Visualize and quantify factors that most contribute to customer decision

#1
To the Direct Marketing department I have to to visualize and quantify the factors that most contribute to a customer’s decision to buy any group of product and any specific product. (Eg. Living in Scotland contributes with 18% to the conference attendance.
I have a range of data that I know about people who are our customers (eg. age, gender, region, interests in specific topics, company size, job level) I also know the transactions for these people (how much they paid for different products the company offers and what they bought).

I was thinking about doing a factor analysis in SPSS (principal component analysis with varimax rotation) and set the number of factors to the number of variables I use in the analysis (I don't want to create groups just to see contribution of each). I want to run this analysis on people who were customers of the product. As I deal with a large number of nominal variables I would set these to dummy variables (0/1). I would also standardize all variables (even the dummies). How can I tell which variable contribute to the decision making? Would the Total variance explained table, % of variance tell me the figure I'm after? Is this the right method to approach the issue? I'm open to any suggestions if I started with the wrong methodology.

Many thanks!
 
#2
I have only limited experience with factor analysis so perhaps someone with more experience can correct me, but one of the main points of factor analysis is to take a large number of variables and to condense it down into a much smaller and more manageable number of broaders constructs. If your goal is to simply look at each individual variable, it would seem to me that defeats the point of factor analysis.
 

hlsmith

Less is more. Stay pure. Stay poor.
#3
I also have only used factor analysis a couple of times, but regression analysis seems a potentially better fit given this description. I am unsure if you can standardize nominal variables (or at least have meaningful information from it). I would be interested if someone wanted to provide more information on that topic.
 
#4
Yes, it would seem as though some type of regression would be more of what you are interested in.

Regarding the standardizing, are you perhaps talking about making sure each question goes in the "same direction". For example, if question 1 asks about "if someone agrees" with something while question 2 asks about "if someone disagrees" with something, then these 2 questions are going in opposite directions. Basically a STRONGLY DISAGREE for question 1 would be equivalent to a STRONGLY AGREE for question 2. Is this what you are talking about for standardizing the questions?
 
#5
Thank you both for your answers! Originally I was thinking about logistic regression too, but I'm not sure that I can achieve % contribution to the decision making with that method. (could I use the odds somehow?) The reason why I was thinking about factoring (despite it defeats the point of the methodology) that the total variance explained table gives percentages. However I have to admit that I'm thinking about this for a couple of weeks now and don't get any closer to the solution.
 
#6
Actually if I rethink what I wrote probably standardizing is an extra unnecessary step. I got confused with cluster analysis, so please ignore that bit.
 

hlsmith

Less is more. Stay pure. Stay poor.
#7
You would be able to use the beta coefficients, standard errors, and p-values to understand the variables and attempt to rank them. The odds ratios would be controlled for the other variables and would provide the increased odds given the presence of the variable compared to its absence or reference level. You would also want to look at potential interactions between the variables. Given your data you may also want to look into hierarchical (multi-level) logistic models to better explain your data.
 
#8
So seems like I should give a try to regression. I'm not familiar with hierarchical (multi-level) logistic models though, so have to do some reading before I start. Is this methodology something that SPSS can do? How can I rank the variables once the model is ready? are there specific rules to follow? Please could you give an example how can I explain the results to marketing? As I don't have any analysis done so far some random figures would do. (Eg. how living in Scotland/London has an impact on the conference attendance). Thanks!
 

hlsmith

Less is more. Stay pure. Stay poor.
#9
Ranking, you could use the previous measures that I mentioned. Many people abhor stepwise models, but perhaps after you determine your finalized model you could also rerun it using forward stepwise variable selection and see which order the program selects, then read the operating manual and you could understand why the program ranked the variables in that particular order.

I don't know if SPSS runs multi-level models, probably.