tag:blogger.com,1999:blog-590043897961646114.post2502613238597205219..comments2024-02-06T02:06:06.364-08:00Comments on Engaging Market Research: The MaxDiff Killer: Rank-Ordered Logit ModelsUnknownnoreply@blogger.comBlogger5125tag:blogger.com,1999:blog-590043897961646114.post-92120485862744666012015-04-28T10:15:10.512-07:002015-04-28T10:15:10.512-07:00I do not advocate complete rank ordering because i...I do not advocate complete rank ordering because it is not something that consumers spontaneously do in the marketplace. Consideration sets contain at most a few of the top performers. However, if you wish to use MaxDiff, you should understand that you can program a computer to complete all possible MaxDiff comparisons using nothing more than respondent's rank ordering of the objects. That is, give me a respondent's rank order of any set of objects, and I can act as their agent to complete any possible MaxDiff exercise. There is no MaxDiff magic. It is nothing but a data collection technique for obtaining a rank ordering.Joel Cadwellhttps://www.blogger.com/profile/14946447393733294251noreply@blogger.comtag:blogger.com,1999:blog-590043897961646114.post-29157505889801331772015-04-28T09:51:06.907-07:002015-04-28T09:51:06.907-07:00With Max-Diff you can scale factors of the objects...With Max-Diff you can scale factors of the objects. So if we are comparing game systems we could include hard - drive space or processor speed. In your example the covariates are by subject. <br /><br />Basically max-diff can perform a conjoint. Can Rank ordered do the same thing. Or only order along one dimension.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-590043897961646114.post-38164027456121662742013-10-07T16:30:00.903-07:002013-10-07T16:30:00.903-07:00Hi Joel, very interesting perspective! In practic...Hi Joel, very interesting perspective! In practical terms, how can the data be collected using Survey Monkey for example? Seems the logic involved does require powerful and expensive survey software to collect the dataAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-590043897961646114.post-16002943569429306642013-02-28T15:42:36.917-08:002013-02-28T15:42:36.917-08:00Thank you for your comments. I too was concerned ...Thank you for your comments. I too was concerned about the ability of respondents to rank order unfamiliar alternatives. However, respondents often complete the ranking exercise, especially when they are instructed to answer even if they are unsure. In this case, the preference ordering can be divided into three sections consisting of the consideration set, the unfamiliar set, and the rejection set.<br /><br />As I noted in my post, extensions to the basic rank-ordered logit attempt to incorporate partial rankings when such distinctions cannot be made. For example, a partial ranking of our six games would allow respondents to rank order only the subset of games with which they were familiar, so their six scores might be 1, NA, 3, 2, NA, and 4. MaxDiff, on the other hand, simply ignores this issue. Alternatives that are seldom selected as best or worst receive MaxDiff scores in the middle of the scale, which is what we see with the complete ranking task. My problem with MaxDiff is not that it produces a ranking. My problem is that MaxDiff is a cumbersome and expensive way to rank a set of alternatives. <br />Joel Cadwellhttps://www.blogger.com/profile/14946447393733294251noreply@blogger.comtag:blogger.com,1999:blog-590043897961646114.post-39060998190597230222013-02-28T09:40:52.968-08:002013-02-28T09:40:52.968-08:00Joel, I enjoy your blog and am happy to say that I...Joel, I enjoy your blog and am happy to say that I have learned a thing or two in reading it. Please keep up the interesting work.<br /><br />In seeming to describing maxdiff as nothing more than a glorified ranking technique, though, I wonder if you might be overlooking one of its primary claimed advantages, particularly in light of your earlier post about constructed vs. recalled preferences. Namely, maxdiff starts from the position that respondents are more likely to be able to identify which pair of alternatives are the furthest apart on some latent scale of importance or desirability than they are to be able to consistently rank all possible possible alternatives, particularly as the number of alternatives grows. In the example you give above, is it reasonable to expect that respondents are able to meaningfully and consistently rank game consoles to which they may be largely indifferent? I would tend to accept the advantage of a maxdiff in this case, but I am curious to know your thoughts.<br /><br />As a disclaimer, I have not (yet) used maxdiff, and I do not use Sawtooth.<br /><br />Anonymousnoreply@blogger.com