Hi everyone,
I was wondering if anyone could put in the right direction about what kind of analysis I would need to undertake for a study I am proposing. In short I am investigating whether different forms of childhood adversity can account for cognitive impairments in later life and whether the association can be explained by reduced GR expression.
Therefore, I have three groups - one early life deprivation group, one early life abuse group and one control. I would sample six different age cohorts ranging from childhood to elderly and measure their cognitive performance using the WCST as well as determine GR expression in three brain regions. Assignment to the three groups will be based on retrospective reports and case files.
My question comes regarding the analysis. At the moment, I am thinking along the lines of a form of multivariate regression where dummy coding would be needed to account for the three categories; however, I am not too sure if this would be appropriate or if something along the lines of a MANOVA would be better.
Any sort of advice would be useful :)
Don't separate subject into different age cohorts. Model age as a continuous variable, perhaps using a flexible spline basis to allow for some non.linearity in the relationship between age and outcome.
You should be clear about the IV and its assumed distribution. I don't know how "cognitive impairment" is operationalized / measured and if it makes sense to assume that such a variable has a normal distribution with constant variance.
Just for curiosity: how do you measure GR expression in brain regions of children?
For each subject you measure GR expression in 3 brain regions. These triples are correlated, but as I understood this is one of the explanatory variables and not the response/outcome variable. I don't think that this is a classical case of repeated-measures (which refers to the response). If you did this to get a better impression of the expression across the brain, you may average these values. If you are interested in finding the region in which the expression correlated with the outcome... well that's a different question. You could use the expression of each region separately, but it might still be useful to consider the correlation with the triples. One could use the differences (or ratios) to the mean of the regions per subject to remove inter-subject variation of the (mean) GR expression. But depending on your model about how GR acts it might also be more instructive to analyze the uncorrected expression levels.
Peter McCabery
, I'm having a hard time finding anything that explains clearly how the Wisconsin Card Sorting Test (WCST) is scored. But some resources suggest that the outcome might be an error count. Is that correct? Thanks for clarifying.
- Badges
- Science topic
- Similar topics
- Biological Science
- Immunology
- Antibodies