2.4 Causal Diagrams

How might you extend your causal diagram to include other potential confounders (or mediators) from this dataset? For example, what happens when we also investigate the potential confounding effect of marriage?

On your causal diagram, which direction does the arrow go between education and marriage? How did you decide this?

#--- Logistic Regression
glm(serostat ~ educat + age.group + married, family = binomial, data = tz) %>% logistic.display() 
## 
## Logistic regression predicting serostat : hiv  positive vs hiv negative 
##  
##                                      crude OR(95%CI)    
## educat: ref.=no education, preschool                    
##    primary                           0.57 (0.32,1.01)   
##    secondary                         0.1 (0.02,0.44)    
##                                                         
## age.group: 20-24 vs 14-19            6.78 (3.42,13.44)  
##                                                         
## married: ref.=never married                             
##    currently married                 10.35 (4.38,24.43) 
##    formerly married                  23.96 (8.71,65.92) 
##                                                         
##                                      adj. OR(95%CI)      P(Wald's test)
## educat: ref.=no education, preschool                                   
##    primary                           0.86 (0.48,1.54)    0.622         
##    secondary                         0.26 (0.06,1.18)    0.081         
##                                                                        
## age.group: 20-24 vs 14-19            2.94 (1.36,6.33)    0.006         
##                                                                        
## married: ref.=never married                                            
##    currently married                 4.58 (1.72,12.19)   0.002         
##    formerly married                  10.36 (3.38,31.73)  < 0.001       
##                                                                        
##                                      P(LR-test)
## educat: ref.=no education, preschool 0.121     
##    primary                                     
##    secondary                                   
##                                                
## age.group: 20-24 vs 14-19            0.003     
##                                                
## married: ref.=never married          < 0.001   
##    currently married                           
##    formerly married                            
##                                                
## Log-likelihood = -248.6921
## No. of observations = 2762
## AIC value = 509.3843

Using your own knowledge and that of your colleagues, or by doing a brief search of the literature, list some potential mechanisms by which education could affect HIV risk in young women in Tanzania. A mechanism is some route by which which change in an outcome is created; the process by which causal effects take place. For example, one mechanism could be education impacting HIV risk because HIV prevention programmes are delivered in schools.

What might the causal diagram look like for each of your identified mechanisms? Are any of your identified mechanisms of action specific to a particular context? Are any of these mechanisms testable from your data?

In the example above, our causal diagram would be amended to include a mediating variable between education and HIV risk. This mechanism would only be relevant in the context of girls who have been to school. We cannot test for this mechanism in the data as we do not have knowledge of whether HIV prevention programmes were implemented or not in each school. Note that we could collect this if we were designing an intervention or a survey - it is key in practical social epidemiology to consider possible mechanisms of action in the design phase. It is often insufficient to ask simply whether there is an effect of a particular construct - we wish to know how and why it exists.