Interpreting means from likert scale data

[Adrian Seabrook]

Hello,

I work for a government department in Australia.

The department runs an annual opinion survey with 5 point likert scale
response items.

The analysis of this data includes calculating a mean of responses using the
following values.

Very dissatisfied - 0
Dissatisfied - 1
Neutral - 2
Satisfied - 3
Very satisfied - 4

So, typical mean scores for any item might be 2.2, 3.3, 3.25 etc etc.

My question relates to the guide to interpretation the department provides
to users of the data.

The guide is as follows -- 0 - 0.8 very dissatisfied, 0.81 - 1.6
dissatisfied, 1.61 - 2.4 neutral, 2.41 - 3.2 satisfied, 3.21 - 4.0 very
satisfied.

My problem with this guide is that it seems to favour the extremities for
example - equal numbers of satisfied and very satisfied responses result in
a mean of 3.5 which according to the guide indicates very satisfied!!

My thinking is that as satisfied is 3 and very satisfied is 4 then 3.5 would
be the critical point in the guide not 3.4.

Is there a standard/accepted way of interpreting this type of data?

Where can I find this information?


Thanks

Adrian

PS. I am aware that there is debate around the legitimacy of carrying out
this sort of data analysis on ordinal data -.. but am interested nonetheless.

[JohnM]

Adrian,

Welcome to the forum.

Unfortunately, no, there isn't a widely accepted standard for interpreting this kind of data. In addition to the dubious scaling that you mention (I especially like the way they set the lower ends of "satisfied" and "very satisfied" at 2.41 and 3.21 respectively ), there is also the huge debate between those who think that Likert scales are ordinal and those who think it's OK to treat them as interval scales.

You may want to find out why they set the scale category ranges the way they do, however. There may be a valid reason.....

Also - find out, in writing, what the exact objectives are for the survey - it may be that the reason the survey is being done is something that won't be adversely affected by the type of analysis run on the data.

Another method, followed mostly in market research, is to report the % of respondents that check the "top box" or "top boxes." In other words, the % that respond as a 3 or 4 on each survey question....just something to think about...

For example, in my job, every November I send out an annual customer survey (Likert scale 1-5, treated as an interval scale) to my internal customers. Treating / analyzing the data as an interval scale has never caused me to reach an incorrect conclusion - I've been pretty successful in using it to identify customer service issues and fixing them, so I'm pretty much a pragmatist on this whole issue....

[quark]

It seems that they set the cutoff points at intervals of length 4/5=0.8 because there are 5 categories and the range of the data is 4.

The article below may be helpful. Here is a quote: "Likert scale data can, in principle, be used as a basis for obtaining interval level estimates on a continuum by applying the polytomous Rasch model, when data can be obtained that fit this model."

http://encyclopedia.thefreedictionar...Likert%20scale

Just my 2 cents.

[JohnM]

Yes, but that makes it possible to score an overall "very satisfied" with an average as low as 3.21 - which would, in most cases, imply that most, or at least many, respondents were NOT "very satisfied."

It just "feels" like they are trying to bias the scale upwards - at the low end of the scale, in most cases, people would be reluctant to give a score of "0."

...and at the upper end, an average score of 3.21 is basically the same as "very satisfied."

[quark]

The intervals are a "convenient average". There should be a better way...