```Date: Mon, 10 Feb 1997 15:42:17 GMT Reply-To: David Nichols Sender: "SPSSX(r) Discussion" From: David Nichols Organization: SPSS, Inc. Subject: Re: Orthogonal designs & duplicate cards In article , James Karr wrote: >In article <5dfnvq\$k1i@netsrv2.spss.com>, nichols@spss.com (David Nichols) >wrote: > >> In article , >> James Karr wrote: >> >I9ve got a question regarding the generation of orthogonal designs for >> >conjoint/trade-off research, and the issue of duplicate cards. >> > >> >SPSS9s conjoint module can generate 3decks2 with the minimum number of >> >cards required to estimate the utility of each factor. It may also, from >> >time to time, produce decks with one or more duplicates (i.e., two cards >> >will have the identical combination of levels for all factors) as part of >> >the orthogonal design. When this happens, it alerts you to the presence >> >of such cards, and allows you to remove them or to generate a new deck. >> > >> >Using an example, now let me ask: If 3 duplicate cards appear as part of a >> >16-card deck, should they be... >> > >> >(a) deleted, because they contain a redundant combination of levels that >> >do not aid in estimating utilities; >> > >> >(b) retained, because to delete them would remove almost 20% of the deck, >> >compromising its orthogonality; >> > >> >or should the researcher (c) simply generate another 16-card deck, and >> >continue to do so until producing a deck with 16 unique cards? >> > >> >To put the question another way: If a 16-card deck has 3 duplicates, does >> >that mean the orthogonal design *really* only needs 13 cards to estimate >> >the utilities for each factor? >> > >> >Any comments would be appreciated. >> > >> >James Karr >> >> I would think rerunning until you get another set of 16 cards with no >> duplicates would be the best way to go. >> >> -- >> ----------------------------------------------------------------------------- >> David Nichols Senior Support Statistician SPSS, Inc. >> Phone: (312) 329-3684 Internet: nichols@spss.com Fax: (312) 329-3668 >> ----------------------------------------------------------------------------- > > >OK, but *why*? The Categories manual also seems to suggest that you might >want to rerun the design until you get a deck with no duplicates, but it >doesn't give any rationale for doing that versus simply deleting the >duplicates and using the remaining cards. > >To take this to the extreme, I could argue that what I should do is try to >generate a design with as *many* duplicates as possible, and then remove >the duplicates before interviewing. This would allow me to save both time >and money (printing costs) at the interviewing stage. And to preserve the >orthogonality of the original design, I would simple re-enter the >duplicate cards into the datafile post interview, using the respondents' >preference ratings for the each card of the duplicate pair they actually >saw (e.g., if card #3 and #12 are duplicates, I would only show #3 to >respondents, and then manually enter #12 and use their rating of #3). > >Thus, at the end I'd still have an orthogonal design, plus the bonus of >cheaper and faster data collection as a result of using a reduced deck for >interviews. What would be wrong with that approach? > >James Karr A full profile conjoint analysis, as implemented in SPSS, is a method that makes a substantial number of assumptions, many of them fairly dubious in a number of contexts (e.g., no interactions among attributes in their effects on evaluations of the desireability of the profile, independence of observations). Using the same score from a single card multiple times certainly destroys any semblance of independence of observations, and reducing the number of different cards you show to people reduces the amount of information you gather. It's up to you; I wouldn't do it. -- ----------------------------------------------------------------------------- David Nichols Senior Support Statistician SPSS, Inc. Phone: (312) 329-3684 Internet: nichols@spss.com Fax: (312) 329-3668 ----------------------------------------------------------------------------- ```

Back to: Top of message | Previous page | Main SPSSX-L page