Hi everyone!
I’m trying to do the real MAXDIFF in LS.
You guys made a great job with that MAXDiff workaround and it looks great but I need an extension.

Basically maxdiff means that I have to choose best or worst option from a sets of statements.

I mean I have 20 statements and in Q1 I need to scale best/worst between randomly chosen 4 of that 20. And so at least 5 times.
And here is my problem. If I use Array by column, I cannot filter rows. Or can I guys?[img]blob:https://forums.limesurvey.org/1267183f-279e-42a4-878b-cc151096497a[/img]

I assumed that I can make 4 rand variables in one group and choose which statement will be insertant (or with a help of ‘if’) in Q1. But it means that there is a chance that the same statement can appear in Q1 twice or three times[img]blob:https://forums.limesurvey.org/83dea1a8-4e91-4716-8b14-4af66ff67ca9[/img]And it’s no good as you understand. Also this method is hard in datapreparation.

The second option I thought about was..

We can create multiple options question (RAnd in my lss) that will randomly choose 4 answers for Q1. In such case the duplicate is not possible. And now I need to solve how to filter that rows... If I use {insertans} I see empty rows. Can we somehow hide them like we do with empty columns in numeric array ([url] framagit.org/SondagePro-LimeSurvey-plugin/hideEmptyColumn [/url])? OR maybe we can filter rows in any other way?

The third option I was thinking

Maybe it’s possible to do maxfiff visualization on numeric array? In such a case we already can filter both rows and columns?

I use LS 3.25.7 at the moment, but hope to be updated in nearest future to LS 5.
The lss is attached.
Would be appreciated for help and questions. 

I wonder why you want to do 4 out of 20 random in MaxDiff.
You usually make a test plan (for most of us the software does it) to ensure
that all your attributes are shown to the respondent in a certain amount of sets.
Depending on how many attributes, interviewtime and respondents you expect to have, you end up with a certain amount of sets and might not show every attribute more than once. The random part in MaxDiff is the position of each attribute in the question and the position of the attribute set to ensure that you don't ask every respondent the set of questions in the same order.

To do that in LimeSurvey you will end up with a lot of copied questions. The lacking feature to do it in a elegant way are equations on answers in "Array by Column" questions. That feature is missing since a long time.

There is no public workaround or plugin to conduct MaxDiff inside LimeSurvey. I have heard of a workaround, but never got my hands on it. When I had to conduct MaxDiff or ChoiceBasedConjoint I used thirdparty survey software and integrated that part inside a LimeSurvey.

My feature request for MaxDiff is from 2009
bugs.limesurvey.org/view.php?id=7377

I'm not aware of an question-type where the relevance is possible on rows and columns. There were some question-types which had the relevance fields but they were without any function. LimeSurvey is still lacking a relevance on single choice answers. Plugins and workarounds are sometimes possible and some activity can be seen in that area. bugs.limesurvey.org/view.php?id=12277

bugs.limesurvey.org/view.php?id=8149

13509: Sub-question relevance does not work in array-by-column questions.
bugs.limesurvey.org/view.php?id=13509
 

Hi, Oksana,
as @jelo: Usually you create a tewst plan.

But here a small example to get 5 times 4 random statements out of 20. And to know which were selected.
And the "old" maxdiff workaround to display it.

Have a look at the javascript to select the statements (in "Pool")
The idea is to enter each statement with the same length. So we can later easily extract the 1st, the 2nd, ... by the function "substr" - and "trim".

And in each of the maxdiff questions there is the script to display them and to avoid the same selection in "BEST" and "WORST"



As the statements selected are stored in "Pool" you may use this for the analysis.

Joffm

@ Joffm   Thanks a lot for your reply. Can you attach an .lss file, so I can see this in work?

Jelo, thanks for your fast reply.
Actually we thought about design such a test combinations by ourself and script it to Limesurvey due to our plan.
As you mentioned, most of the softwares does this testing plan, not a person. And I assumed if I will do it by hands and not a pure random I can affect on the results somehow.
Maybe you have some recommendations where I can find principals how to create those sets? (sorry, it's not LS question anymore)
But I do not understand how many combinations should

Sorry, forgot the most important thing.

 

oksana wrote:

Maybe you have some recommendations where I can find principals how to create those sets? (sorry, it's not LS question anymore)
But I do not understand how many combinations should
 

The field is wide open. In MaxDiff and Conjoint the amount of questions and attributes per question often depend on the analysis part. If you only interested on aggregates results, you need less information than when you want to analyze many segments or on a one person level.

SawtoothSoftware has some fine papers about methods and rules of thumb, which not only apply to their tools.
sawtoothsoftware.com/resources/maxdiff-calculator

To get a feeling for what kind of control a mature MaxDiff-Tool can give, I recommend to take a look at this example:
sawtoothsoftware.com/help/lighthouse-stu...maxdifftutorial.html

When it comes to expectations of customers in the market research field, SawtoothSoftware can deliver.
 

If it may be a bit less expensive than Sawtooth, you may try XLStat, an EXCEL add-on.
[url] www.xlstat.com/en/ [/url]

 

There is a 14 days trial available.

Joffm