In this paper we explain the nonresponse treatment procedure the general characteristics of nonresponse. and the compare effects of treatment methods for the nonresponse situation.
The treatment procedure of nonresponse consists of four steps. First, it is examined why nonresponse occurs in the data collection step. Second, it is determined whether nonresponse is the data analytic stage or not. Third we investigate statistical of nonresponse such as nonresponse type, nonresponse pattern, and nonresponse mechanism if we have to treat nonrespons. Finally, we choose appropriate treatment methods for variable types, and analyze incomplete data.
We investigate the results of treatment methods by using Williamson and Haber(1994)´s data. In these data, one variable is always observed, and the other variable yields nonresponse because of a sensitive question. Thus nonresponse is not missing at random. The nonresponse rate that analyzers use generally tends to be underestimated. In fact. all ad-hoc methods increse estimates in comparison with nonresponse rate. Though nonresponse is not missing at random. MLE in the Mar assumption is overestimated more than MLE of the non-MAR model. On the other hand, the estimates of imputation methods is the largest value because of the selection of incorrect covariates in order to adjust nonresponse. Selecting super-variables is thus required to adjust nonresponse.
have to treat nonresponse. Finally, we choose appropriate treatment methods for variable types,and analyze incomplete data.
study was carried out to investigate the reliability between two retrospective surveys which was initiated with 71 women with spouses in November 1993 first and in December 1994, again. This study evaluated the level of correspondence by comparing pregnancy outcomes reported in the two surveys. The results of the study are as follows :
1) Of 173 pregnancv outcomes that have been investigated by two retrospective surveys. 118 cases(68.2%) were corresponded and 55 cases(31.8%) were contradicted each other in answering.
2) In answering of related to fetal death and induced abortion, a lot of data contradicted each other were observed. But the level of correspondence in live birth was high relatively.
3) In contradicted level the highest case was that after fetal death was answered in first survey answering was dropped in second survey.
4) This study examined levels of correspondance or not and factors related to a different answer, but there isn´t the possible explanation for such a pattern in responding.
|