Two tea estates from each of seven valley circles were selected randomly. Data on monthly production, rainfall, temperature, relative humidity, the quantum of different kinds of fertilizers used, degree of pest and disease infestation were collected from the selected tea estates with the help of a pre-designed structured interview schedule over a period of 12 years starting from 1978 to 1987. In addition, area under plantation was also recorded for each year under study. Average rainfall and dry period of the selected estates has been assumed as the rainfall and dry period of the respective valley. The collected data were first scrutinized and verified whenever needed. After proper scrutinizing estate wise data were processed and stored in the computer for the proper statistical analysis in fitting the best fitted model. The postulated model set for forecasting of annual crop is Pt= f(At, Rti, Dt, Ti,It,Ft).
Where, Pt= Estimated annual production of t th year;
At= Plantation area under tea beginning at the season of the t th year;
Rti= Cumulative rainfall upto i th month of t th year;
Dt= Dry period (days) prevailed before the first rainfall at the beginning of the season in the 7th year;
Ti= Linear time trend (t= 1,2......etc. and 1,2,3......etc. was used for T1 against 1978 T2 against 1979........etc. respectively);
It= % of infestation during the t th year and
Ft= Quality of all sorts of fertilizer used during t th year.
At the very beginning of the analysis attempt has been made to fit forecasting model for each of the selected estates wherefrom data were collected. All possible combinations of the factors such as area, rainfall, dry period, fertilizer, chemicals, degree of infestation and trend have been tried in regression analysis, simple and multiple, in fitting the model. The following criteria have been considered in selection of fitted model of the tea estate:
- The co-efficient of determination [R2 (OLS)]
- The sign and magnitude of the regression co-efficient
The parameters of the model have been estimated by Ordinary Least Square (OLS) methods, if the assumption of classical linear regression models are fulfilled. The OLS estimates are best linear unbiased estimates.