Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 3-32 (2000)

How Has the Japanese Way of Thinking Changed?
A Half Century of the Statistical Survey
of the Japanese National Character

Yoshiyuki Sakamoto
(The Institute of Statistical Mathematics)

A statistical survey of the Japanese national character was first carried out in 1953 by the Research Committee of the Institute of Statistical Mathematics. Since then, a similar statistical survey has been conducted every five years, for a total of ten surveys. The latest tenth nationwide survey was conducted in October 1998. Therefore, these surveys cover most of the latter half of the twentieth century.

The purpose of the present paper is to clarify what aspects of the Japanese way of thinking have changed and what aspects have not changed, based on the analysis of responses to questions asked repeatedly over this half century.

The most notable characteristics of attitude trends over the period 1953 to the 1970's are the following two points:

  1. There were great changes in attitude trends that were related to politics, social problems, lifestyles and so on. In contrast, most of the responses concerning personal relations changed very little over this period.
  2. Until the early 1970's, traditional attitudes decreased and modernized attitudes increased. However, the modernization in the attitude trends slightly reversed or regressed in the middle of 1970's, around just after the first oil crisis, which implied that the attitude trends up to the early 1970's had changed.

However, these new attitude trends did not last until the 1980's, and the attitudes showed new trends after the middle of 1970's. The main findings focused on attitude trends after the middle of 1970's are as follows:

  1. There have been strong inclinations toward family and also toward nature.
  2. Women have been gaining in popularity in contrast with men.
  3. The greatest change in percentage in the latest 1998 survey was a sharp increase of pessimistic views toward Japan in various aspects, such as "economic achievement", "standard of living" and "level of science and technology".
  4. While responses concerning spirituality and personal relations were almost stable until the 1980's, some have been gradually changing in recent years.

Although it is difficult to summarize the above mentioned attitude trends over the latter half of the twentieth century, one point is that 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".

Key words: The Japanese way of thinking, survey of the Japanese national character, longitudinal survey, trend analysis of survey data on attitudes, values, personal relations.

Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 33-66 (2000)

The Future of Quantitative Study on National Character
—To Quantitative Study on Civilization
from Comparative Study of National Character—

Chikio Hayashi
(Emeritus Professor, The Institute of Statistical Mathematics)

The quantitative study on national character implies not only the approach by continuing surveys but also that by cross-national surveys including Japanese Americans and Japanese Brazilians. The former approach is discussed in Sakamoto's paper. The essential idea of methodology of cross-national survey is data-driven and exploratory. Further its research strategy must be developed. We call these CLADS—Cultural Link Analysis and Data Science. In the present paper, these problems are discussed.

In 3. of the contents shown as below, the fundamental ideas of comparative quantitative social research from the international perspectives are mentioned.

First, the conditions of comparativity among the data are discussed and then the outline of design of surveys in the nations selected by CLA is described with questionnaire construction. Finally the data analyses are shown in the following.

a.To reveal the linkage of nations including "Japanese Americans in Hawaii and West Coast in U. S. A." and Japanese Brazilians by quantification method III,
b.Macro-analysis on the scales by APM and clusters of nations,
c.Changing patterns of clusters of nations depending on the groups of questions taken up,
d.J-attitude and Japanese-Americans.
  1. The methodology of continuing survey and ross-national (comparative) survey
  2. Cultural Link Analysis and Data Science
  3. Comparative study on national character
  4. To quantitative study on civilization from the comparison among national characters

Key words: Civilization, comparative study, cross-cultural survey, cultural link analysis, data analysis, data science, national character, quantitative study.

Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 67-76 (2000)

Background of the Survey
on the Japanese National Character

Sigeki Nisihira
(Emeritus Professor, The Institute of Statistical Mathematics)

At the time of the first survey of the Japanese National Character in 1953, we hoped not only to obtain relevant data, but also to establish a model of social research that was based upon our experience. In the following section I will explain the different types of research with which we have been concerned.

  1. In the beginning we primarily used tables of random numbers by Fisher-Yates. There was, however, a flaw in these tables. Moreover it was difficult to import them. After discovering this error we proceeded to make tables with 100 coins, in 1947.
  2. The occupation army had doubts as to whether or not average Japanese people could read and understand documents such as official announcements and newspapers that utilized Japanese and Chinese characters. We therefore examined the literacy of Japanese people (from 16 to 65 years old) in 1947. It was the first nationwide survey and we gained experience with various kinds of survey techniques.
  3. Since 1948 we have received requests for sampling designs from many kinds of organizations.
  4. The concerns raised by the prediction failure of the American presidential election in 1948 spread at once to Japan. However, we felt relatively safe because we used random sampling rather than the quota-system used by the United States. We have experimented several times with the prediction of elections since 1949. In order to examine the validity of opinion surveys, we have compared survey data with election results.
  5. In addition, we conducted the following surveys:
    a. Relation between an announcer's way of talking during a news broadcast and the understanding of listeners
    b. Social-psychological research on several problems of Japanese language
    c. Statistical research on modernologio (i.e. study of behavior in daily life)
    d. The relation between the degree of danger of fire of each house and its insurance premium
    e. Research on questionnaire techniques
    f. Survey on social stratification and social mobility
    g. Survey on the Japanese National Character
  6. Tokyo periodical survey. In order to study the way in which public opinion changes as well as various problems in survey technique, we repeated a survey in Tokyo, every spring and every autumn from 1954 to 1974 and irregularly up until 1982, 52 surveys in all.
  7. We have done many regional studies on different themes.
  8. We participated in the design of labor morale surveys for 3 companies.
  9. The first cross-national survey covering all of Japan was conducted on the knowledge of youth in 1963. One part of the questionnaire had been posed in several countries in Europe, and another part before and during the Second World War in Japan. The second cross-national survey was done in 1968 and dealt with religious issues that had been studied in Europe and the United States.
  10. The Youth Bureau of the Japanese Government has carried out a cross-national youth survey (ages 16 to 24 years old) in eleven countries on general opinion in daily life. It started in 1972 and has repeated about every five years.


In the early days of public opinion surveys, people were asked to respond either "yes (agree)" or "no (disagree)" with regard to specific political or social issues, or to state their degree of satisfaction on some issues. When we composed questions for the first survey of the Japanese National Character, however, as a principle we tried to ask people's attitude toward certain social situations. These questions should be constructed very carefully, but these days too often we find improperly made questions in Japan.

Jean Stoetzel said that he referred to our survey when planning the European Social Value Survey. In the near future I expect that the Institute will organize an international committee for an East Asian Social Value Survey, in order to compare the ways of thinking among the Chinese, the Koreans and the Japanese. As a final step I hope that through the initiative of younger colleagues and scholars of many countries a World Social Value Survey will be carried out.

Key words: Japanese national character, opinion polling, social survey, cross-national study.

Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 77-92 (2000)

Attitudinal Changes in the Last Half of the 20th Century
in Japan: Its Elements and Forms

Yasumasa Kuroda
(University of Hawaii at Manoa)

The present study aims at characterizing the nature of significant attitudinal shifts in the last half of the 20th century as measured by the longitudinal survey of the "Japanese National Character" from 1953 through 1998 in Japan and other related surveys abroad during the last three decades. By "significant change in responses" we meant there is at least a 20% change in responses from 1953 through 1998. I found 21 items on which responses varied at least 20% or more. The analysis makes use of frequency distribution of the survey results and limited cases of cohort analysis results. The scope of analysis is limited to simple attitudinal dynamics and does not go into structural dynamics.

Of these 21 items, 8 of them had to do with democratization, individualism and equality; 7 of them had to do with man-woman equality and relationships, 5 with economic development, nature, environment, and escape from materialism and I dealt with the return to proto-Japanese values. Of all changes, the topic of democratization appears to be where the attitudinal shifts among the Japanese in the last half of the 20th century took place.

One can see the extent to which external pressure—from the Allied Powers' occupational policies of emilitarization and democratization of the defeated Japan to the internationalist waves of the post-cold war era—affected the attitudinal dynamics of the people in Japan. While external pressure plays a significant part in the attitudinal dynamics of the people, internal pressure rising from Japan's own economic prosperity and environmental concerns for example, in a serious way impact the attitudes of the people in Japan. And lastly, when all is settled, there is a trend toward to the people going back to the original point from which they hail, I would characterize these changes to be stochastic dynamics in nature and external in origin, depending upon particular characteristics of period in history.

Key words: Yamazakura and Rashomon model, bilayer model of internationalization, nature-human relations, democracy, individualism, gender equality.

Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 93-119 (2000)

A Bayesian Multinomial Logit Cohort Model for Data Obtained
Using a Multiple Choice Question

Takashi Nakamura
(The Institute of Statistical Mathematics)

This paper presents a Bayesian multinomial logit cohort model to analyze longitudinal social survey data obtained using a multiple choice question. It gives the description of the model, the estimation method of the parameters, the derivation of ABIC (Akaike's Bayesian information criterion), and the model selection scheme. It demonstrates the Bayesian multinomial logit cohort model is identical to the existing Bayesian logit cohort model when the number of response categories is two. A question on religious attitude from the study of the Japanese national character is analyzed both by using the logit model with each category separated, and by using the multinomial logit model on all categories simultaneously, to show the need for the latter model.

Key words: Cohort analysis, multiple choice question, the study of the Japanese national character, Bayesian model, ABIC, religious attitude.

Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 121-145 (2000)

An Application of UNISCAL to the Survey
of the Japanese National Character

Takahiro Tsuchiya
(The Institute of Statistical Mathematics)

This paper proposes UNISCAL (UNI-dimensional SCALing), which is a method to perform simultaneously both optimal scaling of items and selection of uni-dimensional items. The basic idea of UNISCAL was proposed in Tsuchiya (1996). The features of improved UNISCAL in this paper are (1) to provide four variants of \alpha i which serves item selection, in order to treat various kinds of data sets, (2) to control the degree of uni-dimensionality of selected items by changing parameter k and (3) to treat data sets with missing values.

Some artificial data sets are analyzed with both Hayashi's quantification method III and UNISCAL to illustrate the characteristics of UNISCAL in detail. An application of UNISCAL to the tenth nation-wide survey of the Japanese national character is also presented to demonstrate the performance of UNISCAL.

Key words: UNISCAL, survey of the Japanese national character, uni-dimensional scale, Hayashi's quantification method III, homogeneity analysis.

Proceedings of the Institute of Statistical Mathematics Vol. 48, No. 1, 147-178 (2000)

On Sampling Scheme and Sampling Precision
in Five Most Recent Surveys
of the Japanese National Character Study

Tadahiko Maeda and Takashi Nakamura
(The Institute of Statistical Mathematics)

This paper explains the sampling scheme of the five most recent nationwide surveys (from the sixth to the tenth survey) of the Japanese National Character Study and presents the sampling precision of these surveys from two viewpoints: sex and age composition of the samples, and the magnitude of sampling errors. First, the sex and age compositions of both intended samples and achieved samples are compared with those of the estimated Japanese population in each survey year. In order to estimate the magnitude of bias caused by the deviation of the samples, achieved samples were post-stratified by sex and age and back-weighted by the stratum weight in the population to obtain a weighted estimate of the population proportion of the categories of survey items under consideration. Differences between unweighted estimates and weighted estimates by post-stratified samples are examined. The major findings are as follows: (1) Deviation from the population is larger in the achieved samples than in the intended samples. It grows larger in recent surveys. (2) Differences between unweighted estimates and weighted estimates are less than 1% for most of the categories. The differences are relatively larger in the tenth survey. Items with large bias in one survey also tend to show large bias in other surveys. Second, an estimate of the population proportion and its standard error are calculated, assuming that the sampling scheme in each survey was the stratified two-stage random sampling. Major findings are as follows: (3) The average value of standard errors is largest in the tenth survey because it was based upon the smallest sample size. (4) While the standard errors of most items in the sixth and the seventh surveys range between 1.1 and 1.2 times the standard errors assuming simple random sampling, the ratio of these two standard errors ranges from 1.2 to 1.3 in the eighth and the ninth survey. The results of the tenth survey are close to the latter group. (5) The average difference between weighted estimates of proportion and unweighted ones tends to be larger as the response rate becomes lower.

Key words: The Japanese National Character Survey, stratified two-stage random sampling, sampling error, variance of an estimator, ratio of standard errors, bias.

Proceedings of the Institute of Statistical Mathematics Vol.48, No.1, 197-212 (2000)

An Active Learning Algorithm
Using an Information Criterion
for the Maximum Weighted Log-likelihood Estimator

Takafumi Kanamori
(The Graduate University for Advanced Studies)
Hidetoshi Shimodaira
(The Institute of Statistical Mathematics)

We suppose that outputs of a system are generated from a random mechanism having the conditional distribution given inputs. In this paper we study the estimation or identification of the system when the observer can select appropriate inputs to the system. We call such method of estimation Active Learning. We suppose that the statistical model does not correctly specify the system. When the statistical model is not correct, the active learning using the maximum likelihood estimator does not enjoy consistency. Here consistency means the convergence to the optimal parameter fitted by the Kullback-Leibler divergence. Hence we suggest an active learning algorithm using maximum weighted log-likelihood estimator (mwle). The algorithm is shown to enjoy of consistency. Moreover we point out that there are estimators which are better than the consistent estimators when the number of the data is finite. Considering such result we construct another active learning algorithm which selects an appropriate mwle using our information criterion. We give some experiments by computer simulation and explore the effect of the proposed algorithms.

Key words: Key words: Active learning, maximum weighted log-likelihood estimator, information criterion, statistical asymptotic theory, risk.

Proceedings of the Institute of Statistical Mathematics Vol.48, No.1, 213-227 (2000)

How the Scheduled Macroeconomic Announcements
Influence Foreign Exchange Markets

Yoichi Kuwana
(Hitotsubashi University)
Masayuki Susai
(Nagasaki University)
Yoshinori Kawasaki
(The Institute of Statistical Mathematics)

This paper examines the impacts of macroeconomic news releases on the foreign exchange market. Our analysis focuses on yen-dollar exchange rate and macroeconomic announcements made by the U.S. authorities. Utilizing the high frequent data, we examine not only the volatility but also the frequency of the occurrence of quotes, which differentiates our research from preceding works in this field. After establishing a statistical model for the price change and the quote occurrence, we derive likelihood ratio tests for homogeneity between news days and no-news days. In most cases, it turns out that the announcements of macroeconomic indicators have significant influence on the market. The price indices, however, have almost no effect on the drift and volatility parameters, while we can observe significant difference in terms of the average time interval of quotes.

Key words: High frequent data, foreign exchange market, microstructure, volatility, quote, likelihood ratio test.

Proceedings of the Institute of Statistical Mathematics Vol.48, No.1, 229-251 (2000)

Microdata Disclosure Risk Assessment Based on
the Expected Number of Predicted Individuals

Shido Sai
(Okayama Shoka University)

In the microdata, if the combination of categories on the key variables of some individual is different from all other individuals', it is called unique. When the microdata which is sampled from the population is disclosed, the number of unique individuals both in the microdata and the population is usually used to assess the risk of disclosure. However, most individuals other than unique ones are not safe, so it may be necessary to consider the comprehensive index including uniqueness. In this article, new index, the expected number of the individuals which is predicted when the investigator tries to connect all individuals in the microdata to ones in the population using key variables, is proposed, and some properties about it are considered. The Poisson-gamma model is used as a superpopulation model for the key variables. Furthermore, the extension to the general index is carried out.

It is called prediction disclosure that microdata enables the investigator to predict the categories of the sensitive variables for individuals with some degree of confidence. In this article we investigate the expected number of individuals in which the categories of a sensitive variable is predicted, when the investigator tries to estimate them for all individuals in the microdata. To add to the Poisson-gamma model for the key variables, the multinomial distribution and the Dirichlet-multinomial models are adopted for the sensitive variable as a superpopulation model.

Several types of indexes proposed in this article are applied to the sample microdata set of Census in U.S. to make sure of usefulness of them and to find the problems to be solved.

Key words: Microdata, key variable, sensitive variable, prediction disclosure, Poisson-gamma model, Dirichlet-multinomial model.