Hello, and sorry if this question seems very basic.
I am analysing data collected from 3 questionnaires with different scales, two 5-point and one 11-point (0-4, 1-5, 0-10).
Do I need to convert the data to one scale/is there any advantage to doing this? I need to correlate the data and am also carrying out linear regression with a mediator. Thank you
Note that statistically, "standardize" and "normalize" are two different things. Standardization is calculated using the mean and standard deviation of the original data, whereas normalization typically just rescales between two values (such as 0 and 1).
So to answer your question, standardization/normalization is important when different variables have different units and ranges. Otherwise, large values may get a bigger weighting in your data model than small ones, etc. This is quite a good article on the subject: https://towardsai.net/p/data-science/how-when-and-why-should-you-normalize-standardize-rescale-your-data-3f083def38ff
Dear Caroline,
I hope I am understanding your question. Your basic question is "Do you need to convert the data to one scale?" I might rephrase this as "can I convert the data to one scale?" The answer depends on what question(s) you are trying to answer. You indicate that you need to correlate the data. You could do that and even "equate" the scales if the data for all three questionnaires are available for the same respondents and can be assigned to each respondent. In fact, that is a limiting condition. If the data from the three questionnaires were collected from different samples or you cannot align the responses by respondent, it will be very difficulty, if not impossible, to look at any relationships among the three questionnaires. If you have common questions (for the two 5-point responses), you could compare the two populations.
The point is your ability to correlate or equate the three scales depends upon your ability to align the responses of the three scales to each individual respondent.
Good luck,
Jim
James E. McLean
David F. Nettleton Thank you David, I did mean normalise and will look at the article you kindly linked in your answer. I looks like I need to consider both standardisation and normalisation.
James E. McLean
Dear James, Thank you. The data for all questionnaires were collected from the same sample. All respondents answered all three and I am able to align the responses to each individual respondent. Apologies. I should have specified that.
Normalizing can have a few meanings, but often it is to make closer to the normal (Gaussian) distribution. van der Waerden showed in the 1950s how this improved power for things like t-tests, and here (Preprint A Robust Alternative to Pearson's Correlation for Testing As...
) is looking at this for correlations and SEM. But, if you are looking for the relationship among the three, I would start with scatterplots with the untransformed variables. Putting on a similar scale is not necessary for all research questions. And neither is the typical standardizing method which leaves the mean and sd for each to be the same. It will be useful if you spell out in more detail what your research questions are.
Daniel Wright Thank you Daniel. I am looking at the relationship between self-compassion and subjective well-being in the caregiving/informal carer population, whether any relationship between them is mediated by emotion regulation, and, if it is, the extent of that mediation. I have run all the analysis, using SPSS and Andrew Hayes Process macro, but without normalising / equating the scales. As you may have guessed I am not a statistician!
Equating the scales is a different thing too. Rather than using the words normalizing, standardizing, equating, etc., it will be easier for people to address your question is you write down the equation that you are thinking of using. Also, when examining mediation (i.e., analysis of covariance) it is often valuable to draw the a set of constructs [ones you measured and otherwise] with arrows denoting directed relations. The issue of transformation is somewhat tangential to this, however.
I think that the data should be standardized in order to compare the questionnaires
Dear Caroline,
I often develop various ML & ANN models, and I had such tasks as yours. And as my experience shows, there is no strong benefit from standardization (normalization, MinMax scaling et.) for your data if you want to build models based on ML & ANN.
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere
- Air