After taking inventory of all your potential candidate variables, it makes sense to start with single variable analysis. Why complicate things by looking all at multiple variables at once when you can start looking at them one at a time? Often, the results of modeling will suggest immediate elimination of a variable for inclusion in a model, such as one with a high percentage of missing values, or one with data quality issues. Eliminating variables early is often the best course, rather than carrying them through an analysis only to discard them later on. This is especially true if you find two variables which measure the same thing.