At 11:58 in the morning on September 1, 1923, a great earthquake struck the Kanto region of Japan, and the fires that followed burned large parts of Tokyo and Yokohama. Within hours the telegraph and telephone plant that the government relied on for reports was largely gone. Information still moved, by naval wireless, by ship radio and by people on foot. Rumors moved faster than any of those channels, and some of them fed the killing of Korean residents. This piece examines what happened when the official channels disappeared and what a modern agency should take from it.

September 1, 1923: the shaking, the fires and a government between cabinets

The earthquake struck just before noon local time. Its magnitude is usually given as about 7.9, which is the figure the Japan Meteorological Agency uses, and some later estimates run somewhat higher. The rupture lay beneath Sagami Bay and the land around it, southwest of Tokyo, so Yokohama, which sat closer to the source, was shaken severely, while Tokyo took heavy damage that was worst in its low-lying eastern districts. The shaking came close to midday, when cooking fires were burning in homes and restaurants across both cities, so fires broke out in many places at once. Strong winds then drove them through districts of closely packed wooden buildings. By most accounts the fires in Tokyo were not fully out until September 3.

The national government was poorly placed to respond. Prime Minister Kato Tomosaburo had died in late August, and Yamamoto Gonnohyoe was still forming a new cabinet when the earthquake hit. The ministers responsible for the first day were therefore a caretaker government, and a newly installed one took over on the second. On September 2 the government placed Tokyo under martial law by emergency imperial ordinance and extended it to neighboring prefectures over the following days, which gave the Army a large share of the security response. Officials made those decisions with very little reliable information about the scale of the damage, because the systems that normally carried reports to the ministries had been destroyed along with the districts they served.

Japan had no radio broadcasting in 1923, and regular broadcasting did not begin until 1925. The public got its news from newspapers, posted notices and other people. Most of the major Tokyo newspapers lost their offices or printing plants to the shaking and the fire. For the first days, ordinary residents had almost no source of information that carried official weight, and the officials themselves were not much better placed than the people they were trying to govern.

The death toll: two figures and who publishes each

Two figures circulate for the dead and missing, and both come from serious sources. The older figure, roughly 142,000, comes from counts compiled in the years after the disaster, and it appeared in standard references for most of the twentieth century. The newer figure, roughly 105,000, comes from a re-examination of the original records published in the 2000s by the seismologist Takemura Masayuki and colleagues. They concluded that the earlier tallies had counted a substantial number of victims more than once, for example when the same person appeared as missing in one record and as dead in another. Japanese government material on the lessons of past disasters has since used the revised figure, and many recent English-language references give it as well. Older books and many popular retellings still use the higher number.

I give both figures because the disagreement is real and neither should be presented as a tidy settled number. What a reader most needs to know is the direction of the revision: the more careful count is lower, and the gap of more than 30,000 reflects how hard it was for officials working in a destroyed region, among scattered families, to keep consistent records. Most analyses agree that fire killed the large majority of the dead, far more than building collapse did. The single worst site was the open ground of a former Army clothing depot in Honjo, in eastern Tokyo. People had gathered there with their belongings to escape the fires, and a fire whirlwind swept through the crowd. The figure usually given for deaths at that site is about 38,000, and the accounts vary.

For a communications reader, the count is itself a lesson about records. Every disaster figure is somebody’s count at some particular time, built from reports that traveled through the same damaged channels as everything else, and the Kanto figures shifted by tens of thousands over eighty years of re-examination. When I write a situation report I attach a source and a time to every figure. The Kanto toll is the clearest historical case I know for why that habit matters to whoever reads the report later.

Attribute every figure, especially the early ones

The Kanto death toll moved from roughly 142,000 to roughly 105,000 after scholars went back to the original records, which shows how much a careful count can differ from an early one. In your own situation reports and briefings, write every casualty, damage or shelter figure with the name of the source and the time it was given, so that a number from the first hour can be recognized as a first-hour number when someone repeats it on the third day.

The wire plant: telegraph and telephone inside the burn zone

In 1923 Japan’s telegraph and telephone were state services. The Ministry of Communications ran them, along with the post. Telephone service in Tokyo and Yokohama went through manual exchanges, where operators at switchboards connected each call. The long-distance telegraph circuits that linked the capital to the prefectures, and to the international cable and radio stations, ran out of central offices in the city. That design put the critical equipment in a small number of buildings in dense urban districts. It worked efficiently day to day, but it became a single point of failure once a fire burned across those districts.

The earthquake damaged the system in two stages. First, the shaking brought down poles, broke overhead lines and damaged the buildings that housed the equipment. Then fire reached exchanges, telegraph offices and ministry buildings in the burned districts and destroyed equipment the shaking had spared. Contemporary and later accounts agree that wired communication into and out of Tokyo and Yokohama was effectively lost for the first days. I am not giving a building-by-building list of which offices burned and when, because I have not verified those details to the standard this piece needs. Basic service took weeks to restore and the full network took much longer, and rebuilding the telephone system became part of the larger reconstruction of the capital.

The practical result was that the people responsible for the response could not hear from the people and places they were responsible for. Prefectures outside the damage area could not get accurate reports from the capital, and ministries in Tokyo could not reach their own field offices. Yokohama, the second city of the stricken region, was cut off from Tokyo by wire for a time, and broken roads and rail lines added to the isolation. The information that did reach decision-makers arrived through whatever channel happened to survive, and it contained only what the person at the far end of that channel had seen or been told.

Naval wireless and ship radio: the channels that still worked

The Imperial Japanese Navy operated a wireless transmitting station at Funabashi, in Chiba Prefecture east of Tokyo, outside the area that burned. With the wire plant down, that station became one of the government’s main links to the prefectures, to naval bases and to ships during the first days, carrying official traffic in Morse code. The Navy was well placed to fill the gap. It already ran long-range radio for its own purposes, and it could move ships, people and messages along the coast while roads and railways were broken. Navy ships came into Tokyo Bay with relief supplies and personnel, and ships fitted with their own wireless served as communications points along a waterfront where nothing else was working.

Ships in Yokohama harbor carried wireless as well, including merchant ships on trans-Pacific routes. Each had a working set, its own power plant and operators aboard. Accounts of how news of the disaster first reached the outside world generally credit messages from ships at Yokohama and the Ministry of Communications station at Tomioka in Fukushima Prefecture, known as the Iwaki station, which handled trans-Pacific radio traffic. The retellings differ in their details, including the times, the ships named and the parts played by individual operators, so I am not repeating any one version as settled. The broad outline is clear, though: news of the disaster traveled abroad by radio during the same days when residents of Tokyo had little besides word of mouth.

Wireless in 1923 had sharp limits, and those limits shaped what it could do. A Morse circuit carried text only as fast as a skilled operator could send, and a handful of stations carried the traffic of an entire government. Every message had to be written out, handed over, keyed by the operator and copied by hand at the far end. Under those conditions a station could send only what someone chose to hand it, and the officials who wrote the messages decided what to send and how confidently to state it. The fact that the channel worked therefore told the receiving end nothing about whether the content was accurate.

A working channel does not verify its traffic

The operators at Funabashi sent what officials gave them. A message that arrives over the one path still working gains authority simply because it arrived, and the same thing happens today when a satellite phone, an HF net or an amateur relay becomes the only link to a cut-off area. Require every relayed message to carry its original source, the time, and a plain statement of whether the content is confirmed or unconfirmed, and require relaying stations to pass all three along unchanged.

Runners, refugees and posted notices: moving information without wire

Inside the burned area, for the first days, information moved at walking speed. Officials sent messengers on foot and by bicycle, and by car or truck where the streets were clear of debris and refugees. Once martial law gave the Army a security role, police and soldiers carried reports and orders between posts. I have not found a detailed record of courier routes or message volumes that I would trust enough to cite. The general picture in the histories is one of improvised messengers rather than an organized relay system with logs and receipts.

Posted notices became the main way to reach the public. Authorities put up handwritten and printed notices at gathering points and refuges, and the newspapers that could produce anything at all put out small editions and posted bulletins. Survivor accounts commonly describe people writing notes about missing relatives on walls, boards and scraps of paper where refugees gathered, which served as an informal locator system for families. A notice reached only the people who walked past it, and it said whatever its writer believed, which was sometimes accurate and sometimes not.

The largest flow of information out of the region traveled with the refugees themselves. People left the cities on foot, by ship and on whatever trains could run. They carried their own accounts into towns and prefectures that had no other source, along with whatever they had heard on the way. A courier or a refugee passes a message along without the distortion a damaged line introduces. However, each one carries only one person’s knowledge and that person’s version of what others said, so a story passed through several such carriers loses any record of where it began. That loss of provenance is the condition in which rumor travels most easily, because nobody along the chain can tell who first said a thing or on what basis.

Rumors in the gap: what spread and how it traveled

Within about a day of the earthquake, false rumors were spreading through Tokyo, Yokohama and the surrounding prefectures that Koreans were setting fires, poisoning wells and organizing attacks. Those reports were not true. The record does not support organized Korean arson or well-poisoning, and within the first week the authorities themselves were issuing notices that the reports were largely unfounded and warning residents against violence. Other rumors circulated too, including warnings of a larger earthquake or a tsunami still to come, but the rumors about Koreans are the ones that led to killing.

These rumors spread in conditions shaped by colonial rule. Japan had annexed Korea in 1910 and had suppressed the March First independence movement of 1919, and many Koreans were working in the Kanto region, largely as laborers. Fear and prejudice toward them already existed before the earthquake. The disaster added fires burning everywhere with no visible explanation, frightened people crowded into refuges, and no authoritative voice saying what had actually happened. In that setting a false explanation that named a group of people spread through a population with no means of checking it.

The rumors moved by word of mouth, with the refugees and through the posted notices, and in some places police officers and local officials repeated them. They also moved through the official channel that still worked. In the first days, a Home Ministry message to prefectural governors went out through the Funabashi naval station. Historians of the massacre cite it from Japanese government records as describing Koreans as committing arson and other crimes, and as directing strict control of them. I am summarizing the content of that message rather than quoting it. Its effect was that the national ministry responsible for the police, speaking over the one channel that still worked, told prefectures far from the fires that the rumor was established fact.

The correction came later and traveled more slowly. Once authorities moved to counter the rumors, they used the same notices, couriers and refugee flows that had carried the rumors in the first place, and police and soldiers took many Koreans into custody, which protected them in some places and failed to in others. A correction issued on the fourth or fifth day had to overtake a story that had been circulating, with official backing, since the first or second, and it arrived after many killings had already taken place.

The killings of Koreans: what is established and what is disputed

Some facts are well established. In the days after the earthquake, neighborhood vigilante groups known as jikeidan set up checkpoints in Tokyo, Yokohama and the surrounding prefectures and killed Korean residents. Survivor testimony and later court records describe police and soldiers taking part in some of the killings. Survivor accounts describe vigilantes making passers-by pronounce Japanese words that were hard for native Korean speakers to say, and Japanese people with regional accents were among those mistaken for Koreans and attacked. Chinese workers were also killed. Some vigilantes were tried in the months that followed, and accounts of those trials generally describe the sentences as light.

The security response under martial law reached beyond the rumors as well. Military police killed the anarchist Osugi Sakae, his partner the writer Ito Noe, and Osugi’s six-year-old nephew, and Captain Amakasu Masahiko was court-martialed for those deaths. Labor activists were killed while in custody at the Kameido police station. These cases are documented, and they belong in any account of the period because they show the climate in which security forces were operating while the official information channels were carrying rumor.

The number of Korean dead is disputed, and I am giving the range together with who stands behind each end of it. Japanese government counts compiled at the time put the figure in the low hundreds. The Korean Provisional Government in Shanghai published a figure of about 6,600 in its newspaper, the Dongnip Sinmun, compiled from an investigation that relied on Korean survivors and students. Other estimates made by Japanese and Korean writers, then and since, fall between those two. The gap reflects who did the counting, what access each counter had, and what each had reason to record, as well as the ordinary difficulty of counting the dead in a burned region.

Responsibility is also disputed, specifically how far officials originated the rumors, how far they passed them on, and how far they simply failed to stop the violence. The question remains politically contested in Japan, and statements by Japanese officials in recent years about what government records can establish have themselves drawn criticism. This is not my field. Readers who want the full history should go to the scholarly literature on the massacre rather than to a communications article, and my concern here is narrower: the mechanism by which an official channel gave weight to a false report.

Treat a rumor that names a group as a hazard in its own right

At Kanto, a false report that named a group of people moved from rumor to official message to killing within a few days. If an unconfirmed report blaming a group of people for an incident reaches your center or your EOC, label it unconfirmed everywhere it appears, assign someone to confirm or refute it quickly, and if it is false, correct it publicly through every channel that community actually uses, in the languages its members read.

When the official channel carries the rumor: lessons for a modern EOC

Kanto is usually told as a story of infrastructure failure, and the wire plant did fail. The communications failure with the worst consequences, though, happened on a channel that was working. A national ministry sent a message over the Navy’s wireless that gave official weight to a rumor and delivered it to prefectures with no independent way to check it, and those prefectures passed it down to their police and their towns. Every link in that chain performed as designed, which is why the failure is so instructive: the system carried a false statement faithfully, and nothing in it was built to ask whether the statement was true.

The same mechanism operates in a modern EOC, only faster. When a caller reports something, the call-taker enters it as reported, and a line in the situation report later summarizes that entry before the report goes to the state, where someone may include it in a briefing. By the third hop the qualifier has dropped away and the report reads as a finding. Incident management software makes copying and forwarding easier than it was in 1923, and so it makes this drift easier too. The defense is procedural, and it means a status field at the point of entry, carried with every relay and copy, and a named person with the authority to change that status once a report is confirmed or refuted.

The other lesson concerns silence. For days, the residents of Tokyo and Yokohama had no authoritative voice telling them what was known, and rumor filled that space. A public information function that issues early, scheduled updates, even updates that say only what is known and what is not yet known, gives people a source they can expect to hear from again. Silence gives them no reason to wait for one. In practice that means the PIO or the joint information center sets an update schedule in the first hours, keeps to it, and publishes in the languages the community actually reads, with particular attention to communities that a rumor could target or leave out.

The Kanto authorities did correct the rumors, and the correction reached people only after the violence was well under way, because it had to travel through the same broken channels and against a story the ministry itself had already endorsed. A correction works best when people already trust the agency issuing it and when it goes out early over the channels the affected community relies on. Those conditions are set before the incident by planning, staffing and relationships with community organizations, and none of them can be arranged after a rumor has started.

What to do at your agency

  • The public information officer should add a one-paragraph rumor-control procedure to the agency’s existing public information or joint information center annex, naming who logs rumors, who confirms or refutes them and which channel carries the correction, and finished means the paragraph is in the annex and each named role has read and initialed it.
  • The communications center manager should check whether the center’s message form and CAD narrative template include fields for the source of a report and the time it was received, and finished means either both fields exist or a written change request to add them is with the person who maintains the template.
  • The emergency management director should write one false rumor, tied to a specific location, into the inject list for the next scheduled exercise and have the evaluators track how far it travels, and finished means the after-action report states how many positions repeated it and whether any of them repeated it as confirmed.
  • The COML or radio system administrator should list the outward paths that still work if local telephone, cellular and internet service fail together, such as satellite phones, HF amateur stations and state systems, and finished means a one-page list naming the receiving contact at the state EOC for each path is filed with the communications plan.
  • The dispatch supervisor should add a sentence to the existing radio procedures SOP requiring that unverified information passed on the air be stated with its source and prefaced with the word “unconfirmed,” and finished means the sentence is in the SOP and has been briefed on every shift.
  • The public information officer should ask the county’s language access coordinator or interpreter contact which languages a public alert or rumor correction would need in your jurisdiction and how long an after-hours translation takes, and finished means the languages, the contact and the turnaround time are written into the public information annex.

Takeaways

  • The 1923 Great Kanto earthquake and its fires destroyed most wired telegraph and telephone service in Tokyo and Yokohama, because the critical equipment was concentrated in central offices inside the districts that burned.
  • The death and missing toll is given as roughly 142,000 in older counts and roughly 105,000 in a later re-analysis by Takemura Masayuki and colleagues, and the disagreement is real enough that both figures should be presented with their sources.
  • The Navy’s wireless station at Funabashi and the radios aboard ships became the main links out of the region, and the channel’s reliability said nothing about the accuracy of the messages officials handed it.
  • Couriers, refugees and posted notices carried information without equipment distortion, but they lost track of where a story began, which is the condition in which rumor travels most easily.
  • False rumors that Koreans were setting fires and poisoning wells spread within about a day, and a Home Ministry message sent over naval wireless gave them official weight in prefectures far from the fires.
  • Japanese government counts at the time put the Korean dead in the low hundreds, while the Korean Provisional Government’s newspaper published about 6,600, and the number and the degree of official responsibility both remain disputed.
  • A modern EOC can repeat the Kanto failure whenever an unconfirmed report loses its qualifier as it is copied from log to situation report to briefing, so status labels have to travel with every relay.
  • Early scheduled public updates, in the languages the community reads, and prompt public correction of any rumor that names a group of people are the parts of the Kanto lesson an agency can actually plan and staff before the next incident.
Questions or a different view?

Reach me through the contact page. I read every message.