At 04:20 PM 12/3/02 -0600, Olsen Chris wrote:
Dennis and All --i am sorry i even mentioned this issue of the having access to the unlisted ... being somewhat (IMHO) a breach of ethics on the part of gallup ... let's forget that one, i am willing
If my random digit dialing generates numbers with equal probability, and assuming both L's and UL's would answer the phones with equal likelihood, would I not get "answers" from Listed and NonListed in the same relative proportion as they are in the population? Then, if they aren't listed they would be UNlisted and have a phone.
i think jill's original issue is FAR more important than mine ... and, her expansion on what is the target population in this case being "likely voters" is the one we should really pursue
in my stat classes when we talk (and only for a short time unfortunately) about sampling ... i stress that the CONCEPT of a population is rather easy to grasp hold of BUT, the implementation of a sampling PLAN to accomplish a SRS (for example) from that population is a different matter entirely
i say to them ... let's say you want to estimate the average height of all penn state students ... of which, we happen to have about 82,000 system wide? we all can see in our heads what the population is ... that is no problem
what IS the problem is IF we are to implement a sampling plan that would result in a SRS ... from THAT population ... in practice, how are we going to put all names (we can't even know the full list of 82,000) in a big hat and take out say ... an SRS of 200? it is impossible
so, when we come to an even more nebulous concept of "likely voter" ... our problems are 10 times worse
to define a "likely voter" we have to figure out what "likely" means ... what does it mean here? that .... given 10 opportunities to vote, he/she votes at least 5 times? 7 times? ???
if we CAN define what we mean by a likely voter ... then, what sampling plan will get us an SRS from it?
what we know from the gallup url given earlier is that 1221 persons were interviewed BY phone ... so, right off the bat ... we are only talking to people by phone ...
we also know that you don't take a sample of 1221 and then call and find everyone home ... and EVEN if you did ... have them volunteer to talk to you ... so, how many initial phone #s did it take ... to be able to FIRST ... get an answer to the call ... BY a person of voting age ... who is a "likely voter" ... and who is willing to even discuss this with you .. that equals 1221???
what it seems that we have in this case is:
a sample of "selected" phone numbers who either in a once or twice attempt ... we found at home AND, if we were able to find someone home that is also of voting age AND be willing to talk to gallup about their potential voting actions, THAT is our sample ...
so, my question is: is THAT sample to be considered a SRS from "likely voters"?
i really really doubt it
all the formulas for finding the %chance error +/- that are bandied about in typical polls (for even just ONE question) are based on the SRS assumption of YOUR ONE SAMPLE ...
if we have concerns about the sampling plan ... being SRS from the defined TARGET population ... then, we are not in any good position to "know" the accuracy of the results
good sampling is very very tough ...
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
