Variable agreement in Brazilian Portuguese is subject to social stigma. It was brought to the general public's attention in 2011 in a heated nationwide sociolinguistic debate triggered by TV Globo, the principal network. In order to isolate objective factors underlying this debate, we examine the variable education in a trend study of Rio de Janeiro speech. Relying on logistic model relative weights and their corresponding ranges, we use polarization to designate the magnitude of these ranges. Polarization of the education variable in 1980 was moderate, whereas in 2000, polarization becomes extreme in an increasingly uneven social distribution of standard marked forms. The increased effect of education is also reflected in a comparison with the statistical effect of mass media: in 1980, media contact had a stronger effect than education did, whereas in 2000, education exhibits a stronger effect. The results reveal objective correlates to speakers' subjective reactions, suggesting that public discussion of linguistic prejudice would be useful to the community, which can profit from results of sociolinguistic research in a humanistic and emancipatory way, as foreseen by Sankoff (1988a).