Selection bias

In this article, I would like to discuss a situation we encounter very often in daily life: the problem of selection bias. While preparing this article, I also benefited from Fethi Tekyaygil’s article in Medyum. 

First, let us explain what kind of situation selection bias is. Let us assume there are people with different education levels living in a region. Suppose we want to learn what these people think about a specific political view. If we try to reach everyone and ask them to answer our question, it would be a very time-consuming and costly task. Furthermore, it might be impossible to reach some people, or some might refuse to answer. In this case, instead of reaching everyone, reaching a group among them and getting their opinions is a generally accepted method. The most important point we must pay attention to here is trying to select a group of people who can represent the entire region in terms of the characteristic we have determined (education level). We call this group a sample. If the sample lacks representativeness, it means we have acted with bias in our selection. 

As Tekyaygil mentions, selection bias can occur in many different ways. The first of these is Coverage Bias. In this bias, the data has not been selected in a representative manner. For example, if we ask only primary, middle, and high school graduates in this region what they think about a specific political view, and do not ask what university students or postgraduate degree holders think, we would be acting with bias in terms of coverage. 

Another type of bias is called Non-response Bias or Participation Bias. For example, let us assume we know that among the people we direct our questions to, those who do not want to answer are generally illiterate. If we do not include them in the sample at all because they would not want to answer anyway, we fall into the error we call non-response or participation bias. 

The third type of bias is named Sampling Bias. It is a type of bias that can emerge during the data collection phase. It is also encountered in academic studies. It refers to the failure to meet the condition we call randomness or random selection. For example, there is a bazaar or square, a university campus, and a park area in the region in question. If we only go to the university campus and specifically select the people we will direct our questions to from among final-year university students, we eliminate the probability of the people selected for the sample being chosen randomly. Thus, the responses given will not be reliable in terms of representing the entire region. 

I would also like to explain what I mean by saying it is also encountered in academic studies. When students writing theses at university need to conduct surveys to collect data, they usually direct their questions to the circles they can access most easily. In fact, a very appropriate name has been given to this method in the literature. They call it “convenience sampling.” Okay, it might be a good method in terms of collecting data quickly, but what about representativeness? Does the sample they have selected possess the characteristics to sufficiently represent the main group? If not, the results derived from data obtained from such a sample will not reflect the truth. In other words, we can say that “there is no shortcut to the truth.”

To make a sound decision on any subject, many more criteria need to be considered, but if we pay attention to avoiding these three types of bias I have mentioned, especially when making a selection, it becomes possible for us to make more reliable decisions.