Error

Various sources of error may inaccurately impact the results of a study; thus, it is important to evaluate these sources of error to ascertain a study’s results. Errors can be systematic or random.

Systematic Errors

Bias

Bias is a systematic error that is a result of the actual experiment design or conduct. Bias is mainly categorized into two categories: selection bias and information bias.

Image

Selection bias is an error in how study subjects were selected. Common selection biases include control selection bias, self-selection bias, differential surveillance/diagnosis/referral, differential losses to follow-up, and the healthy worker effect. Control selection bias occurs when control subjects are selected in a way that is unrepresentative of the source population that produces the cases. Self-selection bias occurs when the willingness of participants to participate in a study is lopsided by control and case groups.

Information bias, on the other hand, is an error in the collection of data from the individuals, after they have been selected. Common information biases include recall bias, interviewer bias, and misclassification. Recall bias occurs when the groups being studied differ in the accuracy with which they remember the data requested by the study. Interviewer bias occurs when the person asking questions to collect data from the participants differs in behavior between the two groups, leading to false differences between groups. Misclassification occurs when a participant’s exposure or disease status is classified incorrectly, skewing the study results.

Bias can either skew the results towards the null or away from the null, depending on the specific bias. The magnitude with which bias can affect the results also differs by study. In any way, bias can create false associations or make true ones, and it’s minimization is necessary to be more certain of study results.

Confounding

Confounding is a systematic error that occurs due to an outside variable that distorts the exposure-disease relationship. Positive confounding occurs when confounding overestimates the true association, while negative confounding occurs when confounding masks the true association. An example of confounding (in fact, one of the most common confounders) is age. In a study where both the exposure and disease at question are influenced by age, there may be a false association between the exposure and disease when there isn’t really one.

Image

Random Errors

Random errors occur when the association between exposure and disease is inaccurate as a result of chance, which is an uncontrollable force that distorts this relationship. Random errors can be the result of measurement errors (incorrectly classifying an individual by exposure/disease) or the result of sampling variability (an unrepresentative sample due to chance.

Precision and Accuracy

Precision and accuracy, while seemingly similar, are two distinct concepts, each of which signals a particular characteristic of data. Accuracy is the lack of systematic errors, while precision is the lack of random errors. In a way, accuracy can be thought of as the “trueness” of a result, while precision can be thought of as the “closeness” of a result. Data can be just precise, just accurate, both precise and accurate, or neither precise nor accurate.

Image

Effect Modifiers

Effect modifiers are not errors; rather, effect modifiers are variables that, when data is stratified based on the variable, the association between exposure and disease. This is part of effect measure modification, which is not a problem, but rather a scientific phenomenon to be investigated further. These are significant because they can clue researchers into knowing more about the relationship between exposure and disease, and, more specifically, other variables that are involved in this relationship. An example of an effect modifier is sunscreen use in a study looking at how the exposure of having an occupation where one is exposed to the sun for an extended period of time may impact the disease of skin cancer. When stratifying the data between individuals with high sunscreen, low sunscreen, and no sunscreen use, the researchers may find differing measures of association. This may inform the researchers that sunscreen use may have some sort of association with skin cancer.

Question: In the example given, how would you design a study to further investigate the potential association between sunscreen use and skin cancer?

Page updated

Report abuse