I am designing a questionnaire and am using a mixture of multiple choice and Likert scale questions. For most of my questions, a 4-point Likert scale is the most suitable but I have two questions where I think a 5-point Likert scale would be more appropriate as the questions need the option of a neutral answer.
So, my question is whether it is ok to use both a four point and five point Likert scale in the same questionnaire?
I don't know of any problem with using Likert-scored items of different lengths, but if you are working with single items (rather than multi-item scales), you will need to use non-parametric statistics for your analyses.
Hi Georgina Ph. What type of analysis are you intending to perform? If you are focusing on performing strictly item analysis, then no, the number of rating options is not an issue. However, if you are intending to add the item rating scores to produce an overall Likert score that falls along a Likert Scale, then, yes, it is probably a problem. This could be addressed using advanced IRT techniques, however.
David Sul Why would having different item lengths for multi-item scales be a problem? Wouldn't it make sense to just standardize each of them to a mean 0 and standard deviation of 1?
Hello David L Morgan, the Likert ratings adhere to an ordinal level of measurement where the mode is the appropriate measure of central tendency. For a practical example, let's consider how the Olympic medal results are shared with the public. In the Olympics, the first, second, and third place winners receive a gold, silver, and bronze medal, respectively. The published results of medals won by nation do not include the average medal rating. Rather, the mere count of each medal rating is provided, country by country, (typically listed in descending order according to the overall total number of medals won).
To address this issue head on, in this instance, where 1=gold, 2=silver, and 3=bronze, what would having a mean of 2.3, mean, so to speak? The Likert item rating options (e.g., strongly disagree to strongly agree) are of this same form. We assign numbers to these rating options but the numbers we assign to them really represent the ranked order of the rating options.
This is the significance of Prof. Likert's scale development work. By summing groups of item ratings, Likert found a way to convert the ordinal-level information into interval-level information. Within the interval-level environment is where the use of means and standard deviations is more appropriate.
David Sul Yes, of course the individual items are ordinal, but multi-litem scales with sufficient reliability are typically treated as interval. The two item scale would be problematic in this regard, but still not all that unusual.
- Badges
- Science topic
- Similar topics
- Psychology