conspiracy theory, and deliberate deception spread across social media in the aftermath of Sunday’s Bondi shooting.
Some people are clinging to the idea that Google Trends data shows a spike in searches from Tel Aviv or other locations for “Naveed Akram”, the name of one attacker, before the shooting. This is interpreted as proving that Akram was an Israeli spy.
After the attack on US National Guard members and US right-wing activist Charlie Kirk in November, similar stories circulated.
What’s the story? Google informed ABC that Google Trends can sometimes show searches even when none have occurred due to “statistical turbulence”.
In my research I have extensively studied Google Trends and can confirm that this is true. The “noise”, especially when looking for searches of unusual terms or from small areas, can produce strange results.
What is Google Trends?
Google Trends provides information on what people are searching at different times and places. It uses what statisticians refer to as a “time-series”, but these data are unique in two ways.
You can easily choose different time scales such as minute by minute and year by year.
The data is only a sample of Google’s true volume. Time series usually contain all the data available (such as this statistics on hospitalisations).
The <a href="https://support.google.com/trends/answer/4365533?hl=en&ref_topic=6248052#:%7E:text=While%20only%20a%20sample%20of%20Google%20searches%20are%20used%20in%20Google%20Trends%2C%20this%20is%20sufficient%20because%20we%20handle%20billions%20of%20searches%20per%20day. The Google Trends Help page explains it as follows:
Google Trends uses only a small sample of Google searches, but this is enough because we process billions of Google searches every day.
Rare searches and statistical noise
However, I have found that searches for terms that aren’t widely searched (such “Naveed Akram”, prior to the shooting), or in smaller geographical areas (where there are fewer searchers) can produce a variety of results.
The variation is exacerbated by many of the social media posts that are misleading. They show the Trends results for a very small area (such as just the city of Tel Aviv). This high variation creates a pattern of near-zero or zero values, with isolated large spikes. The post below is a good example.
This is often due to “statistical noise”, which is small fluctuations in data that are smoothed when we examine a large number of events. This is evident when you compare searches with high volume.
Google Trends Results Change Over Time
A second misconception is that the data are based on time. Some posts have mentioned how the results displayed seem to change as you move from one page to another. Google Trends data is exactly what you can expect.
The reason for this is that Google only uses a small sample of data. To obtain accurate results, you must aggregate several samples from Google Trends.
This presents a challenge. Google continuously updates its results in real-time for short-term data, such as the type of data used in social media posts. Google adds only one sample per day for longer-term data (although we’ve developed methods to work around this).
What Google Trends numbers really mean
The third mistake is to think that the numbers on Google Trends charts represent the number of searches made for a particular term. The Google Trends Help explains, however, that the values are normalised to the location and time. They are then “scaled to a range from 0 to 100”.
The time in the series that has the most searches will be set at 100. All other points will then be scaled according to this. If the highest number of searches is ten, then it will show as 100. And if three searches were made at another point, that would be 30.
The number at each time point is a measure of the probability that the search terms specified would be found in the location and time indicated.
It is not true that a blog post about the search trends for Charlie Kirk’s alleged killer claimed there were “Less Than 1 in 1 BILLION Odds” of it occurring.
This is a highly probable pattern: If “Lance Twiggs”, the partner of “Tyler James Robinson”, had 40 searches and “Tyler James Robinson”, 30 searches (if scaled up to 100), then this would be exactly what you see (if scaled down to 75).
Common sense is a powerful tool
Some common sense will help, even if you don’t understand all the information on Google Trends. There are many people with the name Naveed, such as a Pakistani soccer player named Muhammad Naveed Acram.
It is not surprising that there were a few searches before the 14th of December for “Naveed Acram”. Google Trends will return any searches containing the query. “Naveed Akram”, for example, will also return “Muhammad Naveed Akram”.
Google Trends data is a great way to understand events that are happening in real-time. It has, for example, been used to predict the results of election as well as referendum with a margin error.
To do this properly and avoid perpetuating fiction, it is important to interpret and understand the data. Google Trends does not reveal anything about Naveed Akram and the Bondi terror attacks.
Jacques Raubenheimer has not disclosed any relevant affiliations other than their academic appointment. He does not consult for, own shares or receive funding from, or work for any company that would benefit from the article.
