I created a, perhaps a bit more involved, survey that I ran over the weekend (attached in this post). The structure is (roughly):

1) Demographics (relevance = 1)
2) Misc (relevance = 1)
3a) Survey with slider (relevance = SAVEDID % 2 == 0 ; questions in randomization group A)
4a) Hidden items for 3a (relevance = 0 ; questions in randomization group A)
3b) Survey with yes/no (relevance = SAVEDID % 2 == 1 ; questions in randomization group
4b) Hidden items for 3b (relevance = 0 ; questions in randomization group

3a has 6 questions and 4a has 18, both in the same randomization group; this should choose 6 questions at random out of the 24 provided. The same goes for 3b and 4b.

The survey is actually not randomizing the items across question groups. Instead, only the order of the items within a question group is randomized. Why?

Additionally, - and this is much more of a problem - the responses are not recorded properly. I answered the survey 5 times myself, responding to each item; however, the .csv export shows that I didn't respond at all (only saving like 3 of the items). Could the data be stored in the database somewhere and this is just an export problem or are the responses lost completely?

Does anybody know what could be the cause of this behavior? Could it be a bug? Any pointers are much appreciated.

I am not quite sure what you are doing there. What is % supposed to do?

SAVEDID % 2 == 0

What should this do?

The actual relevance equation is

(SAVEDID) - (floor((SAVEDID) / 2) * 2) == 0

since modulo (%) is not implemented in the survey tool. I thought it might be easier to understand if I write SAVEDID%2==0 instead. Sorry for the confusion.

What is the LimeSurvey version you use?
Everything can be a bug. So it is important to know the exact version.

I am using LimeSurvey Version 3.15.9+190214

Hi,
IMO you shouldn't do the survey "All in One".

If you do it "group by group" everything seems to be fine.
Unfortunately I did not find Olle's post when he introduced this solution.

But nevertheless:
I answered it three times and got a correct randomization and all answers.

Best regards
Joffm

Hmm, I didn't consider the way it is presented as a potential source of the problem.

Unfortunately, it had to be "All in One". We were giving the survey to people at a public event / exhibition. We tried "group by group" and people would only answer the first group, hand us our tablets back and move on. "All in One" made them respond to everything.

Which version of LimeSurvey are you using that doesn't have any problems? I tried it on both, our online server and our local system; both show the same problem (same version).

The issue with SAVEDID and all in one page.

1. First construct the survey without SAVEDID
2. Start to show survey, update the SAVEDID
3. The page is outdated : SAVEDID start at null, and was set to real final SAVEDID after.

Maybe : move whole question in one group, show welcome page, and survey group by group.

And look at this commit for better unerstanding the issue with SAVEDID : github.com/LimeSurvey/LimeSurvey/commit/...55efc8a50c8b7409cdd9
(this commit are not in last release).

I've just tried changing from "All in One" to "Group by Group" and it is indeed working fine that way.

Since we already had some data collected with the "All in One" survey, I assume that data is lost?

As a sidenote: I got the modulo operator with SAVEDID to work just fine in the "All in One". I added a small javascript snippet to the first demographic question

<script>console.log({SAVEDID})</script>

and with that the branching started to work properly ... unfortunately not the other randomizations and data storage.