The validity of derived abundances depends on the quality of the data and on the method of analysis. Typical quoted values for the uncertainties are 0.1 - 0.25 dex for ratios such as O/H, N/H, Ne/H, a little more for S/H, A/H, C/H, a little less for N/O, Ne/O and a few % for He. The optimism of the investigator is an important factor in the evaluation of the accuracy. This section comments on the various sources of uncertainties in abundance determinations.
3.1. Atomic data
Reviews on atomic data for abundance analysis have been given by Mendoza (1983), Butler (1993), Storey (1997), Nahar (2002). On-line atomic data bases are available from different sites. For example http://plasma-gate.weizmann.ac.il/DBfAPP.html provides links to many sites of interest, including the site of CLOUDY. The XSTAR atomic data base, constructed by Bautista & Kallman (2001) and used in the photoionization code XSTAR can be found at http://heasarc.gsfc.nasa.gov/docs/software/xstar/xstar.html.
The OPACITY and IRON projects (Seaton 1987, Hummer et al. 1993) have considerably increased the reliability of atomic data used for nebular analysis in the recent years. In the following, we simply raise a few important points.
3.1.1. Ionization, recombination and charge exchange
Until recently, photoionization cross sections and recombination (radiative and dielectronic) coefficient sets used in photoionization computations were not obtained self-consistently. Photoionization and recombination calculations are presently being carried out using the same set of eigenfunctions as in the IRON project (Nahar & Pradhan 1997, Nahar et al. 2000). The expected overall uncertainty is 10 - 20%. Experimental checks on a few species (see e.g. Savin 1999) can provide benchmarks for confrontation with numerical computations.
Concerning charge exchange, only a few detailed computations are available (see references in the compilation by Kingdon & Ferland 1996). Coefficients computed with the Landau-Zeener approximation are available for most ions of interest. They are unfortunately rather uncertain. Differences with coefficients from quantal computations, which are available for a few species only, can be as large as a factor 3.
Due to the uncertainties in atomic parameters, the ionization structure predicted by photoionization models is so far expected to be accurate only for elements from the first and second row of the Mendeleev table.