On 19 Jun 2002 09:26:55 -0700, [EMAIL PROTECTED] (Ryan, George W.) wrote: > Colleagues, > > A presentation on the use and prevalence of child safety seats and > restraints at a recent seminar contained an analysis based on a so-called > "convenience sample"; i.e., data which does not constitute a random sample > from the target population. (The speaker was upfront in acknowledging this > fact.) My question to the group concerns any possible usefulness of such > "convenience samples". In particular, can the point estimates (of > proportion of vehicles conforming to laws requiring proper use of safety > devices in automobiles) computed from such samples have any validity? And > what about the computation of lower and upper bounds on the point estimates?
I am a little confused about your status. Your tag identifies you as being from the CDC (CDC&P?) in Atlanta. Assuming that is the famous, national CDC.... Doesn't your organization have, somewhere, some of the most excellent statements about the use of samples, of one sort and another? - that is just my expectation. "In particular" - we validate convenience samples by collecting *several* of them with a variety of characteristics, (And: look within one sample from a variety of angles.) Especially, we try to vary across characteristics that are thought to matter. So if someone says, "It is the cheap, old cars that will not have safety devices," we could draw a new sample that way from the DMV. Finally, we monitor how hard it is to complete the sample: If the hard-to-find cases are different from the early ones, that warns us to worry more about the MISSING ones. > > Given the shortcomings of convenience samples, does one have to forego any > type of meaningful analysis? Or, can an analysis be conducted provided an > emphatic statement is included by the researcher about the shortcomings of > convenience samples? > It sounds like someone has been scaring you unduly about convenience samples. In practice, possible biases should not be ignored, but there are a lot of surveys where no one starts out with much concern. Or the concern would not change the conclusions much, anyway. If 90% of the autos are estimated to have particular shortcomings, or 90% without, the scope of your actions might be pretty well set, despite caveats about precision. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
