SOCR Simulated HELP Data Activity - Revision history

Dinov: /* Multiple imputation */

2014-09-24T17:04:47Z

‎Multiple imputation

2014-09-14T21:10:07Z

‎Multiple imputation

2014-09-14T21:09:50Z

‎Propensity score modeling

2014-09-14T21:06:12Z

‎Multiple imputation

2014-09-14T20:34:11Z

‎Hierarchical clustering

2014-09-14T20:27:54Z

‎Linear discriminant analysis

2014-09-14T20:07:07Z

‎Recursive partitioning

2014-09-14T19:51:52Z

‎Recursive partitioning

2014-09-14T19:48:24Z

‎Factor analysis

2014-09-14T19:42:28Z

‎Bayesian Poisson regression

← Older revision		Revision as of 17:04, 24 September 2014
Line 525:		Line 525:

	===[[SMHS#Chapter_IV:_Special_Topics\|Multiple imputation]]===		===[[SMHS#Chapter_IV:_Special_Topics\|Multiple imputation]]===
		+
		+	Also see the [[SMHS_MissingData\| Missing Values section of the Scientific Methods for Health Sciences EBook]].

	miss_data <- with(help_sim_data, data.frame(homeless, female, i1, sexrisk, indtot, mcs, pcs))		miss_data <- with(help_sim_data, data.frame(homeless, female, i1, sexrisk, indtot, mcs, pcs))

@@ Line 544: / Line 544: @@
   IMP_no_noise <- mi(miss_data.transformed, n.iter=6, add.noise=FALSE)
-  # check convergence, should get FALSE here because only n.iter is small
+ # check convergence, should get FALSE here because only n.iter is small
   converged(IMP, check = "data")
   # converged(IMP, check = "coefs")

← Older revision		Revision as of 21:09, 14 September 2014
Line 551:		Line 551:

	===Propensity score modeling===		===Propensity score modeling===
		+
		+	# Propensity score (PS) techniques may be applied for analysis of
		+	# observational and registry data. The PS is the conditional probability
		+	# of a certain treatment given patient’s covariates. PS may be used to
		+	# eliminate imbalances in baseline covariate distributions between
		+	# treatment groups and to estimate marginal effects.

	==References==		==References==

← Older revision		Revision as of 21:06, 14 September 2014
Line 525:		Line 525:

	===[[SMHS#Chapter_IV:_Special_Topics\|Multiple imputation]]===		===[[SMHS#Chapter_IV:_Special_Topics\|Multiple imputation]]===
		+
		+	miss_data <- with(help_sim_data, data.frame(homeless, female, i1, sexrisk, indtot, mcs, pcs))
		+	summary(miss_data)
		+	# library(Hmisc)
		+	# na.pattern(miss_data)
		+
		+	install.packages("mi")
		+	library(mi)
		+	inf <- mi.info(miss_data)
		+	# run the imputation without data transformation
		+	IMP <- mi(miss_data, info=inf, check.coef.convergence=TRUE, add.noise=noise.control(post.run.iter=10))
		+
		+	# run the imputation with data transformation
		+	miss_data.transformed <- mi.preprocess(miss_data, inf)
		+	IMP_T <- mi(miss_data.transformed, n.iter=6, check.coef.convergence=TRUE, add.noise=noise.control(post.run.iter=6))
		+
		+	# Impute without noise
		+	IMP_no_noise <- mi(miss_data.transformed, n.iter=6, add.noise=FALSE)
		+
		+	# check convergence, should get FALSE here because only n.iter is small
		+	converged(IMP, check = "data")
		+	# converged(IMP, check = "coefs")
		+	# visually check the imputation
		+	plot(IMP); plot(IMP_T)

	===Propensity score modeling===		===Propensity score modeling===

← Older revision		Revision as of 20:34, 14 September 2014
Line 499:		Line 499:

	===[[SMHS_HLM\|Hierarchical clustering]]===		===[[SMHS_HLM\|Hierarchical clustering]]===
		+
		+	# To group similar variables or cluster observations a hierarchical structure
		+	# may be constructed. hclust() and kmeans() are examples of such clustering
		+	# packages
		+
		+	cor_mat <- cor(cbind(mcs, pcs, cesd, i1, a15a, homeless, sexrisk), use="pairwise.complete.obs")
		+	hier_clust <- hclust(dist(cor_mat))
		+	plot(hier_clust)
		+
		+	km_clust <- kmeans(cor_mat, 5, nstart = 25)
		+	plot(cor_mat, col = km_clust$cluster)
		+	points(km_clust$centers, col = 1:5, pch = 8)

	===[[SMHS_ROC\|ROC curve]]===		===[[SMHS_ROC\|ROC curve]]===

← Older revision		Revision as of 20:27, 14 September 2014
Line 486:		Line 486:

	===Linear discriminant analysis===		===Linear discriminant analysis===
		+
		+	# Fisher's linear discriminant analysis (LDA) identifies linear combinations of
		+	# variables splitting the data into separate classes. LDA can help us
		+	# identify clusters homelessness, using a uniform prior
		+
		+	library(MASS)
		+	ngroups <- length(unique(homeless))
		+	lda_model <- lda(homeless ~ age + cesd + mcs + pcs, prior=rep(1/(ngroups-1), (ngroups-1)))
		+	print(lda_model)
		+	plot(lda_model)
		+	plot(lda_model, dimen=1, type="both")

	===[[SMHS_HLM\|Hierarchical clustering]]===		===[[SMHS_HLM\|Hierarchical clustering]]===

@@ Line 467: / Line 467: @@
   # Recursive partitioning creates a decision tree to classify the observations based on
-  # categorical predictors. Here we classify subjects based on their homelessness status,
+  # categorical predictors. Here we classify subjects based on their race,
   # using the following predictors: gender, drinking, primary substance, RAB sexrisk, MCS, and PCS.
   library(rpart)
   subst <- as.factor(substance)
-  homelessness.rpart <- rpart(homeless ~ female + i1 + subst + sexrisk + mcs + pcs, method="class", data=help_sim_data )
+  homelessness.rpart <- rpart(racegrp ~ female + i1 + sub + sexrisk + mcs + pcs, method="class", data=help_sim_data)
   printcp(homelessness.rpart)
 ===Linear discriminant analysis===

← Older revision		Revision as of 19:51, 14 September 2014
Line 465:		Line 465:

	===Recursive partitioning===		===Recursive partitioning===
		+
		+	# Recursive partitioning creates a decision tree to classify the observations based on
		+	# categorical predictors. Here we classify subjects based on their homelessness status,
		+	# using the following predictors: gender, drinking, primary substance, RAB sexrisk, MCS, and PCS.
		+
		+	library(rpart)
		+	subst <- as.factor(substance)
		+	homelessness.rpart <- rpart(homeless ~ female + i1 + subst + sexrisk + mcs + pcs, method="class", data=help_sim_data )
		+	printcp(homelessness.rpart)

	===Linear discriminant analysis===		===Linear discriminant analysis===

← Older revision		Revision as of 19:48, 14 September 2014
Line 451:		Line 451:

	===[[SMHS_PCA_ICA_FA\|Factor analysis]]===		===[[SMHS_PCA_ICA_FA\|Factor analysis]]===
		+
		+	# Factor analysis explains the (observed) variability of a multivariate dataset in terms of
		+	# various underlying unobservable factors. The observed data elements are expressed as
		+	# linear functions of the factors and some random error.
		+	# This is an example of a maximum likelihood factor analysis of "cesd"
		+	# (Center for Epidemiologic Studies–Depression) scale.
		+
		+	fact_ana <- factanal(~ f1a + f1b + f1c + f1d + f1e + f1f + f1g + f1h +
		+	f1i + f1j + f1k + f1l + f1m + f1n + f1o + f1p + f1q + f1r +
		+	f1s + f1t, factors=3, rotation="varimax", na.action=na.omit,
		+	scores="regression")
		+	print(fact_ana , cutoff=0.45, sort=TRUE)

	===Recursive partitioning===		===Recursive partitioning===

@@ Line 438: / Line 438: @@
   # Fit a Cox proportional hazards regression model
   coxph(formula = Surv(dayslink, linkstatus) ~ treat + age + female + cesd, data = help_sim_data, method = "breslow")
 ===[[SMHS_Cronbachs|Cronbach’s $\alpha$]]===