Hi,

MLR = Multiple Linear Regression?... Or... Multinomial Logistic Regression? :-)

Running linear regression in your case with dv = days and iv = exposure, the corresponding coefficients will give you the additional (or reversely) days due to the influence of one factor, all other variables kept unchanged (check for the statistical significance of the model). Somehow useful, can lead to hazardous conclusions.

However, you may want, if feasible*, to group days in intervals, create dummy variables for each and run logistic regression instead - will give you probabilities (odds ratios) for the exposure to trigger a group.

*Careful: if dv is censored, regression is not an option!!

I don't recommend multinomial logistic regression in this case since it is computationally complicated and results (RRR instead of OR) are harder to interpret.

Cox (proportional hazards) regression is something different, being the choice for survival type analysis. It predicts nothing, but it's explaining the relationship between the time variable and the explanatory variables - "time to event". Cox's unit is 1/time (log(Hazard rate)). Regression has no unit.

For better understanding of the model, fit the Cox with the hazard-ratio parameters displayed instead of B and plot the corresponding Kaplan-Maier curve, for intuition.

Regards,

A good reference is by Frank Harrell. See the link below for the cite:

https://www.amazon.com/Regression-Modeling-Strategies-Applications-Statistics/dp/3319194240/ref=sr_1_1?s=books&ie=UTF8&qid=1518656527&sr=1-1&keywords=frank+harrell

Prof Popa has given a good but terse summary. The book by Harrell Gives more details and examples. I would look there first. Notes from the book are at:

http://biostat.mc.vanderbilt.edu/tmp/course.pdf

These are very good.

Best, D. Booth

Thank you Prof Cristian and Prof David for your response! They are truly helpful.