Likert data question

#1
In working with Likert data on a survey when you encounter a double entry on one question how do you handle that response?

my hunch is to throw out that response, but to keep the other valid responses from that survey subject.

similarly, how about blank responses?

thanks
jerry
 

JohnM

TS Contributor
#2
If the double entry is for adjacent response categories (i.e., they couldn't decide between a 4 or a 5), then I code it as the midpoint between them (4.5 in this case).

It's not really invalid, it's just that their attitude, from their point of view, fell in between two choices - and this is going to happen eventually - someone won't be satisfied with the precision afforded by the response choices.

If they're not adjacent, like maybe a 2 and a 4, then that could be interpreted in any number of ways, and should be thrown out (i.e., do they mean a "3" or did they choose one, change their mind, and forget to erase the other response?). Of course, a double-choice of 4 and 5 could be interpreted like this as well, but I feel that this isn't as likely with adjacent choices.

Blanks are blanks - code it as a "non-response." If you really need it, take the sum of the responses on the other questions and try to correlate that sum on everyone else's to the response on the missing question.
 
#3
thanks john,

i'll take a look at some of the forms and see what marks are there. we survey every student in every class every semester so there are lots of data. I'm trying to streamline the management of the data and get it into a useful format. we are using a scantron type system to convert the marks to data for a database/analysis.

with that in mind, suppose that i teach four classes in a particular semester two algebra and two calculus. what is your thinking on combining the dat versus keeping it separate? the surveys ask about student satifaction with instructors in regards to general classroom issues, eg fair grading, timeliness of feedback, quality of handouts etc.

thanks
jerry
 

JohnM

TS Contributor
#4
I can see where it would be difficult to deal with "stray" or "duplicate" entries when the surveys are optically scanned - I assume the scanner can be configured to handle duplicates whichever way you want? Maybe not?

I would keep them separate - you can always combine them later, but not necessarily the other way around:) ....you can answer more questions about the data when you have it separated out by different variables.

I guess it all depends on how you want to look at it.
 
#5
JohnM said:
I can see where it would be difficult to deal with "stray" or "duplicate" entries when the surveys are optically scanned - I assume the scanner can be configured to handle duplicates whichever way you want? Maybe not?
QUOTE]

Yes to some extent you can configure the scanner to handle the duplicates in a couple of different ways. I'm not sure if i could get duplicates to be reported in a way that would allow some further work, i don't know the scanner software that well just yet. i might give it a try.

i will definately keep the responses separated by class, but was just thinking about combining classes where the class sizes are small. we often have classes of 10-12 if one or two are absent on survey day then you have a sample that may not be useful to provide input as to that one class, but if several classes are combined to provide a look at an instructor overall it might provide some useful information.

thanks
jerry