Split Survey Design

#1
I will appreciate any advise/critique.

The problem is as follows:

I have 21 objects which define a context. For each object, I would like to ask questions on 14 attributes. With O(bjects)=21 and A(ttributes)=14, the number of question for each respondent would be 21*14 = 294, excluding about 10 demographic questions. Thus, we are considering a split questionnaire design. I am learning it as I go.

After thinking a bit (while waiting for the books/articles on such missing designs to arrive), we have the following heuristic plan. The attached spreadsheet summarizes it nicely, in color coded cells. In words,

1. Have say, 5 samples of subjects (about 50 subjects each).
2. Sample1 subjects rates 5 objects* 14 attributes. Sample2 rates ONE of the five objects rated by Sample 1 and 4 others. Sample3 rates ONE of the objects rated by Sample 2 and four others ....
3. Our primary interest is in the correlation matrix R(attr, attr) matrix, aggregated across subjects and objects. This feeds into an exploratory factor analysis procedure.

With the above design, I am guessing that all the information necessary for computing the correlation matrix is imputable, thru an appropriate missing values procedure.

Once again, the attached spreadsheet summarizes the scheme nicely.

Please let me know if I am in error - also, I appreciate any advise or readings.

thanks in advance,

Statpuzz
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
Interesting, why can't everyone complete all of the questions? And question assignment will be randomized? Will demographics be used to verify comparability of samples? If added up, do all of these samples represent the entire population? Can you have all subjects complete one exact same question?
 
#3
Good questions!

Re- why can't everyone complete all of the questions? The number would of questions would be about 300. I expect severe fatigue effects and very low data quality.
Re- question assignment will be randomized? didn't think of this, but will consider this.
Re - will demog be used to verify comparability of samples? Yes, and if necessary (i.e., they are predictive of correlations), may be used for imputation as well.
Re - if added, will the samples represent population - Yes
Re- Can you have all subjects complete one exact same question? Didn't think of it, but very interesting. As of now, the set of questions on demographics (D) are common.

On the last item ("exact same question?) - the blocking variable is Objects. If all subjects were to answer rate the SAME object (e.g., O1) ... Let me think. Thanks for the ideas.

Statpuzz