Flux-limited and volume-limited galaxy samples are constructed from theSloan Digital Sky Survey (SDSS) data releases DR4, DR6 and DR7 for statistical analysis. The two-point correlation functions ξ(s), monopole of three-point correlation functions ζ0, projected two-point correlation function ωp and pairwise velocity dispersion σ12 are measured to test if galaxy samples are fair for these statistics. We find that with the increment of sky coverage of subsequent data releases in SDSS, ξ(s) of the flux-limited sample is extremely robust and insensitive to local structures at low redshift. However, for volume-limited samples fainter than L at large scales s ≥ 10 h-1 Mpc, the deviation of ξ(s) from different SDSS data releases (DR7, DR6 and DR4) increases with the increment of absolute magnitude. The case of ζ0(s) is similar to that of ξ(s). In the weakly nonlinear regime, there is no agreement between ζ0 of different data releases in all luminosity bins. Furthermore, ωp of volume-limited samples of DR7 in luminosity bins fainter than -Mr,0.1 = [18.5, 19.5] are significantly larger and σ12 of the two faintest volume-limited samples of DR7 display a very different scale dependence than results from DR4 and DR6. Our findings call for caution in understanding clustering analysis results of SDSS faint galaxy samples and higher order statistics of SDSS volume-limited samples in the weakly nonlinear regime. The first zero-crossing points of ξ(s) from volume-limited samples are also investigated and discussed.