Current scul-tutorial.Rmd replication and potential discrepancies when juxtaposed with the main SCUL Tutorial webpage.

[tomasbajcar]

Hello,

I would like to confirm if running the latest version of scul-tutorial.Rmd would yield identical results as those on the main SCUL Tutorial webpage https://hollina.github.io/scul/articles/scul-tutorial.html. I am concerned about the accuracy of the current implementation.

I suspect that SCUL() now behaves differently because I got different results in SCUL.output (visible especially in coef.exact). I noticed when table_for_hux and PlotShareTable(), which now displays only shares with non-zero coefficients (?), returned different results:

The change is then also visible in the smoke plot:

Furthermore, I would like to address one warning and one error I experienced when running scul-tutorial.Rmd:

Warning:

• StandardizedDiff$naive_prediction_1 <- abs(data_to_plot_scul_vary_lambda$naive_prediction_1 - data_to_plot_scul_vary_lambda$actual_y)/PreTreatmentSD Warning: Coercing LHS to a list. Fix: data.frame(abs(SCUL.output$y.actual-SCUL.output$y.scul)/PreTreatmentSD). (enclosed code within the data.frame() function).

Error:

• Error in SCUL.output$y.actual[1:(SCUL.input$TreatmentBeginsAt - 1), ] : incorrect number of dimensions. Fix: sd(unlist(SCUL.output$y.actual[1:(SCUL.input$TreatmentBeginsAt-1)])) (removed comma)

Similar happens in the second part of the same code chunk (can send my fixed .Rmd).

Then there are also 2 errors on lines 800 and 820 regarding PlotNullDistribution(): nrow(SCUL.output$y.actual) should be length(SCUL.output$y.actual) but I see it fixed on the webpage.

I would be very thankful if you could run the latest version of scul-tutorial.Rmd and compare your results with those on the webpage. Any suggestions or possible explanations regarding the discrepancies are appreciated.
Thank you!