6 Discussion

6.1 Spectrophotometric accuracy

The spectrophotometric accuracy can be estimated by adding individual contributions to the total error in quadrature.

The main sources of errors that affect the spectral shape include: the fitting of the sensitivity function, the published standard star fluxes, the adopted atmospheric extinction curve (Jansen et al. 2000). The error in the fit of the sensitivity function is estimated using the residuals of individual standard stars from the mean calibration. These residuals are dominated by systematic differences in the sensitivity function fitted to the different stars, and are less than 5%. Standard star fluxes are accurate to better than 3% (Massey et al. 1988). Considering this 3% uncertainty, we find that errors in our sensitivity function fits are likely to be less than 5%. Application of the BAO mean atmospheric extinction curve to correct our data introduces an error in the continuum slope of the spectra. Because most galaxies were observed at low airmasses, we expect the error in the continuum slope to be less than 5% at any wavelength.

The main sources of errors that affect restricted ranges in wavelength include: the flat field variations, wavelength calibration, the sky subtraction and residuals due to cosmic ray hits (Jansen et al. 2000). Differences between flat fields taken on different nights within a run are small. The contribution of the read noise to the error in the flat field is negligible. Errors in the wavelength calibration introduce small errors in the inferred flux densities on scales comparable to the distance between individual calibration lines. Errors in the dispersion solution are less than 0.3 Å. These dispersion errors produce spectrophotometric errors of at most 3%. Sky subtraction errors dominate the total error on small scales. Because BCGs are compact objects, errors from sky subtraction are less than 5%. Cosmic ray residuals introduce large errors in the extracted spectra only near emission lines, where the steepness of the local background renders a clean fit difficult. Residuals in continuum or sky portions in the spectra are smaller than 2% of the local background. Errors may be as large as 10% per extracted pixel.

This analysis shows that the spectrophotometry is accurate to better than 10% over small wavelength regions, and about 15% or better on large scales.

6.2 Internal checks of the spectrophotometry

In Table 3 we list the total number of observations of each galaxy; 39 galaxies have been observed twice. To check how consistent our calibration procedure is from night to night, including whatever errors exist in the adoption of a mean extinction curve, we can compare the final spectrophotometry to each individual observation. To illustrate this level of accuracy, we show in Figure 5a. duplicate spectra for three kinds of galaxies: an emission line galaxy (VIIZw153), an absorption line galaxy (IIZw35), and a Seyfert galaxy (Mrk 335). The dates at which the spectra were obtained, the exposure times, and the effective airmasses during the observations are indicated. As expected, the deviations are rarely larger than 10% from 4500 to 6800 Å. We see that the average spectral energy distribution precision of our data is good. No systematic differences are seen.

Figure 4: a) A check of the internal consistency of our spectrophotometry using the spectra of the 3 galaxies that we observed during two different times. VII Zw 153: emission line galaxy; II Zw 35: absorption line galaxy; Mrk 335: Seyfert galaxy. All spectra were normalized to the average level in interval 5475-5525 Å. The spectra in each panel have been offset for clarity.

6.3 Comparison with other spectra

Our sample of blue compact galaxies contains two galaxies that were also observed by Kennicutt (1992): Haro 3 and Mrk 201. We compare our spectra for these objects with those obtained by Kennicutt (1992) in Fig. 5b.

In the 4500-6800 Å region the continuum of our spectra match Kennicutt's to better than 10% over small ranges. Bluewards of 4500 Å differences tend to become larger, up to about 30%. In addition, our spectra show a bluer optical spectrum and stronger emission lines than do Kennicutt's. The aperture used by Kennicutt is much larger than ours (45 $^{\prime\prime}$ circular versus 3 $^{\prime\prime}$ slit). Since BCGs have a central burst of star formation, the difference in aperture size is most likely the cause of the difference in continuum shape and emission lines. The increased aperture size allows for a greater contribution to the flux by older stars surrounding the central brust.

Figure 5: b) Comparison of our spectrophotometry with Kennicutt (1992): Haro 3 and Mrk 201. The spectra have all been normalized to the flux centered on 5500Å and have been offset for clarity.