need help with the questions in the pic

There's something I've always wondered about: Let's say we've done a survey where we asked respondents a bunch of questions, either using a written questionnaire or over the telephone. So the method for collecting every response is the same. When we do a regression analysis we designate one question (or a scale constructed from several questions) as the dependent or response variable (D), and we designate other questions (or scales) as the independent or explanatory variables (X1. X2, etc.). But then in ordinary least squares regression we consider ] to be stochastic (measured with error), yet the X's are considered to be fixed and known (measured without error). Since the data collection method - the method of measurement - was the same for all variables, this seems contradictory. Moreover, a different researcher could let one of the first researcher's X's be her Y , so how can it be right that the same variable is treated as a random variable by the first researcher and treated as fixed and known by the second researcher? What possible statistical justification is there for this practice