2.1. Initial engagement with the Government

In March 2020, 1 week after the first cases of COVID-19 were detected in Ghana, the GSS approached the Minister of Planning to propose how CDR data could inform the Government’s response to the pandemic. The Minister for Planning immediately approached the Minister for Health and Minister for Communication to convene a meeting with the GSS team, while he stepped back from further in-person engagements.

A meeting then took place between the GSS, the Ministry of Health, and Ghana Health Service. Frequent partnerships between the GSS and Ghana Health Service, including the existing CDR project, had fostered a strong working relationship between these institutions. This facilitated quick liaison between the institutions, following the submission of initial letters of invitation. During this engagement, GSS described the available CDR dataset and its potential use cases. This prompted concerns over the representativeness of the available data, which comprises only CDRs from Vodafone Ghana, since other major telecommunications agencies are also prevalent across the country. The Ministry for Planning proposed that the Minister for Health could approach the Minister for Communication, who oversees the National Communications Authority, with regard to encouraging other mobile network operators to participate in the initiative in order to improve the representativeness of the available data.

The intention of the GSS in this engagement was to highlight the potential of using anonymized, aggregated CDRs to provide near real-time information about the emerging situation on the ground. However, the Ministry of Health enquired whether the data could be used to track individual phone numbers to assist with enhanced contact tracing. This was an exercise that the Government was conducting for the 20,000 passengers who had entered Ghana by international flight in the 2 weeks before the borders were closed. The GSS explained that the available data did not include phone numbers or any other personal identifiers, and also that the collaboration agreement explicitly disallowed the use of individual-level data, therefore meaning that contact tracing fell outside both the feasible and legal scope of the project. Since the priority of the political leadership at the time was to engage mobile network operators to facilitate contact tracing, efforts to coordinate with the Ministry of Communication did not proceed, and the Ministry of Health demonstrated only limited interest in the idea of analysis based on the anonymized, aggregated CDR data.

The GSS and Flowminder decided to proceed with the analysis anyway, in order to be able to concretely demonstrate the type of information that it was possible to produce. They worked together to determine the metrics of interest to include in the analysis, and published two reports. The first was published on 3rd April—4 days after the introduction of lockdown restrictions in parts of Ghana, and the second on 15th May—3 weeks after lockdown restrictions had been lifted. The analyses are described in the following section.

2.2. Analysis

There are two fundamental indicators that form the basis of most mobility analyses using CDR data. They are the number of subscribers recorded to have been present at a specified location within a specified time period (“subscriber presence”), and the number of subscribers recorded to have travelled between two specified locations within a specified time period (commonly called “origin-destination matrices” or “transition matrices”). A detailed technical discussion about these indicators, and others that are relevant to the study of mobility and social-distancing interventions, can be found in Kishore et al. (Reference Kishore, Kiang, Engø-Monsen, Vembar, Schroeder, Balsari and Buckee2020). In the context of COVID-19, studying how these two indicators changed on a daily basis, as social-distancing and mobility restrictions were introduced and then lifted, provided insights into how restrictions were affecting the population’s behavior. Insights could be obtained for the entire country, and with a time lag of just a few days. The initial report was published just 4 days after lockdown restrictions were introduced. This geographical coverage and timeliness would not have been possible from traditional data collection methods, that is manual surveys.

The first report (Flowminder, Ghana Statistical Service, & Vodafone Ghana, 2020a) included analyses on subscriber presence within regions (administrative Level 1) and districts (administrative Level 2) each day, and travel between regions and between districts each day, for the time period starting 4 weeks before initial social-distancing and mobility restrictions were introduced, until a few days after lockdown measures had been introduced. This showed how population movements changed in response to each set of restrictions. The report is available from GSS’s website (https://statsghana.gov.gh/gsspublications.php?category=MTkwMDE4MjI2Ny4xMDg=/webstats/90r897632o; accessed 31 August 2020). A second report (Flowminder, Ghana Statistical Service, & Vodafone Ghana, 2020b) was released 3 weeks after lockdown measures had been lifted. This contained similar analyses to the first report, but extending to the post-lockdown period, and including several comments from the Government Statistician regarding policy implications. The report is available from GSS’s website (https://statsghana.gov.gh/gsspublications.php?category=MTkwMDE4MjI2Ny4xMDg=/webstats/90r897632o). The reports also include technical details about the analysis and calculation of the studied indicators.

When interpreting the analyses, it is important to acknowledge that the data have known limitations. Firstly, it is known that the dataset will only include information about people that use a Vodafone Ghana SIM card. Therefore, the entire population is not represented in the dataset, and so it is not possible to make precise statistical inferences about the whole population using only that CDR dataset. Secondly, because CDRs are an actively collected telecommunications dataset—meaning that a record is generated only when a subscriber performs an action with their phone, for example makes a call—there are data available only when subscribers are active. If we observe that overall activity levels have remained constant across the period of study, then it is a reasonable assumption that any observed changes in mobility are due to actual changes in mobility. If, however, we observe that activity levels have changed during the period of study, then it is important to check whether those changes in activity are correlated with observed changes in mobility, as this would indicate that the observed changes in mobility may be artefacts of the changes in activity levels (e.g., more activity means that more movement will be observed). If so, this needs to be taken into account when drawing conclusions from the analysis. In the current case, it was found that there were no significant correlations between activity levels and observed mobility levels.

In spite of the inherent limitations of the data, the insights provided by the data in this instance could not be produced at the same geographic scale, or in as timely a manner, by any other data source. Deeper studies, including the analysis of survey data, can indicate the degree to which the movements of the entire population are represented by the insights drawn from a CDR dataset. This will be specific to each dataset and context—see, for example, Wesolowski et al. (Reference Wesolowski, Eagle, Noor, Snow and Buckee2013) and Arai et al. (Reference Arai, Fan, Matekenya and Shibasaki2016).

2.3. Dissemination

Several channels were used in the dissemination of the two reports. A remote press launch resulted in 9 Ghanaian media outlets publishing 13 articles about the reports online and one major Ghanaian newspaper (Ghanaian Times) published an interview with the Government Statistician in print. The Global Partnership for Sustainable Development Data also published a case study on the analysis (Global Partnership for Sustainable Development Data (2020a)).

The reports were circulated widely through direct channels to the GSS’ Sustainable Development Goals Secretariat’s network of partners, which includes staff from the Ghana Health Service, the Ministry of Health, and other Government MDAs, as well as development partners, academics, and mobile network operators. The response was positive. In particular, staff from Ghana Health Service noted the retrospective value of the insights. For example, they compared travel behavior with case numbers, and observed a link between a spike in travel between the regions of Greater Accra and Oti preceding several local outbreaks in Oti—a region where there had previously been no cases. Questions arose about the possibility of applying a gendered lens to the analysis, but this was not possible within the scope of the current partnership agreement since no information on gender was available.

Finally, the GSS actively approached key Government partners. They delivered a printed copy of the report to the Minister of Health’s office, and a presentation to the Office of the President. There were some difficulties arranging in-person dissemination meetings due to the number of Ministry of Health staff, including the Minister, having to withdraw into quarantine after testing positive for COVID-19.

4.1. Legal and technical challenges

This case study demonstrates how CDR data were used to produce a timely response to the COVID-19 pandemic in Ghana. This was possible because the necessary legal framework and technical infrastructure were already in place, prior to the pandemic. Some communications activities had also been conducted. Having these foundations in place enabled analyses to be conducted and disseminated in a timely manner.

In order to reach this point, the partners had to overcome several challenges during the initial stages of the project. They were:

• • Constructing a project framework that is compliant with all the relevant data privacy regulations, including the GDPR, and that is also practical from an operational perspective. This was achieved through consultation with a number of legal and technical experts, and engaging with the Ghana Data Protection Commission.

• • Designing technical solutions that satisfy all the necessary legal and privacy requirements, and that are feasible to operationalise in the given context. The implemented solution is one in which all personal data are stored and processed, physically, within Vodafone Ghana’s premises. Only anonymized, aggregated data are released, and only to approved parties.

• • Working across the cultural differences between private-sector and public-sector organizations. It is important to recognize that organizations from different sectors have very different structures, incentives, and requirements, and to accommodate those.

• • Managing the perceptions of stakeholders and the public about how the data are being used. This was achieved through careful and transparent communication, to transmit the message that it is aggregated, anonymized data, and not data about any individuals, that are used to provide insights.

If the challenges listed above had had to be dealt with at the stage when the COVID-19 pandemic began, the response would have been much slower—taking months instead of days.

These types of challenges are common in public–private data partnerships, and are discussed as part of the larger public–private data-sharing ecosystem in Global Partnership for Sustainable Development Data (2020b).

4.2. Challenges with Government engagement

Several challenges arose during policy engagement: awareness of the data, bureaucratic processes, and the effects of the pandemic.

Firstly, a lack of awareness within the Government of the existence of indicators on population mobility, derived from aggregated and anonymized CDRs, meant that the GSS had to actively approach ministries to communicate the potential of the data. There was limited time available for engagements with policy makers, and some of that was taken up by communications to sensitise the audience, and outline the scope and limitations of the data. Furthermore, among the ministries who were already aware of the data, they primarily expressed an interest in contact tracing. However, contract tracing activities were not possible with the available data or within the existing project agreement, and correcting these false expectations resulted in a lack of overall interest in what was feasible. A wider sensitisation of ministries, ahead of time, may have prevented some of this confusion and disappointment.

A further institutional challenge was in the protocols of communication between ministries. It is common for new engagements between ministries to be initiated through physical signed letters, as the first step in building trust. This can result in delays due to waiting for letters to be signed, because letters have gone missing, and because of the need for in-person follow-ups to book appointments. The lengthy bureaucratic process meant that during the dissemination of the two analysis reports, the press releases that had originally been planned to follow the ministerial engagements came concurrently, or even before, a face-to-face ministerial dissemination was possible.

Finally, an unexpected challenge was the effect of public health policies on the availability of key stakeholders. The quarantining of two ministers and some staff from the Ministry of Health stilted communications between the GSS and the end-user institutions. This stilting was compounded during the partial lockdown because all remaining engagements were conducted remotely, and the transition to online meetings caused disruption and delays.

4.3. Suggestions to improve preparedness in the future

While the existing setup enabled a reasonably quick response to the need for dynamic sources of data to inform the COVID-19 response, there are three key approaches which would improve future preparedness.

Firstly, further efforts should be invested in sensitising Government MDAs to the potential applications of CDR data. Through a targeted dissemination of existing reports and the production of educational materials, such as videos for a Governmental audience, familiarity with the data can be developed in advance of any emergency situation. This could lead to proactive data requests, clearer understanding of the limitations of the data, and the wider adoption of aggregated, anonymized CDRs as a policy-relevant data source in the future.

Secondly, cautiousness about the public’s reaction to the use of mobile phone data led to a lengthy approval process prior to any reports being published. Some initial public communications had been conducted prior to the pandemic, but only to a limited audience. Sign-off from the senior management level was therefore required from the collaboration partners, due to the need to manage reputational risks. To manage the public reaction and to build public trust for future releases, the partners agreed to produce resources, for a range of audiences, describing the privacy measures taken to protect personal data. These resources will be made widely available, and will accompany future media releases and sensitisation exercises.

Finally, the data used in the analysis originated from a single mobile network operator in Ghana. An inherent limitation of the data is therefore that it is unlikely to be representative of the full population, and that any insights about the mobility behavior of Vodafone Ghana subscribers are not necessarily representative of the behavior of the entire population. This was highlighted within the publications and during press engagements. To produce higher quality insights in the future, efforts should be made to onboard other mobile network operators onto the initiative, as well as to incorporate survey data about phone usage behavior in order to mitigate the biases in the current dataset.