Figure 4.

(a) The first principal component of frequency of use predicts the subjective rank-order rate of change for 110 words as judged in a worldwide sample of 14 language families [11], correlation r = 0.22, p < 0.0263. The language families are described in figure 1. (b) Regression model including the first principal component loadings and parts of speech R² = 0.31, p < 0.0001). Regression lines were allowed to have separate intercepts but were constrained to have the same slopes. Allowing separate slopes did not significantly increase the overall R².