In questionnaire analysis, for example, four aspects of the research issue each are measured by asking the opinion of about 45 respondents with a series of related questions have 5 point Likert scales measured as "very little", "little", "normal", "much", "too much".
I got such data from someone already entered in the SPSS datasheet and I asked my help to confirm the methods used to analyze such data, and the following steps have been done
1) I see "very little", "little", "normal", "much", "too much" are entered with 1,2,3,4,5 in SPSS.
Then response scores for every aspect are measured with averaging (summing up his/her answers to let's say 5 related questions and dividing by the number of questions) to get a score that varies from a minimum of 1 to a maximum of 5.
2) then normality of the aspects (resulted averaged scores) is tested using Kolmogorov Smirnov, and the related p-values for all aspects are all less than 0.05, so the normality assumption has been rejected
3) To measure the correlation between the aspects and test its significance, the Pearson correlation coefficient has been used.
I would be appreciated it if you criticize and let me know if
a) Using 1,2,3,4,5 for measuring "very little", "little", "normal", "much", "too much" is logical or instead -2,-1,0,1,2 shall be used? (Or it makes no difference regarding further analysis asked in the following?)
b) Summing up then averaging the ordinal scaled measurement and assuming it as a continuous variable then testing its normality is correct?
c) Supposing the answer of the continuity assumption in question2 would be “yes” then, of course, using the Pearson correlation to measure the relation between aspects seems to be correct but for testing the significance what shall be done if the normality of aspects were rejected regarding the KS test. Here is it reasonable to use Spearman correlation?
(remember the number of variables used in averaging aspects, is less than 5 for some aspects and let's say 36 for other aspects).
Name please some applied reference books learning how to analyze the questionnaire.
Many thanks in advance.
Hamid
Hi Hamid,
Answering some of your questions below, noting that some people with more expertise may go into more detail in some aspects.
a) 1-5 or -2 to 2 will not make a difference in this case.
b) Likert scales are sometimes assumed to be continuous. In my experience the extent to which this is accepted is partially discipline specific.
c) Pearson checks for linear correlations while Spearman seeks to identify monotonic correlations. Both should identify and correlation and the same directin but will provide slightly different results depending on the data. For likert scales, since the distance between the items cannot be assumed to be the same I would use Spearman.
As for applied texts I really like Andy Field's books (learning statistics with SPSS may be particularly helpful here)
Best of luck with your project
In my opinion and from experience, sum the Likert Scale responses. The scoring of 1,2,3,4,5 is fine. Add scores per person, do not average. Let's say the first 6 items on the questionnaire is measuring a variable (let's say appetite). Prepare a person-by-item matrix where each person obtains a score between 1 and 5 per item. Add the scores of items 1 to 6 to obtain a central appetite score.
Pearson's correlation assumptions does not include normality. However, the data must have evidence of linearity (examined using scatter plot). There should be related pairs and there should be no outliers. Once these rules are met, Pearson correlation can be performed regardless of normality results.
If the data fails linearity, then consider the Spearman rank correlation.
- Badges
- Science topic
- Similar topics
- Correction