5.3. Robustness of results
The main residual worry about accepting the above conclusions is probably whether the assumption of linear bias can really be valid. In general, concentration towards higher-density regions both raises the amplitude of clustering, but also steepens the correlations, so that bias is largest on small scales, as discussed below. We need to be clear of the regime in which the bias depends on scale.
One way in which this issue can be studied is to consider subsamples with very different degrees of bias. Colour information has recently been added to the 2dFGRS database using SuperCosmos scans of the UKST red plates (Hambly et al. 2001), and a division at rest-frame photographic B - R = 0.85 nicely separates ellipticals from spirals. Figure 7 shows the power spectra for the 2dFGRS divided in this way. The shapes are almost identical (perhaps not so surprising, since the cosmic variance effects are closely correlated in these co-spatial samples). However, what is impressive is that the relative bias is almost precisely independent of scale, even though the red subset is rather strongly biased relative to the blue subset (relative b 1.4). This provides some reassurance that the large-scale P(k) reflects the underlying properties of the dark matter, rather than depending on the particular class of galaxies used to measure it.
Figure 7. The power spectra of red galaxies (filled circles) and blue galaxies (open circles), divided at photographic B - R = 0.85. The shapes are strikingly similar, and the square root of the ratio yields the right-hand panel: the relative bias in redshift space of red and blue galaxies. The error bars are obtained by a jack-knife analysis. The relative bias is consistent with a constant value of 1.4 over the range used for fitting of the power-spectrum data (0.015 < k < 0.15h Mpc-1).