Internet Connectivity as GDP nowcasting indicator#
This notebook evaluates multiple metrics derived from Ookla Open Data (Speedtest Intelligence) as nowcasting indicator.
The underlying hypothesis is that internet speed, coverage, and adoption track economic activity. Areas with faster, more widespread, and more evenly distributed connectivity are expected to show higher economic output.
About Ookla Open Data#
Fixed vs. mobile#
Every metric below is split by network type, either fixed or mobile (plus a pooled combined where noted). These network type are defined as:
Fixed broadband is delivered to a specific physical location, where the end user “must remain basically at the same location to use the broadband service because the access is associated with a specific physical location.” It reaches that location by a wireline connection (copper DSL, cable, or fiber-optic), a fixed wireless link, or satellite.
Mobile broadband is usable while moving, “from any physical location subject to coverage by its mobile broadband provider.” In this region it is delivered mainly over 3G (UMTS/HSPA, roughly 14 Mbps down / 6 Mbps up) and 4G (LTE, roughly 100–300 Mbps down / 50–75 Mbps up) networks.
How Ookla builds this dataset#
Ookla’s Speedtest apps (Android and iOS) run hundreds of millions of consumer-initiated tests a month. Each test measures download speed, upload speed, and latency against one of Ookla’s crowdsourced test servers (operators run their own hardware and register it on the platform). The client picks the nearest servers by IP geolocation, then uses whichever has the lowest ping. Results then filtered to results containing GPS-quality location accuracy and aggregated into ~610.8m × 610.8m tiles (zoom level 16 web Mercator) and released quarterly as the public dataset.
Sources:
Metrics#
The candidate metrics are listed below in the order they appear:
Speed trends: average download and upload speed, by network type
Tile coverage: extent of Syria reporting test activity
Population-weighted coverage: share of population in covered tiles
Speed tier population: population distribution across speed brackets
Any-type coverage: fixed only, mobile only, both, or neither
Urban vs. rural gap: coverage split by settlement type
Speed trends#
Definition: Test-weighted average download and upload speed (Mbps), by quarter, split by network type (fixed vs. mobile).
Construction: For each quarterly raw-tile file, each tile’s avg_d_kbps / avg_u_kbps is weighted by its tests count, summed, and
divided by total tests. This produces one figure per quarter per network type.
| Network | Quarter | Tests | Download (Mbps) | Upload (Mbps) | Latency (ms) |
|---|---|---|---|---|---|
| fixed | 2019 Q1 | 33,495 | 5.36 | 4.88 | 98.8 |
| fixed | 2019 Q2 | 31,869 | 5.26 | 4.80 | 101.8 |
| fixed | 2019 Q3 | 107,218 | 7.11 | 9.02 | 55.6 |
| fixed | 2019 Q4 | 237,296 | 7.20 | 9.97 | 46.1 |
| fixed | 2020 Q1 | 26,580 | 4.34 | 4.00 | 83.2 |
Quarter-over-quarter speed change#
Definition: percent change in test-weighted average download/upload speed vs. the previous quarter, by network type.
Construction: pct_change() on avg_down_mbps / avg_up_mbps from the quarterly summary above.
Tile coverage#
Definition: the number of ~600m tiles (zoom level 16) that reported any test activity in a given quarter, by network type.
Construction: Row count of the quarterly raw-tile file, grouped by network type. A tile is counted as covered if it recorded at least one test that quarter, regardless of speed or test volume.
| Network | Quarter | Tiles reporting |
|---|---|---|
| fixed | 2019 Q1 | 2,619 |
| fixed | 2019 Q2 | 2,692 |
| fixed | 2019 Q3 | 8,474 |
| fixed | 2019 Q4 | 11,903 |
| fixed | 2020 Q1 | 2,277 |
New tiles reaching coverage#
Definition: tiles (quadkeys) reporting test activity for the first time in a given quarter (“newly-appeared”), and the cumulative count of distinct tiles ever observed through that quarter, by network type.
Construction: for each network type, take every quadkey’s first quarter of recorded activity across the full raw archive, then count first-appearances per quarter (bars) and the running total (line).
This is a different question from “Tile coverage” above. n_tiles there counts tiles active in a given quarter, and can rise or fall quarter to quarter with testing volume. A tile counted here as “new” is counted exactly once, in the quarter it was first observed, and stays in the cumulative total even if it goes quiet later — so the bottom panel is monotonically non-decreasing and reads as “how much of Syria has Ookla ever measured,” while the top panel reads as “how fast is that footprint still growing.”
| Quarter | New tiles | Cumulative tiles |
|---|---|---|
| 2024 Q4 | 1,240 | 33,486 |
| 2025 Q1 | 4,725 | 38,211 |
| 2025 Q2 | 6,197 | 44,408 |
| 2025 Q3 | 5,678 | 50,086 |
| 2025 Q4 | 5,779 | 55,865 |
Population-weighted coverage#
Definition: The share of Syria’s population living in a tile with any test activity that quarter, by network type.
Construction: each populated quadkey is matched to WorldPop constrained population, using one raster per year (2019–2025), so every quarter of a given year is weighted by that same year’s population (population itself does not change within a year in this series).
Additional data sources: WorldPop constrained population rasters (2019–2025). The WorldPop’s constrained layers use satellite-derived settlement/building-footprint data to restrict population to pixels that are actually built-up, then redistribute a country’s projected total onto only those pixels (Random Forest dasymetric weighting). See WorldPop’s own comparison: Top-down estimation modelling: Constrained vs. Unconstrained.
Limitation: the population series is model-projected, not necessarily reflected the actual number of population.
| Network | Quarter | Population covered |
|---|---|---|
| combined | 2019 Q1 | 32.3% |
| combined | 2019 Q2 | 33.7% |
| combined | 2019 Q3 | 42.7% |
| combined | 2019 Q4 | 46.1% |
| combined | 2020 Q1 | 33.9% |
By the end of Q4 of 2025, more than 70% of Syria population has access to an internet speed of 17.2 Mbps.
Speed tier population#
Definition: Population, by quarter and network type, grouped into speed tiers. The headline cut is the download-speed tier below, which bins each tile by its test-weighted average download speed. Two further cuts are kept for reference: the FCC tier, which applies paired download/upload regulatory thresholds, and the activity-based tier.
Construction: each tile’s population is assigned to the tier matching its measured speed, then summed nationally per quarter.
Additional data sources: WorldPop constrained population (same source and limitations as metric 3).
FCC speed tier (reference)#
Definition: Population, by quarter and network type, grouped into five speed tiers: no observed activity, below minimum (<25/3 Mbps), mid (25/3–100/20 Mbps), meets the FCC broadband standard (100/20–1000/100 Mbps), and gigabit (1000/100+ Mbps).
Construction: each tile’s population is assigned to the tier matching its measured download/upload speed, then summed nationally per quarter.
Additional data sources: WorldPop constrained population (same source and limitations as metric 3). Tier thresholds follow fixed FCC/industry benchmarks and are not calibrated to Syria specifically.
The 25/3 and 100/20 Mbps cut points are based on FCC’s own tier boundaries (25/3 was the prior US “broadband” standard and 100/20 is the standard as of March 2024). To make them concrete, the FCC’s Broadband Speed Guide lists the minimum download Mbps needed for common activities, one at a time:
Activity |
Minimum download Mbps |
|---|---|
VoIP calls, streaming online radio |
<0.5 |
General browsing, email, social media |
1 |
Standard personal video call (e.g., Skype) |
1 |
HD personal video call |
1.5 |
Game console connecting online |
3 |
Streaming SD video |
3–4 |
Online multiplayer gaming |
4 |
Streaming HD video |
5–8 |
HD video teleconferencing |
6 |
Telecommuting / student (general professional use) |
5–25 |
File downloading |
10 |
Streaming 4K video |
25 |
Download-speed tier#
The headline cut. A four-bin scheme on download speed alone: below 10 Mbps, 10-30 Mbps, 30-50 Mbps, and above 50 Mbps. Each tile is classified by its test-weighted average download speed, then population is summed by tier per quarter. It uses download speed only, rather than the paired download/upload thresholds of the FCC cut, so a tile is never held back by an upload figure that rests on far fewer tests.
| Network | Quarter | Speed tier | Share of population |
|---|---|---|---|
| combined | 2019 Q1 | 10-30 Mbps | 6.7% |
| combined | 2019 Q1 | 30-50 Mbps | 1.4% |
| combined | 2019 Q1 | <10 Mbps | 23.7% |
| combined | 2019 Q1 | >50 Mbps | 0.5% |
| combined | 2019 Q1 | No observed activity | 67.7% |
Activity-based download-speed tier#
A further reference cut, bucketed by what a download speed actually lets someone do rather than by regulatory compliance. Bin edges are lifted directly from the FCC Broadband Speed Guide table above, which is the general browsing/email (1 Mbps), the floor shared by HD streaming and HD video teleconferencing (5 Mbps), and 4K streaming (25 Mbps). So no new external benchmark is introduced. Built the same way as the download-speed and FCC tiers above (tile classified by its test-weighted average download speed, then population summed by tier per quarter).
Any-type coverage#
Definition: Each populated quadkey is classified into exactly one of four mutually exclusive categories, including fixed only, mobile only, both, or neither, so the shares sum to 100% of Syria’s population.
Construction: pop_covered_share by network type does not indicate whether a person has any connectivity, since someone without fixed access may still have mobile access; this reclassification addresses that directly.
Additional data sources: Same WorldPop constrained population base.
| Quarter | Neither | Mobile only | Fixed only | Fixed and mobile |
|---|---|---|---|---|
| 2019 Q1 | 67.7% | 4.5% | 11.0% | 16.7% |
| 2019 Q2 | 66.3% | 5.2% | 11.3% | 17.1% |
| 2019 Q3 | 57.3% | 4.5% | 19.1% | 19.1% |
| 2019 Q4 | 53.9% | 2.0% | 24.2% | 19.9% |
| 2020 Q1 | 66.1% | 7.7% | 7.3% | 18.8% |
Urban vs. rural connectivity gap#
Definition: Population-weighted coverage share (as in metric 3), disaggregated by urban vs. rural classification, for fixed, mobile, and combined (fixed and mobile pooled), by quarter.
Construction: each quadkey is classified as urban or rural using the JRC GHSL Settlement Model grid, and population coverage is recomputed within each class.
Additional data sources: JRC GHSL Settlement Model grid, applied as a single 2020 epoch across all years.
| Network | Quarter | Settlement | Population covered |
|---|---|---|---|
| combined | 2019 Q1 | rural | 4.4% |
| combined | 2019 Q2 | rural | 4.7% |
| combined | 2019 Q3 | rural | 15.0% |
| combined | 2019 Q4 | rural | 18.0% |
| combined | 2020 Q1 | rural | 5.1% |
Fixed broadband at Governerate Level#
This section aggregates fixed-broadband download speed up to Syria’s 14 governorates (ADM1), using the population-weighted mean across all active tiles in each governorate for a given quarter.
| Governorate | Quarter | Population | Active tiles | Tests | Download, pop-weighted (Mbps) | Upload, pop-weighted (Mbps) |
|---|---|---|---|---|---|---|
| Tartous | 2024 Q4 | 288,019 | 200 | 1,697 | 3.72 | 1.10 |
| Tartous | 2025 Q1 | 289,752 | 226 | 1,812 | 7.10 | 1.79 |
| Tartous | 2025 Q2 | 378,829 | 370 | 3,531 | 17.61 | 3.04 |
| Tartous | 2025 Q3 | 453,735 | 586 | 6,298 | 15.31 | 4.30 |
| Tartous | 2025 Q4 | 548,036 | 874 | 12,491 | 12.92 | 4.85 |