Question

A survey analyst remarked: "When covariance analysis is used with survey data, there is a
danger that the treatments may be related to the concomitant variable." What is the nature of
the problem? Does this same problem exist when the treatments are randomly assigned to the
experimental units?

Answer 1

When covariance analysis is used with survey data, there is a danger that the treatments may be related to the concomitant variable is true and this leads to bias in the analysis and hence the conclusions can change because of these kind of variables.

For eg: If we wish to compare the salaries of male and female empolyees in a multinational company, we can note down the salaries of males and females and compare them and come up with a conclusion but it can happen that the salary is dependent on the number of years the employee is been working in that company so this variable is not considered while the analysis of salaries of males and females in that company, so this might land us to a complete different conclusion and hence the study will be of now use.

Here in above example it can be said that the number of years the employee is working in that company is an concomitant variable.

Similarly in survey data it might be the case that some variables are not observed while doing the survey but they do affect the variable under study and hence the conclusion.

No, when the treatments are randomly assigned to the experimental units then this problem of concomitent variable decreases that is it doesn't affect the variables under study a lot this is because of the random assignment of the treatment to different experimental group.

As the assignment is random any concomitant variables afftecting it does not come into picture because any variation occuring between the two can be accounted by the random allocation.