Good morning
I am a Stata newbie and am trying to clean the data set of a survey which I recently conducted using LimeSurvey. One of the questions involved a 19 part psychometric test (Basic Psychological Needs satisfaction scale), with each respondent having to tick one of the following responses for each question:
1 = Not at all true
2
3
4 = Somewhat true
5
6
7 = Very true
The response data arrived into Stata in string form so I destringed it using
encode BPN_1, generate (BPN_1_d)
encode BPN_2, generate (BPN_2_d)
etc for all 21 questions in the scale
While the desstringed version of the response data (i.e. BPN_1_d etc) looks fine on the surface i.e still shows 1= Not at all true or 3 or 5 etc in line with the original data entry, when I click on each cell, I can see from the bar at the top of the page that as part of the destringing process Stata has 'recoded' all the original responses using its own (alphabetical??) system so that an original 1 response is coded by Stata as a 2 etc.
I was planning to fix this by going either of the following routes:
label define BPN_order
label values BPN_1_d BPN_order
or
replace BPN_1_d = 1 if BPN_1_d == 5 | Burnout1_d == "1: Strongly Agree"
but the problem is that Stata appears to have recoded a 3 response as a 4 for some of the 19 sub-questions, as a 5 for others and a 9 for others i.e. there doesn't appear to be any consistency as to how Stata has recoded the original responses when destringing.
Has this happened to anyone before and does anyone have any suggestions as to how I might overcome this problem or maybe de-string the original data in a fashion that might avoid it?
Many thanks
The first thing I used to tell my dissertation students was to find a question worth knowing the answer to and plan how you will attack it. As several military organizations remark, "Proper PRIOR Planning Prevents Poor Performance". Your difficulty has nothing to do with STATA and a lot to do with research Design. If you wish to describe what your research question is and give details of what you have done so far, someone here may be able to advise you. Good Luck, D. Booth
Destringing questionnaire responses is a major irritation when you use a service like Survey Monkey to collect data, so I am well familiar with your problem.
The trick is to install Roger Newson's incredibly useful command sencode.
. ssc install sencode
Next define your label :
. lab def agree 0 "Strongly disagree" 1 "Disagree" 2 "Unsure" 3 "Agree" 4 "Strongly agree", modify
(I put the ,modify option so that when you run this in a do-file it will overwrite any pre-existing label.)
Now time to turn the awesome power of sencode onto the problem :
. sencode q1, label(agree) replace
YOu can embed sencode into a loop:
foreach var of varlist q1 - q24 {
sencode `var', label(agree) replace
}
Note the open and close single quotes around `var' when it appears in the loop.
Note that unless the label exactly matches the text of the responses, sencode will generate new label categories for the unmatched responses.
Check this with
. lab list agree
agree:
0 Strongly disagree
1 Disagree
2 Unsure
3 Agree
4 Strongly agree
5 Strongly Disagree
Note that in this case, at least one question had a capital D in the response for strongly disagree, and sencode (rightly) created a new value label to deal with this.
And finally, I feel your pain!
Ronán Michael Conroy
Hey there, i´m using SPSS and converting strings to numeric Variable is always a pain in the *** , maybe sometimes the order is correct but mostly not.
So do that macro also work for SPSS, or do you know some?
Thanks
Martin Krupinski, note that SPSS has an AUTORECODE command you can use for this. Here's a simple example.
* Generate a small dataset to illustrate.
DATA LIST LIST / q1(A20).
BEGIN DATA
"0 Strongly disagree"
"1 Disagree"
"2 Unsure"
"2 Unsure"
"3 Agree"
"3 Agree"
"3 Agree"
"4 Strongly agree"
"4 Strongly agree"
"4 Strongly agree"
"4 Strongly agree"
"5 Strongly Disagree"
"5 Strongly Disagree"
"5 Strongly Disagree"
"5 Strongly Disagree"
"5 Strongly Disagree"
END DATA.
STRING q2 (A20).
COMPUTE q2 = q1.
EXECUTE.
* End of data generation.
* The bit you need starts here.
RENAME VARIABLES (q1 q2 = q1str q2str).
* Here is the key command for what you want to do.
AUTORECODE VARIABLES=q1str q2str
/INTO q1 q2
/PRINT.
* Check that it worked.
FREQUENCIES q1 q2.
* Show that the original and recoded variables are in the file.
LIST.
* If you no longer need the string variables, you may wish to delete them.
DELETE VARIABLES q1str q2str.
LIST.
Bruce Weaver AUTOREEEEEEECOOOOODE
Yes I know. Thats the Reason I'm asking for a better solution.
Thanks.
Martin Krupinski, what is so onerous about AUTORECODE? Can you upload a small dataset that illustrates your pain? ;-) (If you make it a CSV file, people who use Stata will be able to open it easily too.)
Autorecode doesn't switch a string not that good to a ordinal variable, the order of the scale points is random.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics