Skip to main content
. Author manuscript; available in PMC: 2011 May 1.
Published in final edited form as: Comput Stat Data Anal. 2010 May 1;54(5):1405–1418. doi: 10.1016/j.csda.2009.11.016

Table 3.

Estimates of parameters in θfree, under four different sets of constraints, for the small mammal index example. In 1 β02 reached the lower bound used in the L-BFGS-B algorithm, which resembles the behaviour observed in study 1 associated with degenerate solutions. In 2 the fitted HSC is clearly not degenerate and corresponds to approximately 24% of the observations y. The AIC is reduced when more than one mode is allowed as in 3 although there is little evidence of the need for relaxing the equality constraint on α as in 4.

model AIC l p β 0 β 1 α m
1 3565.3 −1776.7 {0.30,0.70} {−1.33,−15.00} 8.94 56.11 0.41
2 3563.7 −1775.8 {0.76,0.24} −2.52 {0.00,13.32} 76.45 0.38
3 3556.7 −1771.4 {0.36,0.64} −2.68 {3.72,3.87} 80.1 {0.30,0.53}
4 3558.6 −1771.3 {0.35,0.65} −2.68 {3.76,3.85} {80.10,80.11} {0.30,0.53}