Ethically, sharing of data for other purposes other than the intended is un-ethical (Callahan 1998)
To get a best predictor of the number of internet users, the researcher used multivariate linear regression. In this type of methodology, each of the predictor variable is modelled against the response variable, in this case the number of internet users. This process is carried over with different combinations of the explanatory variables and the values of R, coefficient of correlation, and R2, coefficient of determination for the different models are calculated. The model with the highest value of R is normally selected as the best fitting model for the data (Bryman 1992). R2 explains the variations in the response variable readings.
In this case, the researcher used all the explanatory variables in the initial model and used the backwards which eliminates the variables which are not better placed to explain the response variable as anticipated. The only problem with this technique is that it may result in the elimination of explanatory variables even before their effects on the entire model have been determined. As a best practice, I suggest individual simple regression equations to determine the individual effects on the response variable and then stepwise inclusion of the variables (Hinton 1995).