How to read GLM results / choose the best model?

Hi, I am wondering in Matlab, which indicator should be used to choose the best model?

When I use fitglm, in addition to the estimates of coefficients, I get

5802 observations, 5796 error degrees of freedom Estimated Dispersion: 0.474 F-statistic vs. constant model: 647, p-value = 0

What is this "Estimated Dispersion"? Is it equivalent to adjuste-Rsquare? The higher the better the model is or the lower the better?

If I use [b,dev,stats]=glmfit, I get dev (deviance), does that tell how good the model fits?Anything is stats useful in telling how good the model fits?

thanks a lot!!!