Statistical analyses

We report the weighted estimates and weighted percentages that are representative of the whole country’s population, by incorporating sample weights in our analyses.

To get an insight into how having a functional difficulty in one domain and controlling for age, sex, and location, could be correlated with having functional difficulties in other domains, we used logistic regression: the response variable could take only binary values (“difficulty”/“no difficulty”).

Correlations among all the predictors - age, location, sex, and all the functional difficulty-related variables (‘pred_see’, ‘pred_hear’, ‘pred_care’, ‘pred_comm’, ‘pred_mob’, and ‘pred_cogn’) were computed. There was a strong positive correlation (using 0.6 as the threshold) among all the functional difficulty-related variables (pred_*). This explains the motivation behind considering only one functional difficulty-related variable at a time, in our analysis. There was no or a weak correlation (i.e., below a particular threshold of 0.6; if any), among the control variables - age, sex, and location.

A functional difficulty in a given domain (say seeing) is a binary variable taking the value of 0 for “no difficulty” and 1 for “any level of difficulty” (“unable to do” or “a lot of difficulty” or “some difficulty”). Each functional difficulty variable in a given domain is used in turn as a response variable and a predictor. For instance, when seeing is a response variable, hearing, mobility, cognition, communication, and self-care, are predictors. The term ‘core’ encompasses hearing, seeing, mobility, and cognition. Functional difficulty data corresponding to the four core domains is available for all six countries. However, only Mauritius, Morocco, and Senegal, have data for two more domains - communication and self-care. Other predictors include age, sex, and location. Age is a continuous variable while the other two predictors (sex and location) are binary variables.

We report the weighted odds ratios (ORs) obtained from the logistic regression, as they are widely used to compare two groups (living with a functional difficulty in a domain or not).

Logistic regression models look like

Country-wise ORs are obtained using the equation given below:

We have specified the 95% CIs and p-values corresponding to each ORs obtained. We also report the crude-estimates (obtained using the model: res_∗=pred_∗) of each ORs obtained.

R software (version 4.2.1) was used to perform all the data and statistical analyses. Moreover, age-specific (‘18–44’ years and ‘45+’ years) ORs were also computed.

We also computed cross-country correlations by using the country-specific population weights (refer to Table 2A). We used weights based on the population sizes for adults estimated by the UN (https://population.un.org/wpp/).

Overall ORs/cross-country estimates

Population weights

Computed (= n2/n1)

Predictors →

Response var ↓