Hi,

I have created a survey with 112 photos for 40 participants. Each participant is going to be shown 56 photos (112/2=56) out of a bucket of all 112 pictures. I have randomized the photos by manipulating with relevance equations.

I am affraid that the distribution of the photos is going to be uneven - is it accurate and what I can do to prevent this?

I have randomized the photos by manipulating with relevance equations.

How did you do that?

And please consider:
You have only 40 paticipants, but 112! / (56! * 56!) different ways to select 56 photos out of 112.
So it is very probable that the distribution is not even.
And what do you think is uneven? If a photo is shown less than 18 times, less than 16, and others more than 24 times?

And is your survey open or closed (tokens)?

Joffm

Hi Joffm,

Thank you for answer. To adress your questions:

1. How did I randomize the photos by manipulating with relevance equations:

In order to do this I have followed instructions I have found on this forum from the subject: How to pick x random questions from a bucket of y . After I played around with it, it worked perfect for me so i decided to stick with that solution. I assumed that this is the only way around to set kind of randomization I need via LS.

2.

And what do you think is uneven? If a photo is shown less than 18 times, less than 16, and others more than 24 times?

I am not precisely concerned about how many times a particular photo is shown.
Every photo comes with 3 items for participants to assess. I am worried that in result, I am going to have an uneven distribution of answers e.g.:

- Photo a: was shown 10 times hence I collected 10 different responses to 3 mentioned items
- Photo b: was shown 3 times hence I collected 3 different responses to 3 mentioned items

In the analysis, I want to compare responds between sets of photos. If I had a case of uneven distribution (like in example), I could not compare between results from photo a and photo b.

3.

And is your survey open or closed (tokens)?

I am not familiar with the token notion. Participants will come to me to fill and the survey and then leave. I do not distribute it via emails etc.


4.

You have only 40 paticipants, but 112! / (56! * 56!) different ways to select 56 photos out of 112.

I am aware of that issue.

I decided not to show 112 photos per participant due to undeniable fatigue they would experience. Hence, I intend to show 56 photos per participants from the randomized group of 112 (in a way that I described above).

If you have any suggestions on how I can solve that matter with LS in more plausible way - please, let me know!

Best,
Magda

Hi, Magdalena,

In the analysis, I want to compare responds between sets of photos. If I had a case of uneven distribution (like in example), I could not compare between results from photo a and photo b.

t-Test will get a result - no matter of the sample sizes. Of course the standard deviation will be rather large and the F-Test will probably fail.
But on the other side even 20 is not too much.
And doing it your way - which is fine - does not avoid an uneven distribution. Did you throw a dice 40 times.
I did and got the following distribution:
4 8 8 10 4 6


Okay, randomization:

I am not precisely concerned about how many times a particular photo is shown

But of course you are concerned about it.

- Photo a: was shown 10 times hence I collected 10 different responses to 3 mentioned items
- Photo b: was shown 3 times hence I collected 3 different responses to 3 mentioned items

You say, the respondents come to you to fill the survey.
So you can predefine the photos each respondebt sees.
So you can create a participant table with 40 dummy participants.
So each one has a token.

Each participant has to get a unique token attribute, just a number from 1 to 40.
In your survey you have a first question of type "multiple short text" (Q0) with 40 subquestions (SQ001 to SQ040)
As question text you enter a string like:
"012056001110045023067098102009..."
meaning a three digit number of the photo 56 times.
This seems to be a lot of work, but it is easy done in EXCEL with some formulas. and "quick add" in LS.

In your survey you can easily capture the photo by an equation:
1. Get the correct string: Like {if(TOKEN:ATTRIBUTE_1==1,Q0_SQ001.question,if(TOKEN:ATTRIBUTE_1==2,Q0_SQ002.question,if(TOKEN:ATTRIBUTE_1=,Q0_SQ003.question,...,if(TOKEN:ATTRIBUTE_1==40,Q0_SQ049.question)))...))}

2. To show the correct photo you use the substr function
<img src="http:www.myServer.com/path_to_photo/picture{substr(eq1,x,3)}.jpg" >
with x= 0,3,6,... (start of the number, index starts with 0)
and your photos named "photo001,jpg", "photo002.jpg", ...

This way you can predefine that each photo is displayed the same time.

Joffm

Later I will send an example.

To add something:
It is not necessary to use a token attribute.
In your case you can give each paticipant a token from 1 to 40.
Then you say {if(TOKEN:TOKEN==1,Q0_SQ001.question,if...

And here the example.

6 participants, 6 photos. Each one sees three.

If you see the photos, only 1-4 is existing. But there is the question text to show.



Best regards
Joffm

And if there are several questions to one photo you can enter the photo in the group description and ask your questions "group by group"

Did you throw a dice 40 times.
I did and got the following distribution:
4 8 8 10 4 6

You are using a strange dice...

Hello,
this morning I tested the expected distribution.
I created a small EXCEL macro to select randomly 56 out of 112 for each of the 40 participants.
This I repeated 10 times.

I got the following results:
1. Minimum and maximum appearance of a photo


2. Frequency (how many times was each photo displayed)


3. Chart of distribution


Considering the interval of confidence (n=40) - expected value = 20 (13.5 - 26.5), 95%-niveau everything is fine from the statistical point of view.

Now you know which frequencies you can expect, and it's up to you to agree to it (that one photo may be displayed 14 times, another one 25 times)

Joffm