Когда я выполняю следующий CFA в lavaan, используя FIML, кажется, что алгоритм сначала использует EM, а затем NLMINB (квазиньютон).Я думаю, что знаю, как управлять аспектами оптимизации NLMINB, используя аргумент control, но я не знаю, как или могу ли я контролировать часть EM.Вот пример.
library(lavaan)
library(MASS)
set.seed(23)
N = 1000
cov = cbind(c(2,0.5),c(0.5,1))
s = mvrnorm(n=N, mu=c(0, 0), Sigma=cov)
m1 = 0.5 + 1*s[, 1] + rnorm(sd=2, n=N)
m2 = 2 + 1*s[, 2] + rnorm(sd=1.5, n=N)
m3 = -3 + 2*s[, 1] -4*s[, 2] + rnorm(sd=4, n=N)
m4 = -1 + 1.5*s[, 1] + 2*s[, 2] + rnorm(sd=0.5, n=N)
m5 = 3 - 1*s[, 1] + 1*s[, 2] + rnorm(sd=3, n=N)
m6 = 2 + 1.2*s[, 1] + 5*s[, 2] + rnorm(sd=1, n=N)
M = data.frame(cbind(m1, m2, m3, m4, m5, m6))
M[1:N/2, 1] = NaN
M[(N/2):N, 2] = NaN
cfa.model <- '
f1 =~ 1*m1 + m3 + m4 + m5 + m6
f2 =~ 1*m2 + m3 + m4 + m5 + m6
f1 ~ f2
'
fcols = c('f1', 'f2')
K = length(fcols)
J = dim(M)[2]
cfa.fit = cfa(cfa.model, data=M, verbose=T, missing='fiml')
lav_mvnorm_missing_h1_estimate_moments: start EM steps
EM iteration: 0 fx = 18.3174217817
EM iteration: 1 fx = 15.6282551073 delta par = 19.11017973
EM iteration: 2 fx = 15.5451600829 delta par = 1.44838250
EM iteration: 3 fx = 15.5164722286 delta par = 0.76379400
EM iteration: 4 fx = 15.5073993955 delta par = 0.40644568
EM iteration: 5 fx = 15.5047273405 delta par = 0.21650406
EM iteration: 6 fx = 15.5039730261 delta par = 0.11529083
EM iteration: 7 fx = 15.5037645370 delta par = 0.06137652
EM iteration: 8 fx = 15.5037073934 delta par = 0.03267151
EM iteration: 9 fx = 15.5036917538 delta par = 0.01739197
EM iteration: 10 fx = 15.5036874644 delta par = 0.00925903
EM iteration: 11 fx = 15.5036862834 delta par = 0.00492977
EM iteration: 12 fx = 15.5036859567 delta par = 0.00262500
EM iteration: 13 fx = 15.5036858659 delta par = 0.00139788
EM iteration: 14 fx = 15.5036858406 delta par = 0.00074446
EM iteration: 15 fx = 15.5036858335 delta par = 0.00039650
EM iteration: 16 fx = 15.5036858315 delta par = 0.00021118
EM iteration: 17 fx = 15.5036858309 delta par = 0.00011248
EM iteration: 18 fx = 15.5036858307 delta par = 0.00005991
EM iteration: 19 fx = 15.5036858307 delta par = 0.00003191
EM iteration: 20 fx = 15.5036858307 delta par = 0.00001700
EM iteration: 21 fx = 15.5036858307 delta par = 0.00000906
EM iteration: 22 fx = 15.5036858307 delta par = 0.00000482
EM iteration: 23 fx = 15.5036858307 delta par = 0.00000257
EM iteration: 24 fx = 15.5036858307 delta par = 0.00000137
EM iteration: 25 fx = 15.5036858307 delta par = 0.00000073
Sigma:
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 6.0799427 1.454745 4.174588 -1.8771549 5.3453480 0.5556121
[2,] 1.4547451 30.295454 -4.094796 -3.4055107 -14.1147626 -3.3789417
[3,] 4.1745879 -4.094796 11.996882 -1.6843854 19.1101797 2.9004991
[4,] -1.8771549 -3.405511 -1.684385 10.3429869 -0.3690654 -0.2135262
[5,] 5.3453480 -14.114763 19.110180 -0.3690654 36.1675277 5.9716330
[6,] 0.5556121 -3.378942 2.900499 -0.2135262 5.9716330 3.3292615
Mu:
[1] 0.4809515 -3.0504037 -1.0852005 2.9048318 1.9060771 2.0361792
Quasi-Newton steps using NLMINB:
Objective function = 1.7888525861901217
Objective function = 0.9851575525661351
Objective function = 0.7268540538683199
Objective function = 0.2557896906942654
Objective function = 0.7584160209525388
Objective function = 0.2684653920239848
Objective function = 0.1621520991171366
Objective function = 0.1107424088803395
Objective function = 0.1296883105884570
Objective function = 0.0940788568119677
Objective function = 0.0864652417773026
Objective function = 0.0721041765085504
Objective function = 0.0633494325628670
Objective function = 0.0536747272719555
Objective function = 0.0502247477693007
Objective function = 0.0413637133003633
Objective function = 0.0379406459862457
Objective function = 0.0320174820525629
Objective function = 0.0289101640770371
Objective function = 0.0270354297037469
Objective function = 0.0249629540989424
Objective function = 0.0218359256938889
Objective function = 0.0189369273140114
Objective function = 0.0151031226656801
Objective function = 0.0122718907717312
Objective function = 0.0092584807074578
Objective function = 0.0088751763044188
Objective function = 0.0076123912076485
Objective function = 0.0072493252272530
Objective function = 0.0068247176958316
Objective function = 0.0060616745422601
Objective function = 0.0056170211463220
Objective function = 0.0052462215041258
Objective function = 0.0050492301407967
Objective function = 0.0049205710553393
Objective function = 0.0048659503391200
Objective function = 0.0047987479162090
Objective function = 0.0047462742914490
Objective function = 0.0047077661522730
Objective function = 0.0046831461851964
Objective function = 0.0046626232584908
Objective function = 0.0046420155019389
Objective function = 0.0046148542070785
Objective function = 0.0045882334071630
Objective function = 0.0045639326820144
Objective function = 0.0045483001134041
Objective function = 0.0045363191382677
Objective function = 0.0045248956450150
Objective function = 0.0045110841388443
Objective function = 0.0045045565114492
Objective function = 0.0045006776349314
Objective function = 0.0044984370789782
Objective function = 0.0044953995539645
Objective function = 0.0044928442197820
Objective function = 0.0044905504696340
Objective function = 0.0044897178432395
Objective function = 0.0044890460805611
Objective function = 0.0044885013990150
Objective function = 0.0044879390368715
Objective function = 0.0044876121320216
Objective function = 0.0044873608054701
Objective function = 0.0044870882203005
Objective function = 0.0044864608711661
Objective function = 0.0044854094744728
Objective function = 0.0044834872439443
Objective function = 0.0044816758042803
Objective function = 0.0044804928677538
Objective function = 0.0044803025628157
Objective function = 0.0044802822439056
Objective function = 0.0044802750925212
Objective function = 0.0044802743608985
Objective function = 0.0044802732265765
Objective function = 0.0044802730781361
Objective function = 0.0044802729972568
Objective function = 0.0044802729735727
Objective function = 0.0044802729459699
Objective function = 0.0044802729251341
Objective function = 0.0044802729060658
Objective function = 0.0044802728936952
Objective function = 0.0044802728880020
Objective function = 0.0044802728859707
Objective function = 0.0044802728850044
Objective function = 0.0044802728842646
Objective function = 0.0044802728836046
Objective function = 0.0044802728834483
Objective function = 0.0044802728834483
convergence status (0=ok): 0
nlminb message says: relative convergence (4)
number of iterations: 79
number of function evaluations [objective, gradient]: 85 80
Computing VCOV for se = standard ... done.
Computing TEST for test = standard ... done.
Таким образом, ясно, что EM выполняется до NLMINB.Итак, сначала я хотел бы знать, почему.И второе, могу ли я что-нибудь изменить в этом начальном шаге EM?Я пытался использовать em аргументы, такие как em.fx.tol, em.iter.max и em.dx.tol, без какого-либо эффекта.
cfa.fit = cfa(cfa.model, data=M, verbose=T, missing='fiml', em.iter.max=5, optim.method='em')
или
cfa.fit = cfa(cfa.model, data=M, verbose=T, missing='fiml', em.iter.max=5)
Но вв любом случае, итерации EM превышают 5.