This article is also available in Russian.
Under the current COVID-19 crisis, what can we learn from our past experience in collecting information through telephone interviews, when person-to-person interview is not possible? This note provides a few points for statistical organizations to consider when deciding whether to adopt such an approach.
Telephone surveys are a commonly used in market research and political polls. Compared to the traditional face-to-face interviews, telephone interview is usually less costly as household visits are not necessary. Telephone interview has also been used in official statistics, in different ways. Some official surveys are implemented as telephone survey only, for example the annual American Time Use Survey is implemented through computer-assisted telephone interviewing (CATI). Telephone interviewing can also be implemented as one part of the entire survey. For instance, the US National Crime Victimization Survey keeps the selected households in the survey for 3.5 years and the interview is conducted every 6 months. The first interview is conducted face-to-face, followed by subsequent ones through telephone interview.
Telephone interview is also being used in situations where face-to-face interview is not possible. For example, Rapid Emergency Response Surveys were conducted in a number of countries in Africa to assess the level of food security and poverty. The surveys were designed as phone surveys and aimed to deliver rapid assessment results.
Is the use of telephone interviews being considered for a regular survey that has been planned for face-to-face interview? Or the objective is to carry out a small telephone interview to collect information that is related to COVID-19? In the formar case, has the field work been partially completed before the pandemic, with the telephone interview being introduced to continue with the remaining households?
These questions will help framing the magnitude of the exercise. For example, a subsample of a recent survey (such as Labour Force Survey, Demographic and Health Survey or Multiple Indicator Cluster Survey) might be fine to collect data for topics related to COVID-19. However it might not be sufficient for producing data on regular surveys by desired geographical breakdowns. The length of a regular survey questionnaire might also be challenging for a telephone interview.
As with sample selection for a person-to-person surveys, telephone surveys need a frame to select the respondents, with each respondent associated with a known selection probability.
Random digit dialing (RDD) is a method of probability sampling that provides a sample of households, families, or persons via a random selection of their telephone numbers. A major challenge for RDD surveys is developing a complete list of telephone numbers that cover all target population in the country.
A relatively easier and reliable way of getting a list of representative respondents through telephone is to use the telephone numbers collected from a recent household survey in the country. This was done in the high frequency cell phone surveys carried out in Sierra Leone after the Ebola crisis. The survey followed a sample of households for whom cell phone numbers were recorded during the nationally-representative Labour Force Survey conducted in July-August 2014. One additional advantage of surveying households from a recent survey is that similar characteristics can be collected in both surveys, thus allowing a comparison of situations before and after the pandemic.
Many respondents may be reluctant to respond to an unknown caller which could lead to a low response rate and potential biases in responses. Potential steps which could be taken to counter-act this is to send out a SMS message in advance of calling, specifying that the agency running the survey will call at a specific point in time and that all information received will be handled confidentially. Links to the webpage of the survey and contact information could also be provided. Further, as with other data collection, information campaigns could take place to make people aware.
Will the survey allow proxy response when one of the household members can respond to the interview on behalf of other household members? If only self-reporting is allowed, guidance needs to be provided on protocols to follow, such as the number of callbacks allowed to reach the designated respondent.
Is there a specific geographic area(s) that the survey would like to target? Will the survey cover areas that are affected differently by the pandemic? The first round of the Sierra Leone High Frequency survey produced estimates for three broad areas that were affected differently by Ebola. If estimates are to be produced for different regions, representativeness of each needs to be considered.
The questionnaire should be short (20-30 minutes) for telephone interview, for various reasons. (Refer to here for more information) Weak connectivity of phone network in some areas often make long interviews difficult. Respondents also tend to have a shorter attention span over the phone than in the face-to-face interview.
It is a difficult task to balance between the short duration of an interview and the amount of information that the survey would like to cover. Basic sociodemographic characteristics should always be covered. If following directly after a recent survey and using a subset of the sample in that survey, it is preferred to use similar question formulation if the same topic is covered. This facilitates comparison across surveys and allows studies comparing the situations before and after the pandemic. It is important to note that, however, there may be questions which work well face-to-face but are not suitable for telephone interviewing. These may include questions that use showcards or have a long list of response options. For example, when asking about income sources, a face-to-face survey might present a showcard with a list of options on and ask respondents to mention all that apply to them. By telephone, a better approach would probably be to ask about each of the potential income sources one by one as a series of questions.
There are specific training elements that need to be taken into consideration for telephone interview. For example, telephone interview is more challenging for enumerators to build rapport with respondent. Therefore carefully prepared introductory text is very important. More information on the topic is available here.
Data collected in survey operations will be captured eventually by a central database. For countries that have been implementing CAPI such database exists already. However for countries using PAPI in their data collection, decisions need to be made on how data collected through telephone interview will be recorded and stored. It is still possible to have enumerators recording responses on paper questionnaire and then transferred to a database later on but is not desirable. However, if a large proportion of field work on paper questionnaire has already been completed, recording the remaining responses on paper questionnaires can be considered. More guidance on setting up a data centre is available online.
As in all data collection exercises, quality of the telephone survey needs to be assessed and clearly documented. In cases when telephone interview is introduced to cope with the Covid-19 challenge, the impact of interview mode change needs to be studied.