We have compiled a sample of 165 radio galaxies from the literature to study the properties of the extended emission line regions and their interaction with the radio source over a large range of redshift 0<z<5.2. For each source, we have collected radio (size, lobe distance ratio and power) and spectroscopic parameters (luminosity, line width and equivalent width) for the four brightest UV lines. We also introduce a parameter ALyα
measuring the asymmetry of the Lyα line, assuming the intrinsic redshift of the line is the same as that for the He II λ 1640 line, and show that this parameter is a good measure of the amount of absorption in the Lyα line. Using these 18 parameters, we examine the statistical significance of all 153 mutual correlations, and find the following significant correlations: (i) Lyα asymmetry ALyα
with radio size D and redshift z, (ii) line luminosity with radio power, (iii) line luminosities of Lyα, C IV, He II and C III] with each other, and (iv) equivalent widths of Lyα, C IV, He II and C III] with each other. We interpret the correlation between redshift and ALyα
as an increase in the amount of HI around radio galaxies at z>3. The almost exclusive occurrence of HI absorption in small radio sources could indicate a denser surrounding medium or an un-pressurized, low density region, as suggested by Binette et al. (2000A&A...356...23B
). Correlations (ii) to (iv) provide evidence for a common energy source for the radio power and total emission line luminosity, as found in flux density-limited samples of radio sources. The luminosity of the Lyα line relative to the other emission lines and the continuum shows a strong increase at z>3, coincident with the increase in the amount of associated HI absorption. This indicates an increased abundance of hydrogen, both ionized and neutral, which may well be the reservoir of primordial hydrogen from which the galaxy is forming. This metallicity evolution is also seen in the nitrogen abundance, which shows a variation of more than an order of magnitude, with the z>3 radio galaxies occupying only the Z<2Z☉
region. To examine the ionization mechanism of the extended emission line regions in HzRGs, we plot the UV emission line data in line-ratio diagnostic diagrams. The diagrams involving the high ionization C IV, He II and C III] lines seem to confirm previous results showing that AGN photo-ionization provides the best fit to the data. However, these models cannot fit the C II]/C III] ratio, which lies closer to the predictions for the highest velocity shock ionization models. We note that the C II] line is five times more sensitive to shock ionization than the high ionization UV lines, and show that a combination of shock and photo-ionization provides a better overall fit to the integrated spectra of HzRGs. A substantial contribution from shock ionization will show up first in shock sensitive lines like C II] or Mg II . We also confirm the findings of Best et al. (2000MNRAS.311...23B
) that shock ionization occurs almost exclusively in small radio sources, and show that the angular size distribution can indeed explain the differences in three HzRG composite spectra. Because most HzRGs have radio sizes ≲150kpc, their integrated spectra might well contain a significant contribution from shock ionized emission.