Hello everyone,

Im working with a variable set (N=914) that includes almost only ordinal variables except some sociodemographics.It is based on a survey which asked working people about their evaluations in regards to different dimensions effected by digitalization.

Furthermore, the variables do not have the same scale. There are some groups of variables which, for example, asked on a 4-point scale how strongly the employed persons rated the importance of certain skills in the working world of tomorrow.
Many of the variables are very asymmetrically distributed. (E.g., 90% of statements in Agree or Strongly Agree.

Other variables give assessments of the evaluation of digitization overall on a 3-point scale or the level of education on a 7-point scale.

My approach:
I applied PCA based on polychoric matrices for the above groups of variables and estimated the components.

For other variables, I either left them ordinally scaled (if they had a reasonably symmetric distribution) or transformed them into dichotomous variants. I further used LR tests to see if factor variables gave better results than the "reasonably symmetric" ordinal variables.

My problem:

Since I want to answer different questions, I use a variety of DVs.
After my transformations both used DVs and IVs include ordinal, binary and linear (PCA components) variables. Therefore, depending on the DV, I use different regression types.
However, for each hypothesis to be answered, I only use 1-2 statistical tests (f.e. one regressionstype + one non-parametric test), and typically did not try until something was significant, so p-hacking is not a big problem for me, I think.

However, the entire research process now involves a variety of different variable types + statistical methods overall, as I have always tried to address each research question "in the best way possible".

Can this lead to problems?