Most definitely its important to understand a particular market's preferences. The decision to adjust or not, though, should be made for each assignment - not as a general rule of thumb. And the methodology for said adjustment should be quantifying support for the adjustment. To Terrel's point, if adding that variable into the regression results in janky R's or a larger residual, or low t (not testosterone), etc., then fireplace shouldn't be an element of comparison. In Texas - that will generally be the case, unless - again to Terrel's point - you could transform 'quality' into a binary variable and add that to the model. In any case, some data sets may suggest fireplace adjustment and others not. That's part of the analysis.