% age 15+) — -0.6042*** (SE: 0.155).
    - Received digital payments (% age 15+) — -0.4371** (SE: 0.186).
    - Received digital payments, in labor force (% age 15+) — -0.8166*** (SE: 0.180).
  - Marginal effect example: increase in share using digital payments from first quartile (19 percent) to third quartile (41.5 percent) correlated with decline in perceived level that all tax officials are corrupt from 15.5 percent to 13.1 percent (column (1)).
- Mobile money (Table 3)
  - Mostly not statistically significant except column (7): Received payments for agricultural products through a mobile phone (% age 15+) — -8.4596*** (SE: 1.565).
  - Interpretation: mobile-phone-only payments may not offer the same deterrent to corruption perception as broader internet-based services.
- Internet use indicators and combinations (Tables 4–5)
  - Use of internet to pay bills or buy online: strong negative effects:
    - Used the internet to pay bills (% age 15+) — -1.7104*** (SE: 0.599).
    - Used the internet to pay bills or to buy something online (% age 15+) — -4.4076*** (SE: 1.570).
    - Used a mobile phone or the internet to access a financial institution account (% age 15+) — -1.1293*** (SE: 0.326).
    - Used a mobile phone or the internet to check account balance (% age 15+) — -1.1321*** (SE: 0.262).
- Endogeneity and Lewbel (2012) heteroskedasticity-based IV
  - Applied Lewbel method to address potential reverse causality; coefficients associated with digitalization remain consistent with baseline results.

### VI. Non-linearities: Internet shutdowns and government ICT promotion
- Internet shutdowns (Section V; Table 9)
  - Evidence on shutdowns: West (2016) documented 81 shutdowns between July 2015 and June 2016, estimating total cost in excess of $2.4 billion over that period.
  - Deloitte (2016) estimated daily GDP losses per day of full Internet shutdown:
    - High-connectivity country: at least 1.9 percent of its daily GDP per day.
    - Medium-level connectivity country: 1 percent of its daily GDP per day.
    - Low-connectivity country: 0.4 percent of its daily GDP per day.
  - Interaction results (Table 9): interactive variables between digitalization and number of Internet shutdowns are positive and significant in columns (6-9), implying the dampening effect of digitalization on corruption perception is reduced in countries with high number of Internet blackouts.
    - Selected estimates: Digitalization (aggregate): -1.5609*** (0.324); Digitalization*Internet shutdowns examples: 0.2833*** (0.032); 0.4848*** (0.167).
- Government success in promoting ICT (Table 10)
  - Government success index from World Economic Forum (1 = not successful at all; 7 = extremely successful).
  - Interaction results: interactive variables between digitalization and government success are negative and strongly significant in columns 1-4, implying successful government promotion of ICT amplifies the dampening effect of digitalization on corruption perception.
    - Selected estimates: Digitalization (aggregate): -4.3558*** (0.655); Digitalization*government success (aggregate): -1.9468*** (0.468); Government success (aggregate): -1.7782*** (0.403).

### VII. Causal identification: submarine cables as instrument (Section VI; Table 11)
- Instrument: district/region indicator = 1 if connected to a submarine Internet cable (data from Hjort and Poulsen (2019), originally Akamai (2012)); 0 otherwise.
- First-stage: submarine cables are a good predictor of Internet use; coefficient positive and strongly significant.
- IV estimates (Table 11):
  - OLS: Use of Internet coefficient = -0.0131** (0.006); Observations = 23,635; Number of countries = 27; R-squared = 0.109.
  - IV: Use of Internet coefficient = -0.4092*** (0.101); Observations = 23,640; Number of countries = 27; R-squared = 0.135.
  - Hansen p-value = 0.52.
  - Cragg-Donald Wald F statistic = 36.75.
  - Anderson-Rubin Wald test, p-value = 0.00.
- Interpretation: increased Internet usage causally lowers perception of corruption; IV estimate large and strongly significant at 1 percent.

### VIII. Trust in tax officials (Section IV; Table 8)
- Digitalization associated with higher trust in tax officials.
  - Aggregate DAI effect (example): increase in the index from first quartile to third quartile associated with:
    - increase in perceived trust in all tax officials from 14 percent to 17.8 percent;
    - decline in probability that none of the tax officials is trusted from 23.5 percent to 18.9 percent.
- Selected coefficient estimates (Table 8)
  - Digitalization (aggregate): 1.5325*** (0.389)
  - DAI Business: 1.2968*** (0.277)
  - DAI People: 0.7776** (0.381)
  - DAI Government: 0.2786* (0.126)
- Disaggregated indices and e-government measures: positive and highly significant in most specifications; DAI Business coefficient generally higher than government and people sub-indices.

### IX. Summary of empirical takeaways
- Using survey data from more than 23,000 individuals in 26 African countries and several indicators of digitalization:
  - Adoption of digital tools is associated with a lower perception of corruption and higher trust in tax officials.
  - On average, adoption of digital tools is correlated with a reduction of corruption perception in the tax administration by around 3 percentage points (paper reports around 3 percentage points and earlier summary notes "around 4 percentage points" depending on specification).
  - Business-sector digital adoption (DAI Business) shows particularly strong association.
  - Internet use and composite digitalization measures (E-Government, E-Participation, Online Service Index, Open Budget, PFM e-services) show robust negative associations with corruption perceptions.
  - Contextual factors (government service delivery, crime reduction, infrastructure, ethnic fairness, tax complexity, oil presence) materially affect perceptions.
  - Causal evidence from submarine cables supports that increased Internet use lowers perception of corruption.
  - Internet shutdowns undermine digitalization’s anti-corruption effects; government success in promoting ICT amplifies them.

### X. Policy implications and recommendations
- Accelerate digitalization while managing associated risks and challenges to combat corruption.
- Avoid intentional Internet shutdowns given their negative effects on transparency, trust, and economic activity.
- Pursue policies to promote ICT development, including regulations, education, and availability of ICT-related tools, to amplify digitalization’s positive effects on trust and reduce perceived corruption.
- Complement digital tool deployment with cybersecurity capacity and administrative strengthening to mitigate risks of manipulation and cyberattacks.
- Further work recommended to understand payoffs of investments to spur digitalization, starting from:
  - provision of basic infrastructure such as electricity;
  - availability of affordable and reliable internet services;
  - design of relevant applications and services.
- Potential gains if well implemented: improved tax administration and compliance; better targeting of social programs; improved public financial management; use of smart contracts in public procurement; exploration of distributed ledger technology (DLT) to promote transparency and reduce corruption.

*Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020068-print-pdf.pdf*

### References .............................................................................................................

### I.   INTRODUCTION

### Context and motivation
- African policymakers have committed to meet the Sustainable Development Goals (SDGs) by 2030.
- Corruption is identified as a key obstacle to mobilizing domestic revenues to meet the SDGs (African Union Commission, 2019).
- Prior evidence and estimates cited:
  - IMF (2019a) found governments perceived as less corrupt collect more revenue; among emerging and low-income countries, a country in the top 25 percent in terms of control of corruption collects 2¾ and 4 percent of GDP more in revenues, respectively, on average, than a country in the lowest 25 percent.
  - Estimated additional annual spending needs of $2.6 trillion in low-income (and emerging markets) to reach the SDGs by 2030 (IMF, 2019b).

### Digitalization: potential benefits and risks for corruption control
- Potential benefits outlined:
  - Digital tools in tax administrations can reduce bureaucracy, improve process efficiency and service delivery, and reduce opportunities for corruption by reducing human interactions (Gupta et al., 2017; IMF 2018).
  - For countries starting with a high level of corruption, digitalization is associated with better control of corruption (IMF 2019a).
  - Example: adopting digital tools could increase indirect tax collection at the border by up to 2 percent of GDP per year (IMF 2018).
  - Digitalization can promote transparency, accountability, citizen participation, reduce search costs, disseminate information cost-effectively, and reduce moral hazard in public sector monitoring (IMF, 2019a; IMF, 2018).
- Potential risks and new opportunities for corruption:
  - Digital records and public service systems can be manipulated by corrupt officials with high IT skills.
  - Digital systems are vulnerable to cyberattacks, which can disrupt government functions and jeopardize citizens’ digitally stored private information, particularly in countries with limited administrative capacity and underfunded security systems (IMF 2018; World Bank 2016).
  - Net impact of digitalization on corruption is ambiguous and context-dependent.

### Research focus and contribution
- Research gap: paucity of empirical studies on digitalization’s effect on perception of corruption.
- This paper estimates the effect of digitalization on the perception of corruption of tax officials in Africa using individual-level Afrobarometer data and multiple digitalization indicators.
- Key contributions:
  - First paper to focus on digitalization’s effect on perception of corruption and trust in tax officials.
  - Uses multiple, complementary indicators of digitalization (World Bank DAI and components; UN e-Government, e-Participation and Online services indices; World Bank open budget and PFM e-services indices).
  - Exploits exogenous variation from the deployment of submarine cables at the subnational level to identify causal effects of internet use on corruption perception.
  - Assesses non-linearity: impact of intentional Internet shutdowns and government success in promoting ICT.

### II.   DATA SOURCES, EMPIRICAL METHODOLOGY AND STYLIZED FACTS

### Data sources and sample
- Primary data: sixth round of Afrobarometer surveys covering 36 African countries; surveys conducted in 2014/2015.
- Sample used in main analysis: more than 23,000 individuals from 26 African countries.
- Afrobarometer features:
  - Nationally representative samples of individuals aged more than 18 years from rural and urban areas.
  - Multilevel random selection methods produce representative cross-sections.
- Main dependent variables from Afrobarometer:
  - Perception of corruption of tax officials: question Q53F — responses coded as 0 if “None”; 1 if “Some of them”; 2 if “Most of them”; 3 if “All of them”.
  - Trust in tax officials: question Q52D — responses coded as 0 if “Not at all”; 1 if “Just a little”; 2 if “Somewhat”; 3 if “A lot”.

### Digitalization indicators and identification strategy
- Digitalization measures used:
  - World Bank’s Digital Adoption Index (DAI) and components.
  - UN’s e-Government, e-Participation and Online services indices.
  - World Bank’s open budget and Public Financial Management (PFM) e-services indices.
- Causal identification:
  - Exploits exogeneous deployment of submarine cables at the subnational level (second administrative unit: district, province, or county depending on country) to identify the causal effect of internet use on perception of corruption.

### Internet shutdowns and ICT promotion as non-linear factors
- Internet shutdown evidence and economic loss estimates:
  - West (2016) documented 81 shutdowns between July 2015 and June 2016, estimating the total cost of shutdowns to be in excess of $2.4 billion over that period.
  - Deloitte (2016) estimated daily GDP losses per day of full Internet shutdown:
    - High-connectivity country: at least 1.9 percent of its daily GDP per day.
    - Medium-level connectivity country: 1 percent of its daily GDP per day.
    - Low-connectivity country (such as many African countries): 0.4 percent of its daily GDP per day.
- Hypothesized effects:
  - Intentional partial or total Internet outages prevent free access to online information, constrain citizens’ rights to accurate information and fact-checking, and hinder businesses’ ability to operate.
  - Internet blackouts could undermine trust in the Internet and in government actions, potentially increasing perceptions that government is corrupt or hiding information.
  - Government promotion of ICT (policy, education, availability of ICT tools) may amplify the anti-corruption benefits of digitalization.

### Key empirical findings (summary)
- Three main results:
  - Higher digital adoption is negatively associated with the perception of corruption of tax officials: on average, adoption of digital tools is correlated with a reduction of corruption perception in tax administration by around 4 percentage points.
  - Trust in tax officials is significantly higher in countries with higher levels of digital adoption.
  - The dampening effect of digitalization on corruption perception is reduced when the government intentionally shuts down the Internet; conversely, successful government promotion of ICT amplifies the dampening effect.
- Robustness:
  - Findings are robust to changes in econometric model and across different indicators of digitalization.
  - Causal evidence from submarine cable deployment at the subnational level supports the causal effect of internet use on perception of corruption of tax officials.

### Policy implications and recommendations
- African countries should accelerate digitalization while managing associated risks and challenges to combat corruption.
- Specific recommendations implied by findings:
  - Avoid intentional Internet shutdowns, given their negative effects on transparency, trust, and economic activity.
  - Establish and pursue policies to promote ICT development, including regulations, education, and availability of ICT-related tools, to amplify digitalization’s positive effects on trust and reduce perceived corruption.
  - Complement digital tool deployment with cybersecurity capacity and administrative strengthening to mitigate risks of manipulation and cyberattacks.

*Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020068-print-pdf.pdf*

### 2014. The DAI is  a  composite  index  measuring  the  extent  of  spread  of  digital technologies

### wpiea2020068-print-pdf - 2014. The DAI is  a  composite  index  measuring  the  extent  of  spread  of  digital technologies

### Data and digitalization indices
- Digital Adoption Index (DAI): aggregate DAI is the simple average of three sectoral sub-indices (business, people, government), each assigned equal weight.
  - DAI, Business: simple average of four normalized indicators — percentage of business with websites; number of secure servers; download speed; 3G coverage.
  - DAI, People: simple average of two normalized Gallup World Poll indicators — mobile access at home; internet access at home.
  - DAI, Government: simple average of three indicators — core administrative systems; online public services; digital identification.
- World Bank indices used:
  - Open Budget Index: measures public availability of budget data online in editable, reusable, free/legal-open format.
  - PFM e-services Index: measures automation and integration of public financial management digital processes (e-filing, e-procurement, digital payroll, debt management).
- United Nations indices used:
  - E-Government Index: weighted average of online services scope/quality, telecommunication infrastructure, human capital index.
    - Telecommunication infrastructure = simple average of: estimated internet users per 100 inhabitants; main fixed telephone lines per 100 inhabitants; mobile subscribers per 100 inhabitants; wireless broadband subscriptions per 100 inhabitants; fixed broadband subscriptions per 100 inhabitants.
    - Human capital component = weighted average composite of adult literacy rate; combined gross enrolment ratio; expected years of schooling; average years of schooling.
  - E-Participation Index: focuses on online use for government information provision, interactions, and engagement in decision-making.
- Additional World Bank World Development Indicators used in robustness checks: use of digital payment methods; use of mobile phone to pay bills or access a financial institution account; use of internet for online services.

### Controls and survey data
- Individual-level controls from sixth wave of Afrobarometer: age, education (0 to 9 where 0 = no formal education and 9 = some post-graduate education), employment status, gender, living area (urban = 1, otherwise 0), wealth (simple average based on car, water, latrine, roof materials), perceptions on ease of evading taxes, ease of finding which taxes to pay, handling by government of health, education, living standards, infrastructure, fighting crime, perceptions of democracy, treatment of own ethnic group, satisfaction with politicians, media channels used (newspapers, social media).
- Country-level controls: World Bank Doing Business business environment quality; oil deposit dummy (1 if respondent lives in area with petroleum site, 0 otherwise, based on USGS (2014) onshore petroleum site data).

### Empirical strategy
- Model class: multilevel (hierarchical) models to account for individuals nested within countries.
- Outcome: ordinal perception of corruption of tax officials with four categories: ‘None’, ‘Some of them’, ‘Most of them’, ‘All of them’.
- Estimation: multilevel ordered probit model with two levels (individual, country).
- Baseline specification (equation (1)):
  - Corrup tion_ic = π Digitalization_c + β X_ic + μ_c + ε_ic
  - Corrup tion_ic = perception of corruption by individual i in country c.
  - Digitalization_c = country-level digital adoption measure.
  - X_ic = individual-level characteristics.
  - μ_c and ε_ic assumed normally distributed; controls include religion and ethnicity fixed effects.
- Main parameter of interest: π (impact of digitalization on perception of corruption of tax officials).

### Variable construction (selected)
- Difficulty of evading taxes: variable 0–3 constructed from survey question on ease/difficulty of avoiding paying tax on property owned, higher values = more difficult evasion.
- Difficulty of finding which taxes to pay: variable coded 0 if “very easy”; 1 if “easy”; 2 if “difficult”; 3 if “very difficult”.
- Handling of health and education services: composite average of two survey questions, values 0–3.
- Fight against criminality: 0 (“very badly”) to 3 (“very well”).
- Infrastructure: composite average of three questions (water and sanitation, roads, electricity), values 0–3.
- Improving living standards: composite average of three questions (jobs, living standards, food), values 0–3.
- Unfair treatment of own ethnic group: coded 0 (“never”) to 3 (“always”).
- Democracy: coded 0 (“the country is not a democracy”) to 4 (“very satisfied”).
- Satisfaction with politicians: coded 0 (“strongly disapprove”) to 3 (“strongly approve”).
- Media channels: newspapers and social media coded 1 if used to get news, 0 otherwise.
- Oil deposits: dummy 1 if living in area with petroleum site.

### Stylized facts (figures summarized)
- Perception of corruption of tax officials: around 9 out of 10 persons surveyed respond that tax officials are corrupt in some way (Panel A).
- Association between digital adoption and corruption perception: strong negative association (Panel B).
- Averaged digital adoption indices by corruption category: level of digitalization is lowest for categories reporting that all or most tax officials are corrupt.
- Trust in tax department: only 1 out of 5 individuals surveyed have full trust in tax officials (Panel A, Figure 3).
- Association between digital adoption and trust: positive correlation — highly digitalized economies have stronger trust in tax officials; low digital adoption associated with lower trust.
- Averaged digital adoption indices by trust category: the lowest digital adoption when people do not trust at all tax officials.

### Baseline results — key findings (Table 1 and narrative)
- Main finding: coefficient on digitalization is negative and strongly significant at the 1 percent level across specifications — greater digital adoption correlated with lower perception of corruption in tax officials.
- Marginal effect example (aggregate DAI, column (1)):
  - Increase in index of digital adoption from first quartile to third quartile is associated with decline in probability that respondents assess that all tax officials are corrupt from 11.2 to 7.5 percentage points.
  - Simultaneously, probability that respondents assess that any tax officials are corrupt increases from 11.7 to 16.7 percentage points.
- Disaggregated DAI (columns 2–4):
  - DAI Business, DAI People, DAI Government coefficients negative and strongly significant at 1 percent level.
  - Magnitude: coefficient associated with DAI Business is higher than other sub-indices — impact of digitalization on perception of corruption is higher for companies.
- E-Government and E-Participation (columns 5–6): coefficients negative and significant at 1 percent level.
- Online Services sub-index (column 7): negative correlation with corruption perception.
- Open Budget and PFM e-services (columns 8–9): negative correlation between online openness of budget process, use of e-services in public finances, and perception of corruption of tax officials.
- Selected control variable findings (Table 1):
  - Handling of improving living standards: coefficient negative and strongly significant at 1 percent across columns (e.g., -0.0954*** in column (1)).
  - Reducing crime: coefficient negative and strongly significant at 1 percent across columns (e.g., -0.0814*** in column (1)).
  - Infrastructure: negative and significant in many columns (e.g., -0.0399*** in column (1)).
  - Unfair treatment of own ethnic group: positive and strongly significant at 1 percent (e.g., 0.0842*** in column (1)).
  - Difficulty of finding which taxes to pay: positive and strongly significant at 1 percent (e.g., 0.0702*** in column (1)).
  - Satisfaction with politicians: negative and strongly significant at 1 percent (e.g., -0.2034*** in column (1)).
  - Media (newspapers/social media use): positive and strongly significant at 1 percent in column (1) (0.0762***).
  - Oil deposits: positive and statistically significant in several columns (e.g., 0.2682** in column (1); 0.3036*** in column (4); 0.3753*** in column (9)).
  - GDP per capita, log: mixed significance across columns (e.g., -0.0993* in column (1); -0.1287*** in column (9)).
- Sample and clustering details (Table 1):
  - Observations: e.g., 21,866 in columns (1)–(4); 23,752 in columns (5)–(7); 23,007 in columns (8)–(9).
  - Number of regions: 301 in columns (1)–(4); 334 in columns (5)–(7); 310 in columns (8)–(9).
  - Number of countries: ranges from 26 to 32 across columns.
  - Region fixed effects: Yes. Ethnicity fixed effects: Yes. Religion fixed effects: Yes.
  - Standard errors reported in parentheses. Statistical significance: *** p<0.01, ** p<0.05, * p<0.1.

### Interpretation and analysis of baseline results
- Digitalization associated with lower perceived corruption of tax officials across multiple measures of digital adoption (aggregate DAI, sub-indices, E-Government, E-Participation, online services, open budget, PFM e-services).
- Better government performance in improving living standards, reducing crime, and infrastructure provision associated with lower perception of corruption, consistent with fiscal exchange theory.
- Perceived unfair ethnic treatment and complexity in finding which taxes to pay are associated with higher perception of corruption.
- Presence of oil deposits associated with higher perception of corruption in several specifications.

### Robustness checks and additional empirical evidence
- Alternative digital adoption proxies from World Bank Global Findex (Table 2):
  - Digital payments (share of population aged 15+ who received or sent digital payments): negative and significant effect on perceived corruption.
  - Example marginal effect (column (1)): increase in share using digital payments from first quartile (19 percent) to third quartile (41.5 percent) correlated with decline in perceived level that all tax officials are corrupt from 15.5 percent to 13.1 percent.
- Mobile phone payment indicators (Table 3):
  - Various mobile-phone-based payment measures largely not statistically significant except column (7).
  - Interpretation: sole use of mobile phones for payments not yet a strong deterrent of corruption perception; mobile phones often used for simple payments and may not support broader internet-based or fact-checking services.
- Internet use indicators (Table 4):
  - Use of internet (population aged 15+ and 25+) to pay bills or to buy something: negative and strongly significant effect on perception of corruption in all reported columns.
  - Combining mobile phone and internet (Table 5): results remain consistent (negative association).
- Endogeneity concerns and alternative estimator (Lewbel, 2012) (Section II.(ii)):
  - Potential reverse causality: corruption and legitimacy of tax officials may affect digital adoption.
  - Applied Lewbel (2012) heteroskedasticity-based instrumental variable approach to address endogeneity without external instruments.
  - Table 6: coefficients associated with digitalization remain consistent and aligned with baseline results when using Lewbel (2012) method.

### Summary of empirical conclusions
- Across multiple measures, specifications, and robustness checks (including alternative indicators and Lewbel IV method), higher digital adoption is consistently associated with lower perception of corruption in tax officials.
- Business-sector digital adoption (DAI Business) shows particularly strong association.
- Internet use and composite digitalization measures (E-Government, E-Participation, online services, open budget, PFM e-services) show robust negative associations with corruption perceptions.
- Contextual factors (government service delivery, crime reduction, infrastructure, ethnic fairness, tax complexity, oil presence) materially affect perceptions of tax official corruption.

*Source: wpiea2020068-print-pdf - 2014. The DAI is  a  composite  index  measuring  the  extent  of  spread  of  digital technologies (IMF working paper content provided).*

### 1. Thus, using an alternative econometric method does not change the findings of the paper.

### Thus, using an alternative econometric method does not change the findings of the paper.

### Robustness: Additional covariates and alternative methods
- Authors include macroeconomic controls (economic growth, inflation, civil war, poverty rate) and one governance variable—rule of law (World Bank’s Worldwide Governance Indicators).
- Results with these additional controls (Table 7) are "quite consistent with our baseline findings in Table 1."
- Coefficients associated with rule of law are negatively associated with high level of corruption perception.

### Digitalization and trust in tax officials (Section IV; findings from Table 8)
- Hypothesis: digitalization improves trust in tax officials by creating direct feedback/communication channels, reducing opacity of government transactions and promoting trust.
- Main statistical pattern:
  - Coefficients associated with different indicators of digital adoption are positive and highly significant at 1 percent level, except in column (4).
  - Using the aggregate index of digital adoption (column 1): an increase in the index from the first quartile to the third quartile is associated with:
    - an increase in the perceived trust in all tax officials from 14 percent to 17.8 percent;
    - a decline in the probability that none of the tax officials is trusted from 23.5 percent to 18.9 percent.
- Disaggregated indices:
  - Coefficients associated with DAI Business, DAI People, DAI Government and indicators of e-government, e-participation, Online Service Index, Open budget, and PFM e-services are positive and significant (many at 1 percent).
  - The coefficient associated with digital adoption by the business sector is higher than those of the government and people.
- Selected coefficient estimates from Table 8 (standard errors in parentheses; significance as reported):
  - Digitalization (aggregate): 1.5325*** (0.389)
  - DAI Business: 1.2968*** (0.277)
  - DAI People: 0.7776** (0.381)
  - DAI Government: 0.2786* (0.126)
- Sample and fixed effects (Table 8):
  - Observations range from 22,248 to 24,182 across specifications.
  - Number of regions range from 301 to 334; Number of countries = 26 (or 26–26/26/26/26/26/26/26/26 as in table).
  - Region, ethnicity, and religion fixed effects included.

### Internet shutdowns and government success in ICT (Section V)
- Internet shutdowns defined: "intentional disruption of internet-based communications...for a specific population, location, or mode of access."
- Data on number of Internet shutdowns extracted from The Internet Society and NetBlocks (not disaggregated by localized vs nationwide in this paper).
- Effect on digitalization–corruption link (Table 9):
  - Interactive variables between digitalization and number of Internet shutdowns are positive and significant in columns (6-9), implying that the dampening effect of digitalization on perception of corruption of tax officials is dampened in countries with high number of Internet blackouts.
  - Selected estimates (Table 9):
    - Digitalization (aggregate): -1.5609*** (0.324)
    - Digitalization*Internet shutdowns (examples): 0.2833*** (0.032) in one specification; 0.4848*** (0.167) in another specification.
  - Observations in Table 9 range from 21,758 to 23,644; Number of countries reported between 26 and 32 in different columns.
- Government success in promoting ICT (Table 10):
  - Government success index from World Economic Forum (survey question: "In your country, how successful is the government in promoting the use of ICTs? (1 = not successful at all; 7 = extremely successful)").
  - Interactive variables between digitalization and government success are negative and strongly significant in columns 1-4, implying that successful government promotion of ICT amplifies the dampening effect of digitalization on corruption perception.
  - Selected estimates (Table 10):
    - Digitalization (aggregate): -4.3558*** (0.655)
    - Digitalization*government success (aggregate): -1.9468*** (0.468)
    - Government success (aggregate): -1.7782*** (0.403)
  - Observations in Table 10: 19,909 to 21,481; Number of countries: 25–31 across columns.

### Identifying causal effect: arrival of submarine cables (Section VI; Table 11)
- Rationale: deployment of submarine Internet cables provides exogenous variation in Internet access; prior evidence indicates submarine cables increased internet penetration rates by 3-4 percentage points.
- Instrumental variable strategy:
  - Instrument: district/region indicator = 1 if connected to a submarine Internet cable (data from Hjort and Poulsen (2019), originally Akamai (2012)); 0 otherwise.
  - Outcome: individual-level use of Internet (Afrobarometer Q92B coded 0–4).
- First-stage and IV diagnostics:
  - First-stage: coefficient on submarine cables is positive and strongly significant (submarine cables are a good predictor of Internet use).
  - Hansen and Anderson-Rubin Wald tests indicate instrument is exogeneous and not weak.
- Key estimates (Table 11):
  - OLS: Use of Internet coefficient = -0.0131** (0.006); Observations = 23,635; Number of countries = 27; R-squared = 0.109.
  - IV: Use of Internet coefficient = -0.4092*** (0.101); Observations = 23,640; Number of countries = 27; R-squared = 0.135.
  - Hansen p-value = 0.52.
  - Cragg-Donald Wald F statistic = 36.75.
  - Anderson-Rubin Wald test, p-value = 0.00.
- Interpretation: increased Internet usage causally lowers perception of corruption; IV estimate is large and strongly significant at 1 percent.

### Conclusion: empirical takeaways and policy implications (Section VII)
- Empirical summary points:
  - Using survey data from more than 23,000 individuals in 26 African countries (Sixth Afrobarometer round) and several indicators of digitalization, adoption of digital tools is associated with a lower perception of corruption and higher trust in tax officials.
  - On average, adoption of digital tools is correlated with a reduction of corruption perception in the tax administration by around 3 percentage points.
  - The dampening effect of digitalization on corruption perception operates through improved trust in tax officials by increasing transparency and reducing opportunities for bribes and influence.
  - Internet shutdowns by governments undermine trust in the Internet and government actions and raise corruption perception.
  - Government policies that successfully promote ICT strengthen the impact of digitalization in reducing corruption perception.
- Policy recommendations and forward-looking considerations:
  - African countries should step up the adoption of digital tools to combat corruption.
  - Countries should refrain from intentionally shutting down the Internet.
  - Further work needed to understand payoffs of different investments to spur digitalization, starting from:
    - provision of basic infrastructure such as electricity;
    - availability of affordable and reliable internet services;
    - design of relevant applications and services.
  - Potential gains if well implemented: improved tax administration and compliance; better targeting of social programs; improved public financial management; use of smart contracts in public procurement; exploration of distributed ledger technology (DLT) to promote transparency and reduce corruption.

*Source: IMF working paper (excerpt provided).*

### REFERENCES

### wpiea2020068-print-pdf - REFERENCES

### Appendix A1: Descriptive statistics (sample-level)
- Sample size (Obs) for all variables listed: 23,635
- Variables with Mean, Std. Dev., Min, Max:
  - Corruption of tax officials — Mean: 1.5; Std. Dev.: 0.9; Min: 0.0; Max: 3.0
  - Trust in tax officials — Mean: 1.5; Std. Dev.: 1.1; Min: 0.0; Max: 3.0
  - Aggregate DAI — Mean: 0.3; Std. Dev.: 0.1; Min: 0.2; Max: 0.6
  - DAI Business — Mean: 0.4; Std. Dev.: 0.1; Min: 0.2; Max: 0.7
  - DAI People — Mean: 0.2; Std. Dev.: 0.1; Min: 0.0; Max: 0.5
  - DAI Government — Mean: 0.4; Std. Dev.: 0.1; Min: 0.2; Max: 0.8
  - E-government index — Mean: 0.3; Std. Dev.: 0.1; Min: 0.1; Max: 0.5
  - E-participation index — Mean: 0.3; Std. Dev.: 0.2; Min: 0.0; Max: 0.8
  - Online service index — Mean: 0.2; Std. Dev.: 0.2; Min: 0.0; Max: 0.7
  - Open budget — Mean: 0.4; Std. Dev.: 0.2; Min: 0.0; Max: 0.7
  - PFM e-services — Mean: 0.6; Std. Dev.: 0.1; Min: 0.3; Max: 0.7
  - Age, <25 — Mean: 0.2; Std. Dev.: 0.4; Min: 0.0; Max: 1.0
  - Age, >25 — Mean: 0.7; Std. Dev.: 0.5; Min: 0.0; Max: 1.0
  - Urban — Mean: 0.4; Std. Dev.: 0.5; Min: 0.0; Max: 1.0
  - Education — Mean: 3.5; Std. Dev.: 2.2; Min: 0.0; Max: 9.0
  - Wealth — Mean: 0.2; Std. Dev.: 0.4; Min: 0.0; Max: 1.0
  - Employment — Mean: 0.4; Std. Dev.: 0.5; Min: 0.0; Max: 1.0
  - Male — Mean: 0.5; Std. Dev.: 0.5; Min: 0.0; Max: 1.0
  - Difficulty of evading taxes — Mean: 2.2; Std. Dev.: 0.8; Min: 0.0; Max: 3.0
  - Handling of health and education services — Mean: 1.4; Std. Dev.: 0.8; Min: 0.0; Max: 3.0
  - Infrastructure — Mean: 1.3; Std. Dev.: 0.8; Min: 0.0; Max: 3.0
  - Reducing crime — Mean: 1.2; Std. Dev.: 0.8; Min: 0.0; Max: 3.0
  - Handling of improving living standards — Mean: 1.0; Std. Dev.: 0.8; Min: 0.0; Max: 3.0
  - Unfair treatment of own ethnic group — Mean: 0.6; Std. Dev.: 0.9; Min: 0.0; Max: 3.0
  - Democracy — Mean: 2.4; Std. Dev.: 1.0; Min: 0.0; Max: 4.0
  - Satisfaction with politicians — Mean: 1.5; Std. Dev.: 0.8; Min: 0.0; Max: 3.0
  - Difficulty of finding which taxes to pay — Mean: 1.9; Std. Dev.: 0.9; Min: 0.0; Max: 3.0
  - Media — Mean: 0.3; Std. Dev.: 0.4; Min: 0.0; Max: 1.0
  - Oil deposits — Mean: 0.0; Std. Dev.: 0.2; Min: 0.0; Max: 1.0
  - GDP per capita, log — Mean: 8.1; Std. Dev.: 0.9; Min: 6.7; Max: 9.8
  - GDP growth — Mean: 4.6; Std. Dev.: 2.0; Min: 0.6; Max: 8.8
  - Inflation — Mean: 7.0; Std. Dev.: 5.3; Min: 0.4; Max: 28.0
  - Civil war — Mean: 0.4; Std. Dev.: 0.3; Min: 0.0; Max: 1.0
  - Poverty — Mean: 0.5; Std. Dev.: 0.8; Min: 0.0; Max: 2.5

### Key regression robustness results (selected tables and coefficients)
- Table 2: Robustness — use of digital payments (Observations: 23,321; Regions: 323; Countries: 26; Region, Ethnicity, Religion fixed effects included)
  - Dependent variables across columns (1)–(4): 
    - (1) Made or received digital payments (% age 15+) — Digitalization coefficient: -0.5143*** (SE: 0.175)
    - (2) Made digital payments (% age 15+) — Digitalization coefficient: -0.6042*** (SE: 0.155)
    - (3) Received digital payments (% age 15+) — Digitalization coefficient: -0.4371** (SE: 0.186)
    - (4) Received digital payments, in labor force (% age 15+) — Digitalization coefficient: -0.8166*** (SE: 0.180)
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1

- Table 3: Robustness — use of mobile money (multiple dependent variables, varying sample sizes and regional counts)
  - Digitalization coefficients (columns (1)–(10)):
    - (1) Paid utility bills: using a mobile phone (% age 15+) — -0.6597 (SE: 0.512)
    - (2) Paid utility bills: using a mobile phone (% paying utility bills, age 15+) — -0.0859 (SE: 0.153)
    - (3) Received wages: through a mobile phone (% age 15+) — -2.0542 (SE: 1.401)
    - (4) Received private sector wages: through a mobile phone (% age 15+) — -1.3597 (SE: 1.560)
    - (5) Received public sector wages: through a mobile phone (% age 15+) — 0.2589 (SE: 1.225)
    - (6) Paid school fees: using a mobile phone (% age 15+) — -0.8592 (SE: 0.815)
    - (7) Received payments for agricultural products: through a mobile phone (% age 15+) — -8.4596*** (SE: 1.565)
    - (8) Received payments from self-employment: through a mobile phone (% age 15+) — -2.0726 (SE: 1.972)
    - (9) Received government payments: through a mobile phone (% age 15+) — 0.0748 (SE: 0.196)
    - (10) Mobile money account (% age 15+) — -0.0420 (SE: 0.176)
  - Observations vary by column (e.g., 23,321 down to 13,215); region and country counts vary accordingly.
  - Significance: column (7) shows a statistically significant negative coefficient at *** p<0.01.

- Table 5: Robustness — use of internet or mobile phone for online payments (Observations and region/country counts reported)
  - Digitalization coefficients:
    - (1) Used the internet to pay bills (% age 15+) — -1.7104*** (SE: 0.599) — Observations: 23,321; Regions: 323; Countries: 26
    - (2) Used the internet to pay bills, older adults (% age 25+) — -1.8411*** (SE: 0.590)
    - (3) Used the internet to pay bills or to buy something online (% age 15+) — -4.4076*** (SE: 1.570) — Observations: 21,775; Regions: 282; Countries: 26
    - (4) Used the internet to pay bills or to buy something online, older adults (% age 25+) — -3.9614*** (SE: 1.525)
  - Additional two specifications (mobile phone or internet to access financial account; to check account balance):
    - Used a mobile phone or the internet to access a financial institution account (% age 15+) — Digitalization: -1.1293*** (SE: 0.326); Observations: 21,645; Regions: 294; Countries: 26
    - Used a mobile phone or the internet to check account balance (% age 15+) — Digitalization: -1.1321*** (SE: 0.262)

- Table 7: Robustness — including additional macro variables (multiple specifications)
  - Panel A (Aggregate DAI, DAI Business, DAI People, DAI Government; Observations: 21,866; Countries: 26)
    - Digitalization coefficients:
      - Aggregate DAI — -1.3474*** (SE: 0.059); R-squared: 0.129
      - DAI Business — -1.2452*** (SE: 0.051); R-squared: 0.132
      - DAI People — -1.0100*** (SE: 0.052); R-squared: 0.124
      - DAI Government — -0.5618*** (SE: 0.043); R-squared: 0.116
  - Panel B (Expanded set with Digital Adoption Index components and macro controls; Observations vary 19,783–21,355; Countries: 24)
    - Selected Digitalization coefficients (columns (1)–(9)):
      - Aggregate DAI — -1.1174** (SE: 0.511)
      - DAI Business — -2.0942*** (SE: 0.472)
      - DAI People — -0.4181*** (SE: 0.083)
      - DAI Government — -0.5009* (SE: 0.265)
      - E-Government Index — -0.2541 (SE: 0.384)
      - E-Participation Index — -0.5210*** (SE: 0.173)
      - Online Service Index — -0.5405** (SE: 0.230)
      - Open budget — -0.6097*** (SE: 0.215)
      - PFM e-services — -1.5984*** (SE: 0.363)
    - Macro control coefficients shown (selected):
      - GDP growth — coefficients around -0.0016 to -0.0219 (not statistically significant in reported columns)
      - Inflation — coefficients range from -0.0118 to 0.0205**, with 0.0205** (SE: 0.008) in column (9)
      - Civil war — coefficients range from -0.0004 to 0.1774 (not statistically significant in reported columns)
      - Poverty — coefficients range from -0.1403** to 0.1609*** depending on specification
      - Rule of law — consistently negative and statistically significant in several specifications (e.g., -0.2527*** (SE: 0.085) in column (1))

- General estimation details across tables:
  - Region fixed effects, ethnicity fixed effects, religion fixed effects are commonly included.
  - Control variables are included in reported specifications unless noted otherwise.
  - Standard errors reported in parentheses; significance: *** p<0.01, ** p<0.05, * p<0.1
  - Number of regions and number of countries vary by specification (example counts include Regions: up to 323; Countries: up to 26)

*Source: wpiea2020068-print-pdf - REFERENCES (appendices and tables as provided in the source PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020068-print-pdf.pdf_