I am performing an ordinal regression analysis with 10 predictors, reduced from an original set of 50. Sample size: N = 380. One of these predictors is the primary predictor of interest (Safety Hazards [SH], a continuous variable). The rest are potential confounders of the primary predictor or contributors to the model.

My primary outcome is Job Satisfaction (JS).

I distilled the 10 candidates from the original 50 by running ordinal bivariate regressions on each candidate.

a) A crude model [JS <-- (beta)*SH]

b) 49 adjusted models: [JS <-- (beta)*SH + any of the 49 predictors].

If any of the bivariate regressions produced a CIE of 10% or more for the primary predictor, SH, it made the list of 10, regardless of its p- value.

If any of the 50 candidates had a p- value of < = 0.1, regardless of its CIE value, it made it to the list of 10 as well.

I sort the list of 10 by CIE value and use that order to determine which variable to test next in the model.

I then begin the final fitting process to extract…