## wpiea2021123-print-pdf - Section VI examines the combined role of these factors in explaining compliance differences

## Source details

**Canonical URL:** [wpiea2021123-print-pdf - Section VI examines the combined role of these factors in explaining compliance differences](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021123-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021123-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021123-print-pdf.pdf.json)

---

### Data
- Source: Publicly-available COVID-19 behavior tracker survey developed by YouGov and the Institute of Global Health Innovation (IGHI) at Imperial College London.
- Sample coverage:
  - 29 advanced and emerging market economies.
  - All surveys conducted between March and December 2020.
  - Spanning over 420,000 individual observations.
  - Typical surveys: around 1,000 respondents in each round.
  - Balanced sample focus: 15 countries surveyed consistently throughout the sample period.
  - After October 2020, coverage narrowed to 15 countries out of the original 29.
  - Note: In China and India, the surveys are only representative of the online and urban online populations, respectively.
- Individual-level variables:
  - Demographics: gender, age, household size, number of children.
  - Health and employment status; occupation information on whether respondents work outside the home and sector for those who do.
  - Attitudes: individual attitudes towards COVID-19, confidence in government handling of the COVID-19 crisis and ability of national health systems to respond (questions not available in China, Hong Kong SAR, Thailand, Saudi Arabia, and the United Arab Emirates).
  - Weighting variable based on age, gender and region used throughout analysis.
- External time-series inputs:
  - Confirmed COVID-19 cases and deaths from the COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University.
  - Stringency of government lockdowns and adoption dates of government policies from the Oxford Covid-19 Government Response Tracker.

### Construction of compliance measures
- Physical distancing compliance index:
  - Takes value one if individuals report “Always” or “Frequently” complying simultaneously with all of the following behaviors:
    - (i) avoiding close contact with other people, including staying at least 6 feet away from people not living in the same household,
    - (ii) frequent handwashing,
    - (iii) covering coughs and sneezes,
    - (iv) cleaning frequently touched surfaces daily.
  - Avoidance of non-household close contacts proxied with responses on avoiding contact with people who have symptoms or may have been exposed to COVID-19 and avoiding social gatherings to protect themselves or others.
- Mask wearing compliance examined separately (CDC also recommends covering mouth and nose with a face mask when around others).
- Behavioral response categories are self-reported and mostly coded qualitatively: “Always”, “Frequently”, “Sometimes”, “Rarely”, “Not at All.”

### Social distancing since the start of the pandemic — key findings
- Overall trends (balanced sample of 15 countries):
  - The share of people complying with all physical distancing recommendations declined by 20 percentage points between April and September 2020.
  - The share of compliers ticked up again in the last quarter of 2020, coinciding with colder weather in the Northern Hemisphere and reintroduction of mobility restrictions and partial government lockdowns in several countries.
  - Even after the uptick, the share of compliers remained at least 10 percentage points below the April peak.
- Close non-household contacts:
  - Close contacts defined as spending time in close physical proximity (i.e., within 2 meters or 6 feet of distance).
  - On average, close contacts increased steadily after April, and were up by almost 90 percent as of September, before declining significantly from October onwards.
  - Trend in close non-household contacts mirrors the physical distancing index and is strongly correlated with mobility indicators such as Google mobility trends.
  - Mobility indicators show a similar hump-shape pattern, although the decline in mobility both precedes and exceeds the reported decline in close contacts, suggesting declining elasticity of self-reported compliance to mobility over time.
- Non-compliance prevalence:
  - The share of people reporting non-compliance with all recommended behaviors remained low and steady, hovering around 3 percent of all respondents.
- Mask wearing:
  - Large increase in mask wearing outside the home, especially in countries where mask wearing was not common before the pandemic.
  - Many respondents appear to mitigate increased physical proximity risk by wearing masks.
  - Mask wearing rises were especially large in countries that adopted mask mandates earlier; countries that adopted mask recommendations or mandates later saw smaller and more gradual increases.

### Country-level patterns and policy role
- Cross-country differences:
  - Almost all countries observed a significant decline in physical distancing and increased adherence to mask wearing between the first half and second half of 2020.
  - Declines in physical distancing were larger in countries that experienced stricter lockdowns earlier (examples: Italy, Spain, Singapore, United Kingdom).
  - Self-reported mask wearing rose almost everywhere, with especially large increases where mask mandates were implemented earlier (including early adopters with increases in January and February and others like Canada, France, Germany, United Kingdom with mandates a few months after).
  - Countries in Northern Europe that adopted mask recommendations or mandates later saw smaller and more gradual increases.
- Implication:
  - Government policies—mask mandates and mobility restrictions—play an important role in driving changes in social distancing over time and in explaining cross-country differences in compliance.

### Individual determinants of social distancing (empirical specification and controls)
- Empirical model:
  - Linear probability model: Compliance_i = μ_i + β′Group_i + γ′Z_i + C_i (Equation (1)).
  - Compliance = 1 if respondent reports “Always” or “Frequently” complying; two measures emphasized: the physical distancing index and mask wearing outside the household.
  - Regressions condition on country-specific, survey wave fixed effects μ.
  - Controls include household size, number of children, employment status and occupation (including telework), respondent health status, and lagged 7-day average new confirmed COVID-19 deaths (in later specifications).

### Age and gender
- Age profiles (percentage point differences relative to the 18 to 24 year-old male cohort):
  - Compliance with physical distancing increases sharply with age:
    - Increase of 11 percentage points (for men) compared to younger cohorts.
    - Increase of 18 percentage points (for women) compared to younger cohorts.
  - Age profile in mask wearing is flatter and non-monotonic, with typically higher compliance among younger and older cohorts compared to middle-age cohorts.
  - Compliance is consistently higher among women compared to men across all cohorts; the gender gap tends to be larger among older cohorts, especially for physical distancing.
- Robustness:
  - Results unchanged when controlling for employment status, occupation type (including work-from-home), household size, number of children.
  - Gender gap persists when restricting sample to full-time employed respondents.
  - Similar patterns observed across different constructions of the physical distancing index.

### Employment sector
- Sample restricted to full and part-time workers; baseline group = manufacturing workers not able to telework.
- Findings (percentage point differences relative to baseline):
  - Teleworkers: average increase in compliance of over 6 percentage points relative to manufacturing workers.
  - Physical distancing higher among food services workers compared to security and public transport workers.
  - Mask wearing more prevalent in healthcare and client-facing sectors (hospitality, retail food services).
  - Lower mask compliance in security, construction and logistics (male-dominated occupations).
- Interpretation: sector differences may reflect gender and social norms within occupations, not only activity type (client-facing vs group-based).

### Attitudes towards COVID-19 (risk perceptions and beliefs)
- Fear of contracting COVID-19 (comparing “Very Scared” to “Not Scared at All”):
  - 20 percentage points more likely to avoid people exposed to COVID-19.
  - 10 percentage points more likely to wash hands frequently.
  - about 8 percentage points more likely to wear face masks.
  - Report 20 percent fewer contacts with non-household members.
- Beliefs about mask effectiveness (1 = Disagree to 7 = Agree):
  - Mask wearing increases linearly with perceived effectiveness.
  - Example: mask wearing is 25 percentage points higher among those who believe masks are effective compared to those who do not.
- Implication: influencing beliefs about effectiveness of social distancing (e.g., masks) is important to shape compliance.

### Confidence in government response
- Comparing respondents with “a lot” of confidence to those with “no confidence at all”:
  - 6 percentage points more likely to wash hands often.
  - 7 percentage points more likely to avoid contact with people exposed to COVID-19.
  - Report 5 percent fewer close contacts with people outside the household in the two weeks prior to the survey.
- Contingent behavior (isolation if advised by healthcare professional):
  - High confidence: 14 percentage points more likely to isolate if advised.
  - Fair amount of confidence: 10 percentage points more likely to isolate.
- Interpretation: compliance improves gradually with degree of confidence in government response; incremental trust gains are associated with higher compliance.

### Lockdowns, mask mandates, and social distancing (policy impacts)
- Event-study model: Compliance_i,t = α_i + Σ_{τ=-5}^{10} β_τ Policy_{i,t-τ} + γ′Z_i,t + C_i,t (Equation (2)).
- Stay-at-home orders (lockdowns):
  - No significant anticipation effects detected in the 5 weeks before implementation.
  - Stay-at-home orders associated with a gradual build in physical distancing starting at week 0 and continuing over subsequent weeks.
  - Persistent and statistically significant increase in physical distancing of about 5 to 7 percentage points over time.
  - Specific behaviors:
    - Large and significant increases in the share of people avoiding going out and avoiding small social gatherings; responses build over first weeks and persist up to two months after implementation.
  - Workplace closures and internal movement restrictions show quantitatively similar effects, suggesting workplace closures contribute substantially to the increase in physical distancing.
- Mask mandates:
  - Implementation associated with immediate increase in self-reported mask wearing from the first week of implementation.
  - Average increase of about 30 percentage points above the pre-mandate baseline; effect peaks about six weeks after mandates and remains stable thereafter.
  - No significant anticipation effects before mandates.
  - Heterogeneity:
    - Impact concentrated in countries with low baseline mask usage (defined as average mask usage above 50 percent in January 2020 — high baseline group includes only countries in Asia).
    - Stronger responses among individuals who believe masks protect themselves or others; nonetheless, mandates increase compliance even among those skeptical of mask effectiveness.
    - Increases in mask wearing are similar across age groups and across levels of trust in government — mandates can overcome variation in trust.

### What drives regional differences in compliance?
- Model: Compliance_i,t = α + λ′Region_i + β_I′Individual_i,t + β_L′Policies_i,t + γ′Z_i,t + C_i,t (Equation (3)).
- Analysis period for regional comparisons: end-March to end-September (balanced sample).
- Baseline comparator region: Asia.
- Baseline regional gaps:
  - Higher self-reported compliance in the Americas compared to Asia; lower in Europe (with Southern and Northern Europe examined separately).
- Decomposition approach: conditional decomposition per Gelbach (2016) to apportion how much regional gaps are explained by individual factors versus government policies versus residual.
- Key result:
  - Individual factors and government policies typically explain over two-thirds of regional differences in compliance.
  - Relative contributions vary:
    - Government policies are key driver of lower mask wearing in Northern Europe (consistent with mask mandates’ impact).
    - Individual factors (attitudes towards COVID-19 and trust in government) play a larger role in explaining differences in physical distancing.

### Conclusions and policy-relevant implications
- Trends:
  - Adherence to physical distancing declined across most countries, with an increase in non-household contacts not fully captured by mobility data.
  - Mask wearing increased, especially in countries where mask use was uncommon prior to the pandemic.
- Demographics and messaging:
  - Women and older individuals consistently report higher compliance across measures.
  - Occupational and sector differences mirror gender and social norms; targeting public health messaging should account for these gaps.
- Trust and compliance:
  - Higher trust in government responses correlates with higher compliance and greater willingness to isolate when advised.
  - Maintaining and rebuilding public trust requires transparent, well-organized responses, consistent and clear communications, and coordinated health and economic actions.
- Policy effectiveness and equity:
  - Stay-at-home orders and related mobility restrictions reduce social interactions; early lockdowns can curb infections but are costly for economically vulnerable populations.
  - Low-cost non-pharmaceutical interventions such as mask mandates have high social distancing multipliers: low implementation costs relative to high potential benefits.
- Residual variation:
  - Even after accounting for individual factors and policies, residual regional differences may reflect community transmission levels, cultural and social norms, enforcement quality, and public health communication effectiveness.

### Suggested avenues for future research
- Compare self-reported compliance with alternative measures of compliance (direct observation; measure partial/incorrect compliance).
- Examine how compliance relates to economic constraints faced by respondents to inform policies alleviating economic hardship.
- Investigate how government policies influence individual attitudes over time and the role of public health communication in shaping those attitudes.

*Source: IMF Working Paper (section IV–VI content as provided in the PDF).*

### Section VI examines the combined role of these factors in explaining compliance differences

### wpiea2021123-print-pdf - Section VI examines the combined role of these factors in explaining compliance differences

### Data
- Source: Publicly-available COVID-19 behavior tracker survey developed by YouGov and the Institute of Global Health Innovation (IGHI) at Imperial College London.
- Sample coverage:
  - 29 advanced and emerging market economies.
  - All surveys conducted between March and December 2020.
  - Spanning over 420,000 individual observations.
  - Typical surveys: around 1,000 respondents in each round.
  - Balanced sample focus: 15 countries surveyed consistently throughout the sample period.
  - After October 2020, coverage narrowed to 15 countries out of the original 29.
  - Note: In China and India, the surveys are only representative of the online and urban online populations, respectively.
- Individual-level variables included:
  - Demographics: gender, age, household size, number of children.
  - Health and employment status; occupation information on whether respondents work outside the home and sector for those who do.
  - Attitudes: individual attitudes towards COVID-19, confidence in government handling of the COVID-19 crisis and ability of national health systems to respond (questions not available in China, Hong Kong SAR, Thailand, Saudi Arabia, and the United Arab Emirates).
  - Weighting variable based on age, gender and region used throughout analysis.
- External time-series inputs:
  - Confirmed COVID-19 cases and deaths from the COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University.
  - Stringency of government lockdowns and adoption dates of government policies from the Oxford Covid-19 Government Response Tracker.

### Construction of compliance measures
- Physical distancing compliance index:
  - Takes value one if individuals report “Always” or “Frequently” complying simultaneously with all of the following behaviors:
    - (i) avoiding close contact with other people, including staying at least 6 feet away from people not living in the same household,
    - (ii) frequent handwashing,
    - (iii) covering coughs and sneezes,
    - (iv) cleaning frequently touched surfaces daily.
  - Avoidance of non-household close contacts proxied with responses on avoiding contact with people who have symptoms or may have been exposed to COVID-19 and avoiding social gatherings to protect themselves or others.
- Mask wearing compliance examined separately (CDC also recommends covering mouth and nose with a face mask when around others).
- Behavioral response categories are self-reported and mostly coded qualitatively: “Always”, “Frequently”, “Sometimes”, “Rarely”, “Not at All.”

### Social distancing since the start of the pandemic — key findings
- Overall trends (balanced sample of 15 countries):
  - The share of people complying with all physical distancing recommendations declined by 20 percentage points between April and September 2020.
  - The share of compliers ticked up again in the last quarter of 2020, coinciding with colder weather in the Northern Hemisphere and reintroduction of mobility restrictions and partial government lockdowns in several countries.
  - Even after the uptick, the share of compliers remained at least 10 percentage points below the April peak.
- Close non-household contacts:
  - Close contacts defined as spending time in close physical proximity (i.e., within 2 meters or 6 feet of distance).
  - On average, close contacts increased steadily after April, and were up by almost 90 percent as of September, before declining significantly from October onwards.
  - Trend in close non-household contacts mirrors the physical distancing index and is strongly correlated with mobility indicators such as Google mobility trends.
  - Mobility indicators show a similar hump-shape pattern, although the decline in mobility both precedes and exceeds the reported decline in close contacts, suggesting declining elasticity of self-reported compliance to mobility over time.
- Non-compliance prevalence:
  - The share of people reporting non-compliance with all recommended behaviors remained low and steady, hovering around 3 percent of all respondents.
- Mask wearing:
  - Large increase in mask wearing outside the home, especially in countries where mask wearing was not common before the pandemic.
  - Many respondents appear to mitigate increased physical proximity risk by wearing masks.
  - Mask wearing rises were especially large in countries that adopted mask mandates earlier; countries that adopted mask recommendations or mandates later saw smaller and more gradual increases.

### Country-level patterns and policy role
- Cross-country differences:
  - Almost all countries observed a significant decline in physical distancing and increased adherence to mask wearing between the first half and second half of 2020.
  - Declines in physical distancing were larger in countries that experienced stricter lockdowns earlier (examples: Italy, Spain, Singapore, United Kingdom).
  - Self-reported mask wearing rose almost everywhere, with especially large increases where mask mandates were implemented earlier (including early adopters with increases in January and February and others like Canada, France, Germany, United Kingdom with mandates a few months after).
  - Countries in Northern Europe that adopted mask recommendations or mandates later saw smaller and more gradual increases.
- Implication:
  - Government policies—mask mandates and mobility restrictions—play an important role in driving changes in social distancing over time and in explaining cross-country differences in compliance.

### Next analytical steps described
- Subsequent sections (IV and V in the source) examine:
  - How compliance varies with individual factors, including personal beliefs and attitudes towards COVID-19.
  - How compliance responds to government policies such as mobility restrictions and mask mandates.

*wpiea2021123-print-pdf - Section VI examines the combined role of these factors in explaining compliance differences*

### 2020.  Green  (red) arrows indicate  an increase  (decrease)  in compliance  over time.

### IV. INDIVIDUAL DETERMINANTS OF SOCIAL DISTANCING

### Empirical specification and data controls
- Linear probability model estimated: Compliance_i = μ_i + β′Group_i + γ′Z_i + C_i (Equation (1)).
- Compliance = 1 if respondent reports “Always” or “Frequently” complying with recommended social distancing behaviors, and 0 otherwise.
- Two compliance measures emphasized: the physical distancing index and mask wearing outside the household.
- Regressions condition on country-specific, survey wave fixed effects μ.
- Controls include household size, number of children, employment status and occupation (including telework), respondent health status (self/household positive test, COVID-19-like symptoms, high-risk underlying conditions), and lagged 7-day average new confirmed COVID-19 deaths (in later specifications).

### Age and gender
- Estimated age profiles are reported as percentage point differences relative to the 18 to 24 year-old male cohort (baseline impact = 0 by construction).
- Findings:
  - Compliance with physical distancing increases sharply with age:
    - Increase of 11 percentage points (for men) compared to younger cohorts.
    - Increase of 18 percentage points (for women) compared to younger cohorts.
  - Age profile in mask wearing is flatter and non-monotonic, with typically higher compliance among younger and older cohorts compared to middle-age cohorts.
  - Compliance is consistently higher among women compared to men across all cohorts; the gender gap tends to be larger among older cohorts, especially for physical distancing.
- Robustness:
  - Results unchanged when controlling for employment status, occupation type (including work-from-home), household size, number of children.
  - Gender gap persists when restricting sample to full-time employed respondents.
  - Similar patterns observed across different constructions of the physical distancing index.

### Employment sector
- Sample restricted to full and part-time workers; baseline group = manufacturing workers not able to telework.
- Findings (percentage point differences relative to baseline):
  - Teleworkers: average increase in compliance of over 6 percentage points relative to manufacturing workers.
  - Physical distancing higher among food services workers compared to security and public transport workers.
  - Mask wearing more prevalent in healthcare and client-facing sectors (hospitality, retail food services).
  - Lower mask compliance in security, construction and logistics (male-dominated occupations).
- Interpretation: sector differences may reflect gender and social norms within occupations, not only activity type (client-facing vs group-based).

### Attitudes towards COVID-19 (risk perceptions and beliefs)
- Fear of contracting COVID-19:
  - Comparing respondents reporting “Very Scared” to those “Not Scared at All”:
    - 20 percentage points more likely to avoid people exposed to COVID-19.
    - 10 percentage points more likely to wash hands frequently.
    - about 8 percentage points more likely to wear face masks.
    - Report 20 percent fewer contacts with non-household members.
- Beliefs about mask effectiveness (1 = Disagree to 7 = Agree):
  - Mask wearing increases linearly with perceived effectiveness.
  - Example: mask wearing is 25 percentage points higher among those who believe masks are effective compared to those who do not.
- Implication: influencing beliefs about effectiveness of social distancing (e.g., masks) is important to shape compliance.

### Confidence in government response
- Comparing respondents with “a lot” of confidence to those with “no confidence at all”:
  - 6 percentage points more likely to wash hands often.
  - 7 percentage points more likely to avoid contact with people exposed to COVID-19.
  - Report 5 percent fewer close contacts with people outside the household in the two weeks prior to the survey.
- Contingent behavior (isolation if advised by healthcare professional):
  - High confidence: 14 percentage points more likely to isolate if advised.
  - Fair amount of confidence: 10 percentage points more likely to isolate.
- Interpretation: compliance improves gradually with degree of confidence in government response; incremental trust gains are associated with higher compliance.

### Lockdowns, mask mandates, and social distancing (policy impacts)
- Event-study specification used: Compliance_i,t = α_i + Σ_{τ=-5}^{10} β_τ Policy_{i,t-τ} + γ′Z_i,t + C_i,t (Equation (2)).
- Stay-at-home orders (lockdowns):
  - No significant anticipation effects detected in the 5 weeks before implementation.
  - Stay-at-home orders associated with a gradual build in physical distancing starting at week 0 and continuing over subsequent weeks.
  - Persistent and statistically significant increase in physical distancing of about 5 to 7 percentage points over time.
  - Specific behaviors:
    - Large and significant increases in the share of people avoiding going out and avoiding small social gatherings; responses build over first weeks and persist up to two months after implementation.
  - Workplace closures and internal movement restrictions show quantitatively similar effects, suggesting workplace closures contribute substantially to the increase in physical distancing.
- Mask mandates:
  - Implementation associated with immediate increase in self-reported mask wearing from the first week of implementation.
  - Average increase of about 30 percentage points above the pre-mandate baseline; effect peaks about six weeks after mandates and remains stable thereafter.
  - No significant anticipation effects before mandates.
  - Heterogeneity:
    - Impact concentrated in countries with low baseline mask usage (defined as average mask usage above 50 percent in January 2020 — high baseline group includes only countries in Asia).
    - Stronger responses among individuals who believe masks protect themselves or others; nonetheless, mandates increase compliance even among those skeptical of mask effectiveness.
    - Increases in mask wearing are similar across age groups and across levels of trust in government — mandates can overcome variation in trust.

### What drives regional differences in compliance?
- Specification: Compliance_i,t = α + λ′Region_i + β_I′Individual_i,t + β_L′Policies_i,t + γ′Z_i,t + C_i,t (Equation (3)).
- Analysis period for regional comparisons: end-March to end-September (balanced sample).
- Baseline comparator region: Asia.
- Baseline regional gaps (physical distancing and mask wearing):
  - Higher self-reported compliance in the Americas compared to Asia; lower in Europe (with Southern and Northern Europe examined separately).
- Decomposition approach: conditional decomposition per Gelbach (2016) to apportion how much regional gaps are explained by individual factors versus government policies versus residual.
- Key result:
  - Individual factors and government policies typically explain over two-thirds of regional differences in compliance.
  - Relative contributions vary:
    - Government policies are key driver of lower mask wearing in Northern Europe (consistent with mask mandates’ impact).
    - Individual factors (attitudes towards COVID-19 and trust in government) play a larger role in explaining differences in physical distancing.

### Conclusions and policy-relevant implications
- Trends:
  - Adherence to physical distancing declined across most countries, with an increase in non-household contacts not fully captured by mobility data.
  - Mask wearing increased, especially in countries where mask use was uncommon prior to the pandemic.
- Demographics and messaging:
  - Women and older individuals consistently report higher compliance across measures.
  - Occupational and sector differences mirror gender and social norms; targeting public health messaging should account for these gaps.
- Trust and compliance:
  - Higher trust in government responses correlates with higher compliance and greater willingness to isolate when advised.
  - Maintaining and rebuilding public trust requires transparent, well-organized responses, consistent and clear communications, and coordinated health and economic actions.
- Policy effectiveness and equity:
  - Stay-at-home orders and related mobility restrictions reduce social interactions; early lockdowns can curb infections but are costly for economically vulnerable populations.
  - Low-cost non-pharmaceutical interventions such as mask mandates have high social distancing multipliers: low implementation costs relative to high potential benefits.
- Residual variation:
  - Even after accounting for individual factors and policies, residual regional differences may reflect community transmission levels, cultural and social norms, enforcement quality, and public health communication effectiveness.

### Suggested avenues for future research (from source)
- Compare self-reported compliance with alternative measures of compliance (direct observation; measure partial/incorrect compliance).
- Examine how compliance relates to economic constraints faced by respondents to inform policies alleviating economic hardship.
- Investigate how government policies influence individual attitudes over time and the role of public health communication in shaping those attitudes.

*Source: IMF Working Paper (section IV–VI content as provided in the PDF).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021123-print-pdf.pdf_