3.0. Introduction
4.1.5. Normality Tests
Figure 4.15. Normal Q-Q plot of diabetes prevalence.
Figure 4.16. Bar, Box, and Histogram of diabetes prevalence.
4.1.5.2. Kruskal -Wallis Test by Ranks
This is a non-parametric method for testing whether samples originate from the same distribution and is mostly used for comparing more than two independent samples of equal or different sizes. Thus, in performing the comparison between multiple groups, we cannot run an ANOVA test for multiple comparisons if the group does not follow a normal distribution. Instead, the Kruskal-Wallis test is used.
The result gave a Kruskal Wallis value of 35.448 with df = 4 and the p-value of 0.000000379. In comparing the value of median prevalence for the region in Table 4.2 with the Kruskal-Wallis test, we then conclude that there is a significant difference among the country regions. This means that the
prevalence level in the West Africa region can be different from that in the South Africa region. Kruskal Wallis test function is then used to compare the independent diabetes prevalence.
We can then conclude that diabetes is independently distributed across the continent.
4.1.5.3. Spearman Rho Correlation Test
Spearman’s rho correlation test was used to measure statistical dependency of the data set. The Table 4.3 below show the result of the spearman correlation test of diabetes and the risk factors. If the p-value < 𝛼 = 0.05, then we say there is a significant relationship between the variables. We also look at if the rho value is positive (+) or negative (-). Lastly, when
r = 0.1 to 0.29, the strength of correlation is said to be small
r = 0.30 to 0.49, the strength of correlation is said to be medium
r = 0.50 to 1, the strength of correlation is said to be large
Table 4.3: Spearman Rho correlation test of diabetes and risk factors Variables Rho (r) p-value Interpretation
Obesity 0.335 0.013 Significant, positive medium strength correlation
Health expenditure -0.132 0.340 Non-significant, negative, small strength correlation GDP 0.573 0.0000058 Significant, positive, large strength correlation Alcohol -0.015 0.915 Non-significant, negative, small correlation Population age 0.504 0.0001 Significant, positive, large strength correlation Age dependency 0.534 0.000032 Significant, positive, large strength correlation Urban population 0.190 0.169 Non-significant, positive, small strength correlation Physical activity 0.050 0.718 Non-significant, positive, small strength correlation Physical density 0.389 0.0036 Significant, positive, medium
HDI 0.439 0.00090 Significant, positive, medium strength correlation MYSC 0.470 0.00034 Significant, positive, medium strength correlation GNI 0.389 0.0036 Significant, positive, medium strength correlation
The results show that obesity, physician density, HDI, MYSC, and GNI are significantly correlated to diabetes with positive medium strength correlations. GDP, population age, and age dependency ratio are significant and positive with larger strength correlations with diabetes. Health expenditure and alcohol are not significantly correlated with diabetes. Urban population growth and physical activity show a nonsignificant correlation as well.
4.1.5.4. Relationship Between Diabetes and Risk Factors
The scatter plot is the best way to assess linearity between two numeric variables. From a scatter plot, the strength, direction and the form of the relationship can be identified. The analysis was conducted in R using library. Figure 4.17 to 4.20 shows the scatter plots for diabetes prevalence (dependent variable) and its association with the risk factors (independent variables). This is a linear representation of the results showing the relationship between diabetes and the risk factors as well as country particularities.
Figure 4.17. Relationship between diabetes prevalence with Obesity and Urbanisation
The results in the Figure 4.17 show that countries such as Egypt and Seychelles have high levels of diabetes prevalence as well as high percentage of obesity, while countries such as South Arica and Togo have high obesity prevalence, but low diabetes.
Figure 4.18. Relationship between diabetes prevalence with Alcohol consumption prevalence and Physical Activity in Africa
Figure 4.19. Relationship between diabetes prevalence with HDI, GNI, and MYSI
Figure 4.20. Relationship between diabetes prevalence with Age dependency ratio y and Physician density
Figure 4.21. Relationship between diabetes prevalence with GDP, Health expenditure, and Population age
Figures 4.17 – 4.21 visually illustrate the relationship between the significant risk factors at a descriptive level and diabetes prevalence, with a linear association observed in all the variables. It also visually shows the outlier countries within the relationship between each covariate and diabetes prevalence. Countries that have high diabetes prevalence are identified both at the continent at large
and within the regions. Three countries are identified which have a diabetes prevalence rate above 15%, and they are Egypt (16.7%), Seychelles (17.4%), and Mauritius (22.3%). Within the region, East Africa shows Comoros and Djibouti as outliers, as their prevalence rate was above 8%, 9.9%, and 8.4%
respectively, and the highest in the region. In the West Africa region, Togo with a 4.8% diabetes prevalence (a little above 4%) and Benin 0.8% below the mean prevalence of 2 in the region are both outliers compared to other countries in the region. In the Central Africa region, Sao Tome and Principe is an outlier and its diabetes prevalence are below 3%. In the North Africa region, Egypt is the only outlier with a prevalence rate of 16.7%, and the highest in the region. In the South Africa region, Mauritius, Seychelles, and South Africa are outliers with diabetes prevalence rates of 22.4%, 17.4%, and 7.6% respectively, and the highest in the region.
Upon visual inspection, the relationship appears to be linear, with all variables having negative direction and looks moderately strong, except alcohol and health expenditure. The strength of the relationship was quantified using spearman rho correlation test.