Re: Embedded Systems Survey

["Robert I. Eachus" <rieachus@earthlink.net>]

tmoran@bix.com wrote:

>   There's a nice trick for getting statistics on something people won't
> tell you.  Ask, for instance:  "Think of a number between 1 and 10.  If
> your number is less than 5, or you prefer Ada, say Yes.  If it's greater
> than 5 and you don't like Ada, say No."  Calibrate by asking another
> group just the number part and not the Ada part, then the difference in
> percentages tells you the percentage of people preferring Ada, without
> telling about any particular person.

  This proposed method has several serious flaws.  First, as stated,
anyone who answers "No" doesn't like Ada.  Second, having people think
of a number introduces a bias which you attempt to correct for by the
second survey.  But notice that clever readers of the survey will answer
"Yes" if they like Ada without guessing a number.  So there will be a
bias in the answers if those who don't like Ada tend to guess different
numbers than those who do.  If you really need to assure people that the
survey is anonymous, you have to use a method which people will trust. 
The easiest way I know of to do this is a (snail) mail survey.  Best I
know of is to bind a postcard (not a blow-in) into a magazine that you
are using to define the population of interest.  If the card is designed
to be folded and stapled, most people trust the survey to be anonymous.

   Of course, that still doesn't mean that their answers will be
honest.  On "Do you want to be a millionaire?" I am constantly amused by
the numbers in response to the "Ask the audience" lifelines.  Even if
two answers have already been eliminated, there will be votes for those
answers.