Hi Carsten,

Thanks for getting back about this. Yes, it is not group randomization, but randomization among a certain "group of questions". Looking at it a little more in detail, it seems to be useful, as it is basically the "randomize questions" bit of this thread, which is great.

The implementation in that sample survey was just a little confusing, so it didn't come out as if it would be a normal "question randomization", which it actually is. The sample survey is just a pretty complicated construction with 3 different "randomization groups".

As far as I have seen you just have write the "randomization group" in the advance features of the question and it will randomize the questions within this group.

The only problem that I see at the moment is to call this "randomization group" a group, as people might easily mix it up with "question groups". I can't come up with a better word at the moment, though.

Before writing the documentation for this we should really think about a better term for this feature. Currently it's quite irritating/unclear.

Suggestions?

Randomization of items (aka "subquestions" coined by limesurvey, right?)
Randomization of questions
Randomization of groups

Sound pretty clear to me. The implementation looks a bit strange. I just copy the name of the group into the field of the question to make the group randomizing. The logical way would be to allow adding a different type of group. But from a developer pov I can understand the way it is implemented. The feature helps a lot. The workaround was a messy one.

Here's are two situations I find myself in and what I'd really like us to be able to handle (smoothly?).

Situation 1: Client is showing three TV ads in a particular market. In the survey, they want to ask the same series of questions about each. However, they want me to randomize the order of presenting questions about the ads.

In the past, I could use the randomize_groups workaround that was on the documentation site.

Situation 2: Client has three types of product. Each one comes in one of three possible configurations. (Nine products in total.) They want me to show three products in a random order, featuring one product from each group of three.

In the past, I could use the randomize_groups workaround that was on the documentation site coupled with three random number generators with a range of values of 1-3 each and filter as follows:

Group1_rand()
Prod1 (show if RandomA = 1)
Prod2 (show if RandomA = 2)
Prod3 (show if RandomA = 3)
Group2_rand()
Prod4 (show if RandomB = 1)
Prod5 (show if RandomB = 2)
Prod6 (show if RandomB = 3)
Group3_rand()
Prod7 (show if RandomC = 1)
Prod8 (show if RandomC = 2)
Prod9 (show if RandomC = 3)

With this approach, I could show one product from each group and present the groups in random order.

I need a way to get to this as soon as possible because my clients are waiting right now for me to get this figured out and to launch their surveys...tomorrow. (Possible???)

Thanks so much!!!

(P.S. If developers need to see the code, I'd be happy to share.)

Yes, there would be 4 different types of randomization one day:

1. Randomization of items/subquestions
--> This is already available for quite a while, but it would be good to be able to exclude more than one item from the randomization, e.g. other, none, all of the above, etc.

2. Randomization of questions
I think the new feature will cover this quite good.

3. Randomization of question groups
This would be the next step when in comes to the implementation of randomization in Limesurvey. It often happens, that you have to ask the same set of questions for different products, brands, situations, etc. However, you don't want to ask the questions groups in different order, in order to avoid a bias towards the first or the last brand mentioned, for example. This is a quite common necessity in market research.

And then there would be the fourth aspect, which I'd call randomization of subgroups of items/subquestions.

So you have a long list of items, which are grouped into different item groups. it would be good to be able to randomize the item groups (with the header being always on top, of course) and within the subgroups. But I think this will require quite advanced programming.

However, I think randomizing questions is already a good step ahead, next thing should be randomization of question groups. Number for would be great, but is more a nice to have for the moment.

@JoyLaneResearch: I didn't really get situation 1. Do they want you to randomize the groups of questions (but order of questions within the groups is the same) or do they want to present several groups that contain the same questions in a specific order but randomize the questions within the group? I assume the first one, right?

Because the second one would be possible with this approach from the svn. It isn't stable though.

holch wrote:

Because the second one would be possible with this approach from the svn. It isn't stable though.

I interpret

ask the same series of questions about each

as one question per Group. Not all three questions on one page.

Randomization of questions in one group
Randomization of questions in different groups

Or a pagebreak question type is put into a group to get randomization across different pages.

I don't know if LS2 sticks to the questionbyquestion / group/ allonone page concept. How would randomize question in "question by question mode" work? Virtual group names in the question attributes to connect these questions?

@holch: Situation 1 is randomize groups (pages) only.

None of my clients have requested randomizing question order within question groups (so far).

My opinion:

Randomizing the order of question groups (pages) is more urgently needed.

If we can randomize pages but not questions, I can, if necessary place one question on each page to force random question order.

However, if we can randomize questions but not pages, I don't see how I can fake a random page order.

Not sure I said that very clearly. Did it make sense?

jelo wrote:

The implementation looks a bit strange. I just copy the name of the group into the field of the question to make the group randomizing.

How is it strange? It is most flexible and very easy to configure. Just edit the questions you want to be in the same rand group and enter the same freely choosable name in each question for the new randomization attribute.

The advantage is that the so randomized questions can span across different question groups (in fact the implementation is not related to question groups at all).

So for example you can pick question 1a from question group a, question 5b from question group B and question 3c from question group c, just put them into your randgroup1 and these three question will be shown in randomized oredr, but at their destined places in the survey.

@JoyLaneResearch: That is what I thought and this seems not to exist yet. The function that Mazi was referring to is basically the possibility to define a certain "group" of questions (not to be mistaken by the already available question groups). The questions in the defined group of questions can be randomized. This is already a nice thing to have, but then, as you said, randomizing question groups is a more common scenario in market research. Let's hope this will come for one of the next releases.

c_schmitz wrote:

jelo wrote:

The implementation looks a bit strange. I just copy the name of the group into the field of the question to make the group randomizing.

How is it strange? It is most flexible and very easy to configure. Just edit the questions you want to be in the same rand group and enter the same freely choosable name in each question for the new randomization attribute.

The advantage is that the so randomized questions can span across different question groups (in fact the implementation is not related to question groups at all).

So for example you can pick question 1a from question group a, question 5b from question group B and question 3c from question group c, just put them into your randgroup1 and these three question will be shown in randomized oredr, but at their destined places in the survey.

I agree with Carsten. This question randomization is quite simple, but flexible. First I thought it was strange too, because the example questionnaire was a little confusing. I didn't really get what it did. Honestly, at the moment I don't see many situations or research designs, would require this level of flexibility in randomizing questions the way it is shown in the sample questionnaire. But as long as you can define some questions that can be randomized, I am already happy.

I also agree that the randomization of question groups would be the more important one, as there are several survey designs that might require this. But I assume that it is more complicated to randomize question groups, than to randomize questions, otherwise they would probably have implemented it first.

And: while I currently can't really think of survey designs that would require the scenario that Carsten was describing, I can imagine that there are survey designs out there, that need this kind of randomization, I just didn't come across them, but others might need this quite a lot.