[lbo-talk] backlash?

[Wojciech Sokolowski]

> Jon Johanning wrote:
> 
> >I'm not an expert in the science (?) of polling, but I think I've
> >seen several stories recently noting that pollers are getting
> >increasingly frustrated trying to find valid samples, given that
> >they get a lot of answering machines when they try to call folks, or
> >when they do reach them, folks refuse to participate, etc.
> 
> Yup, the refusal rate is up, people are switching from land lines to
> cell phones (which aren't as easily accessible), caller ID means
> people aren't answering the phone, etc. Accuracy doesn't seem to have
> suffered all that much, though.
> 

Another problem is random selection of the sample.  Multiple phones
means greater probability of being selected, which cannot be compensated
by weights if the number of multiple phones is not known.  
To illustrate: if we are select a sample of 10 from a listing of 100
phones, and everyone on that list has only one phone, then the
probability of being selected is 1/100.  If however, someone has two
phones, his probability of selection is 2/100; 3 phones render 3/100
etc.

Thus people with multiple phone lines are likely to be over represented
in samples which may lead to biased results.

Wojtek