This page displays the analysis results for latent growth curve models (LGCM) and growth mixture models (GMM) run in Mplus. To do this, we have imported output files from Mplus and formatted the statistics and figures in R, using the MplusAutomation package.

To cite MplusAutomation in publications use:

Hallquist, M. N. & Wiley, J. F. (2018). MplusAutomation: An R Package for Facilitating Large-Scale Latent Variable Analyses in Mplus Structural Equation Modeling, 1-18. doi: 10.1080/10705511.2017.1402334.

A BibTeX entry for LaTeX users is

@Article{, title = {{MplusAutomation}: An {R} Package for Facilitating Large-Scale Latent Variable Analyses in {Mplus}}, author = {Michael N. Hallquist and Joshua F. Wiley}, journal = {Structural Equation Modeling}, year = {2018}, pages = {1–18}, doi = {10.1080/10705511.2017.1402334}, url = {https://www.tandfonline.com/doi/full/10.1080/10705511.2017.1402334}, }

Latent growth curve models (LGCM)

Model fit statistics

First read in the Mplus models, remember to check which data set you are reading in.

si.growth.all <- readModels(paste0(mplus_growth_traj_clustered_full_output_data_path), recursive = TRUE)

Extract the summary variables from the mplus output files. Assuming there are multiple files in the directory, model summaries could be retained as a data.frame as we have done here.

si_growth_summaries <- do.call("rbind.fill",
                        sapply(si.growth.all,
                               "[", 
                               "summaries"))

Cut offs and best fitting models are:

lowest aBIC
TLI > 0.95
CFI > 0.95
RMSEA < 0.05 indicates close fit, and < 0.08 indicates reasonable fit

We chose the Linear model as the best fitting and most parsimonious model.

#social isolation summaries 
si.growth.summaries <- data.frame(matrix(nrow = 4,ncol = 8)) 

#create column names for the variables you will be adding to the empty matrix of si.growth.summaries
colnames(si.growth.summaries) <- c("Model", 
                                  "Parameters", 
                                  "AIC", 
                                  "BIC", 
                                  "aBIC", 
                                  "CFI", 
                                  "TLI", 
                                  "RMSEA") #or whichever indices you want to compare; do si_summaries$ to see what's in there

#create "Model" variable
si.growth.summaries <- si.growth.summaries %>%
  mutate(
    Model = 
    as.factor(c("Linear growth model",
                "Log linear growth model",
                "Quadratic growth model", 
                "Log quadratic growth model")))

#check 
# si.growth.summaries

#add summary information into data frame 
si.growth.summaries <- si.growth.summaries %>%
  mutate(
    Parameters = 
      si_growth_summaries$Parameters) %>% #parameters
  mutate(
    AIC = 
      si_growth_summaries$AIC) %>% #AIC 
  mutate(
    BIC = 
      si_growth_summaries$BIC) %>% #BIC 
  mutate(
    aBIC = 
      si_growth_summaries$aBIC) %>% #aBIC 
  mutate(
    CFI = 
      si_growth_summaries$CFI) %>% #CFI
  mutate(
    TLI = 
      si_growth_summaries$TLI) %>% #TLI
  mutate(
    RMSEA = 
      si_growth_summaries$RMSEA_Estimate) #RMSEA
  
#check
# si.growth.summaries

knitr::kable(si.growth.summaries) #table for Word

Model	Parameters	AIC	BIC	aBIC	CFI	TLI	RMSEA
Linear growth model	9	26268.18	26319.58	26290.98	0.991	0.989	0.027
Log linear growth model	9	11069.13	11120.53	11091.93	0.993	0.992	0.030
Quadratic growth model	13	26255.23	26329.47	26288.16	0.997	0.984	0.033
Log quadratic growth model	13	11064.84	11139.08	11097.77	0.997	0.984	0.043

Means and variances

# linear
linear.mean <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Means") # filter out the means for I and S

linear.variance <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Variances") # filter out the variance for I and S

# quadratic
quad.mean <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_quad_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Means") # filter out the means for I, S and Q 

quad.variance <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_quad_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Variances") # filter out the variance for I, S and Q

Linear LCGM means and variances

linear.mean.var <- rbind(linear.mean, linear.variance) %>%
  select(`Mean/variance` = paramHeader, 
         `Intercept/slope` = param, 
         `estimate` = est, 
         `standard error` = se, 
         `p value` = pval)
kable(linear.mean.var)

Mean/variance	Intercept/slope	estimate	standard error
Means	I	0.808	0.026
Means	S	0.021	0.005
Variances	I	0.718	0.079
Variances	S	0.020	0.003

Quadratic LCGM means and variances

quad.mean.var <- rbind(quad.mean, quad.variance) %>%
  select(`Mean/variance` = paramHeader, 
         `Intercept/slope/quadratic` = param, 
         `estimate` = est, 
         `standard error` = se, 
         `p value` = pval)
kable(quad.mean.var)

Mean/variance	Intercept/slope/quadratic	estimate	standard error	p value
Means	I	0.806	0.027	0.000
Means	S	0.025	0.015	0.093
Means	Q	-0.001	0.002	0.773
Variances	I	0.788	0.127	0.000
Variances	S	0.090	0.029	0.002
Variances	Q	0.001	0.000	0.002

Plots

Linear and quadratic growth curves

#create empty dataset 
dat_si <- data.frame(matrix(nrow = 4, ncol = 3))

#add column names
colnames(dat_si) <- c("Observed", "Linear", "Quadratic")

#add in model statistics from out files - will need to change the folder names depending on which sample you are looking at

dat_si$Observed <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_full_sample_clustered.out$gh5$means_and_variances_data$y_observed_means$values #observed means - used linear but could be any

dat_si$Linear <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means$values #estimated means

dat_si$Quadratic <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_quad_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means$values #estimated means

#add timepoint column 
dat_si$Timepoint <- c(5, 7, 10, 12)

#convert into long format 
dat_si <- dat_si %>%
  tidyr::pivot_longer(-Timepoint, 
               names_to = "Model", 
               values_to = "score")
#check
dat_si

isolation_growth_curve_clustered_full_sample <- ggplot() + #change name of graph here if using the 3 missing sample 
  geom_line(data = dat_si,
        aes(x = Timepoint, 
            y = score, 
            colour = Model), 
        size = 1) +
  geom_point(data = dat_si,
        aes(x = Timepoint, 
            y = score, 
            colour = Model, 
            shape = Model), 
        show.legend = T) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       y ="Social isolation",
       title = "Growth curves for social isolation") +
  scale_y_continuous(expand = c(0.01,0.01), 
                     limits = c(0.8,0.96), 
                     breaks=seq(0.8, 0.96, 0.02)) +
  scale.x +
  combined.traj.theme

isolation_growth_curve_clustered_full_sample

Linear growth curve only

#create empty dataset 
dat_si_single <- data.frame(matrix(nrow = 4, ncol = 1))

#add column names
colnames(dat_si_single) <- c("Linear growth curve")

#add in model statistics from out files - will need to change the folder names depending on which sample
dat_si_single$`Linear growth curve` <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means$values #estimated means

#add timepoint column 
dat_si_single$timepoint <- c(5, 7, 10, 12)

#convert into long format 
dat_si_single <- dat_si_single %>%
  tidyr::pivot_longer(-timepoint, 
               names_to = "Class", 
               values_to = "social.isolation_score")
#check
#dat_si_single

isolation_growth_curve_single_clustered_full_sample <- ggplot() +
  geom_line(data = dat_si_single,
        aes(x = timepoint, 
            y = social.isolation_score, 
          #  colour = Class,
            shape = Class), 
        size = 1.5) +
  geom_point(data = dat_si_single,
        aes(x = timepoint, 
            y = social.isolation_score, 
           # colour = Class, 
            size = 1), 
        show.legend = T) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       y ="Social isolation",
       title = "Linear latent growth curve",
       color = "") +
  scale.y  +
  scale.x +
  combined.traj.theme +
  guides(shape = FALSE, size = FALSE)

isolation_growth_curve_single_clustered_full_sample

Log-transformed linear and quadratic growth curves

#create empty dataset 
dat_si_log <- data.frame(matrix(nrow = 4, ncol = 3))

#add column names
colnames(dat_si_log) <- c("Observed", "Linear", "Quadratic")

#add in model statistics from out files - need to change file name dependant on which sample
dat_si_log$Observed <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_log_full_sample_clustered.out$gh5$means_and_variances_data$y_observed_means$values #observed means

dat_si_log$Linear <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_linear_log_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means$values #estimated means

dat_si_log$Quadratic <- si.growth.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.growth_trajectories.clustered.full_sample.output..isolation_1traj_quad_log_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means$values #estimated means

#add timepoint column 
dat_si_log$Timepoint <- c(5, 7, 10, 12)

#convert into long format 
dat_si_log <- dat_si_log %>%
  pivot_longer(-Timepoint, 
               names_to = "Model", 
               values_to = "score")
#check
#dat_si_log

isolation_growth_curve_log_scale_clustered_full_sample <- ggplot() +
  geom_line(data = dat_si_log,
        aes(x = Timepoint, 
            y = score, 
            colour = Model), 
        size = 1) +
  geom_point(data = dat_si_log,
        aes(x = Timepoint, 
            y = score, 
            colour = Model, 
            shape=Model), 
        show.legend = T) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       y ="Social isolation",
       title = "Growth curves for log-social isolation") +
  scale_y_continuous(expand = c(0.01,0.01), 
                     limits = c(0.44,0.52), 
                     breaks=seq(0.44, 0.52, 0.01)) +
  scale_x_continuous(expand = c(0.1,0.1), 
                     limits = c(5,12), 
                     breaks= c(5, 7, 10, 12)) +
  combined.traj.theme

isolation_growth_curve_log_scale_clustered_full_sample

Growth mixture models (GMM)

As the linear curve was the best fit when running a single latent growth curve, all GMM models are lienar form.

Model fit statistics

#For class models, read in all output files within your folder that you have all your class models
si.classes.all <- readModels(paste0(mplus_GMM_clustered_full_output_data_path), recursive = TRUE) #Recursive means it reads data in sub-folders too

#extract the summary variables from the mplus output files. Assuming there are multiple files in the above directory, model summaries could be retained as a data.frame as follows:
si_summaries <- do.call("rbind.fill",
                        sapply(si.classes.all,
                               "[", 
                               "summaries"))

# have a look at the summary data for all files
si_summaries <- si_summaries[-c(2,3,4,5,7,8,11,13),]  # remove rows that are output from OPTSEED tests and keep 5 and 6 trajectory models that fix S to zero

#social isolation summaries 
si.class.summaries <- data.frame(matrix(nrow = 5, 
                                         ncol = 8)) 

#create column names for the variables you will be adding to the empty matrix of si.class.summaries
colnames(si.class.summaries) <- c("Model", 
                                  "Parameters", 
                                  "AIC", 
                                  "BIC", 
                                  "aBIC", 
                                  "Entropy", 
                                  "VLMR LRT p-value", 
                                  "Class Proportions") #or whichever indices you want to compare; do si_summaries$ to see what's in there

#check colnames
#si.class.summaries

#create "Model" variable
si.class.summaries <- si.class.summaries %>%
  mutate(
    Model = 
    as.factor(c("Two Class", #change nrow in data frame above when adding more models
                "Three Class", 
                "Four Class", 
                "Five Class",
                "Six Class")))  

#add summary information into data frame 
si.class.summaries <- si.class.summaries %>%
  mutate(
    Parameters = 
      si_summaries$Parameters) %>% #parameters col from original summaries data set
  mutate(
    AIC = 
      si_summaries$AIC) %>% #AIC col from original summaries data set
  mutate(
    BIC = 
      si_summaries$BIC) %>% #BIC col from original summaries data set
  mutate(
    aBIC = 
      si_summaries$aBIC) %>% #aBIC col from original summaries data set
  mutate(
    Entropy = 
      si_summaries$Entropy) %>% #Entropy col from original summaries data set
  mutate(
    `VLMR LRT p-value` = 
      si_summaries$T11_VLMR_PValue) #T11_VLMR_PValue col from original summaries data set

#check
#si.class.summaries

#for each class we want to add in the model estimated class counts - then convert this into a list with the percentage of people in each class

#two classes
si.class.summaries$`Class Proportions`[si.class.summaries$Model == "Two Class"] <- list(c(sprintf("%.0f", si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_2traj_full_sample_clustered.out$class_counts$modelEstimated$proportion*100)))

#three classes
si.class.summaries$`Class Proportions`[si.class.summaries$Model == "Three Class"] <- list(c(sprintf('%.0f', si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$class_counts$modelEstimated$proportion*100)))

#four classes
si.class.summaries$`Class Proportions`[si.class.summaries$Model == "Four Class"] <- list(c(sprintf('%.0f', si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_4traj_full_sample_clustered.out$class_counts$modelEstimated$proportion*100)))

#five classes
si.class.summaries$`Class Proportions`[si.class.summaries$Model == "Five Class"] <- list(c(sprintf('%.0f', si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_5traj_full_sample_clustered_S_fixed_to_zero.out$class_counts$modelEstimated$proportion*100)))

#six classes
si.class.summaries$`Class Proportions`[si.class.summaries$Model == "Six Class"] <- list(c(sprintf('%.0f', si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_6traj_full_sample_clustered_S_fixed_to_zero.out$class_counts$modelEstimated$proportion*100)))

#two classes
si.class.summaries$`Class N`[si.class.summaries$Model == "Two Class"] <- list(c(sprintf("%.0f", si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_2traj_full_sample_clustered.out$class_counts$posteriorProb$count)))

#three classes
si.class.summaries$`Class N`[si.class.summaries$Model == "Three Class"] <- list(c(sprintf("%.0f", si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$class_counts$posteriorProb$count)))

#four classes
si.class.summaries$`Class N`[si.class.summaries$Model == "Four Class"] <- list(c(sprintf("%.0f", si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_4traj_full_sample_clustered.out$class_counts$posteriorProb$count)))

#five classes
si.class.summaries$`Class N`[si.class.summaries$Model == "Five Class"] <- list(c(sprintf("%.0f", si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_5traj_full_sample_clustered_S_fixed_to_zero.out$class_counts$posteriorProb$count)))

#six classes
si.class.summaries$`Class N`[si.class.summaries$Model == "Six Class"] <- list(c(sprintf("%.0f", si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_6traj_full_sample_clustered_S_fixed_to_zero.out$class_counts$posteriorProb$count)))

The three class model is chosen as the best fitting model according to fit statistics and logical evaluation The five and six class models here have fixed the variance of S (the slope) to zero, as the original model produces negative variances and is not interpretable.

#Look at summary table 
kable(si.class.summaries)

Model	Parameters	AIC	BIC	aBIC	Entropy	VLMR LRT p-value	Class Proportions	Class N
Two Class	12	25447.50	25516.03	25477.90	0.951	0.0344	93, 7	2077, 155
Three Class	15	24922.10	25007.76	24960.11	0.952	0.0587	90, 5 , 5	2007, 110 , 115
Four Class	18	24629.97	24732.76	24675.57	0.936	0.8906	4 , 84, 2 , 11	83 , 1874, 36 , 238
Five Class	19	24352.62	24461.12	24400.76	0.942	0.0541	5 , 82, 1 , 1 , 10	110 , 1838, 33 , 20 , 232
Six Class	22	24163.34	24288.98	24219.08	0.928	0.2071	1 , 1 , 14, 77, 3 , 4	31 , 12 , 309 , 1719, 67 , 94

Means and variances

gmm.linear.means.1 <- si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Means" & LatentClass == 1) # filter out the means for I and S

gmm.linear.variances.1 <- si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Variances" & LatentClass == 1) # filter out the variance for I and S

gmm.linear.means.2 <- si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Means" & LatentClass == 2) # filter out the means for I and S

gmm.linear.variances.2 <- si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Variances" & LatentClass == 2) # filter out the variance for I and S

gmm.linear.means.3 <- si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Means" & LatentClass == 3) # filter out the means for I and S
  
gmm.linear.variances.3 <- si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$parameters$unstandardized %>%
  filter(paramHeader == "Variances" & LatentClass == 3) # filter out the variance for I and S

gmm.linear.means.variances <- rbind(gmm.linear.means.1, gmm.linear.variances.1, gmm.linear.means.2, gmm.linear.variances.2, gmm.linear.means.3, gmm.linear.variances.3)%>%
  select(Class = LatentClass,
         `Means/variances` = paramHeader,
         `Intercept/slope` = param,
         estimate = est,
         `standard error` = se,
         `p value` = pval)

kable(gmm.linear.means.variances)

Class	Means/variances	Intercept/slope	estimate	standard error	p value
1	Means	I	0.612	0.031	0.000
1	Means	S	0.012	0.006	0.045
1	Variances	I	0.171	0.055	0.002
1	Variances	S	0.003	0.002	0.125
2	Means	I	1.077	0.202	0.000
2	Means	S	0.541	0.062	0.000
2	Variances	I	0.171	0.055	0.002
2	Variances	S	0.003	0.002	0.125
3	Means	I	3.983	0.369	0.000
3	Means	S	-0.318	0.053	0.000
3	Variances	I	0.171	0.055	0.002
3	Variances	S	0.003	0.002	0.125

Plots

Two Class model

#create two class data frame with means and variances
two_class_si <- as.data.frame(
  si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_2traj_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means)

#name the variables
names(two_class_si) <- c("Class 1","Class 2")

#create timepoint column 
two_class_si$timepoint <- c(5, 7, 10, 12)

#convert data to long format 
two_class_si <- two_class_si %>%
  pivot_longer(-timepoint, 
               names_to = "Class", 
               values_to = "social.isolation_score")

# create factor for Class variable
two_class_si <- two_class_si %>%
  mutate(Class = 
              factor(Class, 
                     levels = c("Class 1","Class 2")))
two_class_si

two_traj_plot_clustered_full_sample <- ggplot(two_class_si,
        aes(x = timepoint, 
            y = social.isolation_score, 
          #  colour = Class,
            shape = Class)) +
          geom_line(size = 1.5) +
  geom_point(aes(size = 1)) +
  scale_fill_manual(values = palette) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       title = "Two trajectory model",
       color = "",
       y = "Social isolation") +
  scale.y +
  scale.x +
  combined.traj.theme + 
  guides(shape = FALSE, size = FALSE)

two_traj_plot_clustered_full_sample

Three Class model

#create three class data frame with means and variances
three_class_si <- as.data.frame(
  si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_3traj_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means)

#name the variables
names(three_class_si) <- c("Class 1","Class 2","Class 3")

#create time point column 
three_class_si$timepoint <- c(5, 7, 10, 12)

#convert data to long format 
three_class_si <- three_class_si %>%
  pivot_longer(-timepoint, 
               names_to = "Class", 
               values_to = "social.isolation_score")

# create factor for Class variable
three_class_si <- three_class_si %>%
  mutate(Class = 
              factor(Class, 
                     levels = c("Class 1","Class 2","Class 3")))
three_class_si

three_traj_plot_clustered_full_sample <- ggplot(three_class_si,
        aes(x = timepoint, 
            y = social.isolation_score, 
          #  colour = Class,
            shape = Class)) +
          geom_line(size = 1.5) +
  geom_point(aes(size = 1)) +
  scale_fill_manual(values = palette) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       title = "Three trajectory model",
       color = "",
       y ="Social isolation") + 
  scale.y +
  scale.x +
  combined.traj.theme + 
  guides(shape = FALSE, size = FALSE)

three_traj_plot_clustered_full_sample

Four Class model

#create four class data frame with means and variances
four_class_si <- as.data.frame(
  si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_4traj_full_sample_clustered.out$gh5$means_and_variances_data$y_estimated_means)

#name the variables
names(four_class_si) <- c("Class 1","Class 2","Class 3","Class 4")

#create timepoint column 
four_class_si$timepoint <- c(5, 7, 10, 12)

#convert data to long format 
four_class_si <- four_class_si %>%
  pivot_longer(-timepoint, 
               names_to = "Class", 
               values_to = "social.isolation_score")

# create factor for Class variable to order the colour selection in the graph
four_class_si <- four_class_si %>%
  mutate(Class = 
              factor(Class, 
                     levels = c("Class 2", 
                                "Class 4", 
                                "Class 1",
                                "Class 3")))
four_class_si

four_traj_plot_clustered_full_sample <- ggplot(four_class_si,
        aes(x = timepoint, 
            y = social.isolation_score, 
          #  colour = Class,
            shape = Class)) +
          geom_line(size = 1.5) +
  geom_point(aes(size = 1)) +
  scale_fill_manual(values = palette) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       title = "Four trajectory model",
       color = "",
       y ="Social isolation") + 
  scale.y +
  scale.x +
  combined.traj.theme + 
  guides(shape = FALSE, size = FALSE)

four_traj_plot_clustered_full_sample

Five Class model

#create five class data frame with means and variances
five_class_si <- as.data.frame(
  si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_5traj_full_sample_clustered_S_fixed_to_zero.out$gh5$means_and_variances_data$y_estimated_means)

#name the variables
names(five_class_si) <- c("Class 1","Class 2","Class 3","Class 4","Class 5")

#create timepoint column 
five_class_si$timepoint <- c(5, 7, 10, 12)

#convert data to long format 
five_class_si <- five_class_si %>%
  pivot_longer(-timepoint, 
               names_to = "Class", 
               values_to = "social.isolation_score")

# create factor for Class variable
five_class_si <- five_class_si %>%
  mutate(Class = 
              factor(Class, 
                     levels = c("Class 2",
                                "Class 5",
                                "Class 1",
                                "Class 4",
                                "Class 3")))
five_class_si

five_traj_plot_clustered_full_sample <- ggplot(five_class_si,
        aes(x = timepoint, 
            y = social.isolation_score, 
           # colour = Class,
            shape = Class)) +
          geom_line(size = 1.5) +
  geom_point(aes(size = 1)) +
  scale_fill_manual(values = palette) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       title = "Five trajectory model",
       color = "",
       y ="Social isolation") + 
  scale.y +
  scale.x +
  combined.traj.theme + 
  guides(shape = FALSE, size = FALSE)

five_traj_plot_clustered_full_sample

Six Class model

#create five class data frame with means and variances
six_class_si <- as.data.frame(
  si.classes.all$X.Users.katiethompson.Documents.PhD.LISS.DTP_Louise_and_Tim.Social.isolation.trajectories_Paper.1.data_analysis.mplus.GMM.clustered.full_sample.output..isolation_6traj_full_sample_clustered_S_fixed_to_zero.out$gh5$means_and_variances_data$y_estimated_means)

#name the variables
names(six_class_si) <- c("Class 1","Class 2","Class 3","Class 4","Class 5","Class 6")

#create timepoint column 
six_class_si$timepoint <- c(5, 7, 10, 12)

#convert data to long format 
six_class_si <- six_class_si %>%
  pivot_longer(-timepoint, 
               names_to = "Class", 
               values_to = "social.isolation_score")

six_class_si <- six_class_si %>%
  mutate(Class = 
              factor(Class, 
                     levels = c("Class 4",
                                "Class 3",
                                "Class 6",
                                "Class 2",
                                "Class 1",
                                "Class 5")))
six_class_si

six_traj_plot_clustered_full_sample <- ggplot(six_class_si,
        aes(x = timepoint, 
            y = social.isolation_score, 
           # colour = Class,
            shape = Class)) +
          geom_line(size = 1.5) +
  geom_point(aes(size = 1)) +
  scale_fill_manual(values = palette) +
  scale_color_manual(values = palette) +
  labs(x = "Age (years)", 
       title = "Six trajectory model",
       color = "",
       y = "Social isolation") +   
  scale.y +
  scale.x +
  combined.traj.theme + 
  guides(shape = FALSE, size = FALSE)

six_traj_plot_clustered_full_sample

Plot for all models

#combine all plots
library(ggpubr) # Needed for combining plots
library(gridExtra)
combined_trajectories_plot_clustered_full_sample <- ggarrange(isolation_growth_curve_single_clustered_full_sample, 
                                        two_traj_plot_clustered_full_sample, 
                                        three_traj_plot_clustered_full_sample, 
                                        four_traj_plot_clustered_full_sample, 
                                        five_traj_plot_clustered_full_sample,
                                        six_traj_plot_clustered_full_sample,
                                        ncol = 2, nrow = 3,
                                        widths = c(2.7,2.7,2.7,2.7), 
                                        heights = c(3,3,3,3),
                                        labels = c("A", "B", "C", "D", "E", "F"),
                                        legend = "none"
                                        )

# combined_trajectories_plot_clustered_full_sample_CENTRE_PRES <- ggarrange(isolation_growth_curve_single_clustered_full_sample, 
#                                         two_traj_plot_clustered_full_sample, 
#                                         three_traj_plot_clustered_full_sample, 
#                                         four_traj_plot_clustered_full_sample, 
#                                         five_traj_plot_clustered_full_sample,
#                                         six_traj_plot_clustered_full_sample,
#                                         ncol = 3, nrow = 2,
#                                         widths = c(2.7,2.7,2.7,2.7), 
#                                         heights = c(3,3,3,3),
#                                         labels = c("A", "B", "C", "D", "E", "F"),
#                                         legend = "none"
#                                         )

combined_trajectories_plot_clustered_full_sample

Work by Katherine N Thompson

katherine.n.thompson@kcl.ac.uk

GMM trajectory statistics and figures

Latent growth curve models (LGCM)

Model fit statistics

Means and variances

Linear LCGM means and variances

Quadratic LCGM means and variances

Plots

Linear and quadratic growth curves

Linear growth curve only

Log-transformed linear and quadratic growth curves

Growth mixture models (GMM)

Model fit statistics

Means and variances

Plots

Two Class model

Three Class model

Four Class model

Five Class model

Six Class model

Plot for all models