SMHS MethodsHeterogeneity HTE - Revision history

Pineaumi: /* Latent growth and growth mixture modeling (LGM/GMM) */

2016-05-23T18:42:37Z

‎Latent growth and growth mixture modeling (LGM/GMM)

2016-05-23T18:40:39Z

‎Latent growth and growth mixture modeling (LGM/GMM)

2016-05-23T18:34:56Z

‎Latent growth and growth mixture modeling (LGM/GMM)

2016-05-23T18:33:22Z

‎Footnotes

2016-05-23T18:32:31Z

‎Latent growth and growth mixture modeling (LGM/GMM)

2016-05-23T18:31:20Z

‎Random Forests

2016-05-23T18:31:00Z

‎Example Fifth Dutch growth study

2016-05-23T18:28:28Z

‎Footnotes

2016-05-23T18:28:03Z

‎Random Forests

2016-05-23T18:13:09Z

‎Example Fifth Dutch growth study

← Older revision		Revision as of 18:42, 23 May 2016
Line 414:		Line 414:


−	model.hlme.lin <- lcmm(ntumors ~ Age*Sex, random=~ Age ,subject='ID', epsY = 0.5, convB = 1e-01, convL = 1e-01, <mark>convG = 1e-01</mark>, maxiter=200, data=hdp); model.hlme.lin$conv	+	model.hlme.lin <- lcmm(ntumors ~ Age*Sex, random=~ Age ,subject='ID', epsY = 0.5, convB = 1e-01, convL = 1e-01, <mark>convG = 1e-01</mark>, maxiter=200, data=hdp); model.hlme.lin$\$$conv

	<mark># Now that we have convergence, we can obtain CI’s!!!</mark>		<mark># Now that we have convergence, we can obtain CI’s!!!</mark>

@@ Line 286: / Line 286: @@
 In this example, the optimal number of latent classes according to the BIC is two (the smallest BIC).  The posterior classification is described with:
-  postprob(model.hlme.2)
+  postprob(model.hlme<mark>.2</mark>)
   Posterior classification:
@@ Line 341: / Line 341: @@
   lcmm(fixed = ntumors ~ Age * Sex, random = ~Age, subject = "ID",data = hdp)
-−
   Statistical Model:
       Dataset: hdp
@@ Line 349: / Line 349: @@
       Number of parameters: 8
       Link function: linear
-−
-Iteration process:
+  Iteration process:
       Maximum number of iteration reached without convergence
       Number of iterations:  100
@@ Line 356: / Line 356: @@
                           : likelihood= 5.5e-10
                           : second derivatives= 1
-−
   Goodness-of-fit statistics:
       maximum log-likelihood: -19915.24
@@ Line 368: / Line 368: @@
       Mean UACV per subject: 0
       Mean discrete LL per subject: 0
-−
   Maximum Likelihood Estimates:
-−
   Fixed effects in the longitudinal model:
@@ Line 378: / Line 378: @@
   Sexmale                   -0.66303
   Age:Sexmale                0.01132
   Variance-covariance matrix of the random-effects:

@@ Line 156: / Line 156: @@
   })
-add a new subject ID column (last column in the data, “ID”), this is necessary for the hmle call
+add a new subject ID column (last column in the data, “ID”), this is necessary for the '''hmle''' call
 hdp$\$$ID <- seq.int(nrow(hdp))
@@ Line 178: / Line 178: @@
   Iteration process:
-  Convergence criteria satisfied
+  '''Convergence criteria satisfied'''
   Number of iterations:  34
   Convergence criteria: parameters= 1.2e-09

@@ Line 434: / Line 434: @@
 ===Footnotes===
-*<sup>1</sup>  http://cran.r-project.org/web/packages/rpart/index.html
+*<sup>1</sup> http://cran.r-project.org/web/packages/rpart/index.html
-*<sup>2</sup>   http://www.mayo.edu/hsr/techrpt/61.pdf
+*<sup>2</sup> http://www.mayo.edu/hsr/techrpt/61.pdf
-*<sup>3</sup>   http://dx.doi.org/10.1371/journal.pone.0027608
+*<sup>3</sup> http://dx.doi.org/10.1371/journal.pone.0027608
-*<sup>4</sup>   http://www.nature.com/pr/journal/v73/n3/abs/pr2012189a.html
+*<sup>4</sup> http://www.nature.com/pr/journal/v73/n3/abs/pr2012189a.html
-*<sup>5</sup>  http://stat-www.berkeley.edu/users/breiman/RandomForests/
+*<sup>5</sup> http://stat-www.berkeley.edu/users/breiman/RandomForests/
 ===[[SMHS_MethodsHeterogeneity_MetaAnalysis|Next see: Meta-Analysis]]===

@@ Line 134: / Line 134: @@
 <b>(v)</b> missing at random data
-The objectives of LGM/GMM models (see <b>Latent Class Mixed Models, lcmm</b> R package) are to extend the linear mixed model estimation to:
+The objectives of LGM/GMM models (see <b>Latent Class Mixed Models, lcmm</b> R package<sup>6</sup>,<sup>7</sup>) are to extend the linear mixed model estimation to:
 <b>(i)</b>	heterogeneous populations (relax (iv) above). Use <mark><b>hlme</b> for latent class linear mixed models</mark> (i.e. Gaussian continuous outcome)

@@ Line 98: / Line 98: @@
 Not much change, as the initial tree is not complex!
-===Random Forests ===
+====(3) Random Forests ====
 Random forests may improve predictive accuracy by generating a large number of bootstrapped trees (based on random samples of variables). It classifies cases using each tree in this new "forest", and decides the final predicted outcome by combining the results across all of the trees (an average in regression, a majority vote in classification). See the <b>randomForest</b> package<sup>5</sup>.

@@ Line 65: / Line 65: @@
 </center>
-(1) Classification Tree
+====(1) Classification Tree====
 Let's use the data frame '''fdgs''' to predict Region, from Age, Height, and Weight.
@@ Line 87: / Line 87: @@
 <center>[[Image:SMHS_Methods3.png|500px]] </center>
-(2) Pruning the tree
+====(2) Pruning the tree====
   pruned.fit.1<- prune(fit.1, cp=   fit.1$\$$cptable[which.min(fit.1$\$$\$$cptable[,"xerror"]),"CP"])

@@ Line 99: / Line 99: @@
 ===Random Forests ===
-Random forests may improve predictive accuracy by generating a large number of bootstrapped trees (based on random samples of variables). It classifies cases using each tree in this new "forest", and decides the final predicted outcome by combining the results across all of the trees (an average in regression, a majority vote in classification). See the <b>randomForest</b> package.
+Random forests may improve predictive accuracy by generating a large number of bootstrapped trees (based on random samples of variables). It classifies cases using each tree in this new "forest", and decides the final predicted outcome by combining the results across all of the trees (an average in regression, a majority vote in classification). See the <b>randomForest</b> package<sup>5</sup>.
   library(randomForest)

@@ Line 67: / Line 67: @@
 (1) Classification Tree
-Let's use the data frame fdgs to predict Region, from Age, Height, and Weight.
+Let's use the data frame '''fdgs''' to predict Region, from Age, Height, and Weight.
   # grow tree
   fit.1 <- rpart(reg ~ age + hgt + wgt,   method="class", data= fdgs[,-1])