Figure 11.

A comparison of the performance of the latent non-random mixing model, the Zheng et al. overdispersion model, and the Killworth et al. scale-up method. In each panel the difference between the estimated degree and the known data-generating degree is plotted against age. Three different sets of names were used: a set of names that do not satisfy the scaled-down condition (bad names), the names used in the McCarty et al. survey, and a set of names that satisfy the scaled-down condition (good names). With the bad names, all three procedures show some age bias in estimates, but these biases are smallest with the latent non-random mixing model. With the McCarty et al. names, the scale-up estimate and the Zheng et al. estimates show age bias, but the estimates from the latent non-random mixing model are excellent. With the good names, all three procedures preform well.