After all the data was generated in SAS, the following programs were run:

Parameter Estimates - Proc Reg generated estimated  coefficients for our variables.  This regression model is the result:

yhat = 1091.48 + 138.45x1 + 400.86x2 + 333.17x3 + 848.24x4 + 683.68x5 - 20.9x6 - 101.92x7 + 104.96x8

By viewing the results, we can see that Friday generates the highest Sales Revenue and Monday is the lowest.

Snowfall decreases Sales by $20.90 if it is present that day.  However, note that snowfall was accounted for even if the least amount of snow was present; for example, flurries.

The results also show that coupons have a negative effect on sales.  However, this value is skewed because we did not use a whole years worth of data.  Also, the only month using coupons is January, which happens to be the slack month.

Daily sales in December are $104.96 more than November and January.  Possibly due to the holiday season.

Coefficient of Determination - R^2 = .7730
The 8 explanatory variables account for about 77.3% of the daily sales. This shows a strong positive relationship.

Severe Multicollinearity - Using proc corr,  a correlation table was created to show us if there is any SMC in our study.  By looking at this table, we can conclude that there was no correlation between our explanatory variables because there were no values close to 1.  We also ran the vif's to double check for SMC and none of the values were close to or over 10.

Heteroscedasticity - Proc Plot showed us the variance of the error terms visually and displayed whether or not heteroscedasticity is present in our study.  By viewing the graphs, it was hard to determine if heteroscedasticity was present. So, we ran a Chi Squared test and it suggested that there is some variance in the error terms. 

Autocorrelation - By using the Durbin Watson test, we could see  if there was any correlation between the successive values of the error terms. The results of the test were inconclusive.  Meaning the generated value of the durbin watson test fell in the gray region between 4-du and 4-dl and it is too close to tell if we should reject the null.

Outliers - Using Proc IML, we found 8 observations that had values higher than 2.5 in the weighted residuals section. Therefore, we deleted the outliers in our study by using the proper commands below the input statement in the SAS program.

Specification Error - Looking at the plots, we concluded that non-linear relationships between y and x's do not exist.