Partners
2.3K members online now
2.3K members online now
Community space to get to know one another, talk about marketing and advertising strategy
Guide Me
star_border
Reply

Inconsistent trends

Visitor ✭ ✭ ✭
# 3
Visitor ✭ ✭ ✭

Hi. I'm just completely confused about the data i get from google trends... i search for any keyword and use some date range containing 200 days (e.g.) since these datas are relative, i can easily get the relative increase/decrease per day (by dividing day(i+1) / day(i)) if i now split the daterange into half, so i take the first 100 days and afterwards the second 100 days, i get more or less completely different results. The reason for me doing this in the first place is to get daily data for more than 269 days (which is the maximum amount of days in a daterange allowed for a daily resolution) simply by splitting my daterange (with more then 269 days) into half. The difference after splitting is not just like a change from +10% to +3% (from day i to i+1) - it is also sometimes a change in the sign - from +3% to -1% e.g. Is this me completely being stupid or the data being inconsistent?

1 Expert replyverified_user

Inconsistencies

Visitor ✭ ✭ ✭
# 1
Visitor ✭ ✭ ✭
Hi.
I'm just completely confused about the data i get from google trends...
i search for any keyword and use some date range containing 200 days (e.g.)
since these datas are relative, i can easily get the relative increase/decrease per day (by dividing day(i+1) / day(i))
if i now split the daterange into half, so i take the first 100 days and afterwards the second 100 days, i get more or less completely different results.
The reason for me doing this in the first place is to get daily data for more than 269 days (which is the maximum amount of days in a daterange allowed for a daily resolution) simply by splitting my daterange (with more then 269 days) into half.

The difference after splitting is not just like a change from +10% to +3% (from day i to i+1) - it is also sometimes a change in the sign - from +3% to -1% e.g.

Is this me completely being stupid or the data being inconsistent?

Inconsistencies

Visitor ✭ ✭ ✭
# 2
Visitor ✭ ✭ ✭

Wrong forum or no reasonable explanation?

Re: Inconsistent trends

[ Edited ]
Top Contributor
# 4
Top Contributor

Hi,

 

The data is being scaled by the system and then averaged by your method. So there will always be variations with your method since different weeks and months bring different traffic patterns within any topic. Google Trends is tracking data hourly, not just daily, so stretching the data beyond the data range will never be totally on point.  

 

Trending Searches highlights queries that jumped significantly in traffic among all searches over the past 24 hours, whereas Top Charts highlights topics with high overall search volume for that month.

And...

 Search results are proportionate to the time and location of a query:

  • Each data point is divided by the total searches of the geography and time range it represents, to compare relative popularity. Otherwise places with the most search volume would always be ranked highest.

  • The resulting numbers are then scaled on a range of 0 to 100 based on a topic’s proportion to all searches on all topics.

  • Different regions that show the same number of searches for a term will not always have the same total search volumes.

 

Although you are getting useful data with your Google Trends usage, I would say that you are maximizing it's abilities and anytime you take something beyond it's limits... there will be varied fluctuation.

 

This is the help file on Google Trends, which may be helpful in other ways, if you are not already using it. 

 

It's great to hear from someone putting this useful tool to work.   

 

Please do let us know what else you discover.

 

Kind Regards,

 

James

_________________________________________________________
Google AdWords Top Contributor | Google Partner | GYBO | Local Guide | My Profile


 


 


 

Re: Inconsistent trends

[ Edited ]
Visitor ✭ ✭ ✭
# 5
Visitor ✭ ✭ ✭

Hi James_Clemens,

thanks for your reply.

The reason for me to stretch daily data beyond the range is to train a neural network with google trends (amongst others) as input.

It's just the desired data in my case. Let's forget about regions for now, since i'm not altering this variable - and take a look at a simple example, since my confusion is not gone yet.

 

So let's say we got 4 days, a trend like "automobile" and on those 4 days we have a search volume of 5,8,4,10 (total searches).

If i search at this daterange

(and if i understand the math behind the help-pages correctly - which i obviously don't because it's giving weird results haha)

i will get the following data (maximum is set to 100, the rest is relative to that):

50, 80, 40, 100.

If i now wanna get the relative change to the following day, i get:

+60%, -50%, + 250%

Now i split the daterange into 2 + 2 days, since the maximum will be allways set to 100, i will get:

62.5, 100 and 40, 100 with relative changes of:

+60%, -60%, +250%.

 

No matter how many days i take and split, the only change (mathematically) should be the knot, where i split the range.

But it should

a) not differ too much (from +3% to +10% is very unlikely)

b) not change the sign (except for the knot)

c) especially not differ on every single data point (since i only split the range once, there is only 1 possible change in the relative change)

 

And i experienced all 3 of those at all random spots (in fact only ~20% of the relative change stays the same after splitting).

So i was thinking, that google trends is not using a proportional kind of "relative data" - maybe they are setting the maximum to 100 and the rest is just parsed to some strictly monotone function (like sqrt(x/max)) or log2(x/max +1) etc etc),

but no matter what function you take, it is not possible (mathematically) to generate a change in the sign of the relative change from day i to day i+1.

 

Maybe it's too much to ask for the math behind the given data by google trends, but i would love to have at least a general idea.

Cheers

Inconsistent trends

Top Contributor
# 6
Top Contributor

Hi @s1nnl0s,

 

You are very welcome. 

 

From my usage and yours... I would agree that Google Trends is not using a proportional kind of "relative data" and likely a custom blended formula that we would not be able to replicate. 

 

However, I am trying to locate more info on this so give me a day or so to do so. If you discover something more, please do share it here. I think Google Trends is a fun & powerful product and you are doing some great things with it. 

 

Kind Regards,

 

James

_________________________________________________________
Google AdWords Top Contributor | Google Partner | GYBO | Local Guide | My Profile


 


 


 

Weekly data of Google Trends since 2004

Visitor ✭ ✭ ✭
# 7
Visitor ✭ ✭ ✭

I've exactly the same issue and I see no solution. Overlapping does not work because the proportions between 2 different time series for the same weeks are not constant. Can someone from Google help us out?!?!!?

Inconsistent trends

[ Edited ]
Visitor ✭ ✭ ✭
# 8
Visitor ✭ ✭ ✭

Hi James. There is an issue with Google Trends you might be the perfect person to confirm. We cannot download weekly data since 2004 until 2016 (there is a 5 year max period). Overlapping does not solve the issue because in 2 different time series (e.g. 2004-2008 and 2007-2010) the scaling factor is not constant across the same dates (e.g. the proportion of 04/01/2007 in both time series and the same in 11/01/2007). I suppose this goes in line with what you say

Search Terms and Excel

Visitor ✭ ✭ ✭
# 9
Visitor ✭ ✭ ✭

There is a new issue with Google trends no one seems to confirm. We cannot download weekly data since 2004 to 2016. Overlapping does not seem to solve the issue because the scaling factor is not constant. Any thoughts?

Inconsistent trends

Top Contributor
# 10
Top Contributor

Hi @João P,

 

This does go inline with what @s1nnl0s and myself were discussing. This tool works best to see "trends" not surgical level statistics. To see the trends, the data only needs to be so perfect. Even with the anomalies discovered, it should suffice for any research needs.

 

I am awaiting some info for @s1nnl0s and will apply back once I have something worthy to report.

 

Cheers,

 

James 

_________________________________________________________
Google AdWords Top Contributor | Google Partner | GYBO | Local Guide | My Profile