Portal field news

Portal field news


🏛 | Liberal Democratic Party split ... "Choshu War" and "Joshu War" Intensifying Power Conflict

Photo Liberal Democratic Party split ... (C) Shukan Jitsuwa Web

Liberal Democratic Party split ... "Choshu War" and "Joshu War" Intensifying Power Conflict

If you write the contents roughly
Mr. Nakasone is said to have overwhelmed Mr. Omi in the party's opinion poll.

Toshihiro Nikai, secretary-general of the Liberal Democratic Party, and former Prime Minister Shinzo Abe have recommended different candidates for the next House of Representatives election, and the party's official dispute under the hood ... → Continue reading

 Shukan Jitsuwa Web

This is the web version of Shukan Jitsuwa, a weekly magazine for men.In addition to articles from Shukan Jitsuwa, we also provide original articles on the Web.
We will deliver unique news and articles from a different perspective than other media.

Wikipedia related words

If there is no explanation, there is no corresponding item on Wikipedia.


poll(Yoronchosa, Seronchosa) is a member of a certain social group.public opinionThe purpose is to clarify the trends ofstatisticsOfSocial survey, Or its investigation technique. Those who are in charge of these tasks or who are active as livelihoodsOpinion pollCalled.

Survey method

Public opinion and statistics

statisticsA public opinion survey is a survey that uses public methods to clarify public opinion. Conversely, it is not a poll that does not use statistically correct survey methods. Regarding the power of statistics, we established a method of public opinion survey using statistics.Gallup OfGeorge GallupAccording toI could prove God statistically(I can statistically prove the existence of God)[1].

Statistical surveys include “sample surveys” and “XNUMX% surveys”. An exhaustive survey is a method of surveying all members of a certain social group, and accurate results can be obtained using this method.JapanWhen conducting a public opinion poll, it is necessary to do this for about 1 million subjects, which is very time consuming and laborious, so a public opinion poll uses a sample survey. In addition, the Japanese government will do it once every five yearsCensusThen, XNUMX% survey is conducted.

In addition, we are conducting a statistical survey in JapanMinistry of Internal Affairs and Communications Statistics BureauBelieves that it is essential for children living in the future to acquire the statistical knowledge and skills necessary for sample surveys.[2], In addition to providing statistics education in the new course of study since 24, you can learn statistics at the elementary and junior high school level on the Internet.I see statistics schoolAnd learn high school-level statistics."I see statistical school high schoolWe have a website called ".

Statistical analysis ensures the accuracy of public opinion polls, but public interest incorporated foundations ensure the accuracy of Japanese opinion polls from an ethical perspective.Japan Public Opinion Survey Association.Fuji Sankei Group,Tokyo Sports NewspaperWith almost all major media except Nikkei,Jiji PressThe system isCentral research company, But the Sankei Shimbun has withdrawn from membership on January 2009, 1.[3]),Cabinet Office,National Life CenterAn administrative agency that conducts polls,Keio UniversityMedia Communication Research InstituteInstitute of Statistical MathematicsAcademic institutions that study public opinion polls are affiliated with each other, and the private, government, and academia cooperate with each other in order to conduct correct public opinion polls.[4].

Specimen survey

Specimen survey(Sampling) is a random sampling of a certain number of people (Specimen) And collect answers,statisticsIt is a survey based on. Sampling investigations inevitably involve sampling errors, but if sampling is done correctly, a statistically correct result can be obtained. In polls, this ispublic opinion"Or"public opinion, Etc.

Many polls that conduct sample surveysCentral limit theoremIs the source of accuracy. Unless the population itself is a set that satisfies special properties, the "sample mean" approaches the "true mean of the population" (Law of large numbers), the variance decreases in proportion to the reciprocal of the sample size.

Sampling frame

When a public opinion survey is conducted in Japan, the population will be Japanese nationals, but it is necessary to first prepare a list (sampling frame) in which the elements representing the population that are the actual targets for sampling are described. is there.

A list that describes the elements that represent a population is, for example,Basic Resident RegisterIs used. In Japan, the family register function is almost completely functioning, so the sampling frame created using the Basic Resident Register almost completely matches the target population (target population) to be estimated from the sampling results. However, it is not necessary that the target population is exactly the same as the target population, as long as the sampling frame is representative of the population. ). For examplePhone bookIf you use as a sampling frame, there will be gaps between the population and theCoverage error) May occur, but if you think that these can be ignored or corrected, you can also use the phone book as a frame population.In modern times, it is easier and more accurate.(English edition(Random digit dialing, RDD) Is widespread, so I don't use the phone book.

Basic Resident RegisterWhen is the extraction frame, the extraction frame is about 1 million (pieces). Also, if the population is limited to Japanese voters in a political survey, etc.Electoral listIs mainly used, and it will be about 1 million (pieces). from herelotteryAnd thousands of people are randomly sampled.


Random sampling

When conducting a statistical survey, the sample must always be "random" from the population. thisRandom sampling(Random sampling). Because random sampling randomly determines the sampleProbability extractionAlso say.

In order to randomly sample samples in a sample survey, it is common to use (method of simply sampling the sample randomly), but in a poll, simply sample randomly from all populations. Rather, it is necessary to see the “will” of each specific group (layer) of the population, such as by region such as municipalities or prefectures, and by age group such as young people and the elderly. "" which is subjected to "stratification" (dividing the population into different populations) and "multistage extraction" (extracting a population from a population and extracting a sample from the population) is mainly used.

For example, in the stratification in the “Public Opinion Survey on National Life” of the Cabinet Office, there are 11 stratifications by districts such as Hokkaido and Tohoku, and 65 stratifications by city scale such as wards and municipalities.[5].

“Randomly” means “probabilistically”, that is, all objects in the population are sampled exactly with a statistical method so that they are sampled with the same probability, and never It does not mean that the "proper" extraction is performed. Several methods for generating random numbers for random sampling have been devised, but if the population exists as electronic data such as "digits of a telephone number", thePseudo random numberIs the easiest to use and is commonly used (RDD etc.).

When performing random extraction using non-digitized data, first apply for viewing data such as the Basic Resident Register, and for example, Nikkei Research, from the head office building in Chiyoda-ku, Tokyo, or branch offices in local cities Needs to go to the local government office with his own feet, turn over the number of paper data one by one, check it, and manually assign a serial number to it and perform "system extraction" work[6]However, even if they go to the surveyed person's home with the same foot from there, they may be rejected, which is very difficult mentally and physically.

Significant extraction

Samples that appear to be representative of the population can also be "randomly" extracted from the population if the investigator considers that random sampling does not yield a sample representative of the population. thisSignificant extraction(Judgment sampling). The sample is not determined stochastically in significance extractionNon-probability extractionAlso called (Nonprobability sampling).

The "action" here includes not only the "action" by the investigator but also the "action" by the survey subject (sample) himself. For example, there are often self-proclaimed "public opinion polls" in the private sector that take such survey methods, in which survey collaborators select themselves as "samples" themselves, apply for surveys, and sample themselves. This creates a bias called the "Self-selection bias", leading to inaccurate results.

If a public opinion survey is conducted by significant extraction, it may be possible to obtain a more accurate “public opinion” subjectively by the surveyed person, but there is a difficulty in the objective accuracy seen by anyone other than the surveyed person. .. The advantage of random sampling is not that accurate "public opinion" is obtained, but the sampling error and the size of the confidence level are stochastically determined, and in that sense statistical accuracy can be guaranteed. On the other hand, with significant extraction, even if accurate "public opinion" is obtained, it is not possible to know how much sampling error there is, and therefore statistical accuracy cannot be guaranteed. In the case where the number of surveyed persons is small, random sampling may rather result in a large sampling error. In some cases, a well-informed investigator may obtain the correct results by selecting a representative sample based on his own experience and perception. In a familiar example, it is used in cases such as when the editor-in-chief of Weekly Shonen Manga Magazine decides from the reader questionnaire that "There is hidden support, so continue serialization" etc. It is a method called "application method" that extracts only the people who have applied for the reader questionnaire from the inside), but it is sometimes used in the polls that are conducted before the large-scale survey in the public opinion survey.[7], It is usually not used for public opinion polls because it may be data that has no objectivity and can be valued only by the people who planned it.

Only significant applicants will be surveyed for significant extraction in statistical surveys.Application method(Voluntary response sampling), by assigning a sampling number according to generation and age, recruiting only the assigned number of collaborators and conducting a surveyAllocation method(Quota sampling), the researcher subjectively sets a "typical" sampling target, and only that typical sample (for example, "typical Japanese"Shimbashi StationExtract the previous office worker etc.)Typical method(Typical case sampling), collecting samples based on the introduction of acquaintancesMachine law(Chain sampling, the sample size swells like a snowball as you connect the introductions.Snowman method(Snowball sampling)".Twitter,Facebook"Surveys on SNS such as is typical), catch people traveling on the street etc. and ask for cooperation and conduct a surveyIntercept method(Intercept survey sampling) (Because the people I met by chance are taken as samplesAccidental lawAlso called (accidental sampling). in JapanShimbashi StationTV stations often go in front of them, and are generally called "street surveys" and "station front surveys"). These methods are easier to perform than random sampling, which requires statistically rigorous methods (in that sense, it is also called "Convenience sampling"), so it is quick and easy. This is often done when you want the data, but both are likely to be statistically inaccurate and therefore not polls.

In the history of public opinion polls, there was a time when the "significant extraction method" was used for public opinion polls. In particular, the "allocation method"1936 United States Presidential ElectionGallupWas first introduced, was spotlighted as a method of deriving "public opinion" with a small number of samplings, and is famous for being introduced by many other research companies. In this election, while Literacy Digest predicts the success of Landon with a large sampling of 200 million reader surveys, Gallup will use the "allocation method" to sample less than 1% of Re magazine's 5000 people. Predicting and hitting the Roosevelt candidate leads to inaccurate results even with large-scale sampling in the case of significant extraction, and with a very small number of sampling if the sampling accuracy is high It has become clear that the above can be derived. However, even with the "assignment method", statistical errors were unavoidable due to the nature of "artificial" selection of sample targets, and the election of the Truman candidate in 1948 could not be predicted. Therefore, as a result of the research committee having set up the opinion poll method itself, the "significant extraction method" such as the "assignment method" was denied as the method of the opinion poll, and the "random sampling method" was denied. ”Was only used. Regarding the accuracy of the "public opinion poll", it is necessary for children to know that through such historical trial and error, a system has been put in place to ensure statistical accuracy as much as possible. Provincial Statistics Bureau is thinking[8].

According to Nikkei Research, “A large sample size does not mean random sampling. Abandoning those who refused to cooperate in the survey, and selecting only those who kindly cooperated in the random sampling. No. Once extracted by a probabilistic procedure, do not substitute for another person."[9]However, I cannot compromise the person who refuses, so in reality I am compromising to some extent. The Cabinet Office calls on the people to "understand and cooperate with the opinion poll" because it causes errors.[10].


The value obtained from the sample survey is called the sample value. Since it is unavoidable that the sample value is accompanied by a sampling error, it is necessary to correct it so that it approaches the population value.

The "population value" in Japanese public opinion polls is the value that would come out if all people in the population (about 1 million people for all Japanese people) were surveyed. That is. If all Japanese people can be surveyed, the true population value will be unknown because we will not conduct sample surveys from the beginning. However, population values ​​can be estimated from other sample surveys.

For example, in a survey by Asahi Shimbun, the distortion of the composition ratio by region, sex, and age is corrected using the actual composition value for each household announced by the Ministry of Internal Affairs and Communications.[11].. It may be corrected using the values ​​of polls of other institutions or the census of the census (however, it is normal for newspaper companies and TV stations to use the values ​​issued by rival companies). Not).

Factor analysis

Analyze the correlation of multiple survey items and aggregate into several factors.

For example, multiple survey items are aggregated into two factors, "maintenance/innovation" and three factors, "support/disapproval/neither".


The term "accuracy" as used herein means "accuracy of opinion polls in a statistical sense", not "credibility of opinion polls". First, it is desirable to understand the scope of "statistics" in high school mathematics. "error"When"biasBe careful not to confuse.

As mentioned above, public opinion polls are conducted as strictly as possible based on statistical theory, and their accuracy is statistically guaranteed. Since the sample survey is not a XNUMX% survey,errorIs statistically unavoidable, but because the range of sampling error is also statistically guaranteed, Japan (Cabinet Office) is conducting a public opinion poll.

Coverage error

The error caused by the existence of samples that are present in the population but not necessarily included in the sampling frame isCoverage error".

For example, if the Basic Resident Register was used as a sampling frame, the Basic Resident Register prior to 2012 did not list foreign residents in Japan, so the actual public opinion of all Japanese citizens including foreign residents in Japan There was a risk that an error would occur between the two. In addition, when using a phone number created by the RDD method as a sampling frame, mobile phones were not included in the RDD method before 2016, so all Japanese citizens including those who only have a mobile phone There was a possibility that an error would occur with the public opinion.

Even if the above points are improved, the coverage error due to the fact that unregistered people and people who do not have fixed-line or mobile phones are not included in the selection frame is inevitable.

Sampling error

The error that is always produced in the sample survey isSampling error(Sampling error)."

First, let us consider how much the sampling error can be tolerated. The allowable error range isallowable error". Statistically, the tolerance is 5%. Next, let us consider at what rate the sampling error should be kept within the allowable error range. this,"Confidence levelOr "confidence factor". Statistically, the confidence level is 95%. Ministry of Internal Affairs and Communications is also based on 95%[12].. This means that in the opinion poll, if 20 surveys are conducted, 19 times the sampling error should be within the allowable error range, which is the standard of Japan (Cabinet Office).

The formula for calculating such sample sizes is(N: number of samples, p: response ratio, d: sample error, λ: confidence level)[12].. Target an infinitely large population (called "infinite population"; the actual number of populations in Japan is about 1 million) with a confidence level of 2000% and a sampling error of 95% or less. The required number of people is: λ=5 (at this time, the confidence level is 1.96%.), p=95% (This number is unknown until a public opinion poll is actually conducted, but if it is set to 50%, n is the maximum. And it is easy to calculate, so it is usually set to 50%), d=50% (sampling error 5%), and applying this formula to calculate, "n≈5", that is, " 384.16 people" can be calculated. In other words, the opinion of Japan is that the accuracy of the poll is statistically guaranteed if the number of samples in the poll is 384 or more (in fact, in order to simplify the calculation, λ =384 is often used, at which time the confidence level will be 2%, and the required sample size will be 95.4). The larger the size of the sample, the smaller the error. If the sample size is 400 or more, the sample error will be ±1066% or less. If the sample survey exceeds 3 people, the error can be suppressed to ±9604% or less, but even if the number of samples is doubled, the error remainsThe cost increases as the size of the sample increases, so considering the balance between the cost of opinion polls and the error, in Japanese opinion polls, sample surveys of about hundreds to thousands Have compromised.

Statistically, even if the sample size is small, some reliable figures can be obtained. For example, if the tolerance is relaxed up to 10%, even if the reliability is 95% and the population is infinite, only 96 samples are sufficient. In other words, statistically, assuming that there is no bias at all, just by conducting a public opinion poll on 96 people, it is possible to see the "people's will" of 95 million people with an accuracy of ±10% with a confidence level of 1%. You can

It can be said that the poll was statistically accurate if the "true approval rating" seen as a result of the election was within the range of the error that could be derived from the formula. Is there a 5% probability that the confidence interval is within the error range?Significance levelIf you have a 95% confidence level, there is a 5% chance of this risk. Even if the reliability is 99.99%, it is dangerous as long as the risk rate is not zero, and it is rare that you lose after winning a "certain win" and making a banzai), or apart from the statistical error. somewherebiasIt can be said that there was.

The “bias” that distorts the accuracy of polls will be discussed later. By the way, regarding this "bias", no matter how large the sample size is, even if it is a XNUMX% survey, it distorts the accuracy of the survey.

Survey form

Note that the "survey form" is not related to the "system for random sampling". For example, although a method (net answer) to collect answers using a system called the Internet from a random sampling sample created in advance using a system other than the Internet has been established, The method of sampling is not established.

As the extraction frame of random sampling in mailing or individual visit,Basic Resident Register,Electoral listAre mainly used. For telephones, the "RDD method" is mainly used as a method of creating an extracted frame without using these. From here in the statistical senselotteryIs performed and a sample is extracted.


The survey contents will be sent to your home by postcard, etc., so you can answer and mail it.

Individual visit interview method

The investigator visits the subject's home directly and listens at the interview. Alternatively, a response survey form may be distributed in advance, asked to fill out the survey subject, and the survey form may be collected at a later date.

It takes a lot of time and labor, but because the investigators directly explain and present physical materials, there is less risk of misunderstanding compared to other methods.Cabinet OfficeIs being carried out monthly.

In major media, Jiji Press and NHK are doing this.

NHK calls it "interview survey" and calls it the "royal road for surveys." However, according to a 2005 NHK report, doing this nationwide would require about 18000 investigators, the cost of the survey would be enormous, as well as recruiting investigators and conducting research briefings. A large amount of work is required for preparation and implementation work such as management of investigators at the time of actual inspection. Also, comparing the data from the 1978 and 2003 interviews conducted by NHK, the number of respondents who refused the survey increased due to heightened awareness of privacy, etc. Has dropped from 78.1% to 61.5%[13].. Therefore, as of 2005, most broadcasters have abolished the interview survey. NHK also uses it in conjunction with the RDD-style telephone poll.


A method of answering through the telephone.Because it is an oral survey onlyDoor-to-door visitIt may cause misunderstanding of the question content, but the result can be obtained more quickly than other methods, soJiji PressMajor Japanese newspapers other thanBroadcaster(Especially in TV stations), it is held monthly.

As a method of random sampling in telephone surveys, in recent yearsComputerbyRDD method(Random number method, random digit dialing,Random digit dialing) Is often adopted. On the computerrandom numberBased on calculationPhone NumberIs generated, a call is made, and a question is asked to the responding person,NTTSuch asPhone bookPhone numbers not listed in can be extracted. Although there is "dialing", to clarify that it is a sampling method rather than a method of dialing a telephone, "RDD SamplingIt is also called. The Mainichi Shimbun Company calls it "RDS method (Random Digit Sampling)".

The telephone opinion poll in Japan was first introduced in 1987 by Nihon Keizai Shimbun (Nikkei Research) in Japan. Initially, it was sampling from the telephone directory, but starting with the Mainichi Shimbun adopting the RDD method in 1997, the Yomiuri Shimbun adopted the RDD method in 2008, and finally the major news companies. Switched to RDD method. RDD method has beenFixed-line phoneHowever, since around 2010, the number of people who own only mobile phones without owning fixed phones has increased since the 2016s, so from around XNUMXMobile phoneRDDs targetingAsahi ShimbunIntroduced mobile RDD from July 2016[11]).

In the RDD method, it is necessary to conduct telephone surveys on both fixed and mobile phones.Dual frame surveySay However, since it is not possible to specify the area code with a mobile phone, surveys targeting specific areas have been conducted only for fixed-line phones as before.

In the RDD method, great care is taken to bring the composition of the respondents closer to the "votive contractor". For example, at the Asahi Shimbun, in order to prevent the survey from being biased toward housewives and elderly people who have a high home-at-home rate (first responding to a phone call first), the number of voters who live in that household after the call is connected is first determined. Listen and select one person from the random numbers to be the subject of the survey. Even if the selected person is absent, the target person once decided is not changed, and the time is changed to make multiple calls. Even if you refuse, once again, please continue to cooperate. The survey will be conducted by 10 pm in principle (until 11:XNUMX pm if you can make a reservation) so that people who come home late for work etc. can also answer. If you have only answering machine or ring tone, make different calls at different times. If a person with a contract for multiple lines is selected, make an adjustment according to the number of phones. Since households living alone have a higher probability of being surveyed than households with many voters living together, adjustments will be made accordingly. Even when summing up the results of both fixed and mobile phones, adjustments are made according to the case where people who have both fixed and mobile phones are selected, and the total number of fixed and mobile phones. In addition, corrections are made to eliminate distortion in composition ratios by region, sex, and age.[14].

When conducting a dual-frame survey, care must be taken with the placement of weights on landlines and mobiles. For example, in the case of a telephone opinion poll using the RDD method in Nikkei Research, the sample population is about 1 million because it is a nationwide voter, and the sampling frame of the sample created by the RDD method is about 2 million for both mobile phones and landlines. 3000 million, of which the size of the sample to actually call is 3000 mobile phones and 2000 fixed phones[15].

Dual frame survey

The “Dual Frame Survey,” which uses both landline and mobile phones to survey public opinion, is the latest method of opinion polling as of 2017.

The conventional telephone survey was conducted only for fixed telephones, but in Japan, the age of voting rights was lowered to 2016 years or older in June 6, increasing the need to survey the awareness of young people. In order to eliminate the margin of error in public opinion polls (coverage error), which is caused by the inability to sample the young people who often use only mobile phones, a dual-frame survey was introduced by major media companies.

The dual frame survey in Japan was jointly conducted by 2014 major media companies (Asahi Shimbun, NHK, Kyodo News, Nikkei Research, Mainichi Shimbun, Yomiuri Shimbun) and the Japan Public Opinion Survey Association from 2016 to 6, and was introduced. It was advanced. This was also interesting to the Behavior Metric Society, so the details of the experiment were reported in detail in "Behavior Metrics" (Journal of the Behavior Metric Society) No. 86 by the Yomiuri Shimbun Tokyo Headquarters Public Opinion Survey Department.[16].

If you omit the theory and write only the results, there are many people who do not answer the phone number that you do not know on the mobile phone, so there was a question that the RDD method using the mobile phone might not give a certain number of answers. However, it was found that a certain number of responses were actually obtained, and it was confirmed that the problem of coverage error was ameliorated by conducting a telephone survey using a mobile phone. It was also confirmed by comparison with the mail survey conducted in parallel with the telephone survey that there is almost no effect of "not absorbing the opinions of "people who do not answer the telephone"."

Regarding the error, we found that the integration of fixed-line and mobile phone samples and the widening of the error due to unanswered adjustments were small enough not to affect the interpretation of the survey results. In addition, a mobile phone survey found that the proportion of women among the respondents was lower than that of men, but it was found that the effect on the margin of error is sufficiently small even when gender is corrected.

As a result, the dual-frame survey turned out to be effective for a better survey, so the Yomiuri Shimbun switched the poll method to the dual-frame survey in April 2016.


In order to conduct a public opinion survey in Japan, it is necessary to randomly sample samples from all Japanese citizens, but it is impossible or impossible to do random sampling of samples over the Internet with the current system. Therefore, using the Internet to investigate the public awarenessInternet pollCannot be done over the Internet.

On the other hand, the response method of the public opinion survey that accepts the response using the Internet for the random sample prepared in advance using the method other than the Internet (Net answer) Has been introduced. For exampleMainichi ShimbunSince 2016, in addition to conventional mailing, we have been accepting online poll responses[17].. On the other hand, the Cabinet Office in Japan believes that "it is very difficult to understand the consciousness of the people through the Internet without any bias", because people who use the Internet and those who do not use the Internet have different consciousness. We do not adopt online answers, but a method in which investigators directly interview the person to obtain answers (visit interview method)[18].

A survey of user awareness using the Internet is not a public opinion poll, so the Japanese Cabinet Office simply callsInternet surveyIs called[19].インターネットWith the spread of usage, the Japanese government is also studying how to use Internet surveys, and there is an attempt to use the results of “Internet surveys” in opinion polls. For example, you can put it in a samplingSample biasIn order to solve the problem of (sampling bias), studies such as weighting samples using a propensity score are being conducted. AlsoCabinet OfficeHowever, if you use the “Survey on the attitude of people's lives on the Internet” (Internet survey) that targets only Internet users registered with a research company, and the simple random sampling method (random sampling) (visit interview method) Assuming an error from the (surveyed) “Awareness survey on national life” (public opinion poll), we are looking for utility in the direction of whether it can be used as a reference for future public opinion polls. However, even when comparing the results of correcting the bias of the sample by the visit interview method and the Internet survey, the tendency of response differs depending on the survey method and the frequency of use of the Internet, so it is necessary to pay attention to that point.[20].

Internet user opinion poll

A survey of the public opinion of all Internet usersInternet user opinion pollAs for "," since there is no system that randomly samples samples through the Internet, it has not been possible because all Internet users cannot sample randomly through the Internet.

A market research company conducted an "Internet audience rating survey" in which a sample was randomly selected from Japanese citizens by the RDD method, and if that person was an Internet user, he was selected as a "monitor" and surveyed their consciousness. ofNielsenHas been introduced by. However, this is also "more accurate than a panel survey in which samples are artificially sampled via the Internet", and it is not a random sample from all net users, so it can be a "sample representing all net users". Absent.

As a low-precision “Internet survey” that targets samples extracted “through the Internet” and “artificially” from Internet users, the survey target, such as a monitor registered with a survey company, is fixed in advance. Survey method (Panel survey) Exists. For exampleNico Nico DougaIs conducting a panel survey called "Monthly Internet Public Opinion Survey" that fixes the survey target to Nico Nico Douga users.[21].. Internet survey methods other than "panel surveys" include a method of guiding users who click on a web banner to the survey site (Open survey), but these are not randomly sampled, so even if they are called “public opinion polls”, they are not effective as public opinion polls, and they do not represent all public users, not even public opinion polls.[22].


Even if you conduct a public opinion survey correctly, the statisticalerrorApart from, distorting the accuracy of pollsbiasThere are some (bias).

Sampling bias

First, as a matter of course,populationResults are inaccurate unless randomly sampled from (). For example, if only the supporters of the A candidate were artificially sampled and a public opinion poll was conducted, the support ratio of the A candidate in the survey results was significantly high, regardless of the true value of the support ratio of the A candidate in the population. Become.

Answer bias

For example, with bias, respondents may respond appropriately to end the survey early, or may give the opposite answer to hide the generally evil thoughts such as racial and sex discrimination. is there.

In particularvoteIn opinion polls related to intentions,Secret votingDue to the characteristic of this method, it is possible to “vote candidates that are difficult to announce to pollers”, and the response bias may make a big difference between the poll results and the poll results.2016 United States Presidential Election,Hidden playing cardsIs an example.

No answer bias

For example, there is a bias of not answering because the questioner's attitude is disliked.

The response rate also depends on the subject of the survey. For example, “Reject the survey for the XX newspaper and accept the survey for the ΔΔ newspaper”. Especially in political issues, the response rate of respondents who are favorable to the survey subject is high, and the response rate of those who are not.

For example,Abolition of capital punishmentSueAmnesty InternationalJapan Chapter of 1996House of Representatives general electionAccording to the questionnaire to the candidates, it was the ruling party at that time.LDPThe response rate of the candidates was low. Generally, public institutions and major companiesMass mediaAlthough the response rate to the survey is relatively high, if the response rate is too low, valid respondents' responses cannot be applied to the entire sample. Called election plannerHiroshi Miura"Even if only one company is inaccurate in the mass media survey, if multiple surveys are combined, the results will be highly accurate."[23]

Question induction

For example, if there is a positive or negative foreword such as “I have a XX candidate that increases expectations for political reform,” “A black candidate is rumored to be a XX candidate” before the question sentence, the answer is also the image It is a bias that is dragged by.

Regardless of intentional intention, it must be carried out so that public opinion guidance such as "the answer is guided by the question sentence" or "the existence of a certain question influences the answers of the following questions (carryover effect)" is not performed. I have to. Furthermore, it is difficult to handle "ambiguous answers" and "no answers/don't know", sostatisticsThere is also a problem in estimating the population.

For people who are not interested in the news, they cannot understand the content of the question before answering the question, so it is necessary to explain the background of the news in detail, such as "what is the problem". There is also the problem of leading to questions.

Repeated listening

This is a type of question induction, and asks the respondents who answered "I don't understand well" or "No answer" over and over, saying, "If you say strongly," "Don't say something," "Yes" or "No." Is a bias that induces an answer to.

For example, among the major Japanese media outlets, the Yomiuri Shimbun and Nikkei Shimbun listen to each other repeatedly, so there are fewer "no answers" and more "yes" and "no". On the other hand, since the Asahi Shimbun and Mainichi Shimbun do not listen to each other repeatedly, there is a bias that there are more "no answer" and "I do not understand" than other companies, and "Yes" and "No" less than other companies. Pointed out in the report[24].. There is no problem when comparing the numbers of polls in the same company, but it becomes a problem when comparing the numbers of polls in the same period in the media.

Range bias

This is the bias of the survey range. For example, if only the owners of fixed-line phones are surveyed, it is not possible to follow young people who often have only mobile phones. On the contrary, if only the owners of mobile phones are surveyed, it is difficult to follow the elderly people who often have only fixed phones.

This causes a coverage error.

Other issues

From the research source or its subsidiaryCall centerEtcRound throwThe labor forceDispatched laborIt has been pointed out that the accuracy of the survey decreases due to the harshness such as being covered by[25][26].. In April 2020Fuji Sankei Group OfFuji Television Network, Inc(FNN)WhenSankei ShimbunJointly conducted between May 2019 and May 5, the survey subcontractor Adams Communication conducted a total of 2020 times, and each time one fictitious answer was entered, hundreds of cases were illegally input. It was discovered that both Fuji TV and Sankei Shimbun and other contractorsNippon TelenetApologized, but there is also an opinion that it was one of the causes that this fabrication occurred due to the poor working environment at the outsourcee and the deterioration in the accuracy of the check system of the investigation source.[26][27][28].

If polls are always conducted under the same conditions, it is possible to make yearly/monthly period comparisons, but in order to make a comparison between options, it is necessary to match the changes in the proportion of age, occupation, etc. in the population (Japanese people). Need to process the data. In a public opinion poll that asks for political party approval ratings and voting destinations, major media publish without such processing, so it is necessary to handle it as statistical information with caution.

In order to know the accuracy of the results of a public opinion poll as a statistical survey (range of error, etc.) or whether or not a public opinion poll is being conducted using a statistically correct method, the sample size and specific survey method For example, it is necessary to disclose information with high transparency, but in some cases only the results of public opinion polls are disclosed.

Many media companies conducting public opinion polls are public interest incorporated foundationsJapan Public Opinion Survey Association, And even rival companies cooperate with each other to carry out correct polls, but if a researcher who receives an order for polls is not a member, the "Japan Public Opinion Survey Society Code of Ethics" Anti-social individuals who are not obliged to comply with regulations such as the “Practice Code” and store personal information such as family structure, political views, religious tendencies, and consumption trends after investigation and are sold and used by manufacturers. There are also cases where information is resold.

Problems of RDD method

Compared with the individual visit interview method, it has the advantage that it can be implemented in a short period of time at low cost, but it does not reflect the answers such as those who respond to personal information only by face-to-face survey, those who do not know how to use the telephone , There is a possibility that the respondent's age and occupation may be biased.

Only households with at least one household member having a landline or mobile phone will be surveyed, so for household members who have no household members or hospitals where telephones cannot be used. It is inevitable that a considerable number of members of the society, especially the vulnerable people, are unavoidable, such as persons with disabilities who are unable to use the telephone because the opinions of vulnerable medical personnel during hospitalization and the like cannot be reflected, and foreigners who cannot speak Japanese. Because it is easy to deviate from the population, there may be a large bias in answers depending on the subject or question.

Media historianTakumi SatoCites two essential problems of the RDD method. The first is that "people who are willing to answer a phone call that suddenly enters the space of their private life deviate from the average image of "people's will".[29]The second is that respondents are not always fully aware of the question.[30]

In addition, the RDD method is intended only for fixed-line phones, is used only during the daytime on weekdays, and samples the people who answered the phone (so many housewives and elderly people who are at home during the day and easily receive calls are sampled. There are some people who misunderstand, such as, but there is no such thing. At least the RDD method of Asahi Shimbun and Nikkei Research Inc. also targets mobile phones, and repeats until the household member answers the phone (at Asahi Shimbun, it is 10 or 11 pm at night). Instead of being a household member, the sampled person is replaced to conduct a public opinion poll.

Gallup Survey

As a representative pollGallup SurveyIs mentioned. The Gallup Survey is the American Institute of Public Opinion, a commercial opinion poll. Gallup (Gallup Organization) is a general term for public opinion polls. The surveyPresidential electionIs particularly famous.[31]

George Gallup (George Horace Gallup) (1901-1984) is an American psychologist and statistician. He founded the Statistical Survey Method of Public Opinion and in 1935 established the US Public Opinion Survey.

Gallup now has offices in more than 30 countries around the world, and many investigators are active. The results of the company's surveys areMass mediaFeatured in.

In the 1936 presidential election,Democratic Party OfFranklin Roosevelt (Franklin D. Roosevelt),Republican Party OfAlfred LandonThere were two candidates. A major magazine, The Literary Digest, predicted Roosevelt's defeat after 2 million polls. In contrast, Gallup, which did much less research, expected re-election and Roosevelt re-elected. Gallup was in the limelight due to its expectations.

"(English editionIt is said that the magazine's expectations were missed because it overlooked the characteristics of public opinion polls using telephones, which was rare at the time. At that time, the telephone penetration rate was 40%, and there was a difference in the penetration rate between the wealthy people, who had been using phones for a long time, and the other people. There are many Republican supporters in the wealthy class, so it was analyzed that there was data that was favorable to Landon. For itGraham WaldenThe bias of the survey results of Li was more than 40 million valid responses to 1,000 million hearings, rather than the survey method (phones with a penetration rate of 230%). It is pointed out that it is due to the bias of the respondents' group due to being a faction.

Controversial poll

  • In the 20th century, a debate poll (deliberative poll)James FishkinAdvocated by
  • In the 21st century, a debate-type public opinion poll was adopted in Japan over future nuclear power policy.

The facts from the poller

A reporter at Hebei Shinposha revealed that only elderly people answer the phone during the daytime on weekdays, and that no journalist likes opinion polls. The reason is that they can hang up silently, and those who keep talking to the reporter regardless of the questions from reporters are stated as "stressful work".[32].

Criticism from surveyees

  • 2020 year 6 month,Donald TrumpThe US PresidentCNNWent2020 US presidential electionRegarding the opinion poll toward the US, he called for the withdrawal of the survey results and an apology, saying that "the aim is to deceive US voters through biased questions and distorted extraction methods." Survey results show that it is the opponent of President TrumpJoe BidenShowed a significant lead. In response, the executive vice president of CNN refused to apologize, saying, "It is the first time that American politicians and election camps have suggested legal action because they do not like the results of CNN's opinion poll."[33].


  1. ^ Upset the US Presidential Election (Part XNUMX)-Bias in Sample Survey (XNUMX) | For teaching statistical learning (for teachers) Ministry of Internal Affairs and Communications Statistics Bureau
  2. ^ Positioning of statistical education in schools | For teaching statistical learning (for teachers) Ministry of Internal Affairs and Communications Statistics Bureau
  3. ^ Japan Opinion Poll Association Bulletin "Yoron" No. 103, March 2009
  4. ^ Japan Opinion Poll Association [Group membership list]
  5. ^ Public Opinion Poll on National Life XNUMX Sampling Method -Cabinet Office
  6. ^ System extraction Nikkei Research
  7. ^ How to extract a sample - Saitama
  8. ^ Upset the US presidential election (part XNUMX)-Bias in sample survey ② | For teaching statistical learning (for teachers) Ministry of Internal Affairs and Communications Statistics Bureau
  9. ^ Random sampling Nikkei Research
  10. ^ Poll-Cabinet Office
  11. ^ a b Public Opinion Poll-News Feature -asahi.com
  12. ^ a b I see, Statistics High School | Number of people required for the survey -Statistics Bureau, Ministry of Internal Affairs and Communications
  13. ^ Current status and issues of interview surveys NHK Broadcasting Culture Research Institute
  14. ^ Opinion Poll What is the "RDD" Method-Politics Asahi Shimbun digital
  15. ^ Survey method Nikkei Research
  16. ^ Masashi Fukuda,Introduction of telephone survey for fixed and mobile phones and evaluation of estimated values ``Behavioral Metrics'' 2017 Volume 44 Issue 1 p.85-94, two:10.2333 / jbhmk.44.85
  17. ^ Japanese Public Opinion XNUMX: First Internet Response In addition to Mail Survey -Mainichi Newspaper
  18. ^ Poll-Frequently Asked Questions -Cabinet Office
  19. ^ Awareness survey on public life via the Internet Cabinet Office
  20. ^ Awareness survey on public life on the Internet Summary of survey results Cabinet Office April 2008
  21. ^ Smiley questionnaire
  22. ^ Internet survey Nikkei Research
  23. ^ Miura "Brainwash Election"Kobunsha Publishing Co., Ltd.Paperbacks, January 2005,ISBN-4 334-93351-3, Page 72
  24. ^ The role and limits of polls Kazuaki Minehisa (Editor of the Asahi Shimbun)
  25. ^ "500 people overworked middle-aged and older letters "public opinion poll" was black labor". Daily Gendai (February 2014, 12). As of February 7, 2014オ リ ジ ナ ルMore archives.2020/6/21Browse.
  26. ^ a b "The deep-seated lesion that created the "public opinion poll fabrication" of Fuji Sankei". Toyo Keizai Shimpo (September 2020, 6). 2020/6/21Browse.
  27. ^ "Sankei/FNN joint opinion poll, outsourced employee is illegal". Sankei Shimbun (September 2020, 6). 2020/6/21Browse.
  28. ^ "Japan Telenet FNN Acknowledgment for illegal input of public opinion poll data "Results of betrayal of trust" "Corruption of some employees"". Sports Nippon (September 2020, 6). 2020/6/21Browse.
  29. ^ Takumi Sato, "Media Society-A Perspective to Read the Present", page 113 (Iwanami Shinsho, 2006)
  30. ^ Takumi Sato, "Media Society-A Perspective to Read the Present Time," pages 113-114 (Iwanami Shinsho, 2006)
  31. ^ 1936-2008 Gallup Poll and Voting Rate Results (However, there are cases where a candidate who lost the voting rate was elected) US Presidential Election News: Gallup
  32. ^ poll2018/3/14
  33. ^ "Biden's lead poll, Trump camp withdraws and apologizes to CNN”. CNN (July 2020, 6). 2020/6/13Browse.


Related item

外部 リンク


Back to Top