Aims. We aim to study the structure and kinematics of the broad line region (BLR) of a sample of 27 gravitationally lensed quasars with up to five different epochs of observation. This sample is composed of ∼100 spectra from the literature plus 22 unpublished spectra of 11 systems.
Methods. We measure the magnitude differences in the broad emission line (BEL) wings and statistically model the distribution of microlensing magnifications to determine a maximum likelihood estimate for the sizes of the C IV, C III], and Mg II emitting regions.
Results. The BELs in lensed quasars are expected to be magnified differently owing to the different sizes of the regions from which they originate. Focusing on the most common BELs in our spectra (C IV, C III], and Mg II), we find that the low-ionization line Mg II is only weakly affected by microlensing. In contrast, the high-ionization line C IV shows strong microlensing in some cases, indicating that its emission region is more compact. Thus, the BEL profiles are deformed differently depending on the geometry and kinematics of the corresponding emitting region. We detect microlensing in either the blue or the red wing (or in both wings with different amplitudes) of C IV in more than 50% of the systems and find outstanding asymmetries in the wings of QSO 0957+561, SDSS J1004+4112, SDSS J1206+4332, and SDSS J1339+1310. This observation indicates that the BLR is, in general, not spherically symmetric and supports the existence of two regions in the BLR, one insensitive to microlensing and another that only shows up when it is magnified by microlensing. Disregarding the existence of these two regions, our estimate for Mg II, R1/2 = 67.3−15.7+3.8 √(M/M⊙) light-days, is in good agreement with previous results from smaller samples, while we obtain smaller sizes for the C III] and CIV lines, R1/2 = 31.0−4.0+1.9 √(M/M⊙) light-days and R1/2 = 15.5−3.9+0.8 √(M/M⊙) light-days, respectively.