low and indicates very good fit for both sites (RMSEA = .05 is considered good fit and RMSEA = .04 in Santa Clara and .01 in Pachuca). Finally, the 2nd eigenvalues were much smaller than the first for both sites (0.518 for Santa Clara and 0.716 for Pachuca) which strongly supports a 1 dimensional solution. On the other hand, for Argentina and Poland, there was clearly only one factor. A second factor in both datasets produced small and non-significant loadings, with small eigenvalues. Most importantly, this second factor had loadings that were almost identical to the loadings of the one factor solution, supporting again a 1 dimensional solution. The lower part of Table 3 shows the factor loadings of the four consumption candidates, assuming a single dimension solution. That is, assuming that the 11 items are considered to be a single factor in the merged sample, the addition of 5+/4+ would produce a factor loading of 0.470. In all sites the alcohol consumption variables loaded in the unidimensional alcohol use disorder trait, ranging from a low of 0.470 to a high of 0.865. The variables 5+/4+ and drunkenness produced the most variable loadings across sites, while 12+ drinks and 5+
