The radiative efficiency of AGN is commonly estimated based on the total mass accreted and the total AGN light emitted per unit volume in the universe integrated over time ( the Soltan argument ) .
In individual AGN , thin accretion disk model spectral fits can be used to deduce the absolute accretion rate \dot { M } , if the black hole mass M is known .
The radiative efficiency \eta is then set by the ratio of the bolometric luminosity L _ { bol } to \dot { M } c ^ { 2 } .
We apply this method to determine \eta in a sample of 80 PG quasars with well determined L _ { bol } , where \dot { M } is set by thin accretion disk model fits to the optical luminosity density , and the M determination based on the bulge stellar velocity dispersion ( 13 objects ) or the broad line region ( BLR ) .
For the BLR-based masses , we derive a mean \log \eta = -1.05 \pm 0.52 consistent with the Soltan argument based estimates .
We find a strong correlation of \eta with M , rising from \eta \sim 0.03 at M = 10 ^ { 7 } M _ { \odot } and L / L _ { Edd } \sim 1 to \eta \sim 0.4 at M = 10 ^ { 9 } M _ { \odot } and L / L _ { Edd } \sim 0.3 .
This trend is related to the overall uniformity of L _ { opt } / L _ { bol } in our sample , particularly the lack of the expected increase in L _ { opt } / L _ { bol } with increasing M ( and decreasing L / L _ { Edd } ) , which is a generic property of thermal disk emission at fixed \eta .
The significant uncertainty in the M determination is not large enough to remove the correlation .
The rising \eta with M may imply a rise in the black hole spin with M , as proposed based on other indirect arguments .