Proceedings of the Institute of Statistical Mathematics Vol. 53, No. 1, 3-33(2005)
A statistical survey on the Japanese national character was first conducted in 1953 by the Research Committee of the Institute of Statistical Mathematics. Since then, a similar statistical survey has been conducted every five years, totaling eleven surveys. The latest, the eleventh nationwide survey, was conducted in the autumn of 2003, which was the 50th anniversary of the statistical survey.
The purpose of the present paper is to describe the three topics, particularly (2) and (3), shown below, based on the analysis of responses to questions asked repeatedly over the last half century:
(1) Notable characteristics of attitude trends from the year 1953 to the 1970s
(2) Main findings related to attitude trends after the 1970s
(3) Attitudes toward religious feelings and personal relations
The main findings focusing on attitude trends, particularly after the mid-70s, are as follows:
(1) There has been a steady inclination toward values that prioritize private life as the most important. In this sense, the point could be expressed, not as “the diversification of values,” but as “the unification of values.”
(2) While most of respondents are fairly satisfied with family life, daily life, and living conditions, they are not satisfied with society.
(3) Women have been gaining in popularity in contrast with men.
(4) While responses concerning religious feelings and personal relations were almost stable until the 1980s, some have been steadily changing in recent years.
(5) The completion rate of the latest survey was the lowest of all the eleven surveys. This implies that many people like to avoid various social affairs.
Key words: Survey of the Japanese national character, the Japanese way of thinking, values, longitudinal survey, trend analysis
Proceedings of the Institute of Statistical Mathematics Vol. 53, No. 1, 35-56(2005)
All the participants in the eleventh nationwide survey on the Japanese national character (abbreviated as KS survey) were contacted again by mail to investigate the reasons for their cooperation to the KS survey. Respondents to the mail survey were classified into two groups according to their reasons for cooperation, i.e., willing participants and reluctant participants. We compared the responses to the KS survey among the willing participants, reluctant participants, and nonrespondents to the postsurvey. The nonrespondents tended to be more indifferent toward social relationships than the respondents. The percentage of “Don't Know (D.K.) ” was high in the nonrespondents, while the percentage of “other” appeared to be high in the respondents. Four methods were used in an attempt to estimate the nonrespondents; to limit the respondents to reluctant ones, to adjust weights based on the demographic characteristics, to match “D.K.” and “other” with the nonrespondents, and to exploit the difference between the willing and reluctant participants. The performance of each method was explored through comparison with the actual nonrespondents.
Key words: Nonresponse, postsurvey, reason for cooperation, weighting adjustment, multidimensional scaling
Proceedings of the Institute of Statistical Mathematics Vol. 53, No. 1, 57-81(2005)
This paper discusses the basic characteristics of mail surveys in comparison with those of a different survey mode: face-to-face interviewing. We conducted a mail survey, to compare it with the eleventh nationwide survey of the Japanese national character study, which was carried out by face-to-face interviewing in almost the same survey period. Additional information from a mail survey carried out three years before, and from another omnibus survey by face-to-face interviewing were also considered.
First, we scrutinized the response rates of different demographic groups in each survey, and compared them. The most distincitve result was the lowest response rate in the younger male group. This tendency was more or less common in the two survey modes. Different response rates among groups resulted in a bias of the achieved sample from the population value with regard to some basic demographic variables. While the sex and age distribution of the sample achieved in mail surveys was more similar to the population distribution than that achieved in the face-to-face mode, educational background seemed to be more biased toward higher education in the mail surveys.
Second, we discuss the difference between the results of the two survey modes with regard to two groups of survey items. In the first group, items in the questionnair have neutral or vague response options. Based on the results of some experimental items in the questionnaire, it is shown that, as a feature of self-administratoin mode, neutral response options of mail questionnaire attracts by far the larger portion of respondents, and that this difference from the face-to-face interviewing cannot be decreased by merely dealing with the wording of the question. In the second group, though items do not have neutral options, we still found remarkable differences between some of the items in the two survey modes. We could not find the reasons for these differences by contemplating the item content.
Based on these results, we point out some characteristics of mail surveys.
Key words: Japanese National Character Survey, mail survey, face-to-face interviewing, response rate, mode effect, demographic variables
Proceedings of the Institute of Statistical Mathematics Vol. 53, No. 1, 83-101(2005)
This paper investigates the relationship between the lengths of telephone interviews and response effects, and explores the practical use of the lengths of interviews. Respondents to two telephone surveys are classified into two groups based on the length of their interviews, and the characteristics of the responses between the short time and the long time groups are compared. Firstly, logistic regression analyses reveal that interviews with older respondents tend to be longer. The respondents' reasons for cooperation with the survey are irrelevant to the interview length, although the existence of such relevance was hypothesized. Secondly, more respondents in the long time group tend to select the option “Other or Don't Know” than those in the short time group. Thirdly, the response effects of changing options in terms of their number, wording, or ordering, are reported based on the results of a split-ballot survey. The magnitude of the response effects appears to be more significant in the short time group than in the long time group. Finally, some potential uses of the time length of the interview for data analyses are discussed on the basis of the results obtained in this paper.
Key words: Telephone survey, survey on the Japanese national character, length of interview, response latency, reason for cooperation
Proceedings of the Institute of Statistical Mathematics Vol. 53, No. 1, 103-132(2005)
This paper reconsiders a Bayesian age-period-cohort (BAPC) model with age-by-period interaction effects. The existing BAPC model is rewritten in terms of difference vectors and matrices. A two-factor interaction effect model based on assumptions on the differences of the effect parameters for a general purpose of analyzing two-way contingency table data is proposed. However, simply adding to it cohort effects is shown to fail to separate the age-by-period interaction and the cohort effects. To resolve this difficulty, a new BAPC model with age-by-period interaction effects is proposed. This model orthogonalizes the linear combination of the column vectors by the interaction effects to that by the cohort effects as well as those by the age and the period effects. It also involves a procedure for excluding a subset of age-by-period interaction effects to save the number of parameters to be estimated. An application to a dataset obtained from the surveys of the Japanese national character study is presented.
Key words: Cohort analysis, family is the most important thing, ABIC, gradually-changing-parameter assumption, Bayesian model
Proceedings of the Institute of Statistical Mathematics Vol. 53, No. 1, 133-157(2005)
This paper deals with a bias correction of Akaike's Information Criterion(AIC) for a selection of covariance structure models under a general nonnormal case, when we assume that an observation is distributed according to the normal distribution. The AIC under the normal assumption has a constant bias that depends on a kurtosis of the true distribution. We correct this bias by partly using the cross-validation method. We verify that our criterion is better than the AIC and TIC by calculating a theoretical bias and conducting a numerical experiment. Further, an illustrative analysis applying factor models with respect to Japanese people's level of satisfaction to an actual data set from the eleventh nationwide survey on the Japanese national character is also demonstrated. Using our new criterion, we search the best model from the candidate ones.
Key words: Akaike's information criterion, cross-validation, factor analysis model, model selection, nonnormality