I am reviewing some papers and am trying to understand why in a study the researchers have chosen to categorise age of injury (into early age 0-9 and later age 9+) when looking at outcomes. Would it not make sense to leave age as continuous variable? If so how would this be analysed? For context this is to see if age of brain injury predicts or results in poor mental health outcomes. The outcomes can be entered as whole number scores or could be categorised into clinical/non-clinical severities.
thanks
You could use multiple or simple linear regression with a quantified mental health outcome variable and age as a continuous independent (predictor) variable.
Probably the reason why categoric age has been used in some studies is that mental health development is not linear with age with inflections and discontinuities related to puberty, education modes etc.
you can use simple/multiple logistic regression analysis if the outcome is categorized into binary (clinical/non-clinical severity) or simple/multiple linear regression analysis if the mental health outcome is in a continuous variable.
the reason behind the 9 years old cut-off is it might be related to the specific central nervous system development and any interruptions below 9 years old might severely affect the mental health outcomes. As a reviewer, you might want the author to specifically mention the reason for 9 years old cut off for their independent variable selection
Logistic regression refers to the outcome - any regression can include a mixture of continuous or categorical predictors. In this context, the question as to whether or not it is better to use age as a continuous or categorical variable will largely depend on the nature of the relationship between age and the outcome. In theory, categorising the variable loses information and so keeping it as a continuous variable is, on the face of it preferable. However, if, as seems likely, the relationship between age and the outcome is not a linear one (because, for example, of developmental differences) categorizing it may be a simple way of dealing with such non-linearity. There are other approaches but categorisation is an easily implemented one that is potentially valid - provided the cut point makes biological sense. I think the arguments about categorizing the outcome are less compelling if it is measured on a continuous scale although if there are clear diagnostic cut points doing so (and thus changing to a logistic regression model) may aid interpretation.
- Badges
- Science topic
- Similar topics
- Publishing