Table 5.

A step-by-step example: impact of wealth and parental education on solid fuel use

Step 1: An a priori conceptual causal diagram based on state-of-the-art knowledge	Few published studies have quantitatively assessed the determinants of cooking practices [58,59]; two systematic reviews of peer-reviewed quantitative studies [60] and peer-reviewed qualitative studies as well as grey literature [61] regarding factors that influence household adoption of cleaner fuels and improved cookstoves are currently underway. Based on this limited evidence base, we postulate that cooking fuel use is influenced by: • Wealth, through financial access to cleaner fuels or more efficient, cleaner-burning stoves; • Maternal and paternal education, through knowledge about the health risks associated with indoor air pollution and prioritization of resources towards solving this problem. • This a priori statement is graphically depicted in Figure 4 and summarized as evidence-based hypothesis 9 (Table 6).
Step 2A: An operational single-dataset causal diagram using a DHS dataset for Benin	As the rural Beninese population almost exclusively relies on solid fuels, hypothesis 9 could only be tested in urban Benin. Univariate and multivariable logistic regression analyses show consistent trends in odds ratios for wealth, maternal education and paternal education (Table 7). These findings provide support for the stated hypothesis, leading to retention of all three arrows in the operational single-dataset causal diagram (Figure 5).
Step 2B: An operational multiple-dataset causal diagram using DHS datasets for Kenya, Ethiopia and Namibia	Equally, the analysis in Ethiopia, Kenya and Namibia concludes that all three socio-economic factors play a role (Table 7). Applying the testing criteria in Figure 2, greater maternal education consistently shows odds ratios below 1 in all six settings, where the hypothesis could be tested, implying strong support in individual settings. Applying the testing criteria in Figure 3, it can be concluded that maternal education decreases the chances of solid fuel use with strong support, which graphically translates as a solid decreasing arrow in the operational multiple-dataset causal diagram (Figure 6). Paternal education, on the other hand, was not part of the model selected by the AIC in rural Kenya and urban Namibia; testing in the other four settings concluded with limited (urban Ethiopia, urban Kenya) or strong (urban Benin, rural Namibia) support for a decreasing relationship. Paternal education may therefore not exert an independent influence everywhere, leading to the inclusion of a setting-specific decreasing arrow. Wealth emerges as important across all six settings; in Namibia, however, better mobility – as one dimension of wealth – increases the chances of solid fuel use, a finding that turns out to be robust in several sensitivity analyses. Based on the testing criteria in Figure 3, a setting-specific decreasing arrow best captures the relationship between wealth and solid fuel use (Figure 6).
Step 3: An integrated causal diagram using a pooled WHS dataset for sixteen African countries	With the exception of paternal education, all variables relevant to hypothesis 9 are available in DHS and WHS and assessed in a comparable way; their population distribution in Ethiopia, Kenya and Namibia is similar. Hypothesis testing confirms the robustness of the links between wealth, maternal education and solid fuel use in the individual WHS datasets, as well as in the pooled WHS dataset (Table 8). Consequently, the relationships derived for Figure 6 are incorporated in the integrated causal diagram (Figure 7).