In one of the most used guide to qPCR (mutuated from Livak and Schmittgen, 2001), "Guide to Performing Relative Quantitation of Gene Expression Using Real-Time Quantitative PCR" (Applied Biosystems), Standard deviation (SD), to be incorporated in 2^-ddCT, is calculated starting from Cts as: SD DeltaCt = {(SD of c-myc)^2 + (SD of GAPDH)^2}^1/2 .

Here is the link with the guide in pdf:

https://www.gu.se/digitalAssets/1125/1125331_ABI_-_Guide_Relative_Quantification_using_realtime_PCR.pdf

However, in this way I think that we are measuring the variability of Cts, which are not so biologically meaningful compared to the variability of dCT. Indeed we usually do statistics (e.g. ANOVA) on dCt and plot directly dCT values or Fold Change (i.e. relative expression calculated from dCT) with error bars in the final graph. Are you agree that SD (or SEM) should be calculated directly on dCT values, rather than on Cts?