Hello Folks, I just joined this list and I encountered this problem in a recent analysis of chemical wastewater data. Now I realize that my data are continuous and that the data in question here are not, but there is quite a bit of literature out there that may have connections to data that isn't continuous. One of the popular techniques is what is mentioned below, i.e. replacing the missing value with an average value. People also make assumptions about the underlying distribution of the data (lognormal in my case of pollution data) and simulate missing values with Monte arlo techniques. If this is true "censored" data where the missin value is reported as below detection limit, then sometimes using half the detection limit is appropriate. it usually depends on what the parameter to be estimated is. Again, these are for continuous data, but there may be references in the papers that also look into interval, ordinal or some combination of data. Two papers are Estimation of Distributional paramters for Censored trace level Water Quality Data, by Gilliom and Helsel in the journal of Water Resources Research, Vol 22, No 2, pages 135-146, Feb 1986 Estimation of Descriptive Statistics for Multiply Censored Water Quality Data by Helsel and Cohn, same journal, Vol 24, No 12, pages 1997-2004, Dec 1988 -- Shawn Philippon Lecturer, Faculty of Forestry, SUNY ESF email: [EMAIL PROTECTED] 1 Forestry Drive voice: 315-470-6676 Syracuse, NY 13210 fax: 315-470-6956 > Date: Mon, 31 Jan 2000 07:50:41 -0500 (EST) > From: Robert McGrath <[EMAIL PROTECTED]> > Subject: Re: scoring semantic differential > > The problem with your proposed solutions is that omitting missing values > can lower scores, while using the neutral point can "neutralize" them. > One possibility available in the major statistical packages is to average > the nonmissing values rather than to sum them. This corrects for the > presence of missing values. > > Bob > - -- > Robert McGrath, Ph.D. > Professor > School of Psychology T110A, Fairleigh Dickinson University, Teaneck NJ 07666 > voice: 201-692-2445 fax: 201-692-2304 [EMAIL PROTECTED] > > On 30 Jan 2000, Grover Proctor wrote: > >> After looking in Osgood, Suci, and Tannenbaum's "The Measurement of Meaning" >> and Snider and Osgood's "Semantic Differential Technique," plus several of >> Osgood's individual articles, I cannot find the answer to this simple >> question: >> >> In scoring the Semantic Differential, does one treat MISSING data (i.e., a >> scale to which a subject failed to give a response) as "null" or do you >> assign it the "middle value" of your scale (i.e, 4 on a 1-to-7 scale, or 0 >> in a -3-to-+3 scale)? >> >> Clearly, Osgood hints that if the response is "not applicable" then the >> middle scale is the answer. And NOT having any response (i.e., null) >> distorts the computation of the Osgood D, which is at the center of my >> research. >> >> Has anyone seen anything in the literature, or has your own work delivered >> any insights, which would cast light on this problem? >> >> Thank you for a quick reply, either here or (better) direct to my e-mail >> address! >> >> Grover Proctor >> Dean >> Northwood University >> [EMAIL PROTECTED] >> =========================================================================== This list is open to everyone. Occasionally, people lacking respect for other members of the list send messages that are inappropriate or unrelated to the list's discussion topics. Please just delete the offensive email. For information concerning the list, please see the following web page: http://jse.stat.ncsu.edu/ ===========================================================================
