How the treatments were assigned determines the analysis
In a completely randomized design (CRD), assign treatments at random across all experimental units. It can suit a pot trial when you have reasonably comparable conditions and no planned groups.
In a randomized complete block design (RBD or RCBD), group similar experimental units into blocks first. Randomize treatments separately within each block, with every treatment represented. In a field with a fertility gradient, you could group nearby plots with similar conditions and repeat the full set of treatments in each block.
A replication is an independently treated experimental unit. Three leaves measured on one treated plant are subsamples of that plant. Keep those measurements together when preparing your analysis.
Worked CRD example with three treatments
Suppose you randomly assign three doses to nine separate pots and measure plant height in centimetres after four weeks. Each dose is applied to three pots. In the calculator, choose CRD, enter three treatments and three replications, and keep the original scale.
| Treatment | Pot 1 | Pot 2 | Pot 3 | Mean |
|---|---|---|---|---|
| Control | 8 | 10 | 9 | 9 |
| Low dose | 11 | 10 | 12 | 11 |
| High dose | 12 | 14 | 13 | 13 |
Paste the three pot measurements for each treatment, one treatment per row. Leave out the labels and the mean column. The treatment mean square is 12 and the error mean square is 1 with six error degrees of freedom. The treatment test gives F(2, 6) = 12.00, p = 0.008. The critical difference at 5% is 1.9979 cm, using t(0.975, 6) × √(2 × 1 / 3).
How the data table changes for RBD
Keep one treatment per row, but give each column a block identity. In every row, column 1 must contain the plot from block 1, column 2 the plot from block 2, and so on. The analysis includes a block term that uses this alignment. Keep the original block labels in your source spreadsheet.
These examples use balanced designs. If you have a missing plot, unequal replications or a different blocking layout, check which model can handle your data. Enter zero only when zero was observed.
Load the worked RBD example with square-root transformation