Kruskal-Wallis test on linearised series and SA series

There is an error in seasonal Kruskal-Wallis test on linearised series exported by RJDemetra : it is not computed on the differentiated series.

y <- RJDemetra::ipi_c_eu[,"FR"]
mod <- RJDemetra::jx13(y)
y_lin <- RJDemetra::get_indicators(
    mod, 
    "preprocessing.model.y_lin")[[1]]
seas_tests <- RJDemetra::get_indicators(
    mod, 
    c("diagnostics.seas-lin-qs", "diagnostics.seas-lin-kw", 
      "diagnostics.seas-lin-friedman"))
seas_tests$`diagnostics.seas-lin-qs`
#> [1] 685.5168   0.0000
#> attr(,"description")
#> [1] "Chi2 with 2 degrees of freedom "
rjd3toolkit::seasonality_qs(rjd3toolkit::differences(y_lin, lags = 1))
#> Value:  685.5168 
#> P-Value:  0.0000
rjd3toolkit::seasonality_qs(y_lin)
#> Value:  713.2462 
#> P-Value:  0.0000

seas_tests$`diagnostics.seas-lin-friedman`
#> [1] 298.3385   0.0000
#> attr(,"description")
#> [1] "Chi2 with 11 degrees of freedom "
rjd3toolkit::seasonality_friedman(rjd3toolkit::differences(y_lin, lags = 1))
#> Value:  298.3385 
#> P-Value:  0.0000
rjd3toolkit::seasonality_friedman(y_lin)
#> Value:  302.2953 
#> P-Value:  0.0000

seas_tests$`diagnostics.seas-lin-kw`
#> [1] 157.9391   0.0000
#> attr(,"description")
#> [1] "Chi2 with 11 degrees of freedom "
rjd3toolkit::seasonality_kruskalwallis(rjd3toolkit::differences(y_lin, lags = 1))
#> Value:  331.9686 
#> P-Value:  0.0000
rjd3toolkit::seasonality_kruskalwallis(y_lin)
#> Value:  157.9391 
#> P-Value:  0.0000

It must be the same for the test on the SA series. I think the code should be modified here : https://github.com/nbbrd/jdemetra-sa-advanced/blob/bb6b62bda313dd3a0132fa4e0bf0bab771b55215/jdr/src/main/java/ec/tstoolkit/jdr/sa/SaDiagnostics.java#L609-L612 https://github.com/nbbrd/jdemetra-sa-advanced/blob/bb6b62bda313dd3a0132fa4e0bf0bab771b55215/jdr/src/main/java/ec/tstoolkit/jdr/sa/SaDiagnostics.java#L862-L865 What do you think @palatej ?

rjdverse / rjdemetra

Kruskal-Wallis test on linearised series and SA series #128