Please help us help you and fill where relevant:
Your LimeSurvey version: LimeSurvey Cloud
Version 5.4.12
Own server or LimeSurvey hosting:
Survey theme/template:
==================

Hey everyone, 

I am trying to implement a survey experiment with random allocation of participants. One half is supposed to get the treatment the other half not ( I initiated this process with the  if
Bedingte Bearbeitung; if(test,result_if_true[,result_if_false = ''])
(is_empty( gleichung ), rand(1, 2),  gleichung ) function). However in the beginning I ask for a demographic variable (nationality) and I want for all participants who indicate they have a Russian nationality to also split these people 50/50 on the treatment. And the people with all other ethnicities to be split 50/50 as well. Is this possible to randomize people based on 2 different characteristic? I hope I clearly described my problem. If you need any additional information, please do not hesitate to point out the missing information.
Thanks in advance!

Nina 

I don't really understand the issue.

Each person has the same chance (50%) to get into the treatment group vs. the control group. So each "Russian" has a 50% chance to get into the treatment group, just like each German, each French or each Chinese would have.

With a big number of respondents from Russia you should get a more or less good distribution. Maybe I am wrong, but I don't think that there would be any difference if you would create a random number for each nationality. The change for each respondent is already 50%, it doesn't matter which nationality. So in the long run, you should get to a more or less even distribution between treatment and control group for each nationality.

However, random is random. This means there is no guarantee that you will get exactly 50% in each group. Because for each random draw the chance is 50/50 again. This means it is totally natural that you might have one group with more, the other with less respondents.

You could use different approaches, but they all have one thing in common. An even distribution is not guaranteed and the smaller the sample size the chance of a big difference is higher.

You would need a solution like "bucket least filled", which is probably quite complicated to implement, especially if you have a lot of nationalities.

How do you recruit your respondents? Do you know them before and invite them via email or is this an open survey and anyone can participate?

Hi, Nina, 
you seem to be German.
So you may have a look into my "Tutorial 4:  Gleichungen, Zu- und andere Fälle" in the German part.

And, as @holch,
If there are 50% Russians in group 1 and 50% of other nations in group 1, there are 50% of your whole sample there.

Joffm

If there are 50% Russians in group 1 and 50% of other nations in group 1, there are 50% of your whole sample there.

I don't really understand what you are trying to say here.

My point: if a russian gets into the survey, the chance that Russian gets into the treatment group is 50%, correct?

If you add a random number between 1 and 2 for each nationality, the chance for each russian is still 50%, correct?

So I don't see how implementing a different random number draw for each nationality will improve the distribution of Russians (or any other nationality) over the two different groups.

Aber vielleicht stehe ich ja gerade auf dem Schlauch.

Sorry, @holch.

I just wanted to say it again, only repeated your statement.

So each "Russian" has a 50% chance to get into the treatment group, just like each German, each French or each Chinese would have.

No difference, if I create two random numbers (Russians vs. Other) or only one. The result will be the same.

Joffm
 

Ahhh, then I read it wrong. Glad to hear that I am not totally crazy now. Hahahaha!

Thank you for your response! I am going to recruit my respondents via my own network, which is why a expect a smaller sample, which is why I want to make sure that I do not end up with all my respondents from a Russian-german background in the control group. This is why I was considering to use another solution and not just the complete randomization version. Because I only want to make sure this group is split evenly all other respondents can just be split between the groups. I hope I explained my problem a bit better now. I was also considering to introduce a quota for the treatment for people who indicated in the previous question that they are Russian German but to be honest I do not know how to combine these two conditions

I hope I explained my problem a bit better now.

Nope, not really.

So you know the people who will participate? How will they receive the link? Will you have a closed survey where you can identify each respondent? In this case you could decide beforehand who falls into the "russian-german" group and then could distribute them evenly into the test group. Just use a dice, or let someone else randomly apply them to the groups.

Oh no hahah. I will try it again, thank you very much for your patience. I will share the link via my personal network but I will also encourage others to share the link so I will not know beforehand who is answering the questionnaire and which group they belong to, that's why I would prefer if I could implement something so Limesurvey decides which group I Allocate people to, but at the same time I want to make sure that not all Russian germans end up in one group which could happen by chance... Thank you very mich in advance

I want to make sure that not all Russian germans end up in one group which could happen by chance.

I think that is a very slim chance. Will have to think about this one and if there is something that can be done.

"Much ado about nothing" or "Viel Lärm um Nichts"
Well, you can calculate the probability that you get 20% of the Russians in group 1 and 80% in group 2.
You will get a small, but positive result.

Yes, it is possible, as it is possible that you roll a perfect dice 10 times and it shows 10 times a "6". The probability is 0.00000165%
Or the probability that the ball will land on red 50 times in a row in roulette = 0.0000000000000888%
So, possible but not really probable.

We all know the Galton-board to show the binomial distribution, which approximates a normal distribution.
And this you can expect in your experiment.

Now distribute your people by a random number and it will be fine.

And if there is an imbalance, you always can fine tune
Though it is sufficient to use "random==1" resp. "random==2" to distribute your respondents, you may add a nationality code (nat) 1=Russians, 2=others (only to be anle to fine tune)

Now you start with
group 1: "random==1 and nat<3"
group 2: "random==2 and nat<3"

Now, if by some reasons there are too many russians in group 1 you may change the conditions t
group 1: "random==1 and nat==2" only "other nations with random number==1"
group 2: "random==2 or nat==1" repondents with random number==2  OR Russians without considering the random number.

If it is balanced again, switch back the conditions.

Joffm

And everything you are pondering about you should simulate throwing a coin (or in EXCEL)