What The Data Wants
Data speaks back. This is a thing one can say in a field biology seminar in a particular tone of voice and be understood; it is also a thing one can say and be misunderstood, and the distance between the two understandings is the subject of this chapter. What I mean by it is narrower than it sounds. I do not mean that data is animate, or that the river is telling me something in a language I can translate. I mean that a dataset of sufficient size and continuity will, if one spends enough time with it, reveal patterns that cannot have been anticipated at the design stage, and those patterns will pose questions the original study was not built to answer. This is a commonplace in any longitudinal science. It is nevertheless a strange commonplace, and I have been trying for eight years to be more precise about what it is.
My coho dataset on the Stillwater, as it stood at the end of last October, contains approximately twenty-two thousand individual observations across eight spawning seasons and fifteen survey reaches. It contains, alongside the observations, ancillary data on flow, temperature, precipitation, habitat condition, passage obstacles, and — when I can get it — dissolved oxygen, turbidity, and some trace water chemistry. It contains, by extension through the volunteers' records, otter sightings, invertebrate counts, and weather observations from eight private yards. It