We apply novel survival analysis techniques to investigate the relationship between a number of the properties of galaxies and their atomic ( M _ { \mathrm { HI } } ) and molecular ( M _ { \mathrm { H _ { 2 } } } ) gas mass , with the aim of devising efficient , effective empirical estimators of the cold gas content in galaxies that can be applied to large optical galaxy surveys .
We find that dust attenuation , A _ { V } , of both the continuum and nebular emission , shows significant partial correlations with M _ { \mathrm { H _ { 2 } } } , after controlling for the effect of star formation rate ( SFR ) .
The partial correlation between A _ { V } and M _ { \mathrm { HI } } , however , is weak .
This is expected because in poorly dust-shielded regions molecular hydrogen is dissociated by far-ultraviolet photons .
We also find that the stellar half-light radius , R _ { 50 } , shows significant partial correlations with both M _ { \mathrm { H _ { 2 } } } and M _ { \mathrm { HI } } .
This hints at the importance of environment ( e.g. , galactocentric distance ) on the gas content of galaxies and the interplay between gas and SFR .
We fit multiple regression to summarize the median , mean , and the 0.15 / 0.85 quantile multivariate relationships among M _ { \mathrm { H _ { 2 } } } , A _ { V } , metallicity , and/or R _ { 50 } .
A linear combination of A _ { V } and metallicity ( inferred from stellar mass ) or A _ { V } and R _ { 50 } , can estimate molecular gas masses within \sim 2.5 - 3 times the observed masses .
If SFR is used in addition , M _ { \mathrm { { H _ { 2 } } } } can be predicted to within a factor \lesssim 2 .
In this case , A _ { V } and R _ { 50 } are the two best secondary parameters that improve the primary relation between M _ { \mathrm { { H _ { 2 } } } } and SFR .
Likewise , M _ { \mathrm { HI } } can be predicted to within a factor \lesssim 3 using R _ { 50 } and SFR .