To see what this error would be during the day, Figure 2shows this regression function in blue, along with the nighttime Rsw data that it is fitting (black). The daytime error is about 2%, which is rather large, though small compared to the energy balance residual.
Note that this error is not exactly correctable, since we don't have a Tdome and Tcase for the PSPs. There are some studies which have installed temperature sensors on the PSPs (Bush et al., 2000 and Haeffelin, 2000). Bush et al. find that the error is linear in Td^4-Tc^4, and thus directly explained by radiative heat transfer. However, they find that the coefficients must be determined for each radiometer. If a fit to Td-Tc were better than Td^4-Tc^4, Tony suggests that this would indicate heat transfer by convection, conduction and/or diffusion, rather than by radiation.
I've also looked at data from OASIS98 (Figure 3), and 80 days of SHEBA (all nighttime!) (Figure 4). A slope of 13 fits all of these data quite well! For CASES97, the Tdome.down was bad so it can't be used. Rsw.up from CASES97 (Figure 5) doesn't fit the slope of 13 very well -- perhaps different ventillation was used.