## 5.2 Technology, Mobility, and Unemployment

## Source details

**Canonical URL:** [5.2 Technology, Mobility, and Unemployment](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020208-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020208-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020208-print-pdf.pdf.json)

---

### Data and measurement
- Labor market: Current Population Survey (CPS); unemployment rate constructed at MSA, state, and national levels.
- Mobility: Google Community Mobility Reports; GPS-based indices by activity type (retail and recreation, workplaces, transit station, parks, etc.); disaggregated to county level for the US; index relative to January–February baseline.
- Lockdown / NPI measures (Keystone): 11 dummy variables — (i) closing of public venues; (ii) ban of gathering size 500-101; (iii) ban of gathering size 100-26; (iv) ban of gathering size 25-11; (v) ban of gathering size 10-0; (vi) full lockdown; (vii) non-essential services closure; (viii) ban of religious gatherings; (ix) school closure; (x) shelter in place; (xi) social distancing. Lockdown index = average across the 11 dummies (100% = all 11 NPIs in place).
- Consumer spending: aggregated anonymized consumer purchase data from Affinity Solutions Inc (Chetty et al.[2020]); used at the state level.
- IT adoption: establishment survey on IT budget per employee (CiTBDs Aberdeen, 2016) covering more than 2,800,000 establishments in all US states. Log IT budget per employee (IT_e) used; regression IT_e = δ + α_g(e) + θ_ind(e) + ≤_i with α_g geographic fixed effects (state or MSA) and θ_ind industry (2-digit) fixed effects.

### Descriptive patterns relevant for 5.2
- Lockdown intensity:
  - Strong in April with some states having all 11 NPIs in place; partially loosened in May and June; loosened further in July.
- Mobility trends:
  - Mobility to retail and recreation declined by 30 percent across the U.S. relative to the January baseline.
  - Strongest responding states saw mobility drop by more than 60% relative to the pre-COVID baseline; least responding states saw drops by around 20%.
  - Mobility began gradual increase since mid-April with large heterogeneity across states (Washington D.C. near lower end, Mississippi near upper end).
  - Some states (e.g., South Dakota) saw mobility return to pre-COVID levels as early as mid-May; others (e.g., DC) had not seen strong increase by July despite many restrictions lifted.
- Timing around policy implementation:
  - Evidence consistent with voluntary social distancing: lockdown implementation appears to have occurred after mobility had already declined; mobility already increased before de jure lockdowns were lifted.
- Infection rates around loosening:
  - Infection rates remained relatively stable immediately after measures were lifted; some states show a spike after around 40 days, others remained at pre-reopening levels.

### State-level empirical findings (Equation 4 and Table 1)
- Regression specification:
  - ∆UR_s = α + β1 ∆Mobility_s + β2 IT_s + β3 ∆Mobility_s * IT_s + X'_s σ + (X_s * Mobility_s)' γ + ε_s
  - IT_s = dummy for above-median IT adoption; β3 tests difference in slope between high- and low-IT states.
- Main patterns:
  - Job losses correlate with mobility declines only in states with relatively low IT adoption; in high-IT adoption states the increase in unemployment shows relatively little relationship to mobility declines.
  - Example: Colorado and Nevada both had mobility decline of (a bit more than) 40%; increase in the unemployment rate was twice as large in Nevada (low-IT) than in Colorado (high-IT).
- Key estimated magnitudes (narrative and Table 1 exact coefficients):
  - Column (1): IT = -0.0180 (0.010); ∆Mobility = -0.148 (0.070); R-squared = 0.0575; N = 51.
  - Column (2): IT = 0.134 ∗∗∗ (0.037); ∆Mobility = -0.505 ∗∗∗ (0.102); R-squared = 0.116; N = 51.
  - Column (3): IT = 0.142 ∗∗∗ (0.033); ∆Mobility = -0.622 (0.377); ∆Mobility×IT = 0.463 ∗∗∗ (0.116); R-squared = 0.478; N = 51.
  - Column (4) (with controls): IT = 0.142 ∗∗∗ (0.033); ∆Mobility×IT = 0.476 ∗∗∗ (0.105); R-squared = 0.598; N = 51.
- Interpreted magnitudes:
  - A state with stronger IT adoption saw a 1.8 percentage points weaker increase in the unemployment rate relative to lower-IT states (narrative).
  - Column (2) average effect: a 10 percentage points stronger drop in mobility is associated with a 1.5 percentage points stronger increase in the unemployment rate (narrative based on regression scaling).
  - Column (3) for low IT adopters: a 10 percentage points larger decline in mobility associated with a 5 percentage points larger increase in the unemployment rate (narrative and example comparing Michigan and Ohio).
  - Interaction coefficient (point estimate) on IT × mobility: 0.463; mobility coefficient = -0.505; sum for high-IT states = -0.505 + 0.463 = -0.042 (near-zero slope for high-IT adopters).
- Robustness: interaction coefficient between IT adoption and mobility remains almost identical after adding controls (GDP per capita, population density, manufacturing share, and interactions with mobility).

### Individual-level evidence (CPS, MSA-level IT and mobility; Equations 5–6 and Tables 2–4)
- Estimation framework:
  - Linear probability model: Unemployed_i,t (April/May 2020) regressed on ∆Mobility_msa(i),t, IT_msa(i), interaction ∆Mobility×IT, individual controls Z_i, MSA controls X_msa(i), state fixed effects α_s(i); standard errors clustered at MSA level; regressions weighted by respondent weight.
- Main individual-level magnitudes (Table 2 narrative and exact coefficients):
  - Table 2 Column (1): ∆Mobility = -0.181 ∗∗∗ (0.031); IT = -0.00697 (0.005); R-squared = 0.00346; N = 7,181.
  - Table 2 Column (2): ∆Mobility = -0.239 ∗∗∗ (0.037); IT = 0.0187 ∗∗∗ (0.007); R-squared = 0.00418; N = 12,718.
  - Table 2 Column (3): ∆Mobility = -0.742 (1.559); IT = 0.0193 ∗∗ (0.009); ∆Mobility×IT = 0.0699 ∗∗∗ (0.023); R-squared = 0.0293; N = 27,181.
  - Table 2 Column (4): ∆Mobility = 0.0236 (1.358); IT = 0.0292 ∗∗∗ (0.011); ∆Mobility×IT = 0.0656 ∗∗ (0.032); R-squared = 0.0384; N = 27,181; FEs included.
- Interpreted individual-level magnitudes:
  - A larger decline in MSA mobility is associated with a higher probability of being unemployed in April–May 2020.
  - A one standard deviation larger decline in mobility (equal to 10 pp) increases unemployment probability by 2.4 percentage points in a low-IT MSA (narrative).
  - A one standard deviation larger level of IT adoption in an MSA reduces that increase by 0.7 percentage points, to 1.7 percentage points (narrative).
  - Results stable after controlling for MSA per capita income, share with a three-year Bachelor’s degree, share of minorities, and February unemployment rate.
- Fixed effects and explanatory power:
  - Including additional fixed effects increases R-squared from 0.418% to 3.8% (reported as 0.418% → 3.8% in narrative; Table 2 reports R-squared values).
- Heterogeneity (Table 4 exact coefficients and narrative):
  - ∆Mobility×IT×Male: 0.0306 ∗ (0.017) column (1); 0.0494 ∗ (0.025) column (2).
  - ∆Mobility×IT×Female: 0.0684 ∗∗∗ (0.019) column (1); 0.0894 ∗∗∗ (0.028) column (2).
  - ∆Mobility×IT×White: 0.0346 ∗∗ (0.017) column (3); 0.0610 ∗∗ (0.027) column (4).
  - ∆Mobility×IT×Non-White: 0.0577 ∗ (0.030) column (3); 0.0909 ∗∗∗ (0.035) column (4).
  - ∆Mobility×IT×High/Med Educ: 0.0520 ∗∗∗ (0.016) column (5); 0.0712 ∗∗∗ (0.025) column (6).
  - ∆Mobility×IT×Low Educ: -0.0324 (0.049) column (5); 0.0122 (0.054) column (6).
  - Interpretation: mitigating impact of IT is present for males, females, whites, non-whites, and high/medium education groups; no mitigating impact found for low-education individuals.

### Mechanisms, robustness, and alternate measures (Tables 5–6 and narrative)
- Work-from-home (WFH) channel:
  - High correlation between share of jobs that can be done from home in an MSA and IT (Figure 12).
  - Replacing IT with WFH in regressions yields similar results: MSAs where WFH is more feasible see lower unemployment probability for a given mobility decline.
  - Including both IT×mobility and WFH×mobility: both remain statistically significant but coefficients decline, implying WFH is an important channel but not the only one.
  - Other channels: online sales, contactless sales, more sophisticated IT systems enabling alternate business models.
- Robustness checks (Table 6 exact coefficients, columns summarized):
  - Column (1) Baseline: ∆Mobility = -0.151 ∗∗∗ (0.023); IT = 0.0186 ∗∗ (0.008); ∆Mobility×IT = 0.0488 ∗∗∗ (0.016); R-squared = 0.0202; N = 7,181; FEs: Yes.
  - Column (2) High-Speed Internet: ∆Mobility = -0.397 ∗∗ (0.159); IT = 0.00136 (0.001); ∆Mobility×IT = 0.00391 ∗ (0.002); R-squared = 0.0201; N = 27,181; FEs: Yes.
  - Column (3) High IT (dummy): ∆Mobility = -0.176 ∗∗∗ (0.029); IT = 0.0185 (0.015); ∆Mobility×IT = 0.0880 ∗∗ (0.037); R-squared = 0.0202; N = 7,181; FEs: Yes.
  - Column (4) PCs/Emp: ∆Mobility = -0.150 ∗∗∗ (0.023); IT = 0.00836 (0.009); ∆Mobility×IT = 0.0388 ∗∗ (0.017); R-squared = 0.0202; N = 27,181; FEs: Yes.
  - Column (5) U6 Unemployment: ∆Mobility = -0.269 ∗∗∗ (0.031); IT = 0.00698 (0.010); ∆Mobility×IT = 0.0457 ∗∗ (0.021); R-squared = 0.0246; N = 27,181; FEs: Yes.
  - Note: robustness maintained across alternative IT measures (high-speed internet share, above-median IT dummy, PCs per employee) and across U-3 and U-6 unemployment measures.

### Counterfactual exercise (Equation 7 and narrative)
- Assumptions:
  - BEA: “since 2010, digital economy real gross output growth averaged 2.5 percent per year.” Labor force growth ~ 0.5 percent per year.
  - Assume IT adoption grows at 2 percentage points per year; therefore IT was approximately 10% smaller 5 years ago. Growth assumed homogeneous across MSAs.
- Counterfactual implementation:
  - Re-estimate Equation 5 without normalizing IT (coefficients in IT expenses per employee). Apply 0.9 factor to IT and IT×mobility terms to simulate 2015 IT adoption (Equation 7 formulation).
- Counterfactual results:
  - Estimated counterfactual unemployment rate (average between April and May 2020) under 2015 IT adoption = 16% versus the observed 14%.
  - Difference = 2 percentage points (or 14.3%) higher than observed.
  - Probit specification yields the same result, alleviating concerns about linear model overestimation.

### Policy implications and conclusions
- Technology adoption mitigates short-term labor-market impacts of pandemic-induced mobility declines:
  - High-IT areas bear lower economic cost from social distancing; cost of social distancing is lower where firms adopt IT more heavily, reducing the trade-off between health measures and economic outcomes.
- Distributional consequences:
  - IT does not shield low-skilled workers; the pandemic reinforces skill-biased technological change and can widen inequality between high- and low-educated individuals.
  - Policy priority: programs to improve digital skills among the less-educated population to promote inclusive growth and well-being.
- Longer-run considerations:
  - Short-term analysis shows mitigation via IT; in the medium/long run, firms may substitute labor with technology, potentially increasing job losses.
  - Conversely, IT production and related products may boost growth and generate advantages for IT-intense areas.
- Overall takeaway:
  - IT adoption is an important mitigating factor in the pandemic shock to employment, works in part through WFH feasibility, and has meaningful policy implications for resilience and distributional outcomes.

*Source: wpiea2020208-print-pdf - 5.2  Technology, Mobility, and Unemployment*

### section 3 we describe the data. In section 4 we illustrate some descriptive patterns. In section 5

### wpiea2020208-print-pdf - section 3 we describe the data. In section 4 we illustrate some descriptive patterns. In section 5

### Related Literature
- Voluntary social distancing argued to be more important than lockdowns in disrupting economic activity [Allcott et al., 2020; Bartik et al., 2020; Kahn et al., 2020; Maloney and Taskin, 2020]; mobility and economic activity in the US contracted before lockdowns (Chetty et al.[2020]); lifting lockdowns led to limited rebound in mobility (Dave et al., 2020) with exceptions (Cajner et al.[2020]).
- Goolsbee and Syverson[2020] and Chen et al.[2020] find small or no robust evidence of lockdowns’ impact on high-frequency economic indicators; Sweden experienced similar (though a bit smaller) declines without strict lockdowns [Anderson et al., 2020; Chen et al., 2020].
- Vulnerable groups more harshly impacted: lower income and educational attainment [Cajner et al., 2020; Chetty et al., 2020; Shibata, 2020], minorities [Fairlie et al., 2020], immigrants [Borjas and Cassidy, 2020], and women [Alon et al., 2020; Del Boca et al., 2020; Papanikolaou and Schmidt, 2020]; evidence of widening inequality [Mongey and Weinberg, 2020; Palomino et al., 2020].
- IT adoption can shield various members of society, regardless of gender or race, from mobility-induced COVID-shock; low-educated individuals are not shielded by IT adoption.
- In high IT-adoption areas, overall increase in inequality can be dampened, but benefits accrue primarily to highly educated individuals; low-educated individuals do not benefit and COVID-induced mobility shocks can raise inequality more in high IT-adoption areas.
- Closest related work: Chiou and Tucker[2020] (high-speed Internet diffusion and self-isolation ability).
- Large literature on IT adoption implications for productivity and wages (multiple cited works); IT seen as skill-biased technological change [Violante, 2008]; past recessions hit less-skilled workers harder reinforcing skill-biased change [Heathcote et al., 2020].

### Data Sources (section 3)
- Labor market: Current Population Survey (CPS); construct unemployment rate at MSA, state, and national levels.
- Mobility: Google Community Mobility Reports; GPS-based indices by activity type (retail and recreation, workplaces, transit station, parks, etc.); disaggregated to county level for the US; reported as index compared to pre-COVID baseline (January-February).
- Lockdown data: Keystone (original sources: state web-pages); 11 non-pharmaceutical intervention (NPI) dummy variables:
  - (i) the closing of public venues
  - (ii) ban of gathering size 500-101
  - (iii) ban of gathering size 100-26
  - (iv) ban of gathering size 25-11
  - (v) ban of gathering size 10-0
  - (vi) full lockdown
  - (vii) non-essential services closure
  - (viii) ban of religious gatherings
  - (ix) school closure
  - (x) shelter in place
  - (xi) social distancing
- Lockdown index construction: for each state on a given day, average across the 11 lockdown dummies so that a lockdown of 100% refers to having all 11 NPIs in place at a given time.
- Consumer spending: Chetty et al.[2020] aggregated anonymized consumer purchase data from Affinity Solutions Inc; used at the state level.
- IT adoption: establishment survey on IT budget per employee by CiTBDs Aberdeen for 2016; data on more than 2,800,000 establishments,e in all states in the US; log of IT budget per employee IT_e used and estimated via regression:
  - IT_e = δ + α_g(e) + θ_ind(e) + ≤_i  (equation as given)
  - α_g is geographic fixed effect (state or MSA) interpreted as average log IT budget per employee conditional on industry; θ_ind is industry (2-digit) fixed effect.
- Industry fixed effects included to prevent regional IT measures being driven solely by industry composition.

### Descriptive Patterns (section 4)
- Lockdown intensity across states: strong in April with some states having all 11 NPIs in place; partially loosened in May and June; loosened further in July.
- Mobility trends:
  - Mobility to retail and recreation declined by 30 percent across the U.S. relative to the January baseline.
  - Mobility started declining in mid of April at same time when lockdown policies were put in place (time series insufficient to disentangle causality).
  - Strongest responding states saw mobility drop by more than 60% relative to the pre-COVID baseline; least responding states saw drops by around 20%.
  - Mobility began gradual increase since mid-April; large heterogeneity across states (Washington D.C. near lower end, Mississippi near upper end).
  - Some states (e.g., South Dakota) saw mobility return to pre-COVID levels as early as mid-May; others (e.g., DC) had not seen strong increase by July despite many restrictions lifted.
- Timing around policy implementation:
  - Figure 5 evidence: lockdown implementation seems to have occurred after mobility had already declined—consistent with voluntary social distancing driving decline.
  - Figure 6: mobility already increased before de jure lockdowns were lifted.
- Infection rates around loosening:
  - Figure 7: infection rates remained relatively stable immediately after measures were lifted; some states show a spike after around 40 days, others remained at pre-reopening levels.

### Results — Mobility and Economic Outcomes (section 5.1)
- Hypothesis: decline in mobility to retail/restaurants implies decline in spending at these places and could raise unemployment where mobility declined more.
- Spending regression estimated:
  - ΔSpending_k^s = α + β_k ΔMobility_s + X′ γ + ≤_s  (equation as given)
  - ΔSpending defined as percentage change in credit card spending by subgroup or category k between April and pre-COVID baseline in state s.
  - ΔMobility is percentage change in mobility between April and pre-COVID baseline in state s.
  - X are state-level controls (GDP per capita, minority share, population density).
- Key spending findings:
  - Strong positive correlation between decline in mobility and decline in spending at aggregate level, with significant heterogeneity across individuals and spending categories.
  - Correlation stronger for high- and medium-income individuals than for low-income individuals.
  - Spending on health care and social assistance most strongly correlated with mobility.
    - Reported estimated elasticity: 0.4 implies that a 10 percentage points strong reduction in mobility was associated with a 5 percentage points stronger reduction in spending on health care and social assistance.
  - Accommodation and food services, general merchandise stores, transportation and warehousing, and entertainment and recreation are also highly responsive categories.
  - Spending on grocery and food stores has a negative elasticity with respect to mobility (individuals avoiding restaurants increased grocery spending).
- Unemployment regression estimated:
  - ΔUR_k^s = α + β_k ΔMobility_s + X′_s γ + ≤_s  (equation as given)
  - UR_k^s is the difference in the unemployment rate in state s for category k between April and February.
- Key unemployment findings:
  - Across all individuals, strong negative correlation between mobility and the unemployment rate.
    - A 10 percentage points stronger decline in mobility was on average associated with a 3 percentage points stronger increase in the unemployment rate.
  - Heterogeneity by group:
    - Low-education individuals (without a high-school degree): in a state where mobility declined by 10 percentage points more, the unemployment rate increased by 7.5 percentage points more.
    - Non-Whites: elasticity around 0.6 (implied: a 10 percentage points stronger decline in mobility associated with ~6 percentage points increase in unemployment? text reports elasticity around 0.6).
    - High-educated individuals: only a very small significantly larger increase in unemployment rates in areas where mobility dropped more.
  - Overall implication: drop in mobility caused by the COVID-19 pandemic widens inequality across races and educational attainment.

_https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020208-print-pdf.pdf_

### 5.2  Technology, Mobility, and Unemployment

### 5.2  Technology, Mobility, and Unemployment

### State-level empirical findings
- Correlation patterns:
  - Job losses are correlated with the decline in mobility only in states with a relatively low level of IT adoption.
  - In high-IT adoption states, the increase in unemployment showed relatively little relationship to the degree to which mobility fell.
  - Example: both Colorado and Nevada experienced a decline in mobility of (a bit more than) 40%. However, the increase of the unemployment rate was twice as large in Nevada (low-IT) than in Colorado (high-IT).
- Regression specification (Equation 4) and main parameter of interest:
  - ∆UR_s = α + β1 ∆Mobility_s + β2 IT_s + β3 ∆Mobility_s * IT_s + X'_s σ + (X_s * Mobility_s)' γ + ε_s
  - IT_s is a dummy indicating above-median IT adoption; β3 tests the difference in slope between high- and low-IT states.
- Key estimated magnitudes (Table 1 narrative):
  - A state with stronger IT adoption saw a 1.8 percentage points weaker increase in the unemployment rate relative to lower-IT states.
  - Column (2): A 10 percentage points stronger drop in mobility is associated with a 1.5 percentage points stronger increase in the unemployment rate (average effect).
  - Column (3): For low IT adopters, a 10 percentage points larger decline in mobility was associated with a 5 percentage points larger increase in the unemployment rate.
    - Example: Michigan mobility declined by around 40% while in Ohio mobility declined by 30%; Ohio saw unemployment rise by around 13 percentage points while Michigan rose by approximately 18 percentage points — a 5 percentage points difference for a 10 percentage points difference in mobility decline.
  - Interaction coefficient (point estimate) on IT × mobility: 0.463.
  - Mobility coefficient (in that specification) = -0.505; sum for high-IT states = -0.505 + 0.463 = -0.042 (reflects near-zero slope for high-IT adopters).
- Controls and robustness at state level:
  - Column (4) adds GDP per capita, population density, manufacturing share, and interactions with mobility.
  - The interaction coefficient between IT adoption and mobility remains almost identical after controls, suggesting omitted-variable bias from these controls is unlikely to explain the mitigating impact of IT.

### Individual-level evidence (CPS, MSA-level IT and mobility)
- Estimation framework (Equation 5):
  - Linear probability model where Unemployed_i,t is a dummy for unemployed (in labor force) in April or May 2020.
  - Regressors include ∆Mobility_msa(i),t, IT_msa(i), interaction ∆Mobility * IT, individual controls Z_i, MSA controls X_msa(i), state fixed effects α_s(i). Standard errors clustered at MSA level; regressions weighted by respondent weight.
- Main individual-level magnitudes (Table 2, Table 3 narrative):
  - A larger decline in MSA mobility is associated with a higher probability of being unemployed in April–May 2020.
  - The increase in the probability of being unemployed associated with a large drop in mobility (one standard deviation, equal to 10 pp) is 2.4 percentage points in a low-IT MSA.
  - A one standard deviation larger level of IT adoption in an MSA reduces the increase by 0.7 percentage points, to 1.7 percentage points.
  - Results remain stable and statistically significant after controlling for MSA per capita income, share with a three-year Bachelor’s degree, share of minorities, and February unemployment rate.
- Fixed effects and explanatory power:
  - Column (4) includes additional fixed effects (individual fixed effects by gender, race, education; state fixed effects).
  - Interaction coefficient between mobility and IT remains stable; R-squared increases from 0.418% to 3.8%, indicating additional controls add explanatory power while the IT effect persists.
- Heterogeneity across demographic groups (Equation 6; Table 4; Figure 11 narrative):
  - Triple-interaction coefficients (β5 and β6) are positive for males, females, whites, non-whites, and high/medium education groups.
  - No mitigating impact of IT is found for low-education individuals.
  - Effect is largest for females and non-white individuals (groups that were heavily hit during the pandemic and where IT had more room to mitigate shocks).
- Interpretation and inequality implications:
  - IT adoption shields many workers but not low-skilled workers; this may widen inequalities between high- and low-educated individuals.
  - Possible mechanism: skill-biased technological change — high-skilled workers have larger complementarities with IT and can switch to work-from-home more easily (Dingel and Neiman[2020] indicate around 1/3 of all workers can do jobs from home, most of them higher-educated).

### Mechanisms, robustness, and alternate measures
- Work-from-home (WFH) as a channel:
  - High correlation between the share of jobs that can be done from home in an MSA and IT (Figure 12).
  - Replacing IT with WFH in regression (Table 5) yields similar results: MSAs where WFH is more feasible see lower unemployment probability for a given mobility decline.
  - Including both IT×mobility and WFH×mobility interactions: both remain statistically significant but coefficients decline, implying WFH is an important channel through which IT shields workers, but not the only channel.
  - Other channels: online sales, contactless sales, more sophisticated IT systems enabling alternate business models.
- Robustness checks (Table 6 summary):
  - Column (2): Replacing IT adoption with share of high-speed internet in the MSA — interaction positive and statistically significant at the 10% level.
  - Column (3): Using an above-median IT adopter dummy — interaction positive and statistically significant.
  - Column (4): Replacing log IT budget per employee with ratio of personal computers per employee (Harte Hanks dataset) — results consistent.
  - Column (5): Replacing U-3 unemployment measure with U-6 definition (broader measure) — robustness maintained.
- Numeric precision preserved in reported estimates and tests.

### Counterfactual exercise
- Assumptions and data:
  - BEA: “since 2010, digital economy real gross output growth averaged 2.5 percent per year.” Labor force growth ~ 0.5 percent per year.
  - Assume IT adoption grows at 2 percentage points per year; therefore IT was approximately 10% smaller 5 years ago. Growth assumed homogeneous across MSAs.
  - Re-estimate Equation 5 without normalizing IT (coefficients in IT expenses per employee).
- Counterfactual formulation (Equation 7) applies 0.9 factor to IT and IT×mobility terms to simulate 2015 IT adoption.
- Counterfactual results:
  - Estimated counterfactual unemployment rate (average between April and May 2020) under 2015 IT adoption = 16% versus the observed 14%.
  - This is 2 percentage points (or 14.3%) higher than observed.
  - Probit specification yields the same result, alleviating concerns about linear model overestimation from large IT changes.

### Policy implications and conclusions
- Technology adoption mitigates short-term labor-market impacts of pandemic-induced mobility declines:
  - High-IT areas bear lower economic cost from social distancing; the cost of social distancing is lower where firms adopt IT more heavily, reducing the trade-off between health measures and economic outcomes.
- Distributional consequences and policy priorities:
  - IT does not shield low-skilled workers; COVID-19 reinforces pre-existing skill-biased technological change and may widen inequality.
  - Policies targeted to improve digital skills among the less-educated population are important to promote inclusive growth and well-being.
- Longer-run considerations:
  - Analysis focuses on short-term effects. In medium/long run, firms may substitute labor with technology, potentially increasing job losses.
  - Conversely, production of IT and related products may boost growth and generate advantages for IT-intense areas.
- Overall takeaway:
  - IT adoption is an important mitigating factor in the pandemic shock to employment, works in part through WFH feasibility, and has meaningful policy implications for resilience and distributional outcomes.

*Source: wpiea2020208-print-pdf - 5.2  Technology, Mobility, and Unemployment*

### References

### References

### Key empirical results, figures and captions
- Figure 1: Plots change in unemployment rate between February and April by state on average change in mobility in retail, recreation and transit station in April. Red diamonds = states where IT adoption is above the median; blue triangles = states where IT adoption is below the median. Red and blue lines show linear fits for high-IT and low-IT states respectively. See section 3 and subsection 5.2 for more details.
- Figure 2: Plots change in unemployment rate between February and April by state on average Lockdown stringency index (Keystone) over the same period. Red diamonds = high-IT states; blue diamonds = low-IT states. See section 3 and subsection 5.2 for more details.
- Figure 3: Plots lockdown intensity by state in the beginning of April, May, June, July. Lockdown intensity = average across various NPI measures. See section 3 and section 4 for more details.
- Figure 4: Plots inverse of weighted average lockdown intensity across states (red) and average mobility across states (black). Shaded blue areas plot the 75th and 25th percentile / 90th and 10th percentile and maximum and minimum in terms of retail, recreation and transit mobility. Maximum and minimum states labelled for last available date. See section 3 and section 4 for more details.
- Figure 5: Mobility around lockdown tightening for retail and recreation (left) and transit station (right). Black line = average state; blue area = states at 75th and 25th percentile. See section 3 and section 4 for more details.
- Figure 6: Mobility around lockdown loosening for retail and recreation (left) and transit station (right). Black line = average state; blue area = states at 75th and 25th percentile. See section 3 and section 4 for more details.
- Figure 7: Daily number of new infections per 100,000 people around when a state loosens its lockdown policy. Black line = average state; blue area = states at 75th and 25th percentile. See section 3 and section 4 for more details.
- Figure 8: Plots coefficient and 90% confidence interval of βk from Equation 2: ∆Spending_ks = α + βk ∆Mobility_s + X′γ + ε_s, where ∆Spending_ks is percentage change in spending between April and pre-COVID baseline for income group k; ∆Mobility_s is change in mobility in April 2020 relative to pre-COVID baseline. X includes GDP per capita, population density and the minority share. See section 3 and subsection 5.1 for more details.
- Figure 9: Plots coefficient and 90% confidence interval of βk from Equation 2 for spending category k. See section 3 and subsection 5.1 for more details.
- Figure 10: Plots coefficient and 90% confidence interval of βk from Equation 4:
  - ∆UR_ks = α + β1 ∆Mobility_s + β2 IT_s + β3 ∆Mobility_s * IT_s + X′s σ + (X_s * Mobility_s)′ γ + ε_s,
  - where ∆UR_s = change in unemployment rate between April and February for category k; ∆Mobility_s = change in mobility in April 2020 relative to pre-COVID baseline. X includes GDP per capita, population density and minority share. See section 3 and subsection 5.1 for more details.
- Figure 11: Plots coefficient and 90% confidence interval of β5 and β6 from Equation 6 (see source for full specification). Unemployed_i,t is a dummy =1 if individual i is unemployed in month t (April/May 2020), zero if employed. ∆Mobility_msa(i),t is change in mobility in month t relative to pre-COVID baseline. IT_msa(i) is average level of IT adoption in the MSA. A_i and B_i are dummy variables for gender, race, and education subgroups. X includes GDP per capita, population density and minority share. See section 3 and subsubsection 5.2.1 for more details.
- Figure 12: Plots level of IT adoption in an MSA (horizontal axis) against share of jobs that can be done from home (vertical axis). Share from Dingel and Neiman [2020]. See section 3 and subsubsection 5.2.1 for more details.

### Key regression table results (exact reported coefficients, standard errors and summary stats)
- Table 1: Dependent variable: ∆Unemployment Rate
  - Column (1): IT = -0.0180 (0.010); ∆Mobility = -0.148 (0.070); R-squared = 0.0575; N = 51; Controls: No.
  - Column (2): IT = ∗0.134 ∗∗∗ (0.037); ∆Mobility = -0.505 ∗∗∗ (0.102); R-squared = 0.116; N = 51; Controls: No.
  - Column (3): IT = 0.142 ∗∗∗ (0.033); ∆Mobility = -0.622 (0.377); ∆Mobility×IT = 0.463 ∗∗∗ (0.116); R-squared = 0.478; N = 51; Controls: No.
  - Column (4): IT = 0.142 ∗∗∗ (0.033); ∆Mobility×IT = 0.476 ∗∗∗ (0.105); R-squared = 0.598; N = 51; Controls: Yes.
  - Note: Results of estimating Equation 4. Robust standard errors in parentheses. *p<0.1, **p<0.05, ***p<0.01.

- Table 2: Dependent variable: Unemployed
  - Column (1): ∆Mobility = -0.181 ∗∗∗ (0.031); IT = -0.00697 (0.005); R-squared = 0.00346; N = 7,181; Controls: No.
  - Column (2): ∆Mobility = -0.239 ∗∗∗ (0.037); IT = 0.0187 ∗∗∗ (0.007); R-squared = 0.00418; N = 12,718; Controls: No.
  - Column (3): ∆Mobility = -0.742 (1.559); IT = 0.0193 ∗∗ (0.009); ∆Mobility×IT = 0.0699 ∗∗∗ (0.023); R-squared = 0.0293; N = 27,181; Controls: Yes.
  - Column (4): ∆Mobility = 0.0236 (1.358); IT = 0.0292 ∗∗∗ (0.011); ∆Mobility×IT = 0.0656 ∗∗ (0.032); R-squared = 0.0384; N = 27,181; Controls: Yes; FEs: Yes.
  - Note: Results of estimating Equation 5. Standard errors clustered at the MSA level. Regressions weighted by respondent weight. *p<0.1, **p<0.05, ***p<0.01.

- Table 3: Probit: Dependent variable: Unemployed
  - Column (1): ∆Mobility = -0.840 ∗∗∗ (0.147); IT = -0.0324 (0.022); ∆Mobility×IT = 0.328 ∗∗∗ (0.105); N = 7,181; Controls: No.
  - Column (2): ∆Mobility = -1.115 ∗∗∗ (0.165); IT = 0.0937 ∗∗ (0.037); ∆Mobility×IT = 0.292 ∗∗ (0.147); N = 12,718; Controls: No.
  - Column (3): ∆Mobility = -4.285 (7.616); IT = 0.0893 ∗ (0.046); ∆Mobility×IT = 0.350 ∗∗∗ (0.128); N = 27,181; Controls: Yes.
  - Column (4): ∆Mobility = -0.555 (6.912); IT = 0.154 ∗∗∗ (0.056); ∆Mobility×IT = 0.350 ∗∗∗ (0.128); N = 27,181; Controls: Yes; FEs: Yes.
  - Note: Results of estimating Equation 5 with Probit. Standard errors clustered at the MSA level. Regressions weighted by respondent weight. *p<0.1, **p<0.05, ***p<0.01.

- Table 4: Dependent variable: Unemployed — heterogeneous effects (∆Mobility×IT interacted with subgroup dummies)
  - ∆Mobility×IT×Male: 0.0306 ∗ (0.017) column (1); 0.0494 ∗ (0.025) column (2).
  - ∆Mobility×IT×Female: 0.0684 ∗∗∗ (0.019) column (1); 0.0894 ∗∗∗ (0.028) column (2).
  - ∆Mobility×IT×White: 0.0346 ∗∗ (0.017) column (3); 0.0610 ∗∗ (0.027) column (4).
  - ∆Mobility×IT×Non-White: 0.0577 ∗ (0.030) column (3); 0.0909 ∗∗∗ (0.035) column (4).
  - ∆Mobility×IT×High/Med Educ: 0.0520 ∗∗∗ (0.016) column (5); 0.0712 ∗∗∗ (0.025) column (6).
  - ∆Mobility×IT×Low Educ: -0.0324 (0.049) column (5); 0.0122 (0.054) column (6).
  - R-squared ranges reported (e.g., 0.0204, 0.0386). N reported as 7,181 / 27,181 / 7,181 / 27,181 / 7,181 / 27,181 across columns. Controls and FEs vary by column. *p<0.1, **p<0.05, ***p<0.01.

- Table 5: Dependent variable: Unemployed — includes Teleworking
  - Column (1): ∆Mobility = 0.0236 (1.358); IT = 0.0292 ∗∗∗ (0.011); ∆Mobility×IT = 0.0677 ∗∗∗ (0.025); Teleworking = 0.237 (0.190); ∆Mobility×Teleworking = 1.100 ∗∗ (0.517); R-squared = 0.0384; N = 7,181.
  - Column (2): ∆Mobility = -0.5530 (1.243); IT = 0.0305 ∗∗∗ (0.011); ∆Mobility×IT = 0.0539 ∗∗ (0.025); Teleworking = 0.164 (0.185); ∆Mobility×Teleworking = 1.002 ∗∗ (0.506); R-squared = 0.0385; N = 27,181.
  - Column (3): ∆Mobility = 0.635 (1.550); IT = 0.0305 ∗∗∗ (0.011); ∆Mobility×IT = 0.0539 ∗∗ (0.025); Teleworking not separately reported beyond above. R-squared = 0.0387; N = 27,181.
  - Note: Teleworking_msa(i) = share of jobs that can be done from home in MSA (Dingel and Neiman [2020]). Standard errors clustered at the MSA level. *p<0.1, **p<0.05, ***p<0.01.

- Table 6: Robustness — Dependent variable: Unemployed
  - Specification columns: Baseline; High-Speed Internet; High IT; PCs/Emp; U6 Unemployment.
  - Column (1) Baseline: ∆Mobility = -0.151 ∗∗∗ (0.023); IT = 0.0186 ∗∗ (0.008); ∆Mobility×IT = 0.0488 ∗∗∗ (0.016); R-squared = 0.0202; N = 7,181; FEs: Yes.
  - Column (2) High-Speed Internet: ∆Mobility = -0.397 ∗∗ (0.159); IT = 0.00136 (0.001); ∆Mobility×IT = 0.00391 ∗ (0.002); R-squared = 0.0201; N = 27,181; FEs: Yes.
  - Column (3) High IT (dummy): ∆Mobility = -0.176 ∗∗∗ (0.029); IT = 0.0185 (0.015); ∆Mobility×IT = 0.0880 ∗∗ (0.037); R-squared = 0.0202; N = 7,181; FEs: Yes.
  - Column (4) PCs/Emp: ∆Mobility = -0.150 ∗∗∗ (0.023); IT = 0.00836 (0.009); ∆Mobility×IT = 0.0388 ∗∗ (0.017); R-squared = 0.0202; N = 27,181; FEs: Yes.
  - Column (5) U6 Unemployment: ∆Mobility = -0.269 ∗∗∗ (0.031); IT = 0.00698 (0.010); ∆Mobility×IT = 0.0457 ∗∗ (0.021); R-squared = 0.0246; N = 27,181; FEs: Yes.
  - Note: Standard errors clustered at the MSA level. Regressions weighted by assigned weight of respondent. *p<0.1, **p<0.05, ***p<0.01.

### Selected cited literature (first entries as presented)
- Akerman, A., I. Gaarder, and M. Mogstad, The skill complementarity of broadband internet, The Quarterly Journal of Economics, 130(4), 1781–1824, 2015.
- Allcott, H., L. Boxell, J. Conway, M. Gentzkow, M. Thaler, and D. Y. Yang, Polarization and public health: Partisan differences in social distancing during the coronavirus pandemic, NBER Working Paper, (w26946), 2020.
- Alon, T., M. Doepke, J. Olmstead-Rumsey, M. Tertilt, et al., This time it’s different: The role of women’s employment in a pandemic recession, Tech. rep., University of Bonn and University of Mannheim, Germany, 2020.
- Alstadsæter, A., J. B. Bjørkheim, W. Kopczuk, and A. Økland, Norwegian and us policies alleviate business vulnerability due to the covid-19 shock equally well, Unpublished Working Paper, 2020.
- Anderson, R. M., H. Heesterbeek, D. Klinkenberg, and T. D. Hollingsworth, How will country-based mitigation measures influence the course of the covid-19 epidemic?, The Lancet, 395(10228), 931–934, 2020.
- Autor, D. H., F. Levy, and R. J. Murnane, The skill content of recent technological change: An empirical exploration, The Quarterly Journal of Economics, 118(4), 1279–1333, 2003.
- Bartik, A. W., M. Bertrand, Z. B. Cullen, E. L. Glaeser, M. Luca, and C. T. Stanton, How are small businesses adjusting to covid-19? early evidence from a survey, Tech. rep., National Bureau of Economic Research, 2020.
- Beaudry, P., M. Doms, and E. Lewis, Should the personal computer be considered a technological revolution? evidence from us metropolitan areas, Journal of Political Economy, 118(5), 988–1036, 2010.
- Béland, L.-P., A. Brodeur, and T. Wright, The short-term economic consequences of covid-19: exposure to disease, remote work and government response, 2020.
- Bessen, J. E., and C. Righi, Shocking technology: What happens when firms make large it investments?, Boston Univ. School of Law, Law and Economics Research Paper, (19-6), 2019.
- Bloom, N., The bright future of working from home, SIEPR blog, 2020.
- Bloom, N., and N. Pierri, Cloud computing is helping smaller, newer firms compete, Harvard Business Review, 2018.
- Bloom, N., R. Sadun, and J. Van Reenen, Americans do it better: Us multinationals and the productivity miracle, American Economic Review, 102(1), 167–201, 2012.
- (Additional references continue in the source.)

*Italic: References and figure/table captions and regression results as listed in the source PDF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020208-print-pdf.pdf_