Whenever we carry out a statistical experiment there are two kinds of variables- dependent and independent variables. We wish to quantify the effect that the independent variable has on the dependent variable. For this, there are many statistical techniques available such as linear regression, ANOVA, ANCOVA, etc. But sometimes it happens that we get wrong information from these models because of the presence of a “confounding” variable.

As the name suggests, a confounding variable is a variable which the researcher did not account for and which hence adversely affects the results of the experimental study. The confounding variable also affects the dependent variable but since we did not consider that variable initially that effect shows up in the variance and increases the error of our model.

For example suppose we wish to study the effect of different kinds on fertilizers on the growth rate of crops. But the growth rate of crops also depends on factors like the quality of soil. If we do not account for the quality of soil in our model then it will act as a confounding variable and the results obtained will be erroneous.

One way to avoid confounding error is to take account of all major hidden variable by doing pre-experimental studies. We should also have a control group in our study. For example if we were to grow two batches of crops in the same plot of land, one with fertilizer and the other without fertilizer then the quality of soil would be eliminated as a confounding variable. This will reduce the errors in our model. To reduce random error, we should randomise our sample and replicate our experiment.