•  
  •  
 

Abstract

The role and importance of the industrial sector in the economic development necessitate the need to collect and to analyze accurate and timely data for exact planning. As the occurrence of outliers in establishment surveys are common due to the structure of the economy, the evaluation of survey data by identifying and investigating outliers, prior to the release of data, is necessary. In this paper, different robust multivariate outlier detection methods based on the Mahalanobis distance with blocked adaptive computationally efficient outlier nominators algorithm, minimum volume ellipsoid estimator, minimum covariance determinant estimator and Stahel-Donoho estimator are used in the context of a real dataset. Also some univariate outlier detection methods such as Hadi and Simonoff’s method, and Hidiroglou-Barthelot’s method for periodic manufacturing surveys are applied. The real data set is extracted from the Iranian Manufacturing Establishment Survey. These data are collected each year by the Statistical Center of Iran using sampling weights. In this paper, in addition to comparing different multivariate and univariate robust outlier detection methods, a new empirical method for reducing the effect of outliers based on the value modification method is introduced and applied on some important variables such as input and output. In this paper, a new four-step algorithm is introduced to adjust the input and output values of the manufacturing establishments which are under-reported or over-reported. A simulation study for investigating the performance of our method is also presented.

Share

COinS