Investigating impact of outliers in both independent and dependent variables on agricultural production data.

The production of high yielding variety (HYV) Boro rice depends on both climatic variables and some other non-climatic variables. Outliers may occur commonly in agriculture data. Regression outliers either in independent variables or in dependent variables pose a serious threat to traditional least squares analysis. The impact of some climatic and non-climatic variables like temperature, rainfall, net solar radiation, humidity and wind speed, lag-price and fertilizer on HYV Boro rice production have been investigated using regression diagnostics and robust regression techniques. In this study, we considered the annual HYV Boro rice production data from 1980 to 2000 for Mymensingh and Dinajpur districts in Bangladesh. We found that there were outliers in both the independent and dependent variables. The outlying observations that were found in the independent variables were corrected by the median of the respective variable series, the outliers in the dependent variables have been corrected by the robust least-trimmed squares (LTS) predicted observations of the HYV Boro production of the selected districts. Hence, the re-weighted least squares (RLS) estimation techniques have been used to judge the impact of outliers. The regression diagnostics for the selected districts were computed by both the OLS and RLS methods. Our study reveals that proper correction of outliers is very important for the regression models and there was improvement in the R-squared values for both the districts.

Preview PDF (Abstract) Investigating impact of outliers in both independent and dependent variables on agricultural production data.pdf Download (84kB) | Preview Official URL or Download Paper: http://world-food.net/

Actions (login required)

View Item