Hi.
Recently, I submitted the manuscript to an journal, and
I got a few comments from the reviewer.
My data included in the manuscript was semi-quantitative scoring data of IHC,
and I handled the data as numerical, but the reviewer said it is ordinal,
I cannot understand it.
It is data ranging from 0 to 6, should it be handled as ordinal variable?
Also, I used linear regression analysis to examine
if there is consistent relation between
two variable (IHC data of A and B) in each sample,
and the reviewer said I should use
Spearman ranks correlation test instead of linear regression analysis.
I also cannot understand this, could anybody can explain to me?
Wish you a good day.
Thank you !
Hi,
It all depends upon on which scale the variable was measured.
Nominal scale: when each of the categories is coded 0 to 6 and these numbers only represent a particular category and nothing more..
Ordinal scale: there is a heirarchy of values in ascending or descending fashion..Here a lower and higher point is not determined
Interval scale: When a subject can assume any of these values from 0-6, the Distance between one point and the next is there..
If your variable was an ordered variable then for correlation Spearman's Rank correlation is appropriate.
Linear Regr. ana is done when the dependent var. is on an interval scale.
The "measuement scale" explainedby Bhogaraju Anand is not the most relevant part. The most relevant aspect is the assumed distribution of the variable you are analyzing.
Linear regression is based on the assumption that the variable has a normal distribution (conditional on the values of the predictors, with an expectation as a function of the predictors and the variance being independent of the predictors).
This cannot be a reasonable assumption for your scores. They are bounded (0-6), but the normal distribution is not. The variance must depend on the mean (if the mean is close to a boundary, the values cannot vary much), but this is not the case for the normal distribution.
Sometimes such issues can be solved using a transformation. In your case (doule-bounded variable in (0,6)) this might be something like logit(Y/6). But one should check this (symmetry, linearity, homoscedasticity).
The reviewer was suggesting a different approach:insteadofmodelling a linear relationship between X and Y he asked to check their monotone association (in how far are values of Y consistently larger (or smaller) when the values of X are larger? -Without making any statement about their functional relationship or about the conditional distribution of Y).
https://www.researchgate.net/post/Can_we_use_Likert_scale_data_in_multiple_regression_analysis
@Kim follow the RG thread 👆🏼
Your data are categorical and you are not admitted to use the methods based on normality. Overall, always do check out the presumptions of any model you want to implement. Finally, the comment of professor Wilhelm is great apply it.
There is a problem with Likert scales which I have addressed and faced by psychologists,as it is regarded as an ordinal by some and on a continuous scale by others.
Based on those the reviewer might have suggested the Rank Correlation test.
Now the researcher has to either agree with the reviewer or give a befitting reply.
I was not talking about statistical distributions or assumptions for the other tests.
- Badges
- Science topic
- Similar topics
- Linguistics
- Philology
- Classical Philology
- Manuscripts