We discuss the advantages of intentionally introducing massive missingness (e.g., 80-90% missing) into data collection designs. Planned or random missingness allows investigators to increase the total number of items explored by giving fewer items to any particular participant but more items to the entire set of participants. For inferences made at the structural level (i.e., the factor structures of temperamental, ability, interest or mood items) this improves the generalizability of the conclusions. We examine the tradeoff between breadth of measurement across sets of items versus precision of measurement for the correlation between any pair of items or for scale scores for any one participant. We discuss one example of this procedure (the SAPA project), the open source software we have used, and the results from more than 200,000 participants on more than 2,000 items.