Multiple Regression

[alex7134126]

Hi all, I am new to Eviews. I am currently trying to build a predictive model to predict the number of working hours (HOUR) needed for a construction project. I have got 16 records of the project sizes (SIZE), project costs (COST) and the development periods (DURATION). Additionally, I have a dummy variable indicating whether the workers have to provide additional service (ADDSERVICE) (i.e. clean up the site after the completion of the project).

The correlation coefficients between HOUR and SIZE, COST and DURATION are all positive and significant. So I build a model: HOUR = beta0 + beta1*SIZE + beta2*COST + beta3*PERIOD + beta4*ADDSERVICE. However I was noted that if I predict the hours using this equation, then I will have a probability of getting negative hours which is considered as unacceptable. So for prediction purpose, I decided to log-transform the hours series to ensure the prediction result will always be greater than 0. However here comes another question: the estimated coefficient of SIZE becomes negative which is considered as non-sensible such that the management won't approve it. In addition, the predicted hours deviate from the observed hours by a great amount (the worst case is the % error is greater than 200%), which will not be approved by the management.

I have tried principal component regression to reduce the number of variables, but the prediction results still deviate much.

Now I am stuck and don't know what is the next step, could anyone provide some help? Attached is the workfile I am working on.

[alex7134126]

Hi can anyone help? :(

[startz]

You don't have enough data to get very good estimates. And while it's hard to judge as an outsider, isn't it a little weird for someone to know what a project is going to cost without knowing how many hours it's going to take???

If you are stuck doing the model this way, you should look at the t-statistics on the coefficients.