hello
i am trying to find missing precipitation from other best correlated stations by using multiple linear regression.
But my regression equation gives negative values (which is not possible for ppt)
So what should I do now?
Plz help
Sample calculation is attached
(same equation is obtained using R, XLSTAT, Excel) (Removing missing observations)
You need to use data transformation techniques. I think, log-transform will work in your case.
Dear Summera
I reworked your obtained coefficients and found very different results. The file is attached.
There seems some error in your calculation. I obtained all positive values although they are farther from the actual data as compared to your results, all are positive.
Secondly, I can see that the P-value in your Statistical Analysis results shows that many of your variables are not significant. This implies that you may need to exclude those variables in your final model which are not significant as per the norms of statistics, but that would leave you with maybe only one variable and would affect your concept and your research badly. So this is the time for you to notice that you are not heading in the wrong direction.
Thirdly, I would recommend you to use time series analysis statistical techniques instead of MLR for time-based data. This is what I have learned through a lot of work on statistics and literature review. For details on this third point, you may read my article at the following link where I have used ARIMA modeling for time series analysis. There are few other time series models as well like GM, ARMA, etc.
Article Hydroelectricity Consumption Forecast for Pakistan using ARI...
Lastly, annual precipitation data for forecasting is considered too unreliably. If not daily, at least try to use monthly data for analysis and forecasting.
I hope it would help.
Regards.
Rehan Jamil The negative sign used in column J makes the intercept positive and results in positive values. Shouldn't the intercept be used as it is?
Yes I overlooked the negative sign. Correction is made.
Moreover, you can create a new model without an intercept as well. I have included it already which shows all positive values. But if you see Sheet 1, I have highlighted P-values. All are higher than 0.05 (5%) which means none of the variables is significant. So I would again recommend you to think about my 3rd and 4th point above. Maybe this model is not reliable to be used for forecasting. Try using time series analysis techniques.
It simply means that linear regression is meaningless for this kind of data. If you want to see a global trend, use a moving average or so. If you want to detect seasonal variations, use a periodical function as a model.
You might try my software and find examples in its manual:
www.lerenisplezant.be/fitting.htm
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression