Table 1.
Summarised results for simulation study 1 in which EM was run from fifty sets of starting values. Solutions were clustered in four sets and cluster means and variances (in brackets) of parameter estimates and the observed data log likelihood are reported here. freq. is the number of times the algorithms stopped close to the reported cluster means. The algorithm returned NAs 32 times in 50.
| freq. | l̄y1,…,yn(θ) | p̄1 | β̄01 | β̄02 | β̄1 | ᾱ | m̄ | |
|---|---|---|---|---|---|---|---|---|
| EM | 50 | |||||||
|
| ||||||||
| Optima 1 | 11 | −467.55 | 0.67 | −13.89 | −1.21 | 6.40 | 20.22 | 0.11 |
| (4.5E-3) | (1.1E-3) | (0.91) | (0.010) | (0.20) | (0.48) | (1.1E-4) | ||
|
| ||||||||
| Optima 2 | 1 | −468.80 | 0.93 | −15.00 | 6.32 | 14.21 | 1.16 | 0.11 |
|
| ||||||||
| Optima 3 | 3 | −469.38 | 0.88 | −2.91 | −1.49 | 2.17 | 8.31 | 0.10 |
| (1.4E-3) | (0.021) | (0.065) | (0.026) | (0.024) | (0.026) | (6.5E-5) | ||
|
| ||||||||
| Optima 4 | 3 | −507.22 | 0.89 | −2.11 | 7.55 | 0.43 | 186.32 | 0.76 |
| (0.047) | (6.2E-3) | (0.051) | (4.90) | (0.36) | (130.00) | (4.7E-3) | ||
|
| ||||||||
| Returned NAs | 32 | - | - | - | - | - | - | - |
| (−) | (−) | (−) | (−) | (−) | (−) | (−) | ||
|
| ||||||||
| θ true | −472.78 | 0.5 | −4 | −2 | 2 | 10 | 0.1 | |