Slashdot Mirror

← Back to Stories (view on slashdot.org)

New Google Tool To Find Trend Correlations

Posted by samzenpus on from the not-causation-tag-in-3-2-1 dept.

Kilrah_il writes "In 2008 Google found correlation between seasonal flu activity and certain search term, a finding that allowed it to track flu activity better and more rapidly than previous methods. Now, Google is offering a new tool, Google Correlate, that allows researched to do the same for other trends. 'Using Correlate, you can upload your own data series and see a list of search terms whose popularity best corresponds with that real world trend.' Of course, Google reminds us that correlation does not imply causation."

3 of 76 comments (clear)

  1. Weird science by cultiv8 · · Score: 3, Insightful
    From the Google Correlate Whitepaper:

    Trends in online web search query data have been shown useful in providing models of real world phenomena. However, many of these results rely on the careful choice of queries that prior knowledge suggests should correspond with the phenomenon.

    Yes, that is how science is done; hypothesis, predict, test, evaluate.

    Here, we present an online, automated method for query selection that does not require such prior knowledge. Instead, given a temporal or spatial pattern of interest, we determine which queries best mimic the data. These search queries can then serve to build an estimate of the true value of the phenomenon.

    So we have a backwards type of science: Evaluate, test, predict, hypothesis. Cuz hey, if there's a correlation, there must be a relation, and if there's a relation, we can build an estimate of the value of the relation, right? The marketing manager is gonna LOVE this....

    --
    sysadmins and parents of newborns get the same amount of sleep.
    1. Re:Weird science by Hatta · · Score: 4, Insightful

      Absolutely correct, this is going to swamp us in false positives. Remember, in order for science to work the way it's supposed to we have to report the negative results as well as the positive results. If 20 groups do the same experiment and only one gets a result significant at p=.05, that "positive" result doesn't mean anything. p=.05 means there's a one in 20 chance of the correlation being random.

      It's the same thing here. If Google goes out looking for positive results, and ignores all the negative results this is going to be so skewed as to be worthless.

      --
      Give me Classic Slashdot or give me death!
    2. Re:Weird science by i+kan+reed · · Score: 4, Insightful

      Yes, that's true, except there's a step before hypothesis. Observe. You're not allowed to use data from your observations that generated the hypothesis to support it, but you are allowed to use data to build a hypothesis in the first place.

      As their comic points out, property values correlate to liposuction searches. That's an interesting fact that you might make a socioeconic hypothesis based on. You could then turn to other avenues of research to validate your hypothesis.

      Not everything in science is a race to conclusions.