This is a VERY preliminary unchecked draft report on a ground breaking study concluded by Electricquakes.org .

Human Time Tagged Twitter Comments on Pet Behavior and Other Terms as Possible Precursors of a California Earthquake - www.electricquakes.org May 19, 2009 Author: George F. Watson

Methodology and Background

With an estimated 7 million unique visitors as of February 2009, at the time of this study on May 19th, 2009, Twitter is the largest spontaneous, time tagged, trackable source of spontaneous public commentary that has ever existed on earth and therefore is ripe for the data mining of spontaneous human observations. On May 17th, 2009 Los Angeles experienced a 4.5 magnitude earthquake which was the first reasonably noticeable earthquake in this large population center since Twitter had become a widespread phenomenon. Thus the earthquake and the Twitter phenomenon offered a previously unobtainable sampling of unbiased, human time tagged comments in the scale of millions to test for recorded human perceptions prior to and after an earthquake. This study examined "Tweets" or mentions of strange pet behavior and human physical symptoms on the huge social network Twitter before a subsequent earthquake to attempt to determine if unbiased pet behavior reports or human physical symptoms provided any predictive indicators of the earthquake. In order to graph these mentions the Twitter graphing service at http://twist.flaptor.com was used. Neither Twist.flaptor nor Twitter are responsible for conclusions of this study. All trademarks are the property of their owners.

Citation information:
Researchers and news media citing this study should cite the title as "Human Time Tagged Twitter Comments on Pet Behavior and Other Terms as Possible Precursors of a California Earthquake - www.Electricquakes.org May 19, 2009 Author: George F. Watson".

Below you see the past (NOT LIVE) record of twitter results in the 30 days leading up to the May 17th,2009 Los Angeles 4.5 earthquake.

Below is a graph of mentions of "Earthquake" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. You can clearly see the spike in the mention of the word "earthquake" which occurs immediately AFTER the earthquake in Los Angeles on the 17th. On Sunday at 20:30 PM during that 20 minute sampling window Twist was measuring fully 5.5% of all tweets at that moment contained the word "earthquake".

Below is a graph of the number of tweets with some phrases that might indicate strange dog or cat behavior in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. There are no significant mentions of strange animal behavior before the quake although there are many AFTER the earthquake occurs which functions as a reasonable control data point for the search terms used

Below is a graph of mentions "cat" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. This provides a baseline during the period for the number of tweets about cats in general. It also serves another purpose. It is a broad search to see if there is an increase in owners noticing or Tweeting about their animals for whatever reason before the quake. It is possible owners might notice their animals more because of the animal's strange behavior yet owners might tweet about other things regarding their animal and not mention the strange behavior specifically. Thus this search may act as a more broad indication of the increase in noticability of the pet by the owner before the quake.

The numbers of tweets about "cat" does rise by a significant percentage in the day prior to the earthquake. However the total number of tweets on that day is peeking at Saturday at 4:30 PM at .36% of all tweets which is significantly above the estimated baseline of .21% for the previous and following periods. Unfortunately the total number of tweets being graphed by "Twist"is unknown therefore it is impossible to determine a statistical significance in the strictest sense. However other searches have produced exact numbers of tweets on the vertical axis by Twist when tweets were as high as 50 therefore it is reasonable to conclude that the number of tweets on the day May 6 th proceeding the quake is higher than 50 total thus raising the confidence in the significance of the increase. Furthermore a search of common terms to try to determine a correlation between percentages and actual numbers of Tweets resulted in one indication that a graph peak at .03% equalled 12 tweets at that time. Thus IF tweet traffic is relatively constant 1% of Twist measured tweet traffic may equal 4000 Tweets or messages from Twitter users. using an estimated total tweet traffic of 40,000 messages then .36% mentioning "cat" means 144 messages.

Of all phrase Tweets measured by this study, this Tweet increase in the mention of "cat" is the most significant of all phrase measurements given both the high percentage increase from the 30 day baseline and the timing which occurred exactly one day before the earthquake struck.

Below is a graph of mentions of "my cat" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. A slight narrowing of the above methodology to attempt to limit tweets to owners of animals. There is another large jump in tweets more than doubling the apparent baseline number of tweets however this occurs four days prior to the quake for this phrase. Additionally the jump is only slightly higher than a jump which occurred almost a full 30 days before the quake indicating perhaps that such jumps are not uncommon. Perhaps significantly though the jump four days prior to the quake is higher than immediately following the quake when one might expect many pet owners would be Tweeting about the reaction of "my cat" to the earthquake.

Below is a graph of mentions of "cat disappeared" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. The jump in the tweets about "cat disappeared is large immediately after the earthquake (from a baseline of 1 to 6 tweets) however a closer examination of some of the actual tweets shows a problem in methodology of simple numerical comparison of the number of tweets. It's clear from reading some of the tweets that a news story broke from The Associated Press on May 18, 2009 MARION, Ark. (AP) about a specific cat named "Bubba" which was trapped in a wall and this story about a single incident of a cat disappearing was "retweeted" by many Twitterers thus artificially raising the presumption that more cats disappeared.

For similar reasons "dog" as a search term is ineffective. This is due to the fact that the term "Dog" has enter slang possibly due to the huge social impact of the show "American Idol" and the judge Randy Jackson who uses the term constantly. As a result Twitter users often use the term in a way that is completely unrelated to pet dogs and even sometimes include the term "dog" in their Twitter name. Twist is unable to distinguish between those using the term dog as part of their Twitter name or using the term in a way related or unrelated to the common pet. As a result a single Twitter with the term dog in his name can radically change the Tweet count for "Dog" even when they are not Tweeting about dogs at all.

Below is a graph of mentions of "cat is acting strange" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. There is exactly one tweet measured Thursday at 4:00 which is indicted by Twist as .09 tweets . Clearly such a specific search of Tweets shows that the number of tweets is too low at this time for specific analysis of very such specific phrases in Twitter.

Below is a graph of mentions of "cat is acting weird" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake.

Below is a graph of mentions of "cat ran" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. Theoretically such a phrase should capture tweets including "my cat ran away" and "my cat ran and hid" etc. however it could also capture tweets with no significance such as "a cat ran to its food bowl".

Below is a graph of mentions of "cat ran away" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake.

Below is a graph of mentions of "dog disappeared" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake.

Below is a graph of mentions of "dog howling" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake.

Below is a graph of mentions of "dog ran" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake.

Below is a graph of mentions of "dog ran away" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. This is a more specific phrase search than above which means the prior search can be seen as a control of sorts.

Below is a graph of mentions of "I found my cat" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. If one presumes that animals may leave home before an earthquake in fear then perhaps owners might be more inclined to tweet about finding their animal if it returns once the earthquake is over. (significant note - as of the publication time of this study the aftershock sequence is still very active including a major 4.1 aftershock which has occurred literally today May 19, 2009 . Thus perhaps not enough time has elapsed for animals which ran away to have returned "when the danger has passed.")

Below is a graph of mentions of "I found my dog" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. A variation with t he same caveats as the cat found tweets above.

Below is a graph of mentions of "positive ions" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. Studies have indicated that positive ions in the air before an earthquake may be the factor which triggers animals to flee the area before an earthquake. This is an attempt to quantify any human research or measurements of positive ions that might be tweeted. As expected the tweet level is very low.

Below is a graph of mentions of "predicting an earthquake" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. In addition to pet tweets the study went further to see if their was a significant increase in the tweets of people either predicting or hearing and Tweeting about other people who were "predicting an earthquake" before the earthquake happened.

Below is a graph of mentions of "prediction earthquake" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. A variation of the search terms indicating possible earthquakes predictions before the earthquake

Below is a graph of mentions of "will be earthquake" in Twitter 30 days before the May 17th Los Angeles 4.5 earthquake. A variation of the search terms indicating possible earthquakes predictions before the earthquake.

Conclusions:

There seems to be only one indication from reports of unusual animal activity on Twitter that animal behavior mentions in Twitter offered a large precursor of the earthquake. That indicator is an large percentage increase of people Tweeting about cats in general rather than any specific behavior being described. As described above a possible explanation is that the cats were noticed more by people. The behaviors which made the people notice them were possibly enough to cause the persons to tweet about them but the behaviors were themselves not mentioned in a categorical way in the tweets. The numbers of tweets mentioning animal behavioral idiosyncrasies before the earthquake did not seem to indicate that pet owners were Tweeting about their pet's behavior enough to act as a reliable precursor or that the animals did not display any behaviors that indicated they were aware of an impending quake. Possible conclusions are that 1. pets were not aware of the impending quake. 2. Pet owners were not a tuned to their pets. 3. Pet owners didn't tweet about pet behavior in enough numbers to be statistically significant. This study is severely limited by some technical limitations of the Twist Twitter graphing service and several other factors.

The graphing service is limited by the single vertical scale when multiple terms are examined at once. One term mentioned with a lot of activity makes it impossible to see changes in activity of terms with much less activity.

Twist phrase search is apparently not all inclusive. Searching for "lost dog" may not pick up a tweet with "my dog is lost" or "I lost my dog" therefore the number of tweets or their significance could be limited by poor choice of search phrases graphed.

Twitter mentions are not geographically isolated. We can only monitor all mentions for Twitter users everywhere rather than those in California near the location of the earthquake. However with the low normal (baseline) mentions of animal activity and the large number of southern California Twitter users one would expect any spike from the California area to be noticeable in the totals.

If the number of tweets are too low Twist will not show a 30 day period.

The maximum length of time Twist will graph tweets is 30 days. It is possible animals might behave strangely more in advance of an earthquake than 30 days and if so those tweets would not be picked up by these graphs.

The number of tweets mentioning behavior are usually less than 10 or 20 in the 30 days graphed which is far below the number needed for a statistically significant group to measure variations.

If the same message is "retweeted" by several people it may skew the presumption of independent reports. For example one story of a cat disappearing which circulated AFTER the earthquake was "retweeted" by several users of Twitter thus increasing the number of tweets with the phrase "cat disappeared" substantially.

While few mentions of strange animal activity occurred before the earthquake the mentions of animal activity AFTER the quake could actually be related to the animals behavior BEFORE the quake. For example after a quake a pet owner might be more inclined to mention that their animal behaved strangely before the quake. Unfortunately such post quake mentions are more susceptible to the owner's selective recollection so are not as dependable as an unbiased report. Lots of people like to think their pets have unusual powers or senses.

The quake's size may have been below a necessary threshold for statistically significant activity. Anecdotal reports of animals sensing impending quakes often are associated with larger quakes. It's possible the size of this quake at a mere 4.5 or so in magnitude simply did not put off enough indication in either strength or in a wide enough area for a large number of animals to sense it which were then further limited by their owner's ability to notice and then take the time or have the access to tweet about it on twitter.

Despite those limitations there seems to be only one indication from reports of unusual animal activity on Twitter that animal behavior mentions in Twitter offered a large precursor of the earthquake. That indicator is an large percentage increase of people Tweeting about their cats in general rather than any specific behavior being described. As described above a possible explanation is that the cats were noticed more by people however the behaviors which made the people notice them were enough to cause the persons to tweet about them but the behaviors were themselves not mentioned in a categorical way in the tweets.

One conclusion is certain however and that is that Twitter and tweets provide the first large scale, time stamped records of human statements to be studied for correlations to events in time including earthquakes.