19th February 2014 at 10:06 am #1211
I am using likert scale to measure the frequency of occurence of causes of ineffective public spending. I chose the 5 scale of very frequently, frequently, occassionally, rare and very rare. I have used very frequently to mean 5 times or more than 5 times and very rare to be one or less than one. However the choice of the number 5 is not backed by any previous methodology. I am therefore requesting for direction on this bit of my questionaire. What measures exist on occurences?22nd February 2014 at 6:54 pm #1219Ingo RohlfingMember
I am not from this field, but it just seems like a subjective decision of the type one makes very often in empirical research. Why don’t you ask people about the number of occurrences? You could only use this info, or you gather it in addition in order to check whether your results derived with the Likert scale are robust to the use of continuous information.24th February 2014 at 4:45 am #1218
Thank you Ingo for taking time to reply to my question. Let me reiterate what you have suggested to see whether we are on the same page. You are suggesting instead of the likert scale i simply ask the respondent to indicate the number of occurence. And compare this results with those provided by respondents that used the likert scale to answer the same question. This is in order to check for the robustness of the likert scale.24th February 2014 at 8:37 pm #1217Ingo RohlfingMember
Basically, that’s what I meant. You may have reasons why you use the Likert scale, but I was wondering why you do not ask for the number of occurrences instead. If you have reasons for using the Likert scale, the question on occurrences could at least help you to check the number of occurrences that fall under the category “vrey frequently”. Maybe you find that for most respondents, the number is 5 or only 6, thus indicating little heterogeneity within this category.25th February 2014 at 5:46 am #1216
To make my questionnaire small i lumped a number of questions together in a table and wanted the respondent to take less time filling the questionnaire.12th March 2014 at 2:51 pm #1215Katie MetzlerParticipant
I think what you’re asking is, what evidence is there to support using a 5 point scale over a 7 point scale or 4 point scale or 10 point scale. Interestingly, this thread came up on a mailing list I am a part of recently and there were some helpful replies, though the consensus on the mailing list seemed to be (and these are mostly experienced survey researchers) that there is not that much robust evidence to suggest that 5 or 7 or 10 are ‘better’ than one another.
http://www.methodspace.com/forum/topics/5-or-7-point-likert-scale-d… This previous post on methodspace might also be useful and there is a link to a paper on this topic included.
Katie13th March 2014 at 4:49 am #1214
Thanks Katie this is helpful.13th March 2014 at 7:24 pm #1213Dave CollingridgeParticipant
I think Ingo brings up an important issue.
Does your scale tell participants that “frequently” represents “more than 5 times” or is that something you’ve assumed? Similarly, does your scale tell participants that “very rare” represents “less than one”? There is an important distinction between telling participants what the levels represent and assuming as a researcher what they represent.
Here are some general issues related to scaling. If the levels of a Likert scale are not continuous (i.e., numbers) then the scale is ordinal in nature. This is what we usually have when our scaling is something like very frequently, frequently, occassionally, rare and very rare. Analysis of data from these sorts of Likert scales should be nonparametric, except in cases where we aggregate questions into single scores in which case it is fine to use parametric statistics.
When working with researchers on surveys, I often steer them toward continuous scaling. One way to accomplish this is to identify the end points of the scale with, for example, “very frequently” and “very rarely”. The middle points are non-defined vertical hash lines (3 hash lines if a 5-point scale and 5 hash lines if a 7-point scale).
On the other hand you could ask for the number of occurrences. That would be a ratio (continuous scale).27th March 2014 at 7:38 pm #1212Steve MoranMember
Put yourself in the respondents shoes. When people describe preferences they intuitively use
league tables 1st 2nd 3rd 4th 5th etc
which is the opposite to 5 having more weight than 4 ..etc….1
this can easily be changed just before statistical analysis.
I agree with the comments about the data NOT being continuous and therefore
use non-parametric methods at the outset at least
After Non-Parametric testing…
One could also argue that the choice of 2 say, is representative of all values
between 1.5 and 2.5. Similarly for all other numbers. In which case it could be further argued that the sample of integers
you collected is a subset of a population of all the real numbers from 0.5 to 5.5. At that point you can use the resample
facility to generate a continuous sample of real numbers not integers to represent the population from which
your sample came. Confirm normality with appropriate tests , use Normal dist and t-tests etc subject to normality..
I’ve also seen a questionnaire where the respondent was asked to put a mark on a straight line between nil and high..
researcher simply measured where the mark was.
Oh yes… when you design the Questionnaire get it proof read several times
- You must be logged in to reply to this topic.