I have three Likert scales type surveys within one survey. One has a 0-4 scale, one has a 1-7 scale and one is a 1-6
The SLQ has a 1-7 Likert scale and is scored by adding and dividing the sum by two. See link (http://people.uncw.edu/nottinghamj/documents/slides6/Northouse6e%20Ch10%20Servant%20Survey.pdf). Scores are continuous data so it would seem like I could use Pearson's here.
The MLQ has a 0-4 Likert scale and is scored by adding o this end, the MLQ measures 5 transformational leadership factors, 2 transactional leadership factors and 2 laissez-faire leadership factors using a set of 45 questions. Each question is ranked and the overall score for a factor is calculated using the average score of the questions related to the factor being measured. This seems like ordinal, where I would need to use Spearman's.
The JSS is a 1-6 Likert scale and the scoring process for this scale is summing scores from the scale. Again, we have continuous data here.
I think I would need to use Spearman's here but I am not sure. I have seen that there is not always consensus in these methods in social science research. Thanks for your time!
Hello Jennifer,
You'd be safest in all instances using the Spearman correlation to determine strength of relationships among the various summated scale scores.
While you can find published instances in which people use summated scores from Likert-type response items as interval-strength data, it's always unknown ahead of time how well summated scores will perform as proxies for more rigorously scaled values (e.g., via polytomous item response models or via factor analysis).
Remember, both Spearman and Pearson are measures of linear association; as well, a Spearman correlation is just a Pearson correlation run on ranked scores. So, even in the best of circumstances, you do not sacrifice much information by opting for Spearman. In the worst of circumstances, you avoid mis-representation about how meaningful a given score difference may be.
Good luck with your work.
Spearman's has lower power than this alternative using van der Waerden's approach: Preprint A Robust Alternative to Pearson's Correlation for Testing As...
. However, the choice between Pearson's and any rank-based approach should be based on what you want to measure. Do you want to measure how well a straight line fits the data, and if so, use Pearson's, and depending on the distribution you might bootstrap the p values/CI.
In such Multidimensional scaling, Pearson may not be necessary again, as the pattern of responses may be affected in a certain way thereby affecting the results of the study. likewise Spearman may not do the job fr you as expected. Apply a Factor Analytic approach to get the factor loadings of these scales and then eliminate the scles with lesser loadings and then you can arrive at a Unidimensional scaling that will make your work a little easier.
To use Pearson in this kind of situation, u must have to work on the nominal level Classical Test Theory (CTT) where you sum all the responses and arrive at a value suitable for a Pearson Statistics.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey