library(psych)
library(lavaan)
## This is lavaan 0.6-14
## lavaan is FREE software! Please report any bugs.
##
## Attaching package: 'lavaan'
## The following object is masked from 'package:psych':
##
## cor2cov
library(semTools)
##
## ###############################################################################
## This is semTools 0.5-6
## All users of R (or SEM) are invited to submit functions or ideas for functions.
## ###############################################################################
##
## Attaching package: 'semTools'
## The following objects are masked from 'package:psych':
##
## reliability, skew
library(semPlot)
library(dplyr)
##
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
##
## filter, lag
## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union
library(tidyverse)
## ── Attaching packages
## ───────────────────────────────────────
## tidyverse 1.3.2 ──
## ✔ ggplot2 3.4.0 ✔ purrr 0.3.4
## ✔ tibble 3.1.8 ✔ stringr 1.4.1
## ✔ tidyr 1.2.1 ✔ forcats 0.5.2
## ✔ readr 2.1.2
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ ggplot2::%+%() masks psych::%+%()
## ✖ ggplot2::alpha() masks psych::alpha()
## ✖ readr::clipboard() masks semTools::clipboard()
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
library(haven)
library(writexl)
data <- read.csv("~/okara004@ucr.edu - Google Drive/My Drive/classes/3 -- spring '23/PSYC 214 -- stats/homework/HW #2/HW2_Data/WLS_HW2_Time.csv")
# gender
gender <- factor(data$sexrsp, levels = 1:2, labels = c("Male", "Female"))
summary(gender)
## Male Female
## 3277 3801
3277 + 3801 ### = 7078 total participants
## [1] 7078
3801/7078 ### = 0.5370161, 53.70% female
## [1] 0.5370161
# age
psych::describe(data$age04) ### mean age in 2004: 64.35yrs; SD: 0.71yrs
## vars n mean sd median trimmed mad min max range skew kurtosis se
## X1 1 7078 64.35 0.71 64 64.34 0 63 67 4 0.48 0.54 0.01
psych::describe(data$age11) ### mean age in 2011: 71.22yrs; SD: 0.92yrs
## vars n mean sd median trimmed mad min max range skew kurtosis se
## X1 1 5534 71.22 0.92 71 71.14 1.48 70 74 4 0.59 -0.02 0.01
# education
data$educ92 <- as.numeric(data$educ92)
psych::describe(data$educ92) ### mean: 13.71 (about 1yr of college); SD: 2.30yrs
## vars n mean sd median trimmed mad min max range skew kurtosis se
## X1 1 6858 13.71 2.3 12 13.32 0 12 21 9 1.05 -0.17 0.03
# isolate only the items of interest for the observed matrices.
time_data <- data[ , (7:14)]
# take a look at the observed correlation matrix.
corMat <- cor(time_data, use = "pairwise.complete.obs")
corMat
## Simil_2004 Digits_2004 LFLU_2004 CFLU_2004 Simil_2011 Digits_2011
## Simil_2004 1.0000000 0.1856831 0.2387166 0.2349456 0.5568106 0.1634219
## Digits_2004 0.1856831 1.0000000 0.1902411 0.1690608 0.1501103 0.3478980
## LFLU_2004 0.2387166 0.1902411 1.0000000 0.3370050 0.2054583 0.2107888
## CFLU_2004 0.2349456 0.1690608 0.3370050 1.0000000 0.2384257 0.2326793
## Simil_2011 0.5568106 0.1501103 0.2054583 0.2384257 1.0000000 0.2095168
## Digits_2011 0.1634219 0.3478980 0.2107888 0.2326793 0.2095168 1.0000000
## LFLU_2011 0.2395870 0.1955691 0.5916502 0.3155063 0.2499542 0.2739231
## CFLU_2011 0.2180568 0.1784553 0.2969536 0.5905128 0.2429890 0.2703994
## LFLU_2011 CFLU_2011
## Simil_2004 0.2395870 0.2180568
## Digits_2004 0.1955691 0.1784553
## LFLU_2004 0.5916502 0.2969536
## CFLU_2004 0.3155063 0.5905128
## Simil_2011 0.2499542 0.2429890
## Digits_2011 0.2739231 0.2703994
## LFLU_2011 1.0000000 0.3493626
## CFLU_2011 0.3493626 1.0000000
### the correlations within the T1 items appear to be about the same as the correlations within the T2 items, although the T2 correlations appear to be ever-so-slightly higher. correlations at T1 range from 0.17 to .34, and correlations at T2 range from 0.21 to 0.35, so a pretty similar range, which is good. each variable correlates with itself across time at about 0.50/0.60, with the exception of digits, which has a correlation of 0.35. it's good that these variables correlate highly with themselves across time, because it suggests that the items are reliable. each item at T1 correlates relatively lowly with the different items at T2 (range of 0.15 to 0.32), and this range is relatively comparable to the range we see within time, so that's another sign that the measure is pretty reliable. letter fluency & category fluency tend to correlate highly with each other compared to other pairs of different items (around 0.30 to 0.35). this makes sense because they appear to be similar measures.
# get the means & SDs.
psych::describe(corMat)
## vars n mean sd median trimmed mad min max range skew kurtosis
## Simil_2004 1 8 0.35 0.29 0.24 0.35 0.05 0.16 1 0.84 1.34 0.22
## Digits_2004 2 8 0.30 0.29 0.19 0.30 0.02 0.15 1 0.85 1.71 1.31
## LFLU_2004 3 8 0.38 0.28 0.27 0.38 0.10 0.19 1 0.81 1.24 0.00
## CFLU_2004 4 8 0.39 0.28 0.28 0.39 0.08 0.17 1 0.83 1.24 0.02
## Simil_2011 5 8 0.36 0.29 0.24 0.36 0.05 0.15 1 0.85 1.34 0.21
## Digits_2011 6 8 0.34 0.27 0.25 0.34 0.06 0.16 1 0.84 1.71 1.35
## LFLU_2011 7 8 0.40 0.27 0.29 0.40 0.08 0.20 1 0.80 1.27 0.09
## CFLU_2011 8 8 0.39 0.28 0.28 0.39 0.10 0.18 1 0.82 1.25 0.04
## se
## Simil_2004 0.1
## Digits_2004 0.1
## LFLU_2004 0.1
## CFLU_2004 0.1
## Simil_2011 0.1
## Digits_2011 0.1
## LFLU_2011 0.1
## CFLU_2011 0.1
### will comment on the means & SDs of the var/cov matrix below, since it's more meaningful to interpret.
# turn matrix into data frame.
corMat_df <- as.data.frame(corMat)
# specify path, create file name for excel file, and save it to a data frame.
correlation_matrix <- "~/okara004@ucr.edu - Google Drive/My Drive/classes/3 -- spring '23/PSYC 214 -- stats/homework/HW #2/correlation_matrix.xlsx"
# this function will put the matrix into an excel spreadsheet.
write_xlsx(corMat_df, path = correlation_matrix)
# COVARIANCE MATRIX
covMat <- cov(time_data, use = "pairwise.complete.obs")
covMat
## Simil_2004 Digits_2004 LFLU_2004 CFLU_2004 Simil_2011 Digits_2011
## Simil_2004 5.657111 1.331131 2.417782 3.346110 3.044747 1.006587
## Digits_2004 1.331131 9.417145 2.510843 3.157940 1.052011 2.762883
## LFLU_2004 2.417782 2.510843 18.940356 8.928069 2.057132 2.397620
## CFLU_2004 3.346110 3.157940 8.928069 37.754144 3.328515 3.679962
## Simil_2011 3.044747 1.052011 2.057132 3.328515 5.384844 1.271686
## Digits_2011 1.006587 2.762883 2.397620 3.679962 1.271686 6.902690
## LFLU_2011 2.327008 2.464364 10.640388 7.860272 2.399484 2.973873
## CFLU_2011 2.993917 3.200140 7.657134 21.265482 3.323114 4.173102
## LFLU_2011 CFLU_2011
## Simil_2004 2.327008 2.993917
## Digits_2004 2.464364 3.200140
## LFLU_2004 10.640388 7.657134
## CFLU_2004 7.860272 21.265482
## Simil_2011 2.399484 3.323114
## Digits_2011 2.973873 4.173102
## LFLU_2011 17.488017 8.685133
## CFLU_2011 8.685133 35.660451
# get the means & SDs.
psych::describe(covMat)
## vars n mean sd median trimmed mad min max range skew
## Simil_2004 1 8 2.77 1.43 2.71 2.77 0.76 1.01 5.66 4.65 0.68
## Digits_2004 2 8 3.24 2.62 2.64 3.24 0.80 1.05 9.42 8.37 1.52
## LFLU_2004 3 8 6.94 5.94 5.08 6.94 4.24 2.06 18.94 16.88 0.83
## CFLU_2004 4 8 11.17 12.35 5.77 11.17 3.75 3.16 37.75 34.60 1.19
## Simil_2011 5 8 2.73 1.38 2.72 2.73 0.94 1.05 5.38 4.33 0.52
## Digits_2011 6 8 3.15 1.86 2.87 3.15 1.57 1.01 6.90 5.90 0.72
## LFLU_2011 7 8 6.85 5.43 5.42 6.85 4.53 2.33 17.49 15.16 0.73
## CFLU_2011 8 8 10.87 11.71 5.92 10.87 4.07 2.99 35.66 32.67 1.14
## kurtosis se
## Simil_2004 -0.48 0.51
## Digits_2004 0.98 0.93
## LFLU_2004 -0.71 2.10
## CFLU_2004 -0.18 4.37
## Simil_2011 -0.86 0.49
## Digits_2011 -0.57 0.66
## LFLU_2011 -0.93 1.92
## CFLU_2011 -0.33 4.14
### means at T1 appear to be a bit larger than the means at T2, which makes sense with our hypothesis that cognitive ability is declining. variance at T1 is only slightly larger than variance at T2, meaning that there is slightly less variability at T2. but the variances really aren't all that different from each other.
# export the matrix to an excel file.
covMat_df <- as.data.frame(covMat)
covariance_matrix <- "~/okara004@ucr.edu - Google Drive/My Drive/classes/3 -- spring '23/PSYC 214 -- stats/homework/HW #2/covariance_matrix.xlsx"
write_xlsx(covMat_df, path = covariance_matrix)
equate one set of lambdas per latent variable and one set of intercepts per latent.
model.CI <- '
#Latent variables
CogT1 =~ NA*Simil_2004 + a*Simil_2004 + Digits_2004 + LFLU_2004 + CFLU_2004
CogT2 =~ NA*Simil_2011 + a*Simil_2011 + Digits_2011 + LFLU_2011 + CFLU_2011
# Residuals
Simil_2004 ~~ Simil_2004
Simil_2011 ~~ Simil_2011
Digits_2004 ~~ Digits_2004
Digits_2011 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2004
LFLU_2011 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2004
CFLU_2011 ~~ CFLU_2011
# Residual covariances
Simil_2004 ~~ Simil_2011
Digits_2004 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2011
# Intercepts
Simil_2004 ~ nu1*1
Simil_2011 ~ nu1*1
Digits_2004 ~ nu11*1
Digits_2011 ~ nu12*1
LFLU_2004 ~ nu21*1
LFLU_2011 ~ nu22*1
CFLU_2004 ~ nu31*1
CFLU_2011 ~ nu32*1
#Latent variances/covariances
#Standardized CogT1
CogT1 ~~ 1*CogT1
#covariance
CogT2 ~~ cov*CogT1
#variance Time 2
CogT2 ~~ NA*CogT2
#Latent means
CogT1 ~ a1*1
CogT2 ~ a2*1
a1 == 0
'
# run the model and see how it fits.
model.CI.fit <- sem(model.CI, data = data, missing = "FIML")
summary(model.CI.fit, fit.measures = TRUE, standardized = T)
## lavaan 0.6.14 ended normally after 107 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 32
## Number of equality constraints 3
##
## Number of observations 7078
## Number of missing patterns 75
##
## Model Test User Model:
##
## Test statistic 28.999
## Degrees of freedom 15
## P-value (Chi-square) 0.016
##
## Model Test Baseline Model:
##
## Test statistic 7086.446
## Degrees of freedom 28
## P-value 0.000
##
## User Model versus Baseline Model:
##
## Comparative Fit Index (CFI) 0.998
## Tucker-Lewis Index (TLI) 0.996
##
## Robust Comparative Fit Index (CFI) 0.998
## Robust Tucker-Lewis Index (TLI) 0.996
##
## Loglikelihood and Information Criteria:
##
## Loglikelihood user model (H0) -95272.349
## Loglikelihood unrestricted model (H1) -95257.849
##
## Akaike (AIC) 190602.698
## Bayesian (BIC) 190801.775
## Sample-size adjusted Bayesian (SABIC) 190709.620
##
## Root Mean Square Error of Approximation:
##
## RMSEA 0.011
## 90 Percent confidence interval - lower 0.005
## 90 Percent confidence interval - upper 0.018
## P-value H_0: RMSEA <= 0.050 1.000
## P-value H_0: RMSEA >= 0.080 0.000
##
## Robust RMSEA 0.016
## 90 Percent confidence interval - lower 0.004
## 90 Percent confidence interval - upper 0.026
## P-value H_0: Robust RMSEA <= 0.050 1.000
## P-value H_0: Robust RMSEA >= 0.080 0.000
##
## Standardized Root Mean Square Residual:
##
## SRMR 0.010
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Observed
## Observed information based on Hessian
##
## Latent Variables:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## CogT1 =~
## Simil_2004 (a) 1.059 0.041 25.767 0.000 1.059 0.445
## Digts_2004 1.105 0.057 19.421 0.000 1.105 0.360
## LFLU_2004 2.491 0.086 28.845 0.000 2.491 0.570
## CFLU_2004 3.496 0.138 25.380 0.000 3.496 0.568
## CogT2 =~
## Simil_2011 (a) 1.059 0.041 25.767 0.000 1.021 0.439
## Digts_2011 1.252 0.073 17.163 0.000 1.207 0.458
## LFLU_2011 2.700 0.142 19.020 0.000 2.604 0.620
## CFLU_2011 3.600 0.205 17.582 0.000 3.472 0.578
##
## Covariances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 ~~
## .Sml_2011 2.117 0.082 25.680 0.000 2.117 0.475
## .Digits_2004 ~~
## .Dgt_2011 1.658 0.125 13.276 0.000 1.658 0.247
## .LFLU_2004 ~~
## .LFLU_201 5.098 0.328 15.523 0.000 5.098 0.431
## .CFLU_2004 ~~
## .CFLU_201 11.024 0.750 14.698 0.000 11.024 0.444
## CogT1 ~~
## CogT2 (cov) 0.883 0.039 22.402 0.000 0.915 0.915
##
## Intercepts:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Sm_2004 (nu1) 6.577 0.029 229.089 0.000 6.577 2.763
## .Sm_2011 (nu1) 6.577 0.029 229.089 0.000 6.577 2.827
## .Dg_2004 (nu11) 7.197 0.042 171.902 0.000 7.197 2.344
## .Dg_2011 (nu12) 7.086 0.055 129.049 0.000 7.086 2.690
## .LFLU_20 (nu21) 11.327 0.059 190.461 0.000 11.327 2.592
## .LFLU_20 (nu22) 11.782 0.100 117.506 0.000 11.782 2.804
## .CFLU_20 (nu31) 20.782 0.103 201.946 0.000 20.782 3.376
## .CFLU_20 (nu32) 20.306 0.155 130.890 0.000 20.306 3.381
## CogT1 (a1) 0.000 0.000 0.000
## CogT2 (a2) -0.281 0.030 -9.391 0.000 -0.292 -0.292
##
## Variances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 4.543 0.102 44.745 0.000 4.543 0.802
## .Simil_2011 4.371 0.103 42.293 0.000 4.371 0.807
## .Digits_2004 8.207 0.181 45.225 0.000 8.207 0.871
## .Digits_2011 5.483 0.145 37.705 0.000 5.483 0.790
## .LFLU_2004 12.890 0.414 31.121 0.000 12.890 0.675
## .LFLU_2011 10.878 0.397 27.434 0.000 10.878 0.616
## .CFLU_2004 25.668 0.922 27.830 0.000 25.668 0.677
## .CFLU_2011 24.018 0.937 25.621 0.000 24.018 0.666
## CogT1 1.000 1.000 1.000
## CogT2 0.930 0.075 12.335 0.000 1.000 1.000
##
## Constraints:
## |Slack|
## a1 - 0 0.000
### the p-value for the chi-squared is significant, but we have a pretty big sample size so we don't need to worry about that too much. the CFI, TLI, and RMSEA all meet the appropriate thresholds, indicating the model fits the data pretty well.
# visualization of the model.
semPaths(model.CI.fit, what = "est")
equate all sets of lambdas and one set of intercepts per latent.
model.WI <- '
#Latent variables
CogT1 =~ NA*Simil_2004 + a*Simil_2004 + b*Digits_2004 + c*LFLU_2004 + d*CFLU_2004
CogT2 =~ NA*Simil_2011 + a*Simil_2011 + b*Digits_2011 + c*LFLU_2011 + d*CFLU_2011
# Residuals
Simil_2004 ~~ Simil_2004
Simil_2011 ~~ Simil_2011
Digits_2004 ~~ Digits_2004
Digits_2011 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2004
LFLU_2011 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2004
CFLU_2011 ~~ CFLU_2011
# Residual covariances
Simil_2004 ~~ Simil_2011
Digits_2004 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2011
# Intercepts
Simil_2004 ~ nu1*1
Simil_2011 ~ nu1*1
Digits_2004 ~ nu11*1
Digits_2011 ~ nu12*1
LFLU_2004 ~ nu21*1
LFLU_2011 ~ nu22*1
CFLU_2004 ~ nu31*1
CFLU_2011 ~ nu32*1
#Latent variances/covariances
#Standardized CogT1
CogT1 ~~ 1*CogT1
#covariance
CogT2 ~~ cov*CogT1
#variance Time 2
CogT2 ~~ NA*CogT2
#Latent means
CogT1 ~ a1*1
CogT2 ~ a2*1
a1 == 0
'
# run the model and see how it fits.
model.WI.fit <- sem(model.WI, data = data, missing = "FIML")
summary(model.WI.fit, fit.measures = TRUE, standardized = T)
## lavaan 0.6.14 ended normally after 92 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 32
## Number of equality constraints 6
##
## Number of observations 7078
## Number of missing patterns 75
##
## Model Test User Model:
##
## Test statistic 33.392
## Degrees of freedom 18
## P-value (Chi-square) 0.015
##
## Model Test Baseline Model:
##
## Test statistic 7086.446
## Degrees of freedom 28
## P-value 0.000
##
## User Model versus Baseline Model:
##
## Comparative Fit Index (CFI) 0.998
## Tucker-Lewis Index (TLI) 0.997
##
## Robust Comparative Fit Index (CFI) 0.998
## Robust Tucker-Lewis Index (TLI) 0.997
##
## Loglikelihood and Information Criteria:
##
## Loglikelihood user model (H0) -95274.545
## Loglikelihood unrestricted model (H1) -95257.849
##
## Akaike (AIC) 190601.091
## Bayesian (BIC) 190779.574
## Sample-size adjusted Bayesian (SABIC) 190696.952
##
## Root Mean Square Error of Approximation:
##
## RMSEA 0.011
## 90 Percent confidence interval - lower 0.005
## 90 Percent confidence interval - upper 0.017
## P-value H_0: RMSEA <= 0.050 1.000
## P-value H_0: RMSEA >= 0.080 0.000
##
## Robust RMSEA 0.015
## 90 Percent confidence interval - lower 0.003
## 90 Percent confidence interval - upper 0.024
## P-value H_0: Robust RMSEA <= 0.050 1.000
## P-value H_0: Robust RMSEA >= 0.080 0.000
##
## Standardized Root Mean Square Residual:
##
## SRMR 0.012
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Observed
## Observed information based on Hessian
##
## Latent Variables:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## CogT1 =~
## Simil_2004 (a) 1.034 0.038 27.566 0.000 1.034 0.435
## Digts_2004 (b) 1.153 0.044 26.035 0.000 1.153 0.374
## LFLU_2004 (c) 2.526 0.077 32.628 0.000 2.526 0.577
## CFLU_2004 (d) 3.457 0.123 28.184 0.000 3.457 0.563
## CogT2 =~
## Simil_2011 (a) 1.034 0.038 27.566 0.000 1.051 0.450
## Digts_2011 (b) 1.153 0.044 26.035 0.000 1.172 0.446
## LFLU_2011 (c) 2.526 0.077 32.628 0.000 2.567 0.612
## CFLU_2011 (d) 3.457 0.123 28.184 0.000 3.513 0.584
##
## Covariances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 ~~
## .Sml_2011 2.113 0.083 25.584 0.000 2.113 0.474
## .Digits_2004 ~~
## .Dgt_2011 1.649 0.125 13.223 0.000 1.649 0.246
## .LFLU_2004 ~~
## .LFLU_201 5.101 0.329 15.525 0.000 5.101 0.430
## .CFLU_2004 ~~
## .CFLU_201 11.011 0.750 14.682 0.000 11.011 0.443
## CogT1 ~~
## CogT2 (cov) 0.932 0.025 37.366 0.000 0.917 0.917
##
## Intercepts:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Sm_2004 (nu1) 6.577 0.029 229.588 0.000 6.577 2.770
## .Sm_2011 (nu1) 6.577 0.029 229.588 0.000 6.577 2.819
## .Dg_2004 (nu11) 7.196 0.042 171.458 0.000 7.196 2.336
## .Dg_2011 (nu12) 7.070 0.053 134.534 0.000 7.070 2.693
## .LFLU_20 (nu21) 11.327 0.060 190.161 0.000 11.327 2.586
## .LFLU_20 (nu22) 11.756 0.095 123.203 0.000 11.756 2.803
## .CFLU_20 (nu31) 20.786 0.103 202.173 0.000 20.786 3.383
## .CFLU_20 (nu32) 20.292 0.152 133.925 0.000 20.292 3.371
## CogT1 (a1) 0.000 0.000 0.000
## CogT2 (a2) -0.289 0.030 -9.544 0.000 -0.285 -0.285
##
## Variances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 4.571 0.099 46.015 0.000 4.571 0.810
## .Simil_2011 4.340 0.102 42.628 0.000 4.340 0.797
## .Digits_2004 8.158 0.177 45.969 0.000 8.158 0.860
## .Digits_2011 5.520 0.143 38.626 0.000 5.520 0.801
## .LFLU_2004 12.796 0.402 31.810 0.000 12.796 0.667
## .LFLU_2011 11.001 0.380 28.968 0.000 11.001 0.625
## .CFLU_2004 25.811 0.891 28.954 0.000 25.811 0.684
## .CFLU_2011 23.884 0.916 26.060 0.000 23.884 0.659
## CogT1 1.000 1.000 1.000
## CogT2 1.033 0.043 24.023 0.000 1.000 1.000
##
## Constraints:
## |Slack|
## a1 - 0 0.000
### p-value for chi-square is still significant, but again, we don't have to worry too much about this because of the large sample size. we gain 3 degrees of freedom going from CI to WI, but our model does not fit significantly worse. AIC and BIC are just slightly smaller for WI than CI. CFI & RMSEA are the same between models, and TLI is ever-so-slightly larger for WI than CI. overall, WI is not all that different from CI in terms of model fit.
# visualization of the model.
semPaths(model.WI.fit, what = "est")
equate all sets of lambdas and all sets of intercepts.
model.StrongI <- '
#Latent variables
CogT1 =~ NA*Simil_2004 + a*Simil_2004 + b*Digits_2004 + c*LFLU_2004 + d*CFLU_2004
CogT2 =~ NA*Simil_2011 + a*Simil_2011 + b*Digits_2011 + c*LFLU_2011 + d*CFLU_2011
# Residuals
Simil_2004 ~~ Simil_2004
Simil_2011 ~~ Simil_2011
Digits_2004 ~~ Digits_2004
Digits_2011 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2004
LFLU_2011 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2004
CFLU_2011 ~~ CFLU_2011
# Residual covariances
Simil_2004 ~~ Simil_2011
Digits_2004 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2011
# Intercepts
Simil_2004 ~ nu1*1
Simil_2011 ~ nu1*1
Digits_2004 ~ nu11*1
Digits_2011 ~ nu11*1
LFLU_2004 ~ nu21*1
LFLU_2011 ~ nu21*1
CFLU_2004 ~ nu31*1
CFLU_2011 ~ nu31*1
#Latent variances/covariances
#Standardized CogT1
CogT1 ~~ 1*CogT1
#covariance
CogT2 ~~ cov*CogT1
#variance Time 2
CogT2 ~~ NA*CogT2
#Latent means
CogT1 ~ a1*1
CogT2 ~ a2*1
a1 == 0
'
# run the model and see how it fits.
model.StrongI.fit <- sem(model.StrongI, data = data, missing = "FIML")
summary(model.StrongI.fit, fit.measures = TRUE, standardized = T)
## lavaan 0.6.14 ended normally after 87 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 32
## Number of equality constraints 9
##
## Number of observations 7078
## Number of missing patterns 75
##
## Model Test User Model:
##
## Test statistic 109.972
## Degrees of freedom 21
## P-value (Chi-square) 0.000
##
## Model Test Baseline Model:
##
## Test statistic 7086.446
## Degrees of freedom 28
## P-value 0.000
##
## User Model versus Baseline Model:
##
## Comparative Fit Index (CFI) 0.987
## Tucker-Lewis Index (TLI) 0.983
##
## Robust Comparative Fit Index (CFI) 0.987
## Robust Tucker-Lewis Index (TLI) 0.983
##
## Loglikelihood and Information Criteria:
##
## Loglikelihood user model (H0) -95312.835
## Loglikelihood unrestricted model (H1) -95257.849
##
## Akaike (AIC) 190671.670
## Bayesian (BIC) 190829.560
## Sample-size adjusted Bayesian (SABIC) 190756.471
##
## Root Mean Square Error of Approximation:
##
## RMSEA 0.024
## 90 Percent confidence interval - lower 0.020
## 90 Percent confidence interval - upper 0.029
## P-value H_0: RMSEA <= 0.050 1.000
## P-value H_0: RMSEA >= 0.080 0.000
##
## Robust RMSEA 0.034
## 90 Percent confidence interval - lower 0.028
## 90 Percent confidence interval - upper 0.041
## P-value H_0: Robust RMSEA <= 0.050 1.000
## P-value H_0: Robust RMSEA >= 0.080 0.000
##
## Standardized Root Mean Square Residual:
##
## SRMR 0.021
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Observed
## Observed information based on Hessian
##
## Latent Variables:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## CogT1 =~
## Simil_2004 (a) 1.046 0.036 28.741 0.000 1.046 0.440
## Digts_2004 (b) 1.182 0.044 27.108 0.000 1.182 0.383
## LFLU_2004 (c) 2.361 0.074 31.954 0.000 2.361 0.542
## CFLU_2004 (d) 3.624 0.120 30.146 0.000 3.624 0.586
## CogT2 =~
## Simil_2011 (a) 1.046 0.036 28.741 0.000 1.069 0.458
## Digts_2011 (b) 1.182 0.044 27.108 0.000 1.208 0.459
## LFLU_2011 (c) 2.361 0.074 31.954 0.000 2.412 0.578
## CFLU_2011 (d) 3.624 0.120 30.146 0.000 3.704 0.610
##
## Covariances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 ~~
## .Sml_2011 2.084 0.082 25.493 0.000 2.084 0.471
## .Digits_2004 ~~
## .Dgt_2011 1.599 0.124 12.852 0.000 1.599 0.240
## .LFLU_2004 ~~
## .LFLU_201 5.563 0.314 17.693 0.000 5.563 0.446
## .CFLU_2004 ~~
## .CFLU_201 10.176 0.746 13.633 0.000 10.176 0.422
## CogT1 ~~
## CogT2 (cov) 0.941 0.025 37.130 0.000 0.920 0.920
##
## Intercepts:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Sm_2004 (nu1) 6.572 0.027 245.557 0.000 6.572 2.766
## .Sm_2011 (nu1) 6.572 0.027 245.557 0.000 6.572 2.816
## .Dg_2004 (nu11) 7.124 0.035 205.432 0.000 7.124 2.309
## .Dg_2011 (nu11) 7.124 0.035 205.432 0.000 7.124 2.706
## .LFLU_20 (nu21) 11.528 0.054 214.558 0.000 11.528 2.648
## .LFLU_20 (nu21) 11.528 0.054 214.558 0.000 11.528 2.761
## .CFLU_20 (nu31) 20.567 0.096 213.699 0.000 20.567 3.323
## .CFLU_20 (nu31) 20.567 0.096 213.699 0.000 20.567 3.389
## CogT1 (a1) 0.000 0.000 0.000
## CogT2 (a2) -0.275 0.017 -16.088 0.000 -0.269 -0.269
##
## Variances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 4.550 0.099 45.980 0.000 4.550 0.806
## .Simil_2011 4.304 0.101 42.642 0.000 4.304 0.790
## .Digits_2004 8.126 0.178 45.630 0.000 8.126 0.853
## .Digits_2011 5.472 0.143 38.216 0.000 5.472 0.790
## .LFLU_2004 13.387 0.383 34.915 0.000 13.387 0.706
## .LFLU_2011 11.608 0.361 32.174 0.000 11.608 0.666
## .CFLU_2004 25.170 0.903 27.881 0.000 25.170 0.657
## .CFLU_2011 23.105 0.923 25.044 0.000 23.105 0.627
## CogT1 1.000 1.000 1.000
## CogT2 1.045 0.044 23.829 0.000 1.000 1.000
##
## Constraints:
## |Slack|
## a1 - 0 0.000
### p-value for chi-square is once again significant, and once again we don't need to worry about it too much because of the sample size. we've gained 3 degrees of freedom going from WI to strong invariance, but our model fits significantly worse. AIC and BIC are a bit worse for strong invariance compared to WI, but it's not too big of a difference. CFI and TLI are a bit smaller for strong invariance compared to WI, but they still easily meet the 0.95 threshold, so still a good fit. RMSEA is a bit larger for strong invariance compared to WI, but it still easily meets the 0.05 threshold as well. overall, a little bit worse than WI, but still a good fit!
# visualization of the model.
semPaths(model.StrongI.fit, what = "est")
equate all sets of lambdas, all sets of intercepts, & all sets of residuals.
model.StrictI <- '
#Latent variables
CogT1 =~ NA*Simil_2004 + a*Simil_2004 + b*Digits_2004 + c*LFLU_2004 + d*CFLU_2004
CogT2 =~ NA*Simil_2011 + a*Simil_2011 + b*Digits_2011 + c*LFLU_2011 + d*CFLU_2011
# Residuals
Simil_2004 ~~ e*Simil_2004
Simil_2011 ~~ e*Simil_2011
Digits_2004 ~~ f*Digits_2004
Digits_2011 ~~ f*Digits_2011
LFLU_2004 ~~ g*LFLU_2004
LFLU_2011 ~~ g*LFLU_2011
CFLU_2004 ~~ h*CFLU_2004
CFLU_2011 ~~ h*CFLU_2011
# Residual covariances
Simil_2004 ~~ Simil_2011
Digits_2004 ~~ Digits_2011
LFLU_2004 ~~ LFLU_2011
CFLU_2004 ~~ CFLU_2011
# Intercepts
Simil_2004 ~ nu1*1
Simil_2011 ~ nu1*1
Digits_2004 ~ nu11*1
Digits_2011 ~ nu11*1
LFLU_2004 ~ nu21*1
LFLU_2011 ~ nu21*1
CFLU_2004 ~ nu31*1
CFLU_2011 ~ nu31*1
#Latent variances/covariances
#Standardized CogT1
CogT1 ~~ 1*CogT1
#covariance
CogT2 ~~ cov*CogT1
#variance Time 2
CogT2 ~~ NA*CogT2
#Latent means
CogT1 ~ a1*1
CogT2 ~ a2*1
a1 == 0
'
# run the model and see how it fits.
model.StrictI.fit <- sem(model.StrictI, data = data, missing = "FIML")
summary(model.StrictI.fit, fit.measures = TRUE, standardized = T)
## lavaan 0.6.14 ended normally after 75 iterations
##
## Estimator ML
## Optimization method NLMINB
## Number of model parameters 32
## Number of equality constraints 13
##
## Number of observations 7078
## Number of missing patterns 75
##
## Model Test User Model:
##
## Test statistic 291.616
## Degrees of freedom 25
## P-value (Chi-square) 0.000
##
## Model Test Baseline Model:
##
## Test statistic 7086.446
## Degrees of freedom 28
## P-value 0.000
##
## User Model versus Baseline Model:
##
## Comparative Fit Index (CFI) 0.962
## Tucker-Lewis Index (TLI) 0.958
##
## Robust Comparative Fit Index (CFI) 0.965
## Robust Tucker-Lewis Index (TLI) 0.961
##
## Loglikelihood and Information Criteria:
##
## Loglikelihood user model (H0) -95403.657
## Loglikelihood unrestricted model (H1) -95257.849
##
## Akaike (AIC) 190845.314
## Bayesian (BIC) 190975.744
## Sample-size adjusted Bayesian (SABIC) 190915.367
##
## Root Mean Square Error of Approximation:
##
## RMSEA 0.039
## 90 Percent confidence interval - lower 0.035
## 90 Percent confidence interval - upper 0.043
## P-value H_0: RMSEA <= 0.050 1.000
## P-value H_0: RMSEA >= 0.080 0.000
##
## Robust RMSEA 0.052
## 90 Percent confidence interval - lower 0.046
## 90 Percent confidence interval - upper 0.058
## P-value H_0: Robust RMSEA <= 0.050 0.280
## P-value H_0: Robust RMSEA >= 0.080 0.000
##
## Standardized Root Mean Square Residual:
##
## SRMR 0.041
##
## Parameter Estimates:
##
## Standard errors Standard
## Information Observed
## Observed information based on Hessian
##
## Latent Variables:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## CogT1 =~
## Simil_2004 (a) 1.091 0.037 29.821 0.000 1.091 0.461
## Digts_2004 (b) 1.208 0.046 26.532 0.000 1.208 0.416
## LFLU_2004 (c) 2.425 0.075 32.527 0.000 2.425 0.565
## CFLU_2004 (d) 3.703 0.121 30.602 0.000 3.703 0.599
## CogT2 =~
## Simil_2011 (a) 1.091 0.037 29.821 0.000 1.046 0.446
## Digts_2011 (b) 1.208 0.046 26.532 0.000 1.158 0.402
## LFLU_2011 (c) 2.425 0.075 32.527 0.000 2.324 0.549
## CFLU_2011 (d) 3.703 0.121 30.602 0.000 3.549 0.583
##
## Covariances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 ~~
## .Sml_2011 2.069 0.083 25.044 0.000 2.069 0.469
## .Digits_2004 ~~
## .Dgt_2011 1.702 0.131 12.992 0.000 1.702 0.244
## .LFLU_2004 ~~
## .LFLU_201 5.569 0.318 17.540 0.000 5.569 0.444
## .CFLU_2004 ~~
## .CFLU_201 10.412 0.754 13.816 0.000 10.412 0.426
## CogT1 ~~
## CogT2 (cov) 0.885 0.022 40.260 0.000 0.923 0.923
##
## Intercepts:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Sm_2004 (nu1) 6.572 0.027 244.621 0.000 6.572 2.776
## .Sm_2011 (nu1) 6.572 0.027 244.621 0.000 6.572 2.801
## .Dg_2004 (nu11) 7.136 0.035 206.075 0.000 7.136 2.459
## .Dg_2011 (nu11) 7.136 0.035 206.075 0.000 7.136 2.476
## .LFLU_20 (nu21) 11.504 0.054 214.995 0.000 11.504 2.681
## .LFLU_20 (nu21) 11.504 0.054 214.995 0.000 11.504 2.716
## .CFLU_20 (nu31) 20.573 0.096 213.681 0.000 20.573 3.331
## .CFLU_20 (nu31) 20.573 0.096 213.681 0.000 20.573 3.380
## CogT1 (a1) 0.000 NA 0.000 0.000
## CogT2 (a2) -0.264 0.016 -16.084 0.000 -0.276 -0.276
##
## Variances:
## Estimate Std.Err z-value P(>|z|) Std.lv Std.all
## .Simil_2004 (e) 4.414 0.083 53.253 0.000 4.414 0.788
## .Simil_2011 (e) 4.414 0.083 53.253 0.000 4.414 0.801
## .Digts_2004 (f) 6.966 0.126 55.373 0.000 6.966 0.827
## .Digts_2011 (f) 6.966 0.126 55.373 0.000 6.966 0.839
## .LFLU_2004 (g) 12.534 0.314 39.936 0.000 12.534 0.681
## .LFLU_2011 (g) 12.534 0.314 39.936 0.000 12.534 0.699
## .CFLU_2004 (h) 24.448 0.762 32.101 0.000 24.448 0.641
## .CFLU_2011 (h) 24.448 0.762 32.101 0.000 24.448 0.660
## CogT1 1.000 1.000 1.000
## CogT2 0.919 0.036 25.423 0.000 1.000 1.000
##
## Constraints:
## |Slack|
## a1 - 0 0.000
### p-value for chi-square is once again significant, and once again we don't need to worry about it too much because of the sample size. we've gained 4 degrees of freedom this time going from strong invariance to strict invariance, but our model fits significantly worse. AIC and BIC are a bit worse for strict compared to strong, but again not too big of a difference. CFI and TLI are smaller for strict compared to strong, but they still meet the 0.95 threshold, so still a good fit. RMSEA is larger for strict compared to strong, but it still meets the 0.05 threshold as well. overall, a little bit worse than strong, but still a good fit! this means that we can confidently say we've established measurement invariance. thus, the T2 mean of -0.264 indicates that on average, cognitive ability declined by -0.264 units from T1 to T2, with a variance of 0.919 units. that's a pretty small change, though.
# visualization of the model.
semPaths(model.StrictI.fit, what = "est")