Marilyn Matevia <[email protected]> asked,
> with many nested skip patterns, and about 2200 respondents. After working
> out the skip codes, I still have many "missing" observations for some items,
> and figure these are due to incomplete surveys. I want to create some kind
> of "completion flag" to see which respondents make it to the end of the
> survey, and - if they don't - to see which question is the respondent's last
> (which is complicated by the skip patterns).
One solution is to create a set of variable that indicate which questions a
respondent should have answered, compare that with results, and then work
backwards.
Step 1: Which questions should have been answered
--------------------------------------------------
Due to the skip patterns, questions occur in blocks. Let's call the
blocks B1, B2, ...
What we are going to do is create a set of dummy varibles, sB1, sB2, ...,
corresponding to the blocks. If a respondent should have answered the
questions in block i, sB<i> will be 1; otherwise, it will be 0.
We start by setting all sB<i> variables to zero. To fix ideas, we will
pretend there are 23 blocks. We start by setting all sB<i> variables to 0:
. gen byte sB1 = 0
. gen byte sB2 = 0
. ...
. gen byte sB23 = 0
or, if you prefer,
. forvalues i=1(1)23 {
. gen sB`i' = 0
. }
Now, it is natural to assume the all respondents should have answered
B1:
. replace sB1 = 1
Who should have answered B2? That depends on answers in B1.
Assume responses to questions are in variables q1, q2, ..., q170.
Perhaps a person should have answered B2 if
. replace sB2 = 1 if q2=="m" & q3>1 & q3<4
We proceed through the survey in this manner. Perhaps the code for the
last block reads
. replace sB23 = 1 if q40==3 & q102=="5"
Step 2: whether they answered a block
--------------------------------------
Let us now create another set of variables aB1, aB2, ..., aB23 that
reflects whether questions were answered in a block:
. gen byte aB1 = (q1!=. & q2!="" & q3!=.)
. gen byte aB2 = (q4!=. & q5!=.)
. ...
. gen byte aB23 = (q167!="" & q168!=. & q169!=. & q170!=.
Step 3: Compare
----------------
A respondent has missing date for B<i> if aB<i>==0 & sB<i>==1.
. forvalues i=1(1)23 {
. gen byte mB`i' = aB`i'==0 && sB`i'==1
. }
Step 4: Cleanup
----------------
We may not need aB<i> and sB<i> variables any more:
. drop aB* sB*
Step 5: Last block answered
----------------------------
. gen lastblock = .
. for i=1(1)23 {
. replace lastblock = `i' if mB`i'==1
. }
-- Bill
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
