Have some more survey data. 4 dichotomous variables involving a deadly
force survey, so vars like resist arrest or not, felony or not, death or
not. And my haunting 7-point Likert scale of Reasonable to really
Unreasonable use of deadly force.
Now it gets worse. The person who designed the study tried to do a
factorial survey but instead of randomizing all the surveys has each
respondent answering the same type of survey for all that particular
respondent's surveys. So, police officer 1 will answers 3 questions each
one having the same mix, perhaps, resist arrest, felony, death resulting
and so on. Fortunately, all the cells have enough responses.
So it's not a randomized factorial survey according to the usual use
although the designer thought it was "random" because each police officer
does not know beforehand his or her particular "mix." I did not have the
heart to tell him that there is a difference between random and bias. And
I also did not know enough to be sure except it surely did not sound like
a proper, randomized factorial survey (where each and every scenario has
the variables scrambled up randomly in the descriptions.)
And of course it's opportunisticly sampled (is that a proper word here?)
and the police were not randomly selected.
So I compared a regression violating probably all known assumptions with
non-parametric tests, some using medians and not, and got about the same
rates of significance, which I found reassuring. Is that appropriate?
I guess I should apologize for such low-level questions, but I am too
tired trying to figure out how to do a real factorial survey and keep the
respondents from getting exasperated with all the scenarios. And I have
asked our local statistical gurus, but they are all tired and busy, so I
throw myself at the mercy of internet strangers.
Any pointers will be helpful; really, any. Even any beginner's texts you
might find helpful here.
Adam Sundor
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
