Dear all,

I've performed some test and it appears that the randomization does not work.

I want to randomize in half participants to two different question groups.

I created a hidden question (equation type) using the following string:

{if(is_empty(Q1),floor(rand(1,2)),Q1)}

So it should randomize in half since the condition is Q1 == 1 for the first group and Q1 == 2 for the second.

I did some tests and in one case I had 6 (1) and 4 (2) (for this I set the quota at 10) and in the other 3 (1) and 7 (1).

Does the IP address have any influence on this?

I will have 100 participants in my survey. How can I be sure that they are split 50% in each question group?

Thank you very much in advance.

Dg

dgiacometti wrote: I did some tests and in one case I had 6 (1) and 4 (2) (for this I set the quota at 10) and in the other 3 (1) and 7 (1).

Your randomization is working fine. A test consisting of 10 tries won't always be 5:5.
So 6:4 after 10 runs won't translate into 60:40 after 100 runs.

dgiacometti wrote: I will have 100 participants in my survey. How can I be sure that they are split 50% in each question group?

Currently there is no way to ensure that 100%, because what's with situations like when the respondent quits the survey during the interview? LS would need to tap into the complete responses to decide which group should be assign the next respondents. There are survey packages which can do these things. But I am not aware of a surveytool that ensures 100% even distribution among the randomized sections. Some tools e.g. qualtics allows you to change the distribution during the survey.
Some allow to override randomization partially and let you edit the counts.
www.qualtrics.com/support/survey-platfor...mizer/#EditingCounts

When you only have two groups to assign, you could use the saveid/response id of every interview.
That is always odd/even. But you still often don't end up with 50:50 after 100 runs. That's because people reload the browser or stop during the survey.

If you already know your exact 100 respondents, you can use tokens and attributes to assign the question groups upfront.

I wouldn't spent that much time on a exact split. These excact split demands normally pop up from university tasks. Which is a bit confusing, since demanding 50:50 split during a field phase means you count every completed response afterwards. Even if the content is nonsense (which happens online from time to time

Thank you very much for very detailed and useful explanation.

What are the passages I have to do to in the platform to assign tokens to question groups?

In that case I would not need anymore the equation randomization question at the beginning? Right?

Thank you again.

Bests,

Dg

dgiacometti wrote: What are the passages I have to do to in the platform to assign tokens to question groups?

You're not assigning tokens to questions groups directly.

You can uses tokens. Which means that you provide e.g. a invitation link to your respondents. This link includes a token. In Limesurvey you can add attributes to every token.
www.limesurvey.org/manual/Survey_partici...ned_attribute_fields

You e.g. add group as a attribute to your tokenlist. Next step would be to enter a value for this attribute for every token. So you will end up with 50% with group=1 and 50% with group=2

In Limesurvey you then can access this token attribute group via conditions. You then can display/hide a questiongroup depending on the value group=1 or group=2.

Sounds a bit complicated but once you got uses to tokens you will see that it is not that much work.

The main question to answer first is if you really able to contact every potential respondent and provide them a token/invitation link.

As Jelo said, randomization doesn't work like this...

If you have a random number between of either 1 or 2, for each draw there is a 50% chance to get one or the other. However, it does not mean that at the end you will have 50% "1" and 50% "2", this is not how chance works.

The larger the number of draws, the closer to a 50/50 distribution you should get, but it hardly ever is 50/50. n=10 is definitely not a large number and even n=100 I wouldn't consider a large number. With n=1000 chances are night that you get relatively close to a 50/50 distribution. Important is the word "close".

If you need 50 in one and 50 in another case, it is best to follow Jelos recomendation of distributing the case for each participant beforehand in the token table as custom attribute. But even there is no guarantee that you will end up with 50 for each case, only if really everyone of the sample answers.