Data for the biodiversity experiment described in Brien et al. (2011, Web Appendix G).

The data is available in the following files that differ in their formats:

An R script that produces the anatomy for the design used is available in the following file:

A two-phase experiment consisted of field and laboratory phases. The field experiment used a randomized complete block design with four blocks to look at the effect of two tillage treatments on bacterial and fungal diversity. For each plot, soil samples were taken at the one place; two samples were taken at each of two different depths (05 cm and 510 cm). The resulting 32 soil samples were taken to the laboratory for analysis with a gas chromatograph. In this laboratory phase there were 64 runs and these were divided into two occasions of 32 runs each. During the first occasion, fractions from each of the 32 soil samples were analysed in a systematic order and then, during the second occasion, another 32 fractions were analysed in the same sample order. The two samples taken at each depth were preprocessed using two different methods (ground versus sieved). Each occasion was divided into two intervals, during each of which the fractions from 16 samples from two of the four blocks were assayed. The order of processing of the 16 samples from two blocks, A and B say, is shown in Table 5 of Brien, Harch, Correll and Bailey (2011). of Also given are the values for the response variable, a Gini coefficient computed from readings from BIOLOG™ plates taken at selected incubation times during a run.

As a means of establishing some realistic assumptions about the variation in experiments like this, an analysis is conducted to investigate the magnitudes of the sources of variation in its laboratory phase. Here we seek only to eliminate the sources of field variation and treatment differences, whose interpretation is obscured in any case by the lack of randomization. To do this, the 32 runs within each level of Occasions are divided hierarchically into five sets of time intervals according to five two-level factors, named Int1 to Int5; the factor Intj divides each interval for Int(j-1) in half. The values of these factors are given in data files. Clearly Int2 corresponds to Methods as each Int2 interval consists of 8 consecutive runs, during which 8 samples from the same method were processed. Similarly, Int4 and Int5 correspond to Plots and Depths, respectively. Int1 corresponds to the difference between Blocks 1 and 2 versus 3 and 4 while Int3 corresponds to the differences within these pairs of Blocks.

The Gini values, and the factors indexing them, are given in the files. The factors are as follows:

  1. Occasion
  2. Int1
  3. Int2
  4. Int3
  5. Int4
  6. Int5
  7. Run
  8. Block
  9. Plot
  10. Depth
  11. Sample
  12. Tillage
  13. Method
  14. followed by the response variable
  15. Gini