Fisher's Exact test was used to test for an association between each scale's improvement and the education variable as this test is suitable for small data sets [ 38].
We chose to use only the education variable for consistency between models, because it was more linear than was the poverty variable and less correlated with the other variables in the model.
The Exp (B) of the education variable increases as the gap between the comparison group and the reference group increases.
The education variable was divided into four levels: 1.
The scaling coefficients vary more widely for the education variables than for LEXP and LGNIc.
The marginal effects for the education variables for the farm women and vocational and post-graduate variables for the farm man (Table 4) are positive and highly significant, indicating a positive relationship between education and off-farm participation.
The bivariate education variable was split between those without a college education and those with any completed years of college.
A comparison among those aged 18 and over at the start of the study period between the average adult household education variable as used in this analysis and the individual level of education showed that, in each case, higher education levels led to a decreased risk of HIV mortality (Supplementary Table 3).
A comparison was made of strata-specific mortality rates between the average adult household education variable and the individual level of education among those over 18 years of age as used in previous studies.
There was no three-way interaction between education, MS, and author-theory, F 1, 821) = 0.13, nor was there an interaction using the full categorical education variable, F 5, 802) = 0.23, nor when treating education as continuous, β = .00, all ns.
