I was glad to see AOGS improving the methodological rigor of its readers with this article on confounding. So I thought I would take a minute and help you improve your methodological rigor by talking about the difference between bias and confounding. I’m prompted to share this because I remember being new to the birth world and new to reading research and thinking that studies that had bias were useless.
The reality is that it is impossible to do a study without bias. The data we use, the variables we calculate, the people willing to participate in studies all contribute to bias in studies. As researchers, we know this, and we talk about the biases in the limitation sections of our papers. Bias does not necessarily mean your results are wrong. Bias may mean your findings only reflect a portion of the population. In general bias can either make it easier or harder for your data to show a significant relationship. So if I, as a researcher, can create my study in a way that it is biased against finding a significant relationship, but the data still shows a significant relationship, I can actually be more confident that what I am seeing is a real phenomenon – not less.
There is a source of bias that does mean I cannot trust the finding of my study, and that bias is confounding. It is important to remember that not all bias is confounding, and confounding is not the only type of bias. Confounding is a special circumstance when something unmeasured affects both the intervention I am studying and the outcome I am measuring. So what does this look like in a study? Here are some examples.
In a study to determine if hydrotherapy in labor helps reduce labor pain, 9o% of the women who did not use hydrotherapy had an epidural and the finding was that hydrotherapy resulted in more labor pain than normal. This is confounding because an unmeasured factor, having an epidural, was associated with the intervention (you didn’t get one if you used hydrotherapy) and also with the outcome (the amount of pain a woman felt in labor.
As researchers, we have many ways we try to ensure confounding doesn’t happen. We also try to design studies in a way that bias will not weaken our confidence in the study outcome. We are not always successful.