Confounding exists when an association between two variables complicates interpreting the relationship between one and another variable. For example, an association between gender and carer responsibilities might confound an association between gender and the rate of cancelling sessions. Of course, that a higher rate of cancelling by women (say) might be down to more carer responsibilities does not mean such an association is not also “real” and to be considered seriously in its own right.
Detail #
There are methods to explore confounding. These all start in principle with simply looking at all three associations rather than just the one. For the example above this would mean looking at the association of carer responsibilities with cancellation rate and the relationship between gender and carer responsibilities not just looking at the relationship between gender and cancellation rate. These methods lead into analyses of moderation and mediation. Of course, these analyses are only possible where all variables involved are measured. The “unmeasured variables problem” describes the very obvious problem that we are very rarely able to measure all plausible confounding variables for any relationship that interests us. This is a reason to be very cautious when interpreting associations: always think what other variables, measured or not, might be impinging on, confounding, any simple bivariate relationship you see reported or in your own data.
Try also … #
Mediation
Moderation
Causality
Chapters #
Nothing here yet!
Further reading #
Nothing here yet!
Online tools #
Nothing yet.
