Hi Alias

You cannot convert a Likert scale to interval - however, you may be able to treat a Likert scale as an interval variable. I say may, because there should be at least 5 points on the scale. For parametric analysis, the usual assumption about Normality needs to apply. However, you can always use (and many researchers say you should always use) non-parametric analyses. 

Adrian

Dear Alias,

If you use SPSS, it would be much easier. When a SPSS data set is opened, there are two tabs in the bottom: Data View and Variable View. Click the "Variable View" tab and then go to the "Measure" column. I assume you will see "Ordinal" there, since your data are entered from a Likert scale (e.g., 1, 2, 3, 4, or 5). You simply choose "Scale" and then save the data file. Once you have done this, you can conduct any parametric analysis.

Cheers,

S-K 

The quality of life QOL section of a clinical trials is analyzed under a standard manner: conversion of the likert scale into a continuous variable. How?

You should list for the likert scale(s) you want to analyze the frequencies for each level. For example, 0 correspond to 19.54, 1 to 23.56, 2 to 32.57, 3 to 7.68 etc. Of course the total % should be max 100.

Since you have this distribution from your data, you will create a new variable for that likert scale which for each subject will take the value 23.56 for 1, then 7.68 for 3 etc.

You may standardize these value and you will get less variance.

The goal is to analyze the new variable as a continuous and much more: close to a normal distribution (you may check using Shapiro-Wilk test for normality, for example). You should do this step with all the likert scale variables that you want to analyze!

You need to pay attention to the expected outcome, and if you have repeated questionnaire in time, with likert scales then you will deal with correlated data...(GEE for example).

I recommend SAS for a statistical package which can handle very clearly the issue. 

The problem with transformation of Likert scales to interval scales is what the length of each interval must be. This is not the same for each scale and very much depends on the number of response options and the labeling of these options. An overview of transformations methods, including a method to assess the intervals that corresponds to each response option in the context of the scale can be found in the following paper.

Article Homogenizing Responses to Different Survey Questions on the ...

Yes Tineke, it is correct what you say. You also wrote a wonderful article. Since the distance between the likert scale values cannot be estimated correctly, i.e. agree and strongly agree, pending always on the type of variable, it is wise to consider the frequencies of each level in the database and then use them as values (interval variables) for the new created variable, the "mirrored" one from the likert scale variable. This is a practical method used in clinical trials and accepted by FDA. Of course we may over use the distance between agree and strongly agree, for example, and create a biased  standardization, but it is a direct way for the analysis. 

Usually some corrections come from the standardized surveys which uses the similar likert scale questions.

If the survey type is a brand new, then we don't have a standardized survey and we should consider the survey results as the above suggested methods, without using any standardization toward a national survey results. 

This is coming from the practice of Clinical Trial.