Association, Causation and Bias

This lesson matters when you write a research note, an article, an M&E report, or a recommendation that someone may act on.

7.5.1 What is association?

An association means two variables show a relationship in the data. Facilities with higher training participation may also have higher reporting completeness. That describes what was observed. It does not prove that training caused the completeness. Better-managed facilities may both train staff and submit reports on time. Facilities with more resources may have better access to training. Size and reporting rules may also differ.

7.5.2 What is causation?

Causation means a factor contributes to producing an outcome. That usually needs a suitable design and a serious look at other explanations. If completeness rises after training, the rise is associated with the training period. New software, extra staff or a change in supervision may have happened at the same time. A simple before-and-after comparison cannot separate those effects.

7.5.3 An example

Suppose facilities using digital reporting have higher completeness than facilities using paper. The finding can support an observed association in the facilities studied. It does not prove that digital reporting alone caused the difference. Ask whether the facilities are comparable, whether training differed, whether connectivity differed, whether the reporting rules were the same, and whether other factors explain the gap. The result can still be useful if it is described accurately.

7.5.4 What is bias?

Bias is a systematic problem in design, collection, selection or interpretation that can distort findings.

• Selection bias: the people or records included differ systematically from those the study intends to understand. A digital-tools survey sent only to staff who use computers may not represent staff with little computer access.
• Information bias: information is systematically misclassified. Facilities that read the same indicator differently may not be comparable.
• Recall bias: people remember past events inaccurately. A survey that asks for the exact number of visits two years ago will miss some of them.
• Reporting bias: information is selectively recorded or presented. A programme report that shows successes and omits relevant problems is an example.

7.5.5 What is a confounding factor?

A confounder can distort the apparent relationship between an exposure and an outcome. If management quality influences both training and reporting, and it is ignored, the link between training and completeness may be misread. You do not need advanced statistics to notice the problem.

7.5.6 How can you reduce these problems?

• Define indicators consistently.
• Use an appropriate sampling method.
• Standardise collection.
• Document missing information.
• Check data quality.
• Consider other explanations.
• Compare like with like.
• Report limitations honestly.
• Do not conclude beyond the design.