My question stems from a recent webcast by Brady Brady. Although he didn't specifically review QR in the recent webcast he did briefly mention it, This caused me to start thinking about its applicability outside of the social sciences and socioeconomics areas. One area where people are extremely interested in predicting extreme percentiles is reliability and life analyses. I always tell my coworkers that we shouldn’t be concerned about the mean time to failure, not MTT(B)F metrics either as they can be misleading, as we’d be out of business by then. We need accurate predictions for the early life quantiles such as the 5th or 10 th percentiles.
I can get my head around QR for social sciences as they are often influenced by ‘latent factors’ that might evidence as differences in the quantiles. At least that’s the justification I’m making in my mind but I’m trying to make a justification for more ‘empirical’ systems. I thinking one of the possible benefits is that I wouldn’t be making a distributional assumption that could be inducing bias.
I’m taking some extra time at work with a current data set to look at quantile regression and compare it to parametric survival methods as well as the Generalized regression with a Weibull personality. Basically, I was just wondering anyone else has experience with this or thinks there's merit to investigating QR for reliability applications or not? I’d appreciate any thoughts.