Hi, I've been trying to a good model to predict the response variable in a data set.

This is for Multiple Linear Regression model selection and diagnostics HW.

sales seats value costgood wages newcap ads ft pt are numerical varibales.
Z1-Z11 are qualitive dummy variables. (ie if x=y, z1=1, otherwise z1=0; variable outlook is 1,2,3,4,5,6 1 being very unfavorable, 6 being very favorable)

This model only compares numerical variables.
model sales = seats value costgood wages newcap ads ft pt / selection=STEPWISE;
Results: value, ft, pt, costgood are left

This model only compares qualitive (dummy) variables
model sales = Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8 Z9 Z10 Z11 / selection = stepwise;
Results = Z1, Z5, Z6, Z11

This model compares only significant numerical variables and all qualitive variables
model sales = ft pt value costgood Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8 Z9 Z10 Z11 / selection = stepwise;
results: value, ft, pt, Z9 are left .

This Model compares significant numerical and qualitive variables
model sales = Z1 Z5 Z6 Z11 ft pt value costgood / selection = stepwise;
Results: value, ft, pt, Z1

This model compares all numerical and qualititve variables
model sales = seats value costgood wages newcap ads ft pt Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8 Z9 Z10 Z11 / selection = stepwise;
Results stepwise: value, ft, pt, costgood, z9 are left
Results Backward: value, ft, pt are left
Results CP: value, costgood, ft, pt, z9 are left

Question: Before I can move onto the next part of my HW, I need a good model (best possible). My question is should my model include costgood?

Also, I've been searching for a good SAS website to help me with my work. If anyone know's a good sites that uses SAS to explain Linear Regression in Statistics, I would greatly appreciate it :)