forum.smartpls.com

I have a categorical variable (six values) that measures the time that it takes for the primary application or service to restore when a service incident occurs (1 More than 6 months, 2 Between one month and six months, 3 Between one week and one month, 4 Between one day and one week, 5 Less than one day, 6 Less than one hour).
I also have a single-item continuous variable (value as a whole percentage) for deployment automation: For the primary application or service you work on, what percentage of your deployments are automated?
All other variables are measured using 7 point Likert scale items. So, I have two questions for analysis:
(i) can I have both 6 point categorical and 7 point Likert scale items?
(ii) do i need to split the continuous variable (value entered as a whole percentage) or can i use it as it as with other Likert scale variables?

Thanks for your response.

Generally, combining different variables with different scales is not a problem in PLS because of the routine standardization of input data.

However, PLS as does regression analysis or any other correlational method assumes metric data.
Most people assume that their Likert data is at least quasi-metric in that it satisfies equidistance between categories. Then it may be used even though it is not optimal, but usually the difference to more sophisticated ordinal variable methods is quite small. If the assumption of metric data are strongly violated then results will also become less trustworthy.

Whether your service incidence variable satisfies equidistance between categories is questionable. Thus, you may need to recode that variable into a dummy scheme or so if you cannot assume equidistance.

What to do incase our endogenous variable is in 4 point scale and exogeneous in 5 point scale. Do we need to do anything or we can run smart PLS with the data ?????

As PLS standardizes all data, differences in scales are not so important. However, the whole analysis rest on the assumption of metric or at least quasi-metric data. With 5 or 7 point Likert scales researchers often assume quasi-metric data (and equidistance of the response categories). However, some researchers are arguing that this assumption is already quite strong for 5 point Likert and that 7 point would be better.
I don't know on what kind of scale your 4 points are based, but if you can make quasi-metric assumptions, it is not problem to include it in PLS.

Thanks for the clarification.