The precision of a survey depends largely on how the data is collected. Random sampling and face-to-face interviews, have been regarded as the best method to collect attitudinal data. Rapid changes in social environment of recent years have made it increasingly difficult to conduct face-to-face interviews of randomly selected respondents. At the same time, there is increasing need to obtain results quickly. As a result, other methods of conducting survey have emerged. This special issue contains articles on various survey methods used in media polls and academic research, focusing on survey methods(sampling methods and how the responses are obtained)and the precision of surveys(representativeness of the sample, valid response rates, and the characteristics of respondents).
The sampling of the ASAHI SHIMBUN telephone survey method was the same as the sampling of the face to face survey method. It was the random selection of samples from the electoral registers. But, as we couldn't know all the telephone numbers of the samples, we couldn't make contact with every sample taken. The respondents to telephone survey were less than those of the face to face survey, especially in young persons and urban dwellers. The representativeness of sample was somewhat inferior to the face to face survey, nevertheless, we think the telephone survey is method effectively measures indicates and the tide of public opinion without delay.
In this article, we are examining the data of5different sets of interview and telephone surveys, which were conducted by Yomiuri Newspaper from May 1994 through January 1995. In each sets, both telephone and interview survey data were collected with the identical sampling design, on the same interviewing dates and at the same interview points. According to our comparative examination, demographic distribution of the samples did not differ significantly from telephone surveys to interview surveys, but the response pattern were clearly different on some questions. Therefore, it can tentatively be concluded that the different response patterns between the two interview methods are most likely due to the interview environment created by telephones as media of interview. As these findings could only be obtained through an empirical examination, we need to continue our empirical examination and comparison of this new survey method.
Exit polls are now taking an important role in Election-Night broad-casting to project the final seats and analyze voter's choice on issues. NHK has done more than 180 exit polls by statistical methods since1993. In our sampling system, polling stations are sorted first by party strength and regional differences and then selected with a probability proportinal to size. We interview voters with a different interval set at each point with taking record of refusals. Interviewing are done for 6-7 hours except early morning, noon and evening assumed to cover about 70% of voters and not to make prediction worse. We weight targeted-interviewers to be proportion to real voters and calculate probability of candidate's win. Within the expected sampling error, past exit polls have shown mostly consitent‘ BIAS’ to specific party and candidates such as under-score for Communist Party. Refusals is a possible cause of BIAS, yet to be solved for improvement.
Today many social researches are made by broadcasting companies in Japan. In this paper, first the purposes of these researches are shown. Second this paper focuses on JNN Data Bank Research which is conducted every year by using random samples and self administration method in Kanto area, and discusses the reliability of it. Finally the problems which confront us now are shown and the measures to conquer them are discussed.
This study introduces the real working situations of mail survey practices in Japan and also examines the effects of a number of mail response inducement techniques and their combinations on rates and speed of response as follows: how many questionnaires were returned as undeliverable by the post office on average; using S-shape curve to estimate response rates; discrepancies between the date filled out questionnaires and the date mailed it; quality of contents answered including questionnaire item omission, who answers questionnaires, stability or consist ency and accuracy of responses; comparing respondents composition of demographic character istics with the distribution of the registers to resident's cards; comparison of demographic characteristics and response quality of early versus late respondents; effects of telephone call and postcard follow-ups on returns; effects of questionnaire appearence on responses including different combinations of postage on outgoing and return envelopes, survey sponsorship labeling, colored questionnaire paper, double-sided printing versus single-sided printing questionnaires, perceived questionnaire size and length.
The purposes of this paper are to discuss 1)sampling design and selection procedures and 2)the task of the interviewer in carrying out household visits, which is an element of survey practice. This discussion is based on continuous cross-national studies conducted by the Institute of Statistical Mathematics, beginning with a survey of Japanese-Americans in Honolulu in the 1970s, up to and including a1993survey in Holland. In cross-national surveys which use social data, we face problems of standardization at each stage of the surveys: the definition of population, sampling itself, the survey method, construction of the questionnaire, the selection of questions, and the translation of selected questions to retain comparability. Although these problems are common in domestic surveys, in cross-national comparison there are additional unique issues. Two of these relate to sampling design and subject selection. For sampling design, we make every effort to select samples to meet theoretical requirements. However, in cross-national surveys, there are some issues which can rarely satisfy these conditions. Here, we present in concrete form outlines of sampling designs for surveys we have conducted several countries and regions for our cross-national studies.
The etiology of mixed connective tissue disease(MCTD)still remains unknown and the disease entities named MCTD is not definitely but empirically constructed by the prescribed combination of the signs and symptoms and laboratory findings(SS)which are frequently observed in so-called collagen diseases. Using a clustering technique, internal structure of MCTD has been studied on a mathematical model as to whether it can be divided into smaller groups. Each MCTD case was represented in the form of multidimensional vector shown by the existence or non-existence of 32 SS. The results are as follows: 1. We could not divide MCTD into several subgroups with good reproducibility. 2. However, we sometimes obtained the subgroups which could be characterized by clinical features. It was concluded that the approach shown in this paper could give a working hypothesis to help the further classification of diverse MCTD patients.