| |
 |
|
|
Science Forum Index » Space - Consult Forum » if no main effect, do pattern analysis anyway?
Page 1 of 1
|
| Author |
Message |
| doc3 |
 Posted: Wed Apr 30, 2008 3:59 pm |
|
|
|
Guest
|
I helped somebody design a large, survey-type experiment. The response
rate was quite poor, so I suggested ways of rerunning the experiment
to get a dependable response rate. The experimenter asked me instead
to look for patterns of response across gender, location, and other
variables supposedly randomized in the design. I didn't want to spend
my time fishing in the data, so I moved on to other projects.
What would you have done?
A related question is that when I do a contingency analysis of gender
vs location, for example, the rows and columns may be nonindependent
for lots of reasons, as you know. I calculate the residual for each
cell but I have to guess which relationship the residuals are pointing
to. Do you know of a paper that exemplifies this task well? I'm not
expecting a step-by-step procedure. |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Thu May 01, 2008 7:07 pm |
|
|
|
Guest
|
On Wed, 30 Apr 2008 18:59:25 -0700 (PDT), doc3 <nhirsh33@gmail.com>
wrote:
Quote: I helped somebody design a large, survey-type experiment. The response
rate was quite poor, so I suggested ways of rerunning the experiment
to get a dependable response rate. The experimenter asked me instead
to look for patterns of response across gender, location, and other
variables supposedly randomized in the design. I didn't want to spend
my time fishing in the data, so I moved on to other projects.
What would you have done?
I'm not sure what you mean by "dependable response rate."
A very large fraction of surveys have a low enough response
rate that they would be *vulnerable* to selection bias, if the
selection is related to the content of the survey. That is a
problem for design of the survey sample, and questions.
Were you objecting to the lack of power to say anything
about the subject of the survey?
I think that a crosstabulation of the "randomization factors"
will usually be in order. It might help to explain the low
response rate and give help for a new design.
Quote:
A related question is that when I do a contingency analysis of gender
vs location, for example, the rows and columns may be nonindependent
for lots of reasons, as you know. I calculate the residual for each
cell but I have to guess which relationship the residuals are pointing
to. Do you know of a paper that exemplifies this task well? I'm not
expecting a step-by-step procedure.
In a 2x2 table, each residual is exactly the same number.
Making sense depends on the content of margins, and the
smallest cell deserves special thought.
I can recommend Agresti's books as an intelligent introduction.
You might use Google Scholar to find other sources.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Thu Jul 24, 2008 4:42 am
|
|
