[OT] Google Trends

[Don Russell]

Craig White wrote:
> On Sat, 2006-05-13 at 07:36 -0700, Don Russell wrote:
>   
>> Aaron Konstam wrote:
>>     
>>> On Fri, 2006-05-12 at 20:19 -0500, Les Mikesell wrote:
>>>   
>>>       
>>>> On Fri, 2006-05-12 at 19:42, Ed Greshko wrote:
>>>>     
>>>>         
>>>>> The folks at Red Hat may want to take a look at
>>>>> http://www.google.com/trends.
>>>>>
>>>>> Entering "red hat, suse, fedora, debian" produces results that may be of
>>>>> interest to them.
>>>>>
>>>>>       
>>>>>           
>>>> Or for more optimistic trends, try ubuntu or centos...
>>>>
>>>> -- 
>>>>   Les Mikesell
>>>>     lesmikesell at gmail.com
>>>>
>>>>     
>>>>         
>>> However these would be more usefull if there were values on the vertical
>>> axis. Or have I been reading "How to lie with statistics" for too long?
>>>   
>>>       
>> The charts compare two or more items (in this case 4).... the value or 
>> scale used doesn't matter, as long as it's the same scale and values for 
>> each item tracked.
>>
>> I found it interesting to see search requests for redhat steadily 
>> declining, while searches for the other three distros are very close... 
>> and almost flat overall.
>>
>> As for raw numbers in this case... I'm not interested... it's trend 
>> analysis.... how does one compare to another over time... maybe this is 
>> telling us to "short sell redhat stock". ;-)
>>     
> ----
> it would be impossible to infer anything of significance beyond the
> trend (google searches for red hat, et. al.) and it takes nothing else
> into account (were the searches for help in which case perhaps the help
> resources are improving to the point of people relying less upon google
> to find help), familiarity with the Red Hat brand/name (less people
> needing to find out who Red Hat is), etc.
>
> FWIW, Microsoft too is trending downward. It doesn't mean anything.
>
> Nice exercise in nothing though
>   

I agree.... but adding a scale to the vertical access doesn't solve 
those points.  And, is the sample size statistically significant?

And, most(?) important ... just because you can show a correlation 
between two things, doesn't mean there is a valid cause & effect