Use of adaptive cluster sampling for identifying the land producing non hybrid crops

Abstract

For estimating rare and cluster characteristics the conventional sampling methods are hard very di_cult to be used, instead an alternative method like adapative cluster sampling is thought to be appropriate for in such situations. Coverage of Hybrid Boro usage in Bangladesh is rare and as well as of cluster pattern, so an Adaptive cluster sampling may be appropriate in estimating proportion of Hybrid Boro use. The applicability of the adaptive cluster sampling for this purpose, that is why, is planned to be visualized by a simulation consisting re-sampling of the Agriculture census (2008) data. Investigation of the suitability of the adaptive cluster sampling method for estimating the proportion of land producing non-hybrid crops is the prime objective of the study. Since only household (HH) level data are available (Agricultural Census, 2008), the proportion of HH producing non-hybrid (having the same sense of proportion of HH producing hybrid) was under investigation. The speci_c objectives of the study included to _nd an estimate of the proportion of HH cultivating Hybrid Boro using simple random sampling and adaptive cluster sampling methods. Also of the interest was to obtain bias and standard error of the estimators for each of the methods using extensive simulation studies and to compare them. The simulation study considered di_erent small sample sizes, namely 100, 200, 300 and 500. The choice of sample size is made arbitrarily keeping in the sense that adaptive cluster sampling is more pro_table for smaller sample sizes. The Monte Carlo absolute percentage relative bias and Monte Carlo standard error of the estimators were calculated for each of the methods for each of the sample sizes. The major _ndings of the study compared in terms of Monte Carlo absolute percentage relative bias and Monte Carlo standard error revealed that the estimator of the proportion of HH cultivating Hybrid Boro using adpative cluster sampling method has higher variance and lower bias than simple random sampling has for all the sample sizes considered in this study. The ultimate sample size realized by the application of the adaptive cluster sampling were also recorded and it is seen that the average ultimate sample size is about 10 to 20 percent higher than the initial sample sizes. The most interesting _nding of using an adaptive cluster sampling method was seen to be its strength of capturing more information. In estimation of proportion of HH cultivating Hybrid Boro, it has been revealed from the simulation that the adaptive cluster sampling method is way far better than the simple random sampling method in terms of chances of avoiding a bad sample containing very small number of targetted characteristics. For simple random sampling method such risk is higher for divisions with smaller true population proportion. The _ndings can be triangulated to the issue that the simple random sampling method may produce more bias than the adaptive cluster sampling method.