My survey has alredy been completed. I have 1 DV and 1 UV and 3 moderating variables,all variables were measured using a 5 point Likert except the DV. I need to transform the DV into a 5 point Likert scale, using Jamovi.
Dear Samantha,
Greetings!
Thank you for highlighting an important issue, especially concerning your study.
To effectively convert a 7-point Likert scale to a more concise 5-point scale in Jamovi, utilise the "Transform" feature to recode the existing variable. This process not only streamlines your data but also enhances its clarity for analysis. Begin by creating a new variable and use specific conditions and calculations within the "Transform" function to accurately map the 7 points to 5. A commonly adopted approach is to collapse categories: merge ratings of 1 and 2 into 1, retain 3 as 2, keep 4 as 3, combine ratings of 5 and 6 into 4, while maintaining 7 as 5. This enhances interpretability for both you and your audience.
To effectively map the 7-point scale to a 5-point scale, you could consider the following conditions:
1. Assign a value of 1 if the original score is less than or equal to 2.
2. Assign a value of 2 if the original score is 3.
3. Assign a value of 3 if the original score is 4.
4. Assign a value of 4 if the original score is either 5 or 6.
5. Assign a value of 5 if the original score is 7.
These conditions can be implemented using the "+" button to add them as required. Please review the new variable to confirm that the values have been accurately recoded. Your attention to this detail will help ensure the integrity of the data.
N.B.: While this method is simple, be aware that collapsing categories may result in some loss of detail in your data.
Thank you for your attention to this matter.
There is no need to transform the variables to a common scale for a regression/moderation analysis. Regression-type analyses do not require the independent or dependent variables to be measured on the same scale. In fact, it would actually be problematic to turn a 7-point scale into a 5-point scale. You would increase the amount of measurement error and loose information. It would also not be clear which categories should be collapsed.
The first reply from Ripon Kumar Sarkar seems to have been written by an AI that doesn't know nearly as much about regression as Christian Geiser does. Sigh...
Dear Professor Morgan and Geiser,
Thank you for your feedback. Although we are communicating virtually, I would be happy to meet with both of you once! You are correct in noting that I relied on AI to help draft my response, but I want to ensure that the content is technically sound. Dear Professor, I am still learning about Likert scales and regression analysis. Therefore, I welcome corrections from Scholars like you. It is universal that collapsing categories may result in some loss of data and increase the amount of errors. I have responded only to the question posted by the researcher, Samantha Thoß. Hence, my suggestion/answer may not be accurate or perfect, and both the author and professionals could also ignore it. In today’s world, resources like AI, YouTube, and Google are an integral part of our lives, making it common to utilise them. However, could you please mention which part of my response was misleading? I would appreciate the opportunity to improve my understanding. I am always eager to learn the best practices in this area and refine my thoughts. As a LEARNER, I welcome feedback on various aspects and would also appreciate guidance to prevent any public confusion.
Thank you for your valuable time and kind consideration.
Best regards
Ripon Kumar Sarkar
Ripon Kumar Sarkar Your AI came up with a way to do the recoding, with unstated assumptions about which categories to collapse. But the main issue is that there is no need for the variables in a regression to share the same response categories. For example, think about using education to predict income, where education is scored as years of schooling and income is in thousands of dollars.
Dear Professor Morgan,
Thank you for your note. I don't think there is much of a difference between your expectations and my suggestion. I also know that predictors don't have to have the same response scales for regression analysis to work. I suggested it, though, because the author needed help changing a dependent variable from a 7-point Likert scale to a 5-point scale. I hope my explanation clarifies the issue.
Best Regards, Ripon
The key point is that the author did not need help changing from a 7-point Likert scale to a 5-point scale because this was entirely unnecessary.
Thanks a lot for your input! What I’ve learned so far is that I don’t necessarily need to transform the variables when running bivariate correlations, since those are scale-invariant. However I’m planning to run a hierarchical moderated regression. Are you sure that I dont need to transform?
In my opinion, if your DV diagnostics (histogram, QQ-plot, etc.) are okay, it's not necessary. Transformation may be necessary only when your DV is significantly skewed, there is a heteroscedasticity issue, and you are also experiencing issues with linearity and normality. Then you could also try GLM. Thank you.
Best, Ripon
Hierarchical regression is no different from ordinary regression, it is just a system for entering the independent variables. The same goes for moderation -- if you are willing to assume that your variables are linear for the purposes of running a regression in the first place, then there is no problem with multiplying them together to form interaction effects.
- Badges
- Science topic
- Similar topics
- Higher Education
- Universities