Skip to content
AgriLab
Navigation

Statistical methods guide

OPSTAT-style RBD: square root and CD.

A worked example of original and square-root transformed treatment means, ANOVA error and critical difference.

Run the RBD example

What this example shows

Students searching for an OPSTAT RBD square-root analysis often need to report treatment means and critical difference (CD) on two scales. This worked example shows why the observations must be transformed and the ANOVA recalculated. The transformed CD is not the square root of the original CD.

The data below are illustrative counts for three treatments in three complete blocks. They are a teaching example, not observations from a field trial. The calculator implements AgriLab’s own analysis; it is separate from the original OPSTAT service.

1. Enter the original observations

Use one treatment per row and one block or replication per column. Set RBD, 3 treatments and 3 replications in the single-factor calculator.

Illustrative original observations
TreatmentBlock 1Block 2Block 3
T1101211
T2151614
T3202119
10	12	11
15	16	14
20	21	19

2. Choose the transformation deliberately

Select Square root: √x for this example. Apply the square root to every individual observation before computing transformed treatment means, sums of squares and error variance. For example, the transformed T1 mean is (√10 + √12 + √11) / 3 = 3.3143. It is not simply √11.

The separate √(x + 0.5) option adds 0.5 to each original observation before taking its square root. Use a transformation appropriate to the response distribution and model diagnostics; the presence of count data alone does not establish that a particular transformation is necessary.

3. Read means and CD on the same scale

RBD results for the illustrative data, using √x
ResultOriginal scaleSquare-root scale
T1 mean11.00003.3143
T2 mean15.00003.8715
T3 mean20.00004.4712
Error degrees of freedom44
Error mean square0.3333330.006675
CD at 5%1.30880.1852

For this balanced RBD, CD at 5% = t(0.975, error df) × √(2 × error mean square / replications). The error mean square is calculated separately for each scale. Here both treatment F-tests are significant at 5%; comparisons should follow the relevant ANOVA and the chosen analysis plan.

Compare an original mean difference with the original CD, and a transformed mean difference with the transformed CD. Taking √1.3088 does not produce the transformed CD of 0.1852. Squaring a transformed CD also does not give a constant difference on the original scale.

Numbers are rounded for display. The example values were checked independently with NumPy and SciPy. Check the design, assumptions and missing observations before applying the formula to another experiment.

CRD, RBD and factorial designs are different analyses

A completely randomized design (CRD) does not include a block term. An RBD accounts for a complete set of treatments within each block. Choose the design used in the experiment, rather than choosing whichever output gives a smaller CD. Factorial experiments also require the appropriate main effects and interaction terms.

Browse OPSTAT Tools · Use the methodology wizard · Open two-factor ANOVA

Method references