Source: CDC Studies, Public domain (Gov) CDC Study: http://www.cdc.gov/mmwr/preview/mmwrhtml/00000206.htm http://www.cdc.gov/cancer/lung/basic_info/risk_factors.htm http://www.cdc.gov/tobacco/data_statistics/sgr/2010/highlight_sheets/pdfs/overview_cancer.pdf
Hi. This tutorial covers establishing causality. So to determine causality, generally, a randomized controlled experiment is needed. So let's just quickly review what causality means. It's the event that variation in a suspected response variable is caused by variation in a suspected explanatory variable.
So this can definitely be shown using a controlled experiment-- a randomized controlled experiment. And generally, causality cannot be shown using an observational study. So in most cases, an observational study is a poor way to gauge a cause-and-effect relationship or to gauge causality. But under certain circumstances, an observational study can show causality.
So in order to do that, you have to follow the following criteria. So do the following if you're interested in showing causality using observational studies. So I'll just run through the five criteria. 1, look for cases when correlation remains while other factors vary. 2, check if the effect is present or absent when the explanatory variable is present or absent. 3, look for evidence that larger amounts of suspected cause produce a larger effect. So when you increase that amount of whatever you think is causing it, you want to see if it has a larger effect. 4, check for other possible causes. And 5, try to determine the physical mechanism for the cause and effect.
We'll come back to that list in a minute. Observational studies have shown a positive correlation between smoking and lung cancer have often been cited as proof of causality. So this link between smoking and lung cancer is something that's been studied quite a bit. And this is one of the cases where they can show that they're trying to prove that there's causality using the observational study.
So basically what we're going to do is go through each of the five criteria. And we're going to look at some different research articles that show each. So for 1, look for cases when correlation remains while other factors vary. So what I have here is an article from the Center for Disease Control. It's a government organization. And I want to-- let's look at the bottom highlighted part here. So two studies found a statistically significant correlation between voluntary smoking and lung cancer risk in non-smoking wives of men who smoked.
So again, we're looking for cases when correlation remains while other factors vary. Well, if we're able to show that even people that aren't smoking, but just inhaling the smoke, are getting lung cancer, I'd say that there's a lot of different factors that are varying between smokers and nonsmokers. And if that lung cancer risk is still there, I would say that that first criteria is met.
2, check if effect is present or absent when the explanatory variable is present or absent. We're going to look at that same article, but look at a different portion here. So that's this first highlighted sentence. Cigarette smokers have total cancer death rates that are two times greater than those for nonsmokers. So I would say the effect is present when the explanatory variable is present. So smokers have higher cancer death rates than nonsmokers. So I would say that's a good way of meeting that criteria.
3, look for evidence that larger amounts of suspected cause produce a larger effect. Again, same article, we're going to read the second sentence now. Heavy smokers-- those who smoke more than one pack a day-- have three to four times greater excess risk of cancer mortality. I'd say that's a very clear example of that third criteria.
The fourth criteria-- check for other possible causes. And we're looking for other causes. So here, exposure to radon, exposure to asbestos, exposure to other chemicals-- so if we can recognize that these are also causes of cancer, we can make sure that when we're looking, when we're doing an observational study, that these factors don't confound our response. So we can make sure that we eliminate any people that have been exposed to these different things where they might be at greater risk for lung cancer.
So that's kind of the fourth criteria. And then the fifth criteria-- try to determine the physical mechanism for the cause and effect. So it says, scientists say a chemical found in cigarette smoke has been found to cause genetic damage. So that gives us pretty good evidence of the physical mechanism that's actually causing the cancer. So I would say, since these five conditions have been addressed, I would say at this point, it is OK to show that-- to make the statement that smoking causes lung cancer. So even though these are all just observational studies, because these five criteria were met, I would say it's OK in this case to show causality using just an observational study.
So that has been the tutorial on establishing causality. Thanks for watching.