Business Activity Trends

Business Activity Trends During Crisis uses data on posting activity from Facebook to measure how local businesses are impacted by and recover from crisis events. Leveraging the broad presence of small businesses on the Facebook platform, this dataset provides timely estimates of business activity at scale—overcoming common challenges of traditional data collection such as delays, limited coverage, and lack of standardization.

This is a crisis-triggered dataset shared by Meta to support the understanding of the crisis impact on poverty in Bangladesh. For more details, visit Meta’s Data For Good page.

---

Data

The Business Activity Trends dataset was made available by Meta through the proposal Using Alternative Data to Understand Crisis Impact on Poverty in Bangladesh, Sri Lanka, and India, under the Development Data Partnership.

This notebook includes information for multiple crisis events affecting Bangladesh, India, and Sri Lanka, using posting activity from local business Pages on Facebook.

• 📍 Bangladesh

1. Cyclone Remal (May 27 – June 10, 2024)

2. Flood-Triggered Crisis (August 25 – September 5, 2024)

3. Political Crisis (August 16 – August 30, 2024)

• 📍 India

1. Monsoon Season Events (May 2024)

2. Cyclone Remal (May 27 – June 10, 2024)

3. Landslides in Kerala (July – August 2024)

4. Flooding Events (August 25 – September 5, 2024)

5. Tropical Cyclone Dana (Cyclone date: 24-25 October)

6. Cyclone Fengal (Cyclone date: Nov 25, 2024 – Dec 4,2024)

• 📍 Sri Lanka

1. Southwest Monsoon Floods (June 7 – June 16, 2024)

2. Cyclone Fengal (December 2 – December 16, 2024)

Population Sample

Each crisis dataset uses a fixed sample of Facebook business Pages, defined as Pages that meet the following criteria:

• Admin Presence: The Page must have an active admin

• Recent Activity: The Page must have shown posting activity by the crisis start date

• Established Page: Created at least 90 days before the crisis

• Geotagged: Includes a physical business location

• Business Classification: Internally tagged as a business and passes Meta’s quality checks (excludes spam, duplicates, large corporations)

• Local Focus: Must represent a local business

Data Aggregation

• Spatial: Aggregated to country, admin-1 (division), and admin-2 (district) levels using GADM boundaries

• Temporal: Daily updates, reflecting activity in the previous 24 hours

• By Business Vertical: Businesses grouped into economic sectors; an “all” vertical aggregates all types except “public good”

• Privacy Filtering: Excludes any geographic/business vertical cell with <10 businesses

Business Verticals

Vertical	Description
All	All local businesses in a region (excluding public good)
Grocery & Convenience	Retailers of food, essentials, pharmacies
Retail	Non-grocery retail (e.g., auto, furniture, big-box)
Restaurants	Food and beverage service businesses
Local Events	Venues and businesses related to entertainment and social events
Professional Services	Medical, legal, financial, and technical services
Business & Utility Services	B2B providers (e.g., marketing, electricity, internet)
Home Services	Plumbing, repair, electrical, landscaping
Lifestyle Services	Salons, gyms, child and elder care
Travel	Transport and hospitality (e.g., airlines, hotels)
Manufacturing	Local producers of goods, usually not customer-facing
Public Good	Nonprofits, government offices, and religious orgs

---

Methodology

Business activity is measured using posting behavior from Facebook business Pages, including posts, stories, and reels. The underlying assumption is that higher posting frequency indicates businesses are open and active, while lower posting suggests disruption or closure.

To detect deviations from normal behavior, Meta compares each business Page’s daily post count to its own historical pattern from a 90-day pre-crisis baseline. The resulting metric is called the activity quantile, and it reflects how unusual a given day’s activity is, relative to the baseline.

How It Works

1. Score each business daily: For each business, today’s post count is compared to its own historical range to produce a score between 0 and 1 (a mid-quantile).

2. Aggregate the scores: Scores are combined across all businesses in a given geographic area and sector. Each business contributes equally, regardless of how often it posts.

3. Adjust for scale and variability: The aggregate score is normalized so it can be compared across regions and over time.

4. Transform the score to a 0–1 range: The result can be interpreted as the probability of seeing this level of activity under normal (pre-crisis) conditions.

5. Smooth over time: A 7-day average is applied to reduce noise from day-to-day fluctuations.

How to Interpret the Metric

Value	Meaning	Interpretation
~0.5	Normal activity	Business activity is consistent with the pre-crisis baseline
< 0.5	Lower than usual	Possible business slowdown, closure, or disruption
> 0.5	Higher than usual	Possible recovery, reopening, or increased crisis-related activity

Note: This is a relative activity measure, not a percentage. A value of 0.5 does not mean 50% activity—it means activity is at the expected “normal” level based on the business’s own pre-crisis history.

Full technical documentation of this methodology is available in the Business Activity Trends Methodology Paper by Meta’s Data for Good team.

---

Limitations

The primary limitation of this dataset is its reliance on data from Facebook users, which may not represent the entire Bangladeshi population evenly (Palen & Anderson, 2016). This dataset uses posts from Facebook business pages and groups to estimate changes in business activity. As such, its findings are best suited for understanding how quickly businesses recover from a crisis, such as the political crisis in this case (Eyre et al., 2020).

The methodology assumes that businesses post more frequently on Facebook when they are open and less frequently when they are closed. By analyzing the aggregated posting activity of a group of businesses over time, the dataset infers their operational status, offering insights into disruptions and recovery patterns. However, these assumptions may not hold true in all contexts. For instance, if businesses are posting more about the ongoing crisis rather than regular business operations, this may be interpreted as increased activity, which could lead to misleading conclusions about their operational status. These nuances should be carefully considered when interpreting the data.