Apparent discrepency between code and documentation for fractional bias and fractional error

[pavlovc2]

The documentation states that the numerator and denominator should be summed separately to calculate fractional bias and error.

However, the code for DomainStats in AQ_Misc_Functions.R appears to calculate these metrics as the sum after the fraction is calculated for each observation/model pair. The code also multiplies the result by a 100 to report as a percentage rather than a fraction.

https://github.com/USEPA/AMET/blob/1.5/R_analysis_code/AQ_Misc_Functions.R#L652

fb <- signif(c(fb,(sum((data.df$mod_val-data.df$ob_val)/((data.df$mod_val+data.df$ob_val)/2)))/length(data.df$stat_id))*100,3)
fe <- signif(c(fe,(sum(abs(data.df$mod_val-data.df$ob_val)/((data.df$mod_val+data.df$ob_val/2)))/length(data.df$stat_id))*100,3)

Am I correct that this is a discrepancy, and can this be updated to be consistent with the correct approach?

[wkappel]

Nathan,

Thanks for your comment. It turns out the definition in the document is the one that is incorrect. The summing should be taking place outside of the brackets (as is done in the actual AMET R code). So, we will note this as a correction needed to this document.

Let me know if you have any further questions. Thanks!

Wyat

[pavlovc2]

Got it, thank you for the quick clarification!