You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Blinding simulations is helpful for registration-driven modeling: You can prepare all your analysis scripts with the blinded simulation output, then preregister, and only use the real output when your methods are all prepared. That helps against hindsight bias, and you are encouraged to actually document and archive every simulation and analysis.
The best way would be to let MMM generate blinded output alongside real output; both are structurally similar and only differ in their semantics.
For the final analysis, one just needs to switch from one to the other.
Consider to obfuscate both average and temporal pattern of each aggregation unit:
Multiply with random constant to obfuscate the average.
Multiply each datum with a random number r. It could be a continuous curve in the form of R = [r₁, r₂, r₃, …] = [r₁, r₁*r, r₂*r, r₃*r, …].
The new TOML option would boolean:output.text_tables.blinding = true|false
I think it would make sense to create separate blinded files alongside the real files, e.g. mass_density_per_hft.blinded.tsv. demo_results.Rmd should then show an example of how to read it, something like:
library(tools)
blinded_filename<- paste(file_path_sans_ext(filename), "blinded", "tsv", sep=".")
if (file.exists(blinded_filename))
filename<-blinded_filenameif (!file.exists(filename))
stop("File does not exist: ", filename)
tbl<- read.delim(filename)
attr(tbl, blinded) <- (filename==blinded_filename)
# ...and in the plot:ggplot::labs(caption= ifelse(attributes(tbl)$blinded, "These data are blinded!", ""))
When the user is ready, they can just delete all blinded files:
find -name '*.blinded.tsv' -delete
The text was updated successfully, but these errors were encountered:
Blinding simulations is helpful for registration-driven modeling: You can prepare all your analysis scripts with the blinded simulation output, then preregister, and only use the real output when your methods are all prepared. That helps against hindsight bias, and you are encouraged to actually document and archive every simulation and analysis.
The best way would be to let MMM generate blinded output alongside real output; both are structurally similar and only differ in their semantics.
For the final analysis, one just needs to switch from one to the other.
Consider to obfuscate both average and temporal pattern of each aggregation unit:
r
. It could be a continuous curve in the form ofR = [r₁, r₂, r₃, …] = [r₁, r₁*r, r₂*r, r₃*r, …]
.The new TOML option would boolean:
output.text_tables.blinding = true|false
I think it would make sense to create separate blinded files alongside the real files, e.g.
mass_density_per_hft.blinded.tsv
.demo_results.Rmd
should then show an example of how to read it, something like:When the user is ready, they can just delete all blinded files:
find -name '*.blinded.tsv' -delete
The text was updated successfully, but these errors were encountered: