What is Sampling Bias?
We can define sample selection bias, or sampling bias, as a kind of bias caused by choosing and using non-random data for your statistical analysis. In survey or research sampling, bias is usually the tendency or propensity of a specific sample statistic to overestimate or underestimate a particular population parameter.
Sampling bias can exist because of a flaw in your sample selection process. As a result, you exclude a subset of your data systematically because of a specific attribute. It is worth noting that the risk of sampling bias is present in nearly all elements of both quantitative and qualitative surveys. This is why it may find its source easily in the survey creator as well as the respondents.
Ideally, you have to select all of your survey participants in a random manner. However, in practice, it can be hard to do a random selection of participants due to constraints such as cost and respondent availability. Even if you do not do a randomized data collection, it is crucial to be aware of the potential biases that could be present in your data. If you are aware of these biases, you can take them into account in the analysis to do bias correction and better understand the population that your data represents.
Types of Sampling Bias
Undercoverage
Undercoverage bias happens when you inadequately represent some members of your population in the sample. One of the classic examples of undercoverage bias is the popular Literary Digest survey, predicting that Mr. Alfred Landon would defeat Mr. Franklin Roosevelt in the crucial presidential election of 1936. This research survey sample was adversely affected by the undercoverage of many low-income voters in the country, who were Democrats.
Observer Bias
Observer bias occurs when researchers subconsciously project their expectations on the research. Did you know that this bias may come in several forms?
Some examples are unintentionally influencing your participants during surveys and interviews or engaging in cherry-picking by focusing on some specific statistics that tend to support your hypothesis instead of those that do not.
Self-Selection/Voluntary Response Bias
Self-selection bias (or volunteer/voluntary response bias) occurs when the research participants exercise control over the decision to participate in the study. A great example of this is call-in radio or TV shows soliciting audience participation in various types of surveys often on controversial and hot topics, such as abortion, gun control or affirmative action.
Those individuals that choose to participate in the study are likely to share some characteristics that distinguish them from the ones that choose not to participate. For instance, people who usually have substantial knowledge or strong opinions might be more likely to spend more time answering a research survey than people who don’t.
As a result, your sample will not represent your entire population and often overrepresents people with strong opinions.
Survivorship Bias
Another common bias in research is survivorship bias. Note that it occurs when a sample concentrates on subjects who passed the selection criteria or process and ignores subjects who didn’t pass the selection process. Survivorship bias can produce overly optimistic results or findings from a study.
For instance, if you use the record of existing companies or organizations as the indicator of the overall business climate, you will ignore the companies that failed and hence no longer exist.
Recall Bias
Recall bias is a common error in interview and survey situations. This happens when a respondent fails to remember things correctly. You should know that it is not about good or poor memory–human beings have, by default, a selective memory.
One way to avoid some of the implications of recall bias is by collecting information when a respondent’s memory is fresh.
Exclusion Bias
This bias results from excluding specific groups from your sample, such as the exclusion of subjects that have migrated recently into the study area. It is worth noting that excluding subjects or participants that move out of the relevant study area can affect your study’s validity.