Planning to use household surveys as a data collection tool? The following suggestions relate to the processes of conducting household surveys, not the analysis of the data. I believe the process of surveying is not critically reflected upon as much as it should.
- Before developing your survey, conduct a series of interviews and focus group discussions on the topic. These will ensure that you are asking questions in the right way, and identity new questions that you should be asking. You might have developed an excellent survey about your topic, but missed a key related factor. For example, you have developed a series of questions about knowledge and access to information about HIV and AIDS, but you were unaware that the local service provider speaks a national language whereas many community members only speak a local language. In this case, qualitative information helps identify the language barriers within service provision, in addition to commonly identified barriers of literacy and language of behavior change communication materials.
- Translate and verify. While your data collectors may be fluent English speakers, you want to avoid any miscommunication or misunderstanding. Having the translated survey also supports surveyors when they are collecting data. Then, make sure that your survey has been translated appropriately (i.e. not overly technical terms or terms unfamiliar to the surveyed group). Verify with members of the group being surveyed.
- Hold a focus group discussion with members of the group being surveyed to discuss the survey. Why? Some metrics may be familiar to you and your data collectors, but will be unfamiliar to those being surveyed. For example, should time be measured in minutes? Should weight be measured in kilograms? In many instances you will need to adjust the questions so that the metrics match common measurement. Doing so in a focus group session will allow for a discussion of which metrics are best. Furthermore, some metrics may not be standardized, so these focus groups can identify where those divergences exist. For example, a local measurement of land size might be 0.25 hectare in one community, and 0.3 in another, but the same descriptive term is used.
- Surveyor selection. Consider the socio-cultural dynamics involved and the influence they have upon the types of questions being asked: language, religion, age, gender, ethnicity, politics, status, employment, local or outsider and so forth. These factors affect the data, you will not avoid all forms of bias and influence, but you should be aware of the bias and influence affecting your survey.
- Training. Provide training on the survey and methodology. This will ensure the process is (mostly) uniform, and enable surveyors to ask clarifying questions.
- Quality control. You want to ensure that your data is valid. Sometimes surveys are not entirely based on surveying. Surveyors often have some foreknowledge of the community within which the survey is conducted (if they are not local, they are unlikely to be from very far away). In addition, surveyors are required to repeat the same survey over and over resulting in knowledge "saturation" being reached relatively early on. After having completed 10-15 surveys it is possible for surveyors to fill in surveys that are within the ranges of what is expected. Here are some suggestions for quality control:
- Have supervisors manage teams of surveyors, and randomly check in.
- Collect information that will enable households to be re-located, such as a phone number of house number, and then have a supervisor or independent surveyor verify random samples of surveys.
- If you are using electronic devices for data collection, you may be able to track GPS coordinates and verify where surveys where conducted and how long each survey took.
- Real-time data entry can identify issues, irregularities, patterns and impossible answers. If these are noticed, by daily analysis of data, additional attention or training might be given to the surveyor(s) identified.
- Verification. Ensure your data is valid by checking it for inconsistent and impossible answers. Also take note of normal answers but patterns of answers in surveys, which might indicate a problem. If you can, input the data yourself to identify these issues. If there is too much data for your own inputting, input some of it at the outset so you can guide your data entry clerks to take note of certain questions and patterns.
Having worked with NGOs, governments and researchers, the process of household surveying requires more attention than it is often given. Data analysis is important, but what is being analyzed is more important.