Looking for proper method to analyse a data sample (n=200) with a huge amount of variables (800) I have a data sample (approx 200) from a population of about 60 000 people. There are around 800 columns/variables in my data (the reason being that I had a few questions for which I applied a multiple choice arrays and the data just exploded). I'm trying to find out whether a certain score that was automatically calculated during taking the survey (it is a numerical value from 18-90) has some correlation/causation with other variables (most of which are qualitative and when imported to my statistical package show as "string" ones). It will also be interesting to be able to control for age, income level, gender of the participants (information that I also have). What methods of analysis do you think I can rely on?
 A: Here's what I would do: 


*

*Convert all variables that are stored as a string into numeric format. For instance, if there is a multiple choice question with four possible answers, then create four dummy indicators (binary flags :0/1) for that question.

*Make sure you don't have any missing data. In most cases, based on your description, I would replace missing values with zeros. 

*Perform a correlation analysis to study the correlation between (a) all of the input variables -- e.g., answers, age, gender -- and (b) your outcome variable, i.e., the "score". The correlation coefficients will help you understand which input variables are strongly correlated with the score.

*The next step is to perform a regression analysis on your data. Once the best model is built, you would study the composition of the model (which variables made it to the final model, and what are their standardized coefficients, etc.) to make inferences about the relationships in this sample. For instance, if you have age in the final model, and also the answer for question #5, then you can make a statement about how much question #5 influences the score after controlling for age.

