Hello,

selecting an exclusive option in a multiple choice question causes other items in the question to be recorded as missing. For example:

Q1. Do you own:
a. a cat
b. a dog
c. none of the above [exclusive option]

If the respondent picks c. (the exclusive option) he will be recorder in the dataset like this:

Q1a | Q1b | Q1c
NA | NA | Yes

Is it possible to avoid this behaviour from LimeSurvey and automatically covert NA's into No's? The respondent gave a valid answer with clear interpretation: he doesn't own neither a cat nor a dog. His answers shouldn't be treated as missing data.

 

To understand this behaviour, you need to know that these are "subquestions" that can be either marked (Yes) or not (no answer given to this specific subquestion).

So it is not a yes/no question. If you want that, you need to create a Matrix with "Yes" / "No" options.

Because the respondent has NOT specifically given an answer to the subquestion, we don't even know if the person has read it. Might not be relevant in your case, but can be relevant in other cases. So I think the "NA" behaviour is actually the "correct" one.

Thank you for your answer, holch. I was afraid that it was the intended behaviour.

I will remember about your suggestion to use a yes/no matrix but I think that it carries a serious risk of annoying the respondent. In my example, there are only two subquestions, but imagine having to answer 10 or 15 of these (I don't own a hamster, a fish, an elephant...). For now I will stick with multiple choice questions and cleaning the answers later.

Two things to add:
In my opinion the "exclusive option" works the same way as the array filter. 
Now these filtered options are not available for selection (either not visible or grayed).
Seems the export doesn't distinguish between these two cases.
2. A "search/replace" is done quickly.

You can file a bug report

Joffm

The really confusing part is that when you export the data (as excel / csv / R file) these subitems ARE treated as yes-or-no questions. If I select a dog but not a cat I will get a "no" in the cat column. But if I select none-of-the-above I will get NA's for other subitems. In other words, the value I get is dependent on values in other columns so the subquestions are not really independent. Now, if the person who is analysing the data is not the same person who created the survey and is not familiar with LimeSurvey, it's perfectly reasonable to assume that the percentage of dog owners is equal to the number of yes's divided by the sum of yes's and no's and forget to correct for the number of none-of-the-above's. This is bound to produce plenty of mistakes.

I didn't see your comment, Joffm, before posting my second response, but it seems we're talking about the same thing. I will file a bug report.

I don't really get the "search/replace" comment. In real life scenarios cleaning this stuff up can be pretty annoying. Different NA's may be present in the data for different reasons. You have to programmatically distinguish which are there for good reasons (filters, non-response etc.) and which are not.

 

I will remember about your suggestion to use a yes/no matrix but I think that it carries a serious risk of annoying the respondent.

I certainly agree with that. However, in my opinion it is a different level if something hasn't been checked or if someone actively marked it as "No". Might be only a small difference, but it is a difference nevertheless.

The really confusing part is that when you export the data (as excel / csv / R file) these subitems ARE treated as yes-or-no questions.

Here I actually agree. This is confusing. Probably this is treated like if a subquestion hasn't been shown e.g. via relevance equation. However, here it is different. They subquestion has been shown but then was rendered unselectable by choosing the exclusive option. So it should not be treated differently to the normal behaviour of the question type.

So this seems to inconsistent to me and I think it is worth a bug report as Joffm says.

Now, if the person who is analysing the data is not the same person who created the survey and is not familiar with LimeSurvey, it's perfectly reasonable to assume that the percentage of dog owners is equal to the number of yes's divided by the sum of yes's and no's and forget to correct for the number of none-of-the-above's. This is bound to produce plenty of mistakes.

Here I don't really agree. The person who analyses the data has the responsibility to understand the question that was asked and how the data is structured, so I don' think this is really a valid "excuse".

I create a plugin long time ago for this : gitlab.com/SondagesPro/altExcludeMultiple

But you need to create the config.xml file to test it in 5 and 6.
www.limesurvey.org/manual/LimeStore_extension_development
(or copy and adapt an exemple : gitlab.com/SondagesPro/QuestionSettingsT...g.xml?ref_type=heads )

Thank you all for your responses. I filed a bug report. 

I create a plugin long time ago for this :  gitlab.com/SondagesPro/altExcludeMultiple

I will check it out. Thank you!

The person who analyses the data has the responsibility to understand the question that was asked and how the data is structured, so I don' think this is really a valid "excuse".

I'm not saying that it's a valid excuse. Just stating the obvious that it increases the risk of mistakes when people are in a hurry, distracted, lazy etc.