Though I would share with the groups a number of items. I was given a reference on what is called "The unbounded Write-In scale for desirability". It is a photocopy and the person didn't write the book it came from (a Marketing Stats book I am certain of). The instructions given are as follows: This section lists some brands. Please tell us how you feel about these brands by writing L's or D's or an N into the boxes next to them. # If you like a brand, write a L or LL or LLL or as many L's as you want (the more you like it, the more L's you should write next to it) The advantages of the scale are these: (and I quote from the book) 1) It has a natural zero point to indicate indifference. 2) It does not require the respondent to use numbers, which some respondents find troublesome, but allows one to give direct behavioral expression to the quantitative aspects of their feelings 3) It is unbounded, that is, open-ended in both directions, so there is no particular ceiling of like or dislike. No matter how much a respondent reports liking or disliking something, they can always like or dislike something else more 4) It is constructed out of tiny increments of effort that restrain teh respondent from indiscriminate excesses and keep the responses within reasonable bounds. (I did not find this to be true in my data) 5) The author indicates that this sort of scale tends to create a more normal distribution than anchors scales (which are usually skewed towards the highend. No disadvantages listed (Not that there are not any, just that the author did not list them). After my original post I began thinking about this scale. Maybe no transformation is necessary because it is in theory, an infinite scale. Just like price, miles, or temperature that would be used as a DV. Although not Interval (Ordinal?) this type of measurement probably goes a step above an anchored 10 point scale. It was suggested that I row center and column standardize this data before factor analyzing. Replies: On Fri, Jun 02, 2000 at 04:57:51AM -0400, Donald F. Burrill wrote: > On Thu, 1 Jun 2000, Lance Hoffmeyer wrote: > > > I have collected data using a questionnaire with 20 rating scales. > > The respondents were asked to use tick marks to rate the answers so > > there were no anchors. > It may not be possible to answer your query > without more details. I am not visualizing what you are describing. > How did they "use tick marks"? Were the tick marks supplied, or did the > respondents make them? What was the response medium? (E.g., a row of > hyphens from edge to edge of the page, or from left margin to right > margin, or containing exactly 51 hyphens; or a row of hyphens with "+" > signs every 10th place, or a row of hyphens with one blank space between > adjacent hyphens; or a row of dots (periods); or columns rather than > rows; or a horizontal line from margin to margin; etc.) > By "there were no anchors" I take it you mean that neither > qualitative nor quantitative values were supplied on the response medium. > How then did you get the quantities you next report? (Measure in cm, or > mm, or tenths of inches, from the edge of the page, or from the left > margin, or from the left end of the line, ...?) > > > ... Some people have minimums of '0' and maximums of '10' while others > > have minimums of '6' and maximums of '68'. I wish to do a factor > > analysis so I need to get everything on the same scale. Initially, I > > thought of row standardizing the data. After thinking about it I am > > not completely certain this solves the scale problem. > > I rather suspect it generates new scale problems, mostly. > > > My first question is whether row standardization puts all respondents > > data onto the same scale? > Sounds to me rather as though it forces > all responses onto an artificially similar scale. Almost certainly it > will not be "the same" scale, because I can see no way of determining > whether in fact (just for instance) the maximum score for one person is > equivalent to the maximum score for another. If this does not bother > you, go ahead. > [OTOH, if you weren't bothered at the outset by supplying > a wholly unanchored, free-floating response medium (as you claim to have > done), it is hard to imagine your being seriously bothered by little > quibbles of this sort this late in the day.] > > > The second question is what would be another technique or a better > > technique to get all data onto the same scale? > > Others on the list may be more imaginative than I am. If a "same scale" > (or reasonable facsimile thereof) was not provided to respondents at the > outset, I do not see how one can blithely assume that they would have > been responding on any kind of common scale, let alone derive any kind > of equivalences for translating different respondents' output to some > kind of synonymy. > Unless, of course, there was something about the context that > gave implicit meaning, possibly even anchors of a sort, to the response > medium. If there was, you haven't so far described it. > -- DFB. Rich Ulrich wrote: > I have collected data using a questionaire with 20 rating scales. The > respondents were asked to use tick marks to rate the answers so there were > no anchors. Some people have minimums of '0' and maximums of '10' while > others have minimums of '6' and maximums of '68'. I wish to do a factor > analysis so I need to get everything on the same scale. Initially, I thought > of row standardizing the data. After thinking about it I am not completely > certain this solves the scale problem. - Well, you are not going to solve "the scale problem" by merely subtracting out the mean. You have eliminated all possibility of looking at between-subject differences in the over-all Means, by not providing anchors; and you can't look at single ranges, either. "Magnitude estimation" would have eliminated zero and then used logs of the scores; that would have preserved a semblance of "range" for each subject, though it still tosses out the mean. > ... My first question is whether row > standardization puts all respondents data onto the same scale? The second > question is what would be another technique or a better technique to get all > data onto the same scale? >From where you are, you can: Standardize by mean-and-SD, or take ranks, or (further) convert the ranks to normal-deviates. Whichever of these you do, the main criticism you receive should be about your data collection which left these options. The far better technique is to ponder and figure out your scaling/scoring questions before you collect the data. What did you hope to gain by not providing anchors? Dick Lehman wrote: > I wish to do a factor >analysis so I need to get everything on the same scale. Nope. You don't need to standardize things. Factor analysis is based on a correlation matrix and concerns relationships among variables. The fact that the variables themselves vary in their univariate characteristics (mean, sd, range, etc) is irrelevant. Lance Hoffmeyer =========================================================================== This list is open to everyone. Occasionally, less thoughtful people send inappropriate messages. Please DO NOT COMPLAIN TO THE POSTMASTER about these messages because the postmaster has no way of controlling them, and excessive complaints will result in termination of the list. For information about this list, including information about the problem of inappropriate messages and information about how to unsubscribe, please see the web page at http://jse.stat.ncsu.edu/ ===========================================================================
