I have an experiment I am trying to analyze using ANOVA (using R). It was set up in a split-plot design, with the main factor as "treatment" and the split-plot factor as "sub_plot." The main treatment had 5 treatments (f-j), and each treatment had 4 split plots (1-4). The response variable was "result." An example of my dataset is provided below. Note that this is not my real data, but an example of what I am working with.

Despite various data transformation attempts, the data will not pass the Shapiro test, the test I am using to test the normal distribution. I have used the log transformation, square root, cubed root, and the Box-cox transformation. None will pass the test for normality.

**What non-parametric alternatives are there to analyze a split-plot design?** For a 1-way ANOVA, I use the Kruskal-Wallace test, but this is mainly for randomized complete block designs. I have not analyzed split-plot designs that do not meet the assumption of normality.