4. Rate of agreement among traditional morphologists
We have no way of estimating the stability of the morphological classification, defined as the fraction of galaxies given the same morphological type when observed twice with the same observational material by the same morphologist. However we can estimate the effect of the subjectivity of the Hubble types' definition and how it is related to the quality of the observational material by measuring the reproducibility of the Hubble type estimate in the cases when only the morphologists differ, when only observations differ and when both differ.
4.1. Different morphologists
We start with the results of a comparative classification exercise performed among expert morphologists using the same images (Lahav et al. 1995, Naim et al. 1995). Six expert morphologists classified 835 (nearby and large) galaxies in 16 classes. All the morphologists looked at exactly the same laser printed images, except for one, who looked at images on a computer screen.
The solid line in the left panel of Fig. 1 shows the relative agreement (in %) among morphologists (grouped by pairs) on the morphological type of a given galaxy. This agreement is measured by the fraction of galaxies (among 835) given the same coarse Hubble type (E, S0, or S+Irr) by a pair of morphologists. We put the galaxies in one of the three bins according to the T value listed in Naim et al. (1995). The fraction of galaxies given the same morphological type ranges from % to %, depending on the pair of morphologists, with a mean of %.
The dotted line in the same panel shows the relative agreement for easily classified galaxies (i.e. for galaxies whose morphological type is not followed by a colon or question mark). This is the fraction of galaxies (among the ones easily classified by both morphologists of a pair) that were given the same coarse Hubble type by both morphologists. The agreement is much better, of the order of 90 %, which is normal, since the task is admittedly easy.
The right panel of Fig.1 shows that this good agreement in fact only concerns a minor fraction, 50 % and often less, of the sample. It does not show how often pairs of morphologists agree on the type of a galaxy, but how often they agree on whether the galaxy is easy to classify or not. This fraction of galaxies which are easily classified by pairs of morphologists (i.e. whose type is not followed by a colon or question mark for either morphologist of the pair) is rather small. On this small fraction the agreement on the type is excellent.
Furthermore, part of the agreement among morphologists is due to chance, since we have reduced the number of bins from 16 to 3.
Dressler and Oemler classified the galaxies of the Abell 851 cluster from the same HST images before refurbishing (Dressler et al. 1994a). Couch, Sharples and Smail did the same for the galaxies of Abell 370 (Couch et al. 1994). In both cases, the disagreement among morphologists on the assignment of the morphological type was about 20 to 25%.
4.2. Different images
We now compare the estimates of a morphologist looking at different images. Dressler et al. (1994b) classified twice the galaxies of the distant cluster Cl 0939+4713 observed with HST, before and after refurbishing. Taking the types of galaxies from their Fig. 4, we find that 20 % of the galaxies have different coarse Hubble types.
Naim et al. (1995) and we find a similar rate of disagreement between the morphological types assigned by de Vaucouleurs (Naim et al. 1995) and by RC3 (de Vaucouleurs et al. 1991) for the sample of 835 galaxies.
4.3. Different morphologists and images
The agreement, when both morphologists and images differ, can be estimated from the published morphological types of galaxies in the Coma cluster. We take Dressler's (1980) morphological type as reference, because it concerns a larger and deeper sample than the others. The agreement between Dressler (1980) on the one hand, and other morphologists (Rood & Baum 1967, Butcher & Oemler 1985) and RC3 (de Vaucouleurs et al. 1991) on the other, on the types of Coma galaxies is shown in the last three rows of Table 1. The agreement, for classifications based on different observational materials, is relatively good, even if our comparison is biased by the fact that low values of the agreement are not permitted because of the small number of classes which naturally induces some chance agreements. We have checked that the chance agreement can be as large as 35 % to 75 % depending on the morphological composition of the sample and on the exact way in which we randomly classify the galaxies (conserving or not the morphological composition of the sample).
Table 1. Agreement on the Hubble type estimates
These three comparisons, when only morphologists differ, when only
images differ and when both differ, show that about 20 % of the
galaxies have an unprecise Hubble type. This inaccuracy arises because
of the quality of the material used for the morphological estimation
and/or because of the subjective nature of morphological
classification. The fraction is even larger if galaxies are classified
from Schmidt plates.
© European Southern Observatory (ESO) 1997
Online publication: July 3, 1998