NLM logo

Finding and Using Health Statistics

Content

Dependent and Independent Variables

In analytical health research there are generally two types of variables. Independent variables are what we expect will influence dependent variables. A Dependent variable is what happens as a result of the independent variable. For example, if we want to explore whether high concentrations of vehicle exhaust impact incidence of asthma in children, vehicle exhaust is the independent variable while asthma is the dependent variable.  

A confounding variable, or confounder, affects the relationship between the independent and dependent variables. A confounding variable in the example of car exhaust and asthma would be differential exposure to other factors that increase respiratory issues, like cigarette smoke or particulates from factories. Because it would be unethical to expose a randomized group of people to high levels of vehicle exhaust,[1] a study comparing two populations with differential exposure to vehicle exhaust would rely on a natural experiment, or a situation in which this already occurs due to factors unrelated to the researchers. In this natural experiment, a community living near higher concentrations of car exhaust may also live near factories that pollute or have higher rates of smoking.

When running a study or analyzing statistics, researchers try to remove or account for as many of the confounding variables as possible in their study design or analysis. Confounding variables lead to bias, or a factor that may cause an estimate to differ from the true population value. Bias is a systematic error in study design, subject recruitment, data collection, or analysis that results in a mistaken estimate of the true population parameter.[2]

Although there are many types of bias, two common types are selection bias and information bias.  Selection bias occurs when the procedures used to select subjects and others factors that influence participation in the study produce a result that is different from what would have been obtained if all members of the target population were included in the study.[2]  For example, an online website that rates the quality of primary care physicians based on patients’ input may produce ratings that suffer from selection bias.  This is because individuals that had a particularly bad (or good) experience with the physician may be more likely to go to the website and provide a rating. 

Information bias refers to a “systematic error due to inaccurate measurement or classification of disease, exposure, or other variables.”[3]  Recall bias, a type of information bias, occurs when study participants do not remember the information they report accurately or completely.  The subject of confounding and bias relates to a larger discussion of the relationship between correlation and causation.  Although two variables may be correlated, this does not imply that there is a causal relationship between them. 

One way to determine whether a relationship between variables is causal is based on three criteria for research design: temporal precedence meaning that the hypothesized cause happens before the measured effect; covariation of the cause and effect meaning that there is an established relationship between the two variables regardless of causation; and a lack of plausible alternative explanations. Plausible alternative explanations are other factors that may cause the dependent variable under observation.[4]. These alternative explanations are closely related to the concept of internal validity.  

[1]Trochim, W.M.K. “Establishing Cause and Effect.” Research Methods Knowledge Base, 10/20/2006. Web 1/24/2017.
[2] “Bias, Confounding and Effect Modification” Stat 507, Epidemiological Research Methods, Penn State Eberly College of Science, 2017 Web 1/24/17.
[3] Aschengrau A. and G.R. Seage. (2014) Epidemiology in public health. 3rd ed. Burlington, MA: Jones & Bartlett Learning.
[4]. Due to a long history of unethical research in health and social sciences, researchers have many ethical obligations when conducting research, particularly with human subjects. These obligations were first codified in the Nuremburg Code in 1946, which specified that the benefits of research must outweigh the foreseeable risks. Ethical obligations continue to evolve to protect human subjects, including confidentiality and anonymity unless waived and informed consent. Increasingly, communities that have a stake in the outcomes of research are involved in its design and informed of the outcomes of the study. All federally funded research in the United States is subject to review by an Institutional Review Board (IRB).
1

Previous Section Next Section