#図6-1の分析

#本書の分析に用いたソフトウェア（Mplus）とは異なるため，

#本書とは適合度がわずかに一致しない。

#図6-1のモデルと図6-3のモデルでは，扱っている変数が異なるため，

#欠測を含む標本のサイズも異なる。そのため，2つのモデルで

#標本サイズが1つ異なり，その影響で共分散行列も異なるが，その違いはわずかである。

#以上の理由で，2つのモデルの分析対象は図6-3に対する共分散行列で統一している。

library(lavaan)

#共分散行列の下三角要素の読み込み

lower <- '

1.064,

0.515,1.006,

0.663,0.591,1.16,

0.525,0.573,0.711,1.067,

0.391,0.41,0.375,0.388,0.976,

0.044,0.125,-0.025,0.019,0.092,0.585,

-0.03,0.165,-0.003,0.07,0.146,0.195,1.413,

-0.003,0.11,-0.017,-0.01,0.08,0.334,0.474,1.052,

0.099,0.215,0.065,0.135,0.148,0.282,0.523,0.467,1.317,

0.066,0.163,0.092,0.089,0.163,0.284,0.423,0.412,0.591,1.14,

-0.036,0.03,-0.012,0.03,0.044,0.117,0.128,0.157,0.117,0.09,0.489,

0.022,0.04,0.072,0.074,0.01,0.118,0.214,0.135,0.212,0.175,0.039,1.093,

0.012,0.058,0.075,0.081,0.013,0.107,0.223,0.116,0.219,0.198,0.033,0.571,1.259,

0.134,0.217,0.084,0.125,0.146,0.134,0.248,0.155,0.259,0.249,0.094,0.162,0.102,0.718,

0.126,0.198,0.088,0.128,0.196,0.236,0.34,0.332,0.393,0.356,0.123,0.268,0.191,0.371,1.069,

0.099,0.219,0.075,0.124,0.158,0.208,0.229,0.254,0.406,0.318,0.127,0.228,0.205,0.345,0.611,0.957,

0.219,0.266,0.155,0.146,0.203,0.216,0.221,0.262,0.387,0.285,0.162,0.178,0.131,0.3,0.487,0.568,0.908,

0.093,0.169,0.061,0.078,0.174,0.177,0.309,0.243,0.363,0.312,0.139,0.229,0.196,0.349,0.562,0.614,0.47,1.031'

#共分散行列の作成

dep6.cov <- getCov(lower,names=c("harm1","harm2","harm3","harm4","harm5","friend1","friend2","friend3","friend4","friend5","parent1","parent2","parent3","dep1","dep2","dep3","dep4","dep5"))

#モデルの指定（図6-1）

model.fig6.1 <- '

f1 =~ 1*harm1+harm2+harm3+harm4+harm5

f2 =~ 1*friend1+friend2+friend3+friend4+friend5

f3 =~ 1*dep1+dep2+dep3+dep4+dep5

f3 ~ f1+f2

'

#auto.var=TRUEで，観測変数の誤差分散と潜在変数の分散を推定する

#auto.cov.lv.x=TRUEで，外生的な潜在変数間の共分散を推定する

fit.fig6.1 <- lavaan(model.fig6.1, auto.var=TRUE, auto.cov.lv.x=TRUE, sample.cov=dep6.cov, sample.nobs=515)

summary(fit.fig6.1, standardized=T, rsquare=T, fit.measures=TRUE)

#====================================================

#図6-3の分析

#本書の分析に用いたソフトウェア（Mplus）とは異なるため，

#標準誤差が求まらなかったり，本書の推定値と一致していないところがある。

#図6-1のモデルと図6-3のモデルでは，扱っている変数が異なるため，

#欠測を含む標本のサイズも異なる。そのため，2つのモデルで

#標本サイズが1つ異なり，その影響で共分散行列も異なるが，その違いはわずかである。

#以上の理由で，2つのモデルの分析対象は図6-3に対する共分散行列で統一している。

library(lavaan)

#共分散行列の下三角要素の読み込み

lower <- '

1.064,

0.515,1.006,

0.663,0.591,1.16,

0.525,0.573,0.711,1.067,

0.391,0.41,0.375,0.388,0.976,

0.044,0.125,-0.025,0.019,0.092,0.585,

-0.03,0.165,-0.003,0.07,0.146,0.195,1.413,

-0.003,0.11,-0.017,-0.01,0.08,0.334,0.474,1.052,

0.099,0.215,0.065,0.135,0.148,0.282,0.523,0.467,1.317,

0.066,0.163,0.092,0.089,0.163,0.284,0.423,0.412,0.591,1.14,

-0.036,0.03,-0.012,0.03,0.044,0.117,0.128,0.157,0.117,0.09,0.489,

0.022,0.04,0.072,0.074,0.01,0.118,0.214,0.135,0.212,0.175,0.039,1.093,

0.012,0.058,0.075,0.081,0.013,0.107,0.223,0.116,0.219,0.198,0.033,0.571,1.259,

0.134,0.217,0.084,0.125,0.146,0.134,0.248,0.155,0.259,0.249,0.094,0.162,0.102,0.718,

0.126,0.198,0.088,0.128,0.196,0.236,0.34,0.332,0.393,0.356,0.123,0.268,0.191,0.371,1.069,

0.099,0.219,0.075,0.124,0.158,0.208,0.229,0.254,0.406,0.318,0.127,0.228,0.205,0.345,0.611,0.957,

0.219,0.266,0.155,0.146,0.203,0.216,0.221,0.262,0.387,0.285,0.162,0.178,0.131,0.3,0.487,0.568,0.908,

0.093,0.169,0.061,0.078,0.174,0.177,0.309,0.243,0.363,0.312,0.139,0.229,0.196,0.349,0.562,0.614,0.47,1.031'

#共分散行列の作成

dep6.cov <- getCov(lower,names=c("harm1","harm2","harm3","harm4","harm5","friend1","friend2","friend3","friend4","friend5","parent1","parent2","parent3","dep1","dep2","dep3","dep4","dep5"))

#モデルの指定（図6-3）

model.fig6.3 <- '

f1 =~ 1*harm1+harm2+harm3+harm4+harm5

f2 =~ 1*friend1+friend2+friend3+friend4+friend5

f3 =~ 1*parent1+parent2+parent3

f4 =~ 1*dep1+dep2+dep3+dep4+dep5

f4 ~ f1+f2+f3

f2 ~ f1

f3 ~ f1

f1 ~~ f2

'

#auto.var=TRUEで，観測変数の誤差分散と潜在変数の分散を推定する

fit.fig6.3 <- lavaan(model.fig6.3, auto.var=TRUE, sample.cov=dep6.cov, sample.nobs=514)

summary(fit.fig6.3, standardized=T, rsquare=T, fit.measures=TRUE)

#====================================================

#図6-5の分析(バージョン1)

#平均構造を含まないモデル

#本書の数値と一致しない

library(lavaan)

#変数名

varnames<-c("friend1","friend2","friend3","friend4","friend5","dep1","dep2","dep3","dep4","dep5")

#高群の共分散行列の下三角要素の読み込み

high <- '

0.785,

0.232,1.62,

0.456,0.528,1.281,

0.398,0.671,0.612,1.513,

0.411,0.495,0.536,0.749,1.428,

0.181,0.303,0.201,0.287,0.271,0.74,

0.306,0.41,0.44,0.473,0.458,0.417,1.131,

0.267,0.346,0.349,0.468,0.408,0.427,0.693,1.128,

0.307,0.334,0.359,0.436,0.345,0.367,0.564,0.644,1.105,

0.223,0.42,0.302,0.423,0.367,0.435,0.662,0.763,0.549,1.174'

#高群の共分散行列の作成

high.cov <- getCov(high,names=varnames)

#低群の共分散行列の下三角要素の読み込み

low <- '

0.344,

0.113,1.156,

0.175,0.385,0.793,

0.096,0.304,0.261,1.014,

0.096,0.291,0.235,0.336,0.77,

0.016,0.117,0.051,0.122,0.13,0.566,

0.081,0.184,0.149,0.176,0.135,0.189,0.842,

0.07,0.034,0.092,0.22,0.118,0.123,0.373,0.63,

0.036,0.018,0.09,0.197,0.102,0.079,0.236,0.323,0.526,

0.059,0.122,0.12,0.185,0.154,0.147,0.317,0.335,0.243,0.763'

#低群の共分散行列の作成

low.cov <- getCov(low,names=varnames)

#各群の標本サイズの指定

N<-list(highG=258, lowG=254)

#各群の共分散行列をまとめる

full.cov<-list(highG=high.cov, lowG=low.cov)

#等値ありモデルの指定（図6-5）

model.fig6.5 <- '

f1 =~ 1*friend1+friend2+friend3+friend4+friend5

f2 =~ 1*dep1+dep2+dep3+dep4+dep5

f2 ~ c(a,a)*f1

'

#auto.var=TRUEで，観測変数の誤差分散と潜在変数の分散を推定する

fit.fig6.5 <- lavaan(model.fig6.5, auto.var=TRUE, sample.cov=full.cov, sample.nobs=N, group.equal=c("loadings"))

summary(fit.fig6.5, standardized=T, rsquare=T, fit.measures=TRUE)

#等値なしモデルの指定（図6-5）

model.fig6.5 <- '

f1 =~ 1*friend1+friend2+friend3+friend4+friend5

f2 =~ 1*dep1+dep2+dep3+dep4+dep5

f2 ~ c(ahigh,alow)*f1

'

#auto.var=TRUEで，観測変数の誤差分散と潜在変数の分散を推定する

fit.fig6.5 <- lavaan(model.fig6.5, auto.var=TRUE, sample.cov=full.cov, sample.nobs=N, group.equal=c("loadings"))

summary(fit.fig6.5, standardized=T, rsquare=T, fit.measures=TRUE)

#====================================================

#図6-5の分析(バージョン2)

#平均構造を含めた場合のプログラム

#本書の数値と一致する

library(lavaan)

#変数名

varnames<-c("friend1","friend2","friend3","friend4","friend5","dep1","dep2","dep3","dep4","dep5")

#高群の共分散行列の下三角要素の読み込み

high <- '

0.785,

0.232,1.62,

0.456,0.528,1.281,

0.398,0.671,0.612,1.513,

0.411,0.495,0.536,0.749,1.428,

0.181,0.303,0.201,0.287,0.271,0.74,

0.306,0.41,0.44,0.473,0.458,0.417,1.131,

0.267,0.346,0.349,0.468,0.408,0.427,0.693,1.128,

0.307,0.334,0.359,0.436,0.345,0.367,0.564,0.644,1.105,

0.223,0.42,0.302,0.423,0.367,0.435,0.662,0.763,0.549,1.174'

#高群の平均ベクトルの読み込み

high.means<-c(0.492,0.992,0.624,1.050,0.818,1.616,1.450,1.174,1.109,1.225)

#高群の共分散行列の作成

high.cov <- getCov(high,names=varnames)

#低群の共分散行列の下三角要素の読み込み

low <- '

0.344,

0.113,1.156,

0.175,0.385,0.793,

0.096,0.304,0.261,1.014,

0.096,0.291,0.235,0.336,0.77,

0.016,0.117,0.051,0.122,0.13,0.566,

0.081,0.184,0.149,0.176,0.135,0.189,0.842,

0.07,0.034,0.092,0.22,0.118,0.123,0.373,0.63,

0.036,0.018,0.09,0.197,0.102,0.079,0.236,0.323,0.526,

0.059,0.122,0.12,0.185,0.154,0.147,0.317,0.335,0.243,0.763'

#低群の平均ベクトルの読み込み

low.means<-c(0.205,0.677,0.374,0.587,0.413,1.094,0.874,0.618,0.496,0.732)

#低群の共分散行列の作成

low.cov <- getCov(low,names=varnames)

#各群の標本サイズの指定

N<-list(highG=258, lowG=254)

#各群の共分散行列をまとめる

full.cov<-list(highG=high.cov, lowG=low.cov)

#各群の平均ベクトルをまとめる

full.means<-list(highG=high.means, lowG=low.means)

#等値ありモデルの指定（図6-5）

model.fig6.5 <- '

f1 =~ 1*friend1+friend2+friend3+friend4+friend5

f2 =~ 1*dep1+dep2+dep3+dep4+dep5

f2 ~ c(a,a)*f1

friend1 ~ c1*1

friend2 ~ c2*1

friend3 ~ c3*1

friend4 ~ c4*1

friend5 ~ c5*1

dep1 ~ c6*1

dep2 ~ c7*1

dep3 ~ c8*1

dep4 ~ c9*1

dep5 ~ c10*1

f1 ~c(fm1h, fm1l)*1

fm1l==0

'

#auto.var=TRUEで，観測変数の誤差分散と潜在変数の分散を推定する

fit.fig6.5 <- lavaan(model.fig6.5, auto.var=TRUE, sample.cov=full.cov, sample.mean=full.means, sample.nobs=N, group.equal=c("loadings", "intercepts"))

summary(fit.fig6.5, standardized=T, rsquare=T, fit.measures=TRUE)

#等値なしモデルの指定（図6-5）

model.fig6.5 <- '

f1 =~ 1*friend1+friend2+friend3+friend4+friend5

f2 =~ 1*dep1+dep2+dep3+dep4+dep5

f2 ~ c(ahigh,alow)*f1

friend1 ~ c1*1

friend2 ~ c2*1

friend3 ~ c3*1

friend4 ~ c4*1

friend5 ~ c5*1

dep1 ~ c6*1

dep2 ~ c7*1

dep3 ~ c8*1

dep4 ~ c9*1

dep5 ~ c10*1

f1 ~c(fm1h, fm1l)*1

fm1l==0

'

#auto.var=TRUEで，観測変数の誤差分散と潜在変数の分散を推定する

fit.fig6.5 <- lavaan(model.fig6.5, auto.var=TRUE, sample.cov=full.cov, sample.mean=full.means, sample.nobs=N, group.equal=c("loadings", "intercepts"))

summary(fit.fig6.5, standardized=T, rsquare=T, fit.measures=TRUE)

#====================================================

#2014.09.22 尾崎幸謙・荘島宏二郎