Confounding variable, also known as confounding factor or lurking variable can be defined as an undesirable adjustable that has an influence on the partnership between the parameters of an experiment. Although they are not the changing of genuine interest (i. e. the indie adjustable), they can effect the outcome associated with an experiment and they are considered to be unwanted as they could add problem to an test. An effective designed test should aim to decrease or control the affect of such confounding variables in order to avoid type 1 error; one that increases a 'phony positive' bottom line that the impartial factors have a informal romantic relationship with the reliant variable. The partnership between your two observed variables is called a spurious romantic relationship, hence a confounding variable is a threat to the validity of inferences made about cause and effect, i. e. the internal validity because the detected impact should be attributed to the independent adjustable rather than to the confounding adjustable.
An example can be illustrated by the partnership between glaciers cream sales and drowning deaths. When these parameters are inserted into a statistical research, they could show a positive and potentially statistically significant relationship. However, it is a blunder to infer a causal romance (i. e. , glaciers cream causes drowning) because an important confounding variable which in turn causes both snow cream sales and an increase in drowning fatalities is not accounted for: i. e. summertime. Although there's a body of literature of criteria for causality, Pearl said that confounding variables cannot be identified in conditions of statistical notions by itself; some causal assumptions are occasionally necessary. For example, when causal assumptions are being described in the form of causal graphs, a simple criterion called backdoor will identify models of confounding factors.
Types of confounding variables
Confounding variables may also be categorized according with their source: the choice of measurement tool, situational characteristics, or inter-individual dissimilarities.
Solution
There are several ways to beat confounding variables in an experimental design by excluding or managing it. This is actually the following:
Case control studies: by assigning the same confounding parameters to both experimental and control group can control for such confounder, for example, if the cause of multiple infarct dementia has been studied, age and sex could be the confounding variables, therefore these factors should be matched up paired between your two participant organizations. In addition, randomization is also another solution as having all confounding variables (whether known or unknown) will be evenly allocated across all groupings.
Cohort studies: this is done by admitting a specific group of participants in to the sampling human population, for example a specific age range which may have an impact on multiple infarct dementia, therefore only a certain group is chosen for the analysis design such as male aged 45-50 years of age. This might limit the amount of matching between the groups and also cohorts can be comparable in regards to the possible confounding variable.
Stratification: in the example of multiple infarct dementia study, physical exercise is hypothesized to be a variable that can prevent this dementia from occurring. With age as a possible confounder. The sampling data will be stratified by age group so that the association between physical exercise and dementia can be analyzed per generation. If different generation produces different risk ratios (this is analyzed by statistical tools called Mantel-Haenszel methods), then time is seen as a confounding adjustable.
Despite alternatives for the controlling and restricting confounding variables, these strategies have limitations too. For example if the participant in the case-control research is a 47year old African-American from Alaska, avid rugby player, vegetarian, working as an engineer and suffer from multiple infarct dementia. Proper matching would need a person of the same characteristics but with the sole difference of being healthy. That is extremely difficult to attain and there is a risk of over- or undermatching of the study population. Also, in a cohort review, too many people may be excluded with this requirements, and in stratification, solitary strata can get too small and contain only a little, non-significant number of samples.
Randomization
One of the most typical known reasons for the living of confounding variable is when the experimental design does not randomly assign members to organizations or some types of individual difference such as capacity, extroversion, height and weight. For instance, studies involving a comparison between men and women are inherently plagued with confounding variables since the communal environment for males and females is completely different to begin with. However, this will not mean that there is no value in gender comparability studies or other studies that will not employ random project but it means that results interpretation should be done cautiously. In total, random task is a good and powerful tool in experimental design. Although it does not decrease the entire amount of extraneous changing in an test, it aims to equalize the problem that might occur as a result of extraneous varying, therefore it can greatly reduce systematic error: mistake that varies within the unbiased variable.
Multivariate analyses
Another method for handling confounding variable is by the use of covariates in multivariate analyses. However, this only offers little information about the strength of the confounding changing in comparison to stratification methods. Furthermore, confounding variables aren't always known or measurable this means residual confounding (term for incompletely manipulated confounding) may appear. Within an experimental design, covariate adjustment can help to reduce the noise in an results variation whilst allowing the manipulation impact to be performed. In amount, successful randomization can decrease confounding factors by bother assessed and unmeasured factors, whereas statistical control addresses only confounding factors that have been measured and can expose more confounding parameters and other biases through inappropriate control.
Mismeasurement and mis-specification
Although it's important to identify confounding parameters in a report there is often a risk of creating a statistically managed but imperfectly measured factor that may confound an association of the parameters. This is termed residual confounding which explains the mismeasurement and a good example is given to demonstrate this. In a study example, it was found that folks with higher rate of vacation is correlated with lower threat of mortality. Several explanations can account for this as vacation mitigates stress, diminishes anger and encourage more exercise. Within the other hand, healthier people might be more more likely to travel so getaway may not be considered a genuine causal factor but only a marker of original health status that naturally predicts longevity. As a result, vacation may continue to be to be a significant predictor even after altering for baseline health status as the covariate. Hence, it is easy to construct some potential confounders but many would lack plausibility. For instance, people with more friends may have significantly more vacations and good friend was indeed the predictor changing instead, low-stress working environment and large selection of food (I. e. completeness of diet) may all feature to prolong life too. However, because plausibility is an extremely subjective factor for considering whether enough potential founders are included. To identify confounders Priori knowledge of the likely causal pathways are essential. The major disadvantage of this is that observational studies imply that the strength of any causal inference will rely upon the biologic plausibility of the putative factor, and the implausibility of uncontrolled potential confounders. In addition, observations contain some judgmental aspect which ranges from experimenters. For instance, vacation may extend longevity because sick and tired people tend to travel less, to cope with this. Measurements of members' preliminary health can be utilized as an adjustment but this however cannot be assessed without problem. In addition, health can be measured in a wide variety of ways rather than all can be included and handled for. This raises increasingly more questions such as: can the utilization of original stress test be used to capture aspects of health confounded by vacation? Is body mass index relevant? As a result, even if the perfect measure of confounder is used it is measured with error and adjustment for it may not get rid of the effects of getaways.
From the statistical analysis perspective, poorly assessed confounding variables triggers more problems as its effect might not be linear, by assuming linearity on the outcome as specified by the model by coming into confounding factors as a covariate in standard regression models may not fully adapt for the confounder effects are not linear on that size.
Mediators and confounders
There is a common turmoil that different causal explanations can be possible when modification is used to lessen or get rid of the predictive electric power of the indie variable. For instance, a confounding variable may sometimes be considered a marker of some causal factors but it isn't directly involved in the causal chain in one variable to another and there's a problem of over-adjustment. Considering an example on the hypothesis that high blood pressure (BP) reaction to stress causes Hypertension. To test this hypothesis, a longitudinal study should be conducted where BP reactivity and relaxing BP degrees of a large band of members should be assessed. Result results should survey that high reactivity to be the risk factor for later hypertension but the situation is reactivity that are a marker for increased BP resting level which is not important per se. consistent with this issue, those participants with higher relaxing BP may correlate with high BP reactivity scores. To regulate for the existing confounding variable, the initial resting BP levels should be tweaked by regression examination which llustrates whether BP reactivity is related to any predictive information beside just the initial resting BP level. This may show that reactivity is no more a very predictive factor and the majority of the variation in the follow-up BP levels may be accounted for by the initial resting levels. However, this does not mean that reactivity is not causally related to future BP position, i. e. if increased reactivity preceded initial increase in relaxing BP level, it could also be accountable partly for the original increase in relaxing BP level. That is a predicament whereby an individual variable may have both confounding and mediating assignments simultaneously. The exemplory case of getaway and mortality is employed to demonstrate this: assuming that people who go on more trips are less inclined to die more than a 5-calendar year longitudinal research, including a factor: preliminary health position in the regression model could eliminate this association. On the other hand, if people in poor health take fewer vacations then this eradication may reflect removing a confounding variable by health position. However, if the participants' tendencies to be on vacation are frequent on the 5 season period then health status will reflect the cumulative health impact of your lifetime's vacation practices. This demonstrates health position will contribute partially as a mediator of vacationing effects. This misunderstandings between a mediator and a confounder will be less clear if the chance factor is not steady over time. For instance if the participant has only started having getaways, then these will not be reflected in the original health status and may have higher possibility to predict following health with primary health position as a covariate in the analysis. However, if these changes become uncontrollable, it can create a quasi-experimental design. For instance, if people take getaways due to change in their company coverage rather than the reason of making friends or have spare time, and other group have less vacation for the same reason. Then in cases like this, you'll be able to assess the effect of vacation individually of original health status.
In total, indiscriminate modification of covariates may lead to erroneous conclusions and many socialdemographic parameters can be mediated by other factors such as low income, unfulfilling careers, no friends etc. furthermore, there may also be other intermediate factors like self-determinations and release of stress human hormones that may have an impact on the results. Considering the wide range of variables posted, any inaccurate options of them can lead to a lowering or elimination of predictive ability. Moreover, by managing a mediator may produce further confounding parameters, which will then increase or reduce the organizations of the unbiased and dependent actions. Furthermore, it may even create a fresh spurious association when in simple fact no effect exists.
In sum, despite the number of constraints reviewed in this critical review, they have got an important role in behavioural research as randomized trials are sometimes found to be impractical and unethical. Regardless of the harmful statistical control of confounding factors will gain perception into special cautions in attracting conclusions and writing research in the future.
