Thursday, May 26, 2005


Google Sets

There's a neat tool that the folks at Google have developed as a product of the "Google Labs". It's called Google Sets. It is a linguistic device that will automatically create larger sets of items from a short list of examples that you provide. For example, if you type in orange, banana, pear . . . it will spit back pear, banana, orange, apple, grapefruit, sweet potato, kiwi fruit, carrots, pineapple, grape, dates, cherries, figs, and strawberries.

You might say, "Nifty, but what's the real-world use of this gizmo?" Indeed, the Google Sets discussion board over at Google Labs is loaded with users complaining that they don't see the actual value of Google Sets, and their level of emotional frustration over this problem is actually kind of entertaining to me. It's a FREE APPLICATION, PEOPLE -- if you don't see a use for it, move on with your life and go outside and play kickball!

I, however, have found a very appropriate use for this tool, in the practice of market research survey design. Oftentimes, a questionnaire will contain a question that, while we don't know the specific answers we're going to get, we can surmise that 90%-95% of all the responses will fall into a set of maybe 5, 6, 7, or 8 predictable answers. The remaining 5%-10% can be captured with an "Other" response code (which may or may not include an interviewer instruction to have the respondent "specify" the response). Setting up a question this way has three main benefits:
  • By forgoing a fully open-ended question (free response), it reduces the interviewer's time spent typing in the text of the respondent's answer
  • It reduces the amount of back-end coding effort, because the vast majority of your response data has been captured immediately into 5 to 8 or so numeric punches
  • It allows the opportunity of basing skip patterns off the coded numeric responses, rather than losing that opportunity due to an open-end not having any programming "intelligence"

So, how might I use Google Sets? Suppose I am writing a questionnaire about overnight cruise lines that travelers may have boarded within the past 2 years. Since I've never been on a cruise, I'm probably not an expert at developing a pre-coded list of cruise lines for my survey -- after all, I can only think (off the top of my head) of Carnival, Holland America, and the Disney Cruise Line. This is hardly a complete list of possible cruise brands that our respondent travelers might mention. Enter Google Sets! By typing in the three brands I know, I get a beautiful automated set of the following brands in return:

  • Carnival
  • Disney Cruise Line
  • Holland America
  • Royal Caribbean
  • Celebrity
  • Norwegian
  • Cunard Line
  • Princess
  • Windjammer Barefoot

Now I have what seems to be a reliable list of 9 major cruise lines that I can be fairly sure will capture at least 90% of the response data for that question. I just need to alphabetize them (to help the interviewers), add my "Other-specify" stub, and we're ready to go on the phones!

Of course, since we're diligent and meticulous market researchers, we will also check our "Other-specifies" after the first night or two of interviewing. If any multiple responses are showing up en masse, we will add them to our pre-listed codes.

1 Comments:

At 8:31 AM, June 12, 2005, Anonymous Anonymous said...

As a former computer programmer,
I am always interested in knowing what logic is used in the search program. Even at blazing speeds, it is hard to imagine that the search program is doing a byte by byte compare on every article on the web - then returning a million hits in a few seconds. I can't fathom it. As someone once said,
it's a miracle.

 

Post a Comment

<< Home