SJ Balaji. I think mathematically you can calculate an average annual growth based on predicted values. However, the acceptability of your calculated outputs may depend on where you will use your data.

Hi Eddie.  I don't get the second line. Could u pls explain (what is "acceptability depends where i use the data).  Could i not use it in publishing a paper?

Using average for prediction is an accepted tool. This is exemplified by "moving averages". Sometimes there are projects that require true or historical data. If your study expects or accepts as an answer to your research problem predicted data, then it is acceptable. They could be used in publishing paper for as long as they answer your research question.

Eddie, could u suggest some literature that support this procedure?  I will be of great help.

S,J, Balaji, I do not know exactly your study but I guess you can refer to feasibility studies particularly the financial planning or analysis portion for your related literature.

Thnx Eddie.

So x is time (years), and you appear to have growth in y over time for which a linear regression would be suitable?  Then it is probably heteroscedastic.  

If you are looking for literature, you could check the long reference lists at the end of the attached and see if anything interests you. 

Article Properties of Weighted Least Squares Regression for Cutoff S...

Thnx James.

S.J.  -  It does not sound like you have a lot of complex data for data analysis re multiple regression.  So be careful regarding finding relationships that do not really exist.  Actually, that is always a useful caution.  

It sounds like linear regression may be your best choice, though for all I know it could be nonlinear.  Time as a regressor sounds a bit iffy to me, in my experience, but it does not sound like you have much else with which you can work.  At any rate, linear regression is often a very useful tool.  I would minimize my reliance on R-square, and drop p-values altogether (functions of sample size) and just compare the regression coefficient to its standard error.  The best tool you have is probably the variance of the prediction error, as shown in econometrics books like Maddala.  The estimated standard error of the prediction error is found, for example, as STDI in SAS PROC REG.  Regression is often very accurate, but of course, it depends on the data and the relationship(s) that exist.   -  Jim  

James and Paul, I think this is one reason why we must recommend that down voting be taken out. It is being used by people with incomplete learning. Anyway, let me give each of you one up vote. Ed

I am guessing I will get a down vote for this comment,

James, and Paul.  Sorry that u had the 'dislike' from some anonymous person.  Of course, it is not the network for entertainment.  Let the outlier to feel guilty of showing him here, but if he can, let him discuss for his behavior. 

It sounds as though you are only trying to provide a descriptive growth rate in a variable that is trending upwards over time.  Regression should be fine for that.  If you use the natural log of the dependent variable and regress that on a constant and time, the coefficient on time will essentially give you the percentage growth rate for the data period you are examining.

As per my comments above, I agree with John, except for using the log.  

Dear  S.J. Balaji, a non-parametric regression would be useful. Non-parametric regression is useful to show the link between two variables without specifying beforehand a functional form (in your case to predict the rate of growth over time). It can also be used to estimate the local derivative of the first variable with respect to the second without having to specify the functional form linking them. So, the effect between the two variables (incomes in log, rates (regardless the scale, absolute or relative value) can be depicted nicely in a linear non-parametric scatter.