Right, and a million monkeys banging away at typewriters could replicate all of Shakespeare's works over the next 10 years.
But come on: a 0.949 correlation between "ibuprofen" searches and "gateway bible" searches over the past 7 years? Sorry, either google's data is fubar'd or something else has yet to be explained. --Doug On Mon, Oct 3, 2011 at 4:24 PM, Greg Sonnenfeld <[email protected]> wrote: > Given an extremely large data set, wherein the long term trend of the > data has a fair correlation (e.g. people may search more in one month > than another ) it seems very likely that you'll get an overlap with > something non-correlated, simply because you're drawing from such a > large set and there are only so many possibilities for the high order > components of the signal. > > I'm betting i could create a set of functions by throwing random > polynomials or sine series for a bound range 10^8 times (similar to > the entropy of a 2 word phrases drawn from a 10k word dictionary ) and > find excellent correlation between 2 of most any normalized signal. > > > **************************** > Greg Sonnenfeld > > > On Mon, Oct 3, 2011 at 10:19 AM, Roger Critchlow <[email protected]> wrote: > > And I fail to understand why searches for "advil" would map to searches > for > > "chai tea latte" and "pappardelle" with r>0.931, guess I need to read the > > Comic Book documentation. > > -- rec -- > > > > On Mon, Oct 3, 2011 at 9:55 AM, Douglas Roberts <[email protected]> > > wrote: > >> > >> Sorry about the delay; busy weekend. > >> Yes, google correlate maps search data, not actual real data, although > you > >> do have the option of loading your own time series and letting google > >> correlate that for you. > >> Still, I fail to understand why searches for "ibuprofen" would map to > >> searches for "gateway bible" with r=0.949. > >> --Doug > >> > >> On Fri, Sep 30, 2011 at 3:10 PM, Robert J. Cordingley > >> <[email protected]> wrote: > >>> > >>> I thought the example Doug sent was a correlation between the frequency > >>> of search terms, not with any actual real data. I always thought > >>> correlating time based data was difficult since as usage rises in > general > >>> one might expect search term frequency to rise together too. Real > >>> statisticians can chime in here. > >>> > >>> 'santa fe' was interesting in showing an annual cycle presumably > >>> corresponding to people's interest in finding info on a tourist > >>> destination. > >>> > >>> Drawing was interesting in that it seems one could always find some > >>> search term frequencies to correlate with any shape curve. > >>> > >>> Conclusion: it was fun. > >>> > >>> Thanks > >>> Robert C > >>> > >>> On 9/30/11 2:36 PM, Douglas Roberts wrote: > >>> > >>> That's the one I was using, Tom. > >>> --Doug > >>> > >>> On Fri, Sep 30, 2011 at 2:34 PM, Tom Johnson <[email protected]> > wrote: > >>>> > >>>> Is this any help? https://www.google.com/trends/correlate/ > >>>> > >>>> -tj > >>>> > >>>> > >>>> > >>>> On Fri, Sep 30, 2011 at 9:53 AM, Douglas Roberts < > [email protected]> > >>>> wrote: > >>>>> > >>>>> Has anybody been able to get anything useful out of that thing? Most > >>>>> of the items I've searched for return totally bizarre results. For > example, > >>>>> searching on ibuprofen with the intent to see any correlations with > >>>>> influenza give this as one of the higher-correlated results: > >>>>> > >>>>> > http://www.google.com/trends/correlate/search?e=ibuprofen&e=gateway+bible&t=weekly > >>>>> WTF? Does this show that on-line religion causes headaches? > >>>>> Seriously, has anybody found this tool to be of any practical use? > >>>>> --Doug > >>>>> > >>>>> -- >
============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org
