1.
Enter
data into the spreadsheet following normal WinORS conventions; that is, several
independent variables and at least one dependent variable.
2.
Enter
names on row one and one-letter
variable types (VType) across row two (I, D, F,
G, or W).
3.
Choose
Menu path: Applications / Statistical Methods / Regression / Ordinary Least
Squares.
If the data is not saved to disk WinORS will force you to File/Save at
this point. If the block range of the
data is not already set, then use the left-facing red
roll-up arrow and do so at this time.
Proceed to generate a solution.
4.
Analyze
the regression results. First check the
validity of the regression parameters.
When multicollinearity is present (see: Variance
Inflation Factors (VIFs) drop or
transform offending variables (see Multicollinearity
- Solving the Problem ).
Upon the elimination of collinear variables go back to step #3 and solve
the revised model.
5.
Upon
reaching a solution with minimal effects of multicollinearity
check the results to determine whether the model meets the following OLS
statistical assumptions: Regression Model Analysis - Linearity
(1); Regression Model Analysis - Outliers & Normality
(2); Regression Model Analysis - Constant Variance (3).
6.
Time
Series Models: Check the two-tailed Durbin-Watson
(d) and Durbin H test. If your time-series model fails this test (Reject Ho) then immediately go to menu tree:
Solution / Current Files (/SC). On the
tree click the plus sign (+) associated with Regression. Open one of the adjustment files: Durbin
Adjusted -or- First Differenced. Which one?
Your choice, but we suggest that the beginning user start with the first
differenced data. After opening this file, do not modify any settings. Simply solve for the OLS estimates. If the adjustment works (not always
guaranteed), then upon reaching a solution to the OLS model on the first
differenced data, the Durbin-Watson test should report Accept. A result of Inconclusive also
allows you to go forward to the next step (7).
If it does not work (fails the Durbin-Watson test), then open the Durbin
Adjusted file from the Solutions menu (/SC).
If that fails as well, then proceed to step 7 and be sure to note this
result in your final write-up.
7.
All
model builders (time-series as well as cross-sectional) need to check for
constant variance in the residual error term.
Review White’s Test for Homoskedasticity and the constant variance graph. If the assumption is violated (see the help file for assistance
in the interpretation of the plotted results), then go to menu tree: Solution /
Current Files (/SC). On the tree click
the plus sign (+) associated with Regression.
Open the file: Wieghted Least Squares. This file contains the data of the current
regression model with one appended column.
The added column is the OLS weight (square of the individual error
terms) with a variable type of W. This is a common adjustment that will assist
in (not guarantee) the removal of non-constant variation in the error term.
8.
After
invoking any combination of the data transformations discussed above,
the resultant model is suitable for the interpretation of causal effects
only. That is, the model cannot be used
for econometric forecasting purposes.
9.
If
the original data (not logarithmic transformation data)
was used to solve a linear-additive model, then
choose the Elasticity Tab to view elasticity
calculations for each observation. The
average elasticity is reported on the last row of this page. If you neglected to check the Elasticity
check box, simply solve the model again with this boxed in the checked
condition.
10.
Write-up
final results.