This tutorial talks about bias. First we'll talk about what it is and then look at a couple of common sources of it. Typically statistical studies are done with diligence. However, as studies get complex, sometimes it can be difficult to interpret things properly. Now one of the things that can arise is bias. And as we conduct more studies and as we can recognize the sources of bias, we can ensure that our studies are done properly.
Bias-- what it is a systematic favoring of certain outcomes. And the key to this definition is the systematic part. There are occasionally going to be errors that are going to arise just from randomness, but with bias, it's the systematic part of it that is the issue. Now bias can be intentional, or it's not-- doesn't really matter, and it can arise in both situations.
The issue with bias is that it causes both inaccuracy and unreliability. You can no longer trust this study because it might be inaccurate. And like I've said a couple of times, there are lots of causes of it, and it can happen at any point during the study. It can happen during the setup of the study, during the execution, or the analysis phase.
Now let's look at a couple of examples. One type of bias is selection bias, which is also called selection effect. As the name would suggest, the error is here. The systematic errors come during the selection part of it. And it's arising when there's a lack of proper procedures in selecting how the participants are chosen. It can result in an unrepresentative sample, and that leads to problems with the reliability of our study. So it's not just how the participants are chosen, but anything that's a lack of proper procedures at that point during the study.
Another type is participation bias. Participation bias is a type of non-response bias. And in participation bias, what happens is the people who respond differ systematically in some important way from those who don't. So there's an issue with the participants choosing whether or not to be part of the study and whether or not to respond, and differences between those two categories. For example, if you mail a survey about life habits to elderly residents, perhaps the residents that replied were healthier and fitter than those that did not. As a result, the two groups, people who replied and those that didn't, differ in a systematic way which is going to affect our results.
Another kind of bias is response bias. That occurs when participants aren't answering truthfully. So their responses aren't accurate. They're not truthful. And this can come from the researcher using leading words, kind of trying to lead the participant to answer in a certain way. And the participant might want to be pleasing the researcher, or there might be moral concerns. For example, if you're asking questions to teenagers about drinking habits, they might be concerned about responding yes and creating problems for themselves.
Now another type is the Hawthorne Effect. The Hawthorne Effect is when participants in the study behave differently just because they know that they're in the study. So for example, I know in my classroom when the students know that the principal comes in, they act much differently. They're being studied. They're being observed. So they change their behavior. This happens in research situations as well.
Another type of bias is the publication bias. The issue here is when researchers not only, but often publish papers with positive results. So a positive result is one where the researcher can conclude that the tested treatment is effective. So they're not necessarily positive because it's a good result, but positive in that the effect is caused by the treatment. And they're more often published because they're favorable, and they're the ones that advance science. Sometimes when you're studying something, and the results come back that your treatment does not have an effect, that's still important to know. However, it doesn't necessarily look as glamorous because it's not telling you what should happen and what does happen.
And the issue with publication bias is it overstates the effects of explanatory variables. If you look at the papers, and you say, oh, all of these treatments have effects, then you have a tendency to believe that explanatory variables do have effects. However, the papers where the explanatory variables don't have an effect just don't get published as often. And that's where publication bias has an issue.
This has been your tutorial on bias.