charleytichenor
New member
- Joined
- Mar 19, 2010
- Messages
- 1
I am developing an ensemble cost forecast for a large project. There are many factors that have been quantified and incorporated into the model, but one part is causing me some hesitation.
Suppose that among several cost forecasting methods used in the ensemble, there are two particular cost forecasting methods of interest here. One forecast results from a multiple linear regression/ANOVA using four independent cost predictor variables. The other consolidates the data so that a simple linear regression is conducted using one independent predictor variable. Both models have a Significance F p-value below .01. The multiple variable model is used to hopefully take advantage of greater detail in the data by using four variables. The simple model is used because there may be some small to moderate violations of the underlying multiple regression assumptions in the data, and a simple model might reduce the magnitudes of those violations.
I hesitate to use a simple (arithmetic) average of the two cost forecasts to arrive at a single cost forecast because more data was used with the four-variable model, so I am considering a weighted average. What factors should I consider in choosing the weighted average approach v. the simple average approach. If the weighted average approach is better, and if I weight the 4-variable model by a factor of 10, what do you suggest I weight the simple model at. If there is no clear justification either way, does that in itself provide a reasonable justification for a simple average.
Suppose that among several cost forecasting methods used in the ensemble, there are two particular cost forecasting methods of interest here. One forecast results from a multiple linear regression/ANOVA using four independent cost predictor variables. The other consolidates the data so that a simple linear regression is conducted using one independent predictor variable. Both models have a Significance F p-value below .01. The multiple variable model is used to hopefully take advantage of greater detail in the data by using four variables. The simple model is used because there may be some small to moderate violations of the underlying multiple regression assumptions in the data, and a simple model might reduce the magnitudes of those violations.
I hesitate to use a simple (arithmetic) average of the two cost forecasts to arrive at a single cost forecast because more data was used with the four-variable model, so I am considering a weighted average. What factors should I consider in choosing the weighted average approach v. the simple average approach. If the weighted average approach is better, and if I weight the 4-variable model by a factor of 10, what do you suggest I weight the simple model at. If there is no clear justification either way, does that in itself provide a reasonable justification for a simple average.