hi, I have a quite simple starting position consisting of seven independent vars (mostly nominal or ordinal), and one response var (metric, or if necessary recodable as binary var). The point that makes me hesitate to apply a multivariate VA (GLM or so) is the fact that my client wants that one should be able in the end to say "what factor is most importantly influencing the RV, which one is the second important,....". Well, the concept "importance" does not exist in stats to my knowledge, and my client surely means factor effects. So, I assume some good old multivariate VA is the proper way. However, individual knowledge is limited, so I would like to be corrected, e.g. if I may better apply an other PROC. Cheers, Chris