CAUSAL INFERENCE - ANSWERS

Causal Inference Problem Bank Answers


Question 1
Temporality


Question 2
Observing a biological gradient between number of cigarettes and risk of lung cancer helped early epidemiologists feel confident that smoking was a causal risk factor for lung cancer. On the other hand, you might not see a biological gradient between body mass index (BMI, kg/m2) and mortality because both those with low BMI and those with high BMI have higher risk of mortality than those with middle-range BMI. **There are many possible answers to this question!


Question 3
Bradford-Hill suggests that consistency between studies is a causal consideration. But there are many reasons that one study might report one finding while another reports something different. The population or time period in the two studies might be very different. Perhaps the earlier study did not accurately assess exposure to the chemical or the investigators used different confounders in their analysis. The association demonstrated in the recent study is not necessarily spurious because it is not consistent with prior research. You can also think of this in terms of causal pies: if the causal complement that operates with the chemical to cause breast cancer is common in the population used in the recent study but not in the population used in the original study, that would explain the lack of consistency between the findings. The counterfactual model would encourage us to examine the exact contrast examined in each study in terms of exposure dose and type for the exposed and unexposed groups. In addition, it might suggest that one study is closer to the ideal experiment in terms of randomizing exposure, minimizing bias, using a very large sample, etc. There is insufficient information available to say anything further on the studies in question.


Question 4

• a. True 
• b. True 
• c. False 
• d. True 
• e. True 
• f. False 
• g. False 
• h. False

Question 5

• a. Analogy 
• b. Biological plausibility and coherence 
• c. Coherence, consistency, and experiment 
• d. Consistency and coherence 
• e. Biological gradient, strength of association

Question 6

• a.False 
• b.True 
• c. - 
• d.False 
• e.True

Question 7

• a. Sufficient cause 
• b. Component causes 
• c. Necessary cause 
• d. Causal complement 
• e. The effect would appear stronger in the US because more people would be able to complete the rest of the sufficient cause. The strength of an effect for one variable will depend on how common the rest of the pie is in that population (how often it gets completed). 
• f. You would eliminate depression in the population. Every sufficient cause includes genetics, so every cause would be prevented.

Question 8

• a. A cause of disease is an event, condition, or characteristic that preceded the disease and without which the disease either would not have occurred or would have occurred later. 
• b. A sufficient cause is a set of conditions without any one of which the disease would not have occurred. 
• c. A necessary cause is a component cause that is a member of every sufficient cause.

Question 9

• a. 4 
• b. C, D, E F, G or H are all component causes, but aren’t necessary because they only appear in some, not all of the sufficient causes. 
• c. U, A, B – these component causes appear in every sufficient cause.
• d. 1 (pie on the bottom left) 
• e. Zero 
• f. Faced with limited public health resources (as is often the case), one approach could be to focus prevention efforts on one or more of the necessary causes (A, B), or to use a prevention program that would simultaneously focus on preventing C and D. If one causal pie represents a disease pathway that is more common, i.e., contributes more to disease prevalence and incidence, it might be a good strategy to focus prevention efforts on component causes in that causal pie, even if they are not present in other pies. Note: Prevention efforts could not be focused on elimination necessary cause U because that component cause is Unknown.

Question 10

• In the ideal experiment, you would take a population of children, have them grow up in a home with lead from conception to age 5 and then measure their cognitive development score. You would then go back in time, take the exact same population of children, have them grow up in a home with low levels of lead from conception to age 5 and then measure their cognitive development score. You would also make sure that every thing else was exactly the same between the two scenarios. 
• a. You cannot ‘go back in time’, and you cannot ethically expose children to higher levels of lead. It would also be very difficult and expensive to collect data on all children in your population (some will move, etc.). Instead, you could take a sample of children from the population, measure the lead in their homes at one or more time points between conception and age 5, and then compare cognitive development scores between children who had high and low lead exposures. However, children exposed to high and low lead are likely to differ in other ways as well. 
• b. Y1 – Y0 or Yhigh lead– Ylow lead