Incorporating time series date into before/after intelligence testing? (n00b alert)

#1
Hello there!

A short introduction: My background is pure maths and computer science however I did not take anything past first year statistics at university. I consider myself fairly able when it comes to learning new mathematical concepts however 'I dont know what I dont know'. Please excuse me if I have made some fundamentally bad assumptions, I am here to learn after all!

I've been tasked with performing a statistical analysis on some data related to a drug trial. Three drugs are being compared. The drug is related to cognitive performance in a number of domains. Students were matched with a control in the same classroom and performed an intelligence test at the start and end of a six month period. The main task here is obviously to determine how each drug affected performance in the intelligence test. I think this can be acheived via a standard M/ANOVA. However there is another issue...

The complicating factor is the side effects of the drugs and how they may have affected cognitive performance or interacted with student's ability to learn. We have time series data for the duration of the trial where each student submits a weekly report of any physiological or pyschological effects they might experience.

Some anecdotal evidence suggests that for a particular drug, the initial side effects (nausua, sleeplessness) can last up to two months (33% of the experimental period). It is hypothesised that this may affect the students' ability to learn while they are suffering from said side effects. Now because the before/after intelligence tests are scaled to the age of the students, and the participants are matched to individual controls who are not taking the drugs, the situation could arise where the drugs *were* helping cognitive performance but because of the 'lost time' due to the physiological effects of the drug, the before/after differences will not be as pronounced as if the side effects of the drug were not existant.


Note that the above does not preclude the possibility of permanent side effects which are affecting a student's ability to learn. In the end this will be correlated with the effectiveness of a drug, and as such is not a part of this particular study. We are not looking to find if nausua affects cognitive performance, we are only interested in the effect of the drugs on cognitive performance.

So in summary, this is what I am looking for:

How can I translate this weekly time series of drug side effects into a 'handicap' (for lack of a better word) when it comes to determining how well a student has progressed over a six month period? To complicate things further, each drug has a different scale for measuring side effects. I dont want to hardcode a handicap value - because the effect may not exist at all!

I would appreciate any assistance with this problem. I hope I have made myself clear in the post, please ask if you would like me to clarify anything further. Like I said, I do not have an extensive stats background but I am quite proficient with mathematicals and am looking to broaden my horizons with this fascinating science. Statistics in social sciences seems to come with a whole new set of problems!
 
#2
hello,
I am not really sure what the problem exactly is, but certainly I think you can make models with structure equation modelling, with or without growth modeling. However, if I only care about whether or not the drug causes learning difficulty, I would probably start with a survival analysis to determine event-free survival. Then I might start to think how to play with the event-free-ness and see if it can be a continuous variable (I am not sure how it can be).
If this is not a good option, I think the complexity in doing it perfect is due to the quadratic like time-series data -- Do you mean that students might recover later and the learning ability returns? It might be worthwhile investigating what curve best describes the time series data. hope that helps.