Goal: Have a questionnaire with 10 and then 30 questions, then a number of formulas (scores) and then route the survey respondee to one of 16 pages / links. 
The MBTI and many other classical tests are using questions and then formulas to gauge, which one of different classes is one most likely to apply. Depending on this one can then offer an explanation of the class, or tips on behaviour etc. 

The issue is that Limequery is fundamentally sequentally organized. With sections skipped, but not with a full parallel design, that after a certain point we have a path dependent path. E.g. if we have a first sets of 30 questions, then an analysis, and then we have 20 questions selected out of 50 dependent on the result from the first path. I.e. a middle of the clustering a result, and then 50 questions which are shown or skipped depending on a variable from before. 

How would you structure this ?

PS: The purpose of this general question is an advice page for lonely people, and e.g. advice is very different for a 16 year old, a 30 single parent or an 80 year old. We have 8-10 pages or clusters preidentified, but need a good router to those pages.

Though it's a bit confusing 
in the first paragraph it is 
10 questions, then 30, then calculation, then 1 out of 16 pages.

In the third paragraph it is:
30 questions, then calculation, then 20 out of 50 questions.

Nevertheless: use ExpressionScript to calculate and conditions to show the appropriate result page.

Better you provide a lss export of a sample.

Joffm 

Hi Joffm

Thanks, it is somewhat clear but not perfectly clear in my head. I have used Limesurvey before in 2019-2020, but that was a linear scenario (one long survey, no major path dependency) and I am currently trying to figure something out which I have in my head.

The extended scenario is: There are say possible 50 recommendations to help lonely people find connection. Out of those for a specific subgroup say 10 recommendations are really useful and 30 are utterly irrelevant and potentially harmful/offensive etc.

This is not exact science but based on a number of research papers, books, etc. one can synthesise a reasonable answer, or more precisely a set of answers.

So the rough process is. Use 10 questions for demographics and then another 20 questions for "intensity of loneliness" etc., use this then to cluster into one of 10 groups (this groups will potentially change over time with additional insight), then use another set of 20 questions (10 to 30, potentially as variable triggered selected subset of 50 global questions), and then based on the insight of the 2nd set prioritise suggestions, select links whatever to helpful resources, etc..

The goal is to rather than have a website that says to the general public - if you are lonely, try these 10 very general suggestions - have something - if you answer this questionnaire, we can identify 5 proposals that are specific to your target group.

I played around with typeform, and jotform, but their marketing lingo is way higher than the actual perfomance of the platform, and I am now returning to Limesurvey, which is remarkably cumbersome, but does have a lot of technical capabilities.

However I am no expert, and deciding on the key feature how I set up this sequencing I am aware I can get a lot right or F*ck things up. Hence my question.

I would love to have visual workflow / process designer feature with a branching functionality for different branches, but that is not (yet) on offer.

My hope is that somebody responds with "you should do this with a single survey and this branching functionality" or "setup 1 key survey and then 10 followup surveys and use a token or URL based handover to link to them.

So I am trying to get advice before locking myself into a certain design. Does that make more sense ?

christianlangkamp wrote:

use this then to cluster into one of 10 groups (this groups will potentially change over time with additional insight), then use another set of 20 questions (10 to 30, potentially as variable triggered selected subset of 50 global questions), and then based on the insight of the 2nd set prioritise suggestions, select links whatever to helpful resources, etc..
 

From your descriptions that still sounds like a linear survey.
You have questions. You then have to use ExpressionScript to anaylze the answers to find out, to which group the respondent can be assigned.
ExpressionScript means that you have a lot IF-Statements and a hidden question where yThenou save the group number.
Then you ask more questions and again have to analyze the answers via ExpressionScript.
The recommonedations are done with ExpressionScript again. If that group and these answers, show this text or page.
I recommend to create a simple survey with just one question and two answers for every step of your process.

I wonder what didn't worked in TypeForm.
 

Dear Jelo,

Thanks for this. What I meant with Jotform/Typeform is that I couldnt see anything that their UI did substantially better than what I had already done with Limesurvey, which is a powerful yet not always intuitive and easy to manage tool.

I would love to have a Alteryx/KNIME like process flow to work with for the branching, or a Matrix like If-Then maintenance structure.

I did in my past survey have both these elements in excel, and used then formulas in excel to turn the logic into copy-paste expressionscript items that I then pasted into the logic part of the questions.

It all "works", but it becomes quite cumbersome to maintain.

But yes, thanks to your responses I think doing it in a single survey with extensive expressionscript logic is then the likely solution.

christianlangkamp wrote:

It all "works", but it becomes quite cumbersome to maintain.
But yes, thanks to your responses I think doing it in a single survey with extensive expressionscript logic is then the likely solution.
 

I'm not sure there is a way to present complex logic in an easy way to maintain. Flowcharts like in KNIME are a great way to present workflows.
But iif you have many IF-Clauses in a graphical view, they still would cause confusion. A structured text was often the easier approach.
If LimeSuvrey would support dynamic lists you would have all recommendations in one list and just remove the ones not qualifed.
 

The difference is if you have 10 or more questionnaires all in parallel, and funnel at the start via a variable (result of previous questionnaire) into the one you want or you have one questionnaire with many, and then use the variable to make a show/nonshow decision. 

I think I somehow did a "if X is element of {A,B,C,D} then show question Y" expression logic statement, but I might have wanted and not managed and then implemented with more cumbersome logic. Yes, supporting dynamic lists and these things would be helpful, as well as if there was an easy exit to process things in R/Python while running the survey for intermediate results. But beggars cant be choosers. 

The main point is that I think it should be feasible to build a data input based "recommender" (If you answered Q1 as A2 and Q2 as B3 then sheet 7 (out of 10) might be relevant to you and also you might try suggestions S1, S7 and S10) using the expression logic and the questionnaire capabilities somehow, and initially I am more concerned that I am capable of implementing the logic and can use it to direct then people to the relevant pages on a website rather than piling them with "this is all anyone might have ever found useful in such a situation". 

And if it gets traction and ends up being more useful with additional functionality or nuancing or branching, I can look for a different solution. 

What I am ultimately after is being able to run a cluster analysis at some point on the input data, and then see whether the hypothesised cluster and characteristics loosely hold up, or whether there are clusters I have not seen and hence need adapted recommendation sheets for, and thus can change up the logic elements (i.e. the show this logic) without changing the major final structure (completely rebuilding the questionnaire).

Hi,
@jelo saved me a lot of writing - during my breakfast.
But

I think doing it in a single survey with extensive expressionscript logic is then the likely solution.

1. If a single survey or a 1:n structure is preferable we only see if you provide an example.
2. ExpressionScript shouldn't be too extensive.
Eg. to calculate the group ( 1 out of 16) of MTBI you only say (one single equation)
{join(if(countif("1".that.Q1.sq_E.NAOK)>countif("1".that.Q1.sq_I.NAOK),"E","I"), if(countif("1".that.Q1.sq_N.NAOK)>countif("1".that.Q1.sq_S.NAOK),"N","S"), if(countif("1".that.Q1.sq_T.NAOK)>countif("1".that.Q1.sq_F.NAOK),"T","F"), if(countif("1".that.Q1.sq_P.NAOK)>countif("1".that.Q1.sq_J.NAOK),"P","J"))}
Then you will get something like "ENTJ"
Of course you can get a numerical result (1-16)

But as long as we do not know which test you really want to use,...?
I do not think it's really MTBI with its really low reliability

I did in my past survey have both these elements in excel, and used then formulas in excel to turn the logic into copy-paste expressionscript items that I then pasted into the logic part of the questions.

This seems to be really cumbersome and obviously not the best solution.

If LimeSuvrey would support dynamic lists you would have all recommendations in one list and just remove the ones not qualifed.

This is a point that @jelo and I would like to be included for many years.

Well, with some small limitations you can "fake" such a list either by an array(text) or even by a question of type "huge text" (where you use stringfunctions like "strpos", "substr", etc. to capture the desired recommendation text.
I often use a hidden container (an array(text)) to store interim results.
This way you can enter all equations in a single question.

But - I repeat - as long as everything is really vague I can't say more.

Joffm

 

The difference is if you have 10 or more questionnaires all in parallel

This usually is a bit confusing
We usually understand the entire survey as "the questionnaire".
I suppose you use it differently.

And don't forget:
LimeSurvey is a data collecting tool, not an analysing tool.

BTW:

I think I somehow did a "if X is element of {A,B,C,D} then show question Y" expression logic statement, but I might have wanted and not managed and then implemented with more cumbersome logic.

I think I showed this within the last month.
Like
Condition for question Y
strpos('#ABCD',X)>0