Distance travelled used for examining impacts of mobility restrictions under COVID-19 in Sierra Leone (Flowminder, 2020)


In the frame of a collaboration between the Sierra Leone Directorate of Science, Technology and Innovation (DSTI), the Massachusetts Institute of Technology (MIT), Flowminder and Africell Sierra Leone, the study was conducted after Sierra Leone introduced countrywide measures to reduce COVID-19 spread during spring 2020. Based on CDR data provided by Africell between February and November 2020, the analysis aimed at evaluating how these measures' introduction, and later their lifting, impacted population mobility across the country.

Insights on this approach

This analysis first showed an overall reduction of mobility of population during Spring 2020, with the average number of chiefdoms visited per day and subscriber on working days reduced by 8% following the introduction of inter-district travel restrictions (9 April until 30 June). This is equivalent to double the mobility decrease observed on a normal Sunday. During the two three-day lockdown periods (5th-8th of April and 3rd-6th of May), the number of chiefdoms visited per subscriber also reduced by circa 25%, which is four to five times lower than a normal Sunday.

Reduced flows to cities during the period of inter-district travel restrictions (11 April until 24 June) were also demonstrated: Freetown saw a median 24% reduction in in-out flow, compared to a 16% median reduction on a normal Sunday, whilst Bo Town, the second largest city after Freetown, saw its in-out flow reduced by 25%, i.e., a reduction 25 times greater than the reduction observed on normal Sunday (1%).

Further, the study revealed there were large increases in movements between Freetown and some chiefdoms between the announcement and implementation of first lockdown. Similar patterns were observed for the main cities in the country, Bo Town, Kenema, and Port Loko.

Finally, results showed evidence of a return to normal mobility patterns once the inter-district movement restrictions were lifted at the end of June, despite lower mobility indicators overall that may result from a decrease in calling frequency following a pricing change from the operator (see challenges below).

Key steps taken for producing statistics

The impact of movement restrictions measures on the actual mobility of the population was demonstrated with four different indicators derived from CDR:

  • "presence" indicators: median subscriber counts in a specific location per 24h period
  • average number of localities visited per subscriber per 24h period
  • "flow" indicators: median counts of subscribers seen at two different locations within 24h. Locations were defined to correspond to intra- and inter-district movements, between chiefdoms (within the same district or between districts), or into and from the two major cities.
  • "distance travelled" indicator: distance between the locations visited by subscribers seen at two different locations (in this case chiefdoms) within 24h (median across subscribers).

Impact was measured as the relative change with baseline (before movement restrictions) for these indicators.

One additional indicator was used to check for the potential bias created in the data following a pricing change by the operator: total count of "events", i.e., calls and SMS.

Areas for improvement and stratified challenges

This analysis is a good example of issues with measurement bias in CDR. In Sierra Leone the operator changed their pricing about a month before movement restrictions were put in place, which resulted in changes in the number of calls made and SMS sent across the subscriber base, which in turn directly resulted in changes in subscriber counts and flows. In such cases, where changes in phone use and mobility restrictions coincide in time, it is not possible to disentangle the effect of each on aggregated indicators of presence and flows, and even at individual level data it would be difficult. However, we note that the 'distance travelled' indicator is less sensitive to changes in phone use and may be used as a more robust indicator than counts of subscribers and flows. Number of flows is the most affected as this particular indicator relies on subscribers making at least two calls or more in under 24 hours, and a decrease in phone use may hide mobility of subscribers. To conclude, we would recommend focusing on distance travelled in periods when significant changes in phone use are observed, and not using other indicators outside of stable periods.

  • No labels