A New Approach of Point Estim ation from Truncated or G rouped and C ensored D ata A hm ed G uellil 1 and T ew fik K ernane 2 29 Dec 2008 D epartm entofProbabil ity and Statistics , Facul ty ofM athem atics U niversity ofSciences and Technol ogy U ST H B , BP 32 El -A l ia , A l geria D epartm entofProbabil ity and Statistics Technol ogy U ST H B Facul ty ofM athem atics U niversity ofSciences
BP 32 El -A l ia A l geria
A New Approach of Point Estim ation from Truncated or G rouped and C ensored D ata A hm ed G uellil 1 and T ew fik K ernane 2 29 Dec 2008 703BF7CDC8AA0796479DC4F3007FE693 arXiv:0802.2155v2[stat.ME] GROBID - A machine learning software for extracting information from scholarly documents

A bstract W e propose a new approach for esti m ati ng the param eters of a probabi l i ty di stri buti on. It consi sts on com bi ni ng two new m ethods ofesti m ati on. T he rst i s based on the de ni ti on of a new di stance m easuri ng the di erence between vari ati ons oftwo di stri buti onson a ni te num berofpoi ntsfrom thei r supportand on usi ng thi sm easureforesti m ati on purposesby them ethod ofm i ni m um di stance. For the second m ethod,gi ven an em pi ri caldi screte di stri buti on,we bui l d up an auxi l i ary di screte theoreti caldi stri buti on havi ng the sam e supportofthe rstand dependi ng on the sam e param eters of the parent di stri buti on of the data from w hi ch the em pi ri cal di stri buti on em anated. W e esti m ate then the param eters from the em pi ri caldi stri buti on by the usualstati sti calm ethods. In practi ce,we propose to com pute the two esti m ati ons,the second based on m axi m um l i kel i hood pri nci pl e of know n theoreti cal properti es,and the rst bei ng as a controlof the e ecti venessofthe obtai ned esti m ati on,and forw hi ch we prove the convergence i n probabi l i ty,so we have al so a cri teri on on the qual i ty ofthe i nform ati on contai ned i n the observati ons. W e appl y the approach to truncated orgrouped and censored data si tuati ons to gi ve the avour on the e ecti veness ofthe approach. W e gi ve al so som e i nteresti ng perspecti ves ofthe approach i ncl udi ng m odelsel ecti on from truncated data,esti m ati on ofthe i ni ti altri alval ue i n the cel ebrate EM al gori thm i n the case oftruncati on and m erged norm alpopul ati ons,a test ofgoodness of t based on the new di stance,qual i ty ofesti m ati ons and data. K ey wordsand phrases:EM al gori thm ,M i ni m um di stance,M odelsel ecti on from truncated data,Poi nt esti m ati on,Truncated data,G rouped and censored data.