Possible error in computing tau in BchronCalibrate?

[ercrema]

There is a possible error in the computation of tau in BchronCalibrate(). This was detected when comparing Bchron to rcarbon and OxCal (via oxcAAR). There are still some differences after the fix but this is presumably due to the use of the t distribution?

library(Bchron)
library(rcarbon)
library(oxcAAR)


x1=BchronCalibrate(4000,40,calCurves='intcal13')
x2=calibrate(4000,40)
quickSetupOxcal()
x3=oxcalCalibrate(4000,40)


plot(x1$Date1$ageGrid,x1$Date1$densities,type='l')
lines(x2$grids[[1]][,1],x2$grids[[1]][,2],col='red',lty=2)
oxcalDF= x3[[1]][6][[1]]
lines(abs(round(oxcalDF[,1])-1950), oxcalDF[,2],lty=3,col='blue')

[andrewcparnell]

Thanks for the comment. The use of the t-distribution in Bchron is deliberate. You don't truly know the age sd for a date so really you should use a t-distribution rather than a Gaussian. I think there's a paper by Andres Christen which discusses this issue but I don't have it to hand. The differences are likely very slight.

[ercrema]

Oh, by no mean I wasn't referring much about the use of the t-distribution but more about how tau was computed (without squaring the error term in the calibration curve).

[andrewcparnell]

Wow, and apologies for not fulling getting your query. Thank you so much for spotting this. I must have introduced this bug back in V4 and I'm now on version 4.5!

Fixed on GitHub and will be updating this ASAP on CRAN.

[ercrema]

No problems (I was a bit unclear...) and thanks for developing the package!