Package 'Hmisc'

Description

%nin% is a binary operator, which returns a logical vector indicating if there is a match or not for its left operand. A true vector element indicates no match in left operand, false indicates a match.

Usage

x %nin% table

Arguments

Value

vector of logical values with length equal to length of x.

See Also

match %in%

Examples

c('a','b','c') %nin% c('a','b')

Description

Computes the mean and median of various absolute errors related to ordinary multiple regression models. The mean and median absolute errors correspond to the mean square due to regression, error, and total. The absolute errors computed are derived from $\hat{Y} - \mbox{median($\hat{Y}$)}$, $\hat{Y} - Y$, and $Y - \mbox{median($Y$)}$. The function also computes ratios that correspond to $R^2$ and $1 - R^2$ (but these ratios do not add to 1.0); the $R^2$ measure is the ratio of mean or median absolute $\hat{Y} - \mbox{median($\hat{Y}$)}$ to the mean or median absolute $Y - \mbox{median($Y$)}$. The $1 - R^2$ or SSE/SST measure is the mean or median absolute $\hat{Y} - Y$ divided by the mean or median absolute $\hat{Y} - \mbox{median($Y$)}$.

Usage

abs.error.pred(fit, lp=NULL, y=NULL)

## S3 method for class 'abs.error.pred'
print(x, ...)

Arguments

Value

a list of class abs.error.pred (used by print.abs.error.pred) containing two matrices: differences and ratios.

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University School of Medicine
[email protected]

References

Schemper M (2003): Stat in Med 22:2299-2308.

Tian L, Cai T, Goetghebeur E, Wei LJ (2007): Biometrika 94:297-311.

See Also

lm, ols, cor, validate.ols

Examples

set.seed(1)         # so can regenerate results
x1 <- rnorm(100)
x2 <- rnorm(100)
y  <- exp(x1+x2+rnorm(100))
f <- lm(log(y) ~ x1 + poly(x2,3), y=TRUE)
abs.error.pred(lp=exp(fitted(f)), y=y)
rm(x1,x2,y,f)

Description

Add Spike Histograms and Extended Box Plots to ggplot

Usage

addggLayers(
  g,
  data,
  type = c("ebp", "spike"),
  ylim = layer_scales(g)$y$get_limits(),
  by = "variable",
  value = "value",
  frac = 0.065,
  mult = 1,
  facet = NULL,
  pos = c("bottom", "top"),
  showN = TRUE
)

Arguments

Details

For an example see this. Note that it was not possible to just create the layers needed to be added, as creating these particular layers in isolation resulted in a ggplot error.

Value

the original ggplot object with more layers added

Author(s)

Frank Harrell

See Also

spikecomp()

Description

Given a data frame and the names of variable, doubles the data frame for each variable with a new category "All" by default, or by the value of label. A new variable .marginal. is added to the resulting data frame, with value "" if the observation is an original one, and with value equal to the names of the variable being marginalized (separated by commas) otherwise. If there is another stratification variable besides the one in ..., and that variable is nested inside the variable in ..., specify nested=variable name to have the value of that variable set fo label whenever marginal observations are created for .... See the state-city example below.

Usage

addMarginal(data, ..., label = "All", margloc=c('last', 'first'), nested)

Arguments

Examples

d <- expand.grid(sex=c('female', 'male'), country=c('US', 'Romania'),
                 reps=1:2)
addMarginal(d, sex, country)

# Example of nested variables
d <- data.frame(state=c('AL', 'AL', 'GA', 'GA', 'GA'),
                city=c('Mobile', 'Montgomery', 'Valdosto',
                       'Augusta', 'Atlanta'),
                x=1:5, stringsAsFactors=TRUE)
addMarginal(d, state, nested=city) # cite set to 'All' when state is

Description

Tests, without issuing warnings, whether all elements of a character vector are legal numeric values, or optionally converts the vector to a numeric vector. Leading and trailing blanks in x are ignored.

Usage

all.is.numeric(x, what = c("test", "vector", "nonnum"), extras=c('.','NA'))

Arguments

Value

a logical value if what="test" or a vector otherwise

Author(s)

Frank Harrell

See Also

as.numeric

Examples

all.is.numeric(c('1','1.2','3'))
all.is.numeric(c('1','1.2','3a'))
all.is.numeric(c('1','1.2','3'),'vector')
all.is.numeric(c('1','1.2','3a'),'vector')
all.is.numeric(c('1','',' .'),'vector')
all.is.numeric(c('1', '1.2', '3a'), 'nonnum')

Description

Works in conjunction with the approx function to do linear extrapolation. approx in R does not support extrapolation at all, and it is buggy in S-Plus 6.

Usage

approxExtrap(x, y, xout, method = "linear", n = 50, rule = 2, f = 0,
             ties = "ordered", na.rm = FALSE)

Arguments

Details

Duplicates in x (and corresponding y elements) are removed before using approx.

Value

a vector the same length as xout

Author(s)

Frank Harrell

See Also

approx

Examples

approxExtrap(1:3,1:3,xout=c(0,4))

Description

Expands continuous variables into restricted cubic spline bases and categorical variables into dummy variables and fits a multivariate equation using canonical variates. This finds optimum transformations that maximize $R^2$. Optionally, the bootstrap is used to estimate the covariance matrix of both left- and right-hand-side transformation parameters, and to estimate the bias in the $R^2$ due to overfitting and compute the bootstrap optimism-corrected $R^2$. Cross-validation can also be used to get an unbiased estimate of $R^2$ but this is not as precise as the bootstrap estimate. The bootstrap and cross-validation may also used to get estimates of mean and median absolute error in predicted values on the original y scale. These two estimates are perhaps the best ones for gauging the accuracy of a flexible model, because it is difficult to compare $R^2$ under different y-transformations, and because $R^2$ allows for an out-of-sample recalibration (i.e., it only measures relative errors).

Note that uncertainty about the proper transformation of y causes an enormous amount of model uncertainty. When the transformation for y is estimated from the data a high variance in predicted values on the original y scale may result, especially if the true transformation is linear. Comparing bootstrap or cross-validated mean absolute errors with and without restricted the y transform to be linear (ytype='l') may help the analyst choose the proper model complexity.

Usage

areg(x, y, xtype = NULL, ytype = NULL, nk = 4,
     B = 0, na.rm = TRUE, tolerance = NULL, crossval = NULL)

## S3 method for class 'areg'
print(x, digits=4, ...)

## S3 method for class 'areg'
plot(x, whichx = 1:ncol(x$x), ...)

## S3 method for class 'areg'
predict(object, x, type=c('lp','fitted','x'),
                       what=c('all','sample'), ...)

Arguments

Details

areg is a competitor of ace in the acepack package. Transformations from ace are seldom smooth enough and are often overfitted. With areg the complexity can be controlled with the nk parameter, and predicted values are easy to obtain because parametric functions are fitted.

If one side of the equation has a categorical variable with more than two categories and the other side has a continuous variable not assumed to act linearly, larger sample sizes are needed to reliably estimate transformations, as it is difficult to optimally score categorical variables to maximize $R^2$ against a simultaneously optimally transformed continuous variable.

Value

a list of class "areg" containing many objects

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[email protected]

References

Breiman and Friedman, Journal of the American Statistical Association (September, 1985).

See Also

cancor,ace, transcan

Examples

set.seed(1)

ns <- c(30,300,3000)
for(n in ns) {
  y <- sample(1:5, n, TRUE)
  x <- abs(y-3) + runif(n)
  par(mfrow=c(3,4))
  for(k in c(0,3:5)) {
    z <- areg(x, y, ytype='c', nk=k)
    plot(x, z$tx)
	title(paste('R2=',format(z$rsquared)))
    tapply(z$ty, y, range)
    a <- tapply(x,y,mean)
    b <- tapply(z$ty,y,mean)
    plot(a,b)
	abline(lsfit(a,b))
    # Should get same result to within linear transformation if reverse x and y
    w <- areg(y, x, xtype='c', nk=k)
    plot(z$ty, w$tx)
    title(paste('R2=',format(w$rsquared)))
    abline(lsfit(z$ty, w$tx))
 }
}

par(mfrow=c(2,2))
# Example where one category in y differs from others but only in variance of x
n <- 50
y <- sample(1:5,n,TRUE)
x <- rnorm(n)
x[y==1] <- rnorm(sum(y==1), 0, 5)
z <- areg(x,y,xtype='l',ytype='c')
z
plot(z)
z <- areg(x,y,ytype='c')
z
plot(z)

## Not run: 		
# Examine overfitting when true transformations are linear
par(mfrow=c(4,3))
for(n in c(200,2000)) {
  x <- rnorm(n); y <- rnorm(n) + x
    for(nk in c(0,3,5)) {
    z <- areg(x, y, nk=nk, crossval=10, B=100)
    print(z)
    plot(z)
    title(paste('n=',n))
  }
}
par(mfrow=c(1,1))

# Underfitting when true transformation is quadratic but overfitting
# when y is allowed to be transformed
set.seed(49)
n <- 200
x <- rnorm(n); y <- rnorm(n) + .5*x^2
#areg(x, y, nk=0, crossval=10, B=100)
#areg(x, y, nk=4, ytype='l', crossval=10, B=100)
z <- areg(x, y, nk=4) #, crossval=10, B=100)
z
# Plot x vs. predicted value on original scale.  Since y-transform is
# not monotonic, there are multiple y-inverses
xx <- seq(-3.5,3.5,length=1000)
yhat <- predict(z, xx, type='fitted')
plot(x, y, xlim=c(-3.5,3.5))
for(j in 1:ncol(yhat)) lines(xx, yhat[,j], col=j)
# Plot a random sample of possible y inverses
yhats <- predict(z, xx, type='fitted', what='sample')
points(xx, yhats, pch=2)

## End(Not run)

# True transformation of x1 is quadratic, y is linear
n <- 200
x1 <- rnorm(n); x2 <- rnorm(n); y <- rnorm(n) + x1^2
z <- areg(cbind(x1,x2),y,xtype=c('s','l'),nk=3)
par(mfrow=c(2,2))
plot(z)

# y transformation is inverse quadratic but areg gets the same answer by
# making x1 quadratic
n <- 5000
x1 <- rnorm(n); x2 <- rnorm(n); y <- (x1 + rnorm(n))^2
z <- areg(cbind(x1,x2),y,nk=5)
par(mfrow=c(2,2))
plot(z)

# Overfit 20 predictors when no true relationships exist
n <- 1000
x <- matrix(runif(n*20),n,20)
y <- rnorm(n)
z <- areg(x, y, nk=5)  # add crossval=4 to expose the problem

# Test predict function
n <- 50
x <- rnorm(n)
y <- rnorm(n) + x
g <- sample(1:3, n, TRUE)
z <- areg(cbind(x,g),y,xtype=c('s','c'))
range(predict(z, cbind(x,g)) - z$linear.predictors)

Description

The transcan function creates flexible additive imputation models but provides only an approximation to true multiple imputation as the imputation models are fixed before all multiple imputations are drawn. This ignores variability caused by having to fit the imputation models. aregImpute takes all aspects of uncertainty in the imputations into account by using the bootstrap to approximate the process of drawing predicted values from a full Bayesian predictive distribution. Different bootstrap resamples are used for each of the multiple imputations, i.e., for the ith imputation of a sometimes missing variable, i=1,2,... n.impute, a flexible additive model is fitted on a sample with replacement from the original data and this model is used to predict all of the original missing and non-missing values for the target variable.

areg is used to fit the imputation models. By default, linearity is assumed for target variables (variables being imputed) and nk=3 knots are assumed for continuous predictors transformed using restricted cubic splines. If nk is three or greater and tlinear is set to FALSE, areg simultaneously finds transformations of the target variable and of all of the predictors, to get a good fit assuming additivity, maximizing $R^2$, using the same canonical correlation method as transcan. Flexible transformations may be overridden for specific variables by specifying the identity transformation for them. When a categorical variable is being predicted, the flexible transformation is Fisher's optimum scoring method. Nonlinear transformations for continuous variables may be nonmonotonic. If nk is a vector, areg's bootstrap and crossval=10 options will be used to help find the optimum validating value of nk over values of that vector, at the last imputation iteration. For the imputations, the minimum value of nk is used.

Instead of defaulting to taking random draws from fitted imputation models using random residuals as is done by transcan, aregImpute by default uses predictive mean matching with optional weighted probability sampling of donors rather than using only the closest match. Predictive mean matching works for binary, categorical, and continuous variables without the need for iterative maximum likelihood fitting for binary and categorical variables, and without the need for computing residuals or for curtailing imputed values to be in the range of actual data. Predictive mean matching is especially attractive when the variable being imputed is also being transformed automatically. Constraints may be placed on variables being imputed with predictive mean matching, e.g., a missing hospital discharge date may be required to be imputed from a donor observation whose discharge date is before the recipient subject's first post-discharge visit date. See Details below for more information about the algorithm. A "regression" method is also available that is similar to that used in transcan. This option should be used when mechanistic missingness requires the use of extrapolation during imputation.

A print method summarizes the results, and a plot method plots distributions of imputed values. Typically, fit.mult.impute will be called after aregImpute.

If a target variable is transformed nonlinearly (i.e., if nk is greater than zero and tlinear is set to FALSE) and the estimated target variable transformation is non-monotonic, imputed values are not unique. When type='regression', a random choice of possible inverse values is made.

The reformM function provides two ways of recreating a formula to give to aregImpute by reordering the variables in the formula. This is a modified version of a function written by Yong Hao Pua. One can specify nperm to obtain a list of nperm randomly permuted variables. The list is converted to a single ordinary formula if nperm=1. If nperm is omitted, variables are sorted in descending order of the number of NAs. reformM also prints a recommended number of multiple imputations to use, which is a minimum of 5 and the percent of incomplete observations.

Usage

aregImpute(formula, data, subset, n.impute=5, group=NULL,
           nk=3, tlinear=TRUE, type=c('pmm','regression','normpmm'),
           pmmtype=1, match=c('weighted','closest','kclosest'),
           kclosest=3, fweighted=0.2,
           curtail=TRUE, constraint=NULL,
           boot.method=c('simple', 'approximate bayesian'),
           burnin=3, x=FALSE, pr=TRUE, plotTrans=FALSE, tolerance=NULL, B=75)
## S3 method for class 'aregImpute'
print(x, digits=3, ...)
## S3 method for class 'aregImpute'
plot(x, nclass=NULL, type=c('ecdf','hist'),
     datadensity=c("hist", "none", "rug", "density"),
     diagnostics=FALSE, maxn=10, ...)
reformM(formula, data, nperm)

Arguments

Details

The sequence of steps used by the aregImpute algorithm is the following.
(1) For each variable containing m NAs where m > 0, initialize the NAs to values from a random sample (without replacement if a sufficient number of non-missing values exist) of size m from the non-missing values.
(2) For burnin+n.impute iterations do the following steps. The first burnin iterations provide a burn-in, and imputations are saved only from the last n.impute iterations.
(3) For each variable containing any NAs, draw a sample with replacement from the observations in the entire dataset in which the current variable being imputed is non-missing. Fit a flexible additive model to predict this target variable while finding the optimum transformation of it (unless the identity transformation is forced). Use this fitted flexible model to predict the target variable in all of the original observations. Impute each missing value of the target variable with the observed value whose predicted transformed value is closest to the predicted transformed value of the missing value (if match="closest" and type="pmm"), or use a draw from a multinomial distribution with probabilities derived from distance weights, if match="weighted" (the default).
(4) After these imputations are computed, use these random draw imputations the next time the curent target variable is used as a predictor of other sometimes-missing variables.

When match="closest", predictive mean matching does not work well when fewer than 3 variables are used to predict the target variable, because many of the multiple imputations for an observation will be identical. In the extreme case of one right-hand-side variable and assuming that only monotonic transformations of left and right-side variables are allowed, every bootstrap resample will give predicted values of the target variable that are monotonically related to predicted values from every other bootstrap resample. The same is true for Bayesian predicted values. This causes predictive mean matching to always match on the same donor observation.

When the missingness mechanism for a variable is so systematic that the distribution of observed values is truncated, predictive mean matching does not work. It will only yield imputed values that are near observed values, so intervals in which no values are observed will not be populated by imputed values. For this case, the only hope is to make regression assumptions and use extrapolation. With type="regression", aregImpute will use linear extrapolation to obtain a (hopefully) reasonable distribution of imputed values. The "regression" option causes aregImpute to impute missing values by adding a random sample of residuals (with replacement if there are more NAs than measured values) on the transformed scale of the target variable. After random residuals are added, predicted random draws are obtained on the original untransformed scale using reverse linear interpolation on the table of original and transformed target values (linear extrapolation when a random residual is large enough to put the random draw prediction outside the range of observed values). The bootstrap is used as with type="pmm" to factor in the uncertainty of the imputation model.

As model uncertainty is high when the transformation of a target variable is unknown, tlinear defaults to TRUE to limit the variance in predicted values when nk is positive.

Value

a list of class "aregImpute" containing the following elements:

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[email protected]

References

van Buuren, Stef. Flexible Imputation of Missing Data. Chapman & Hall/CRC, Boca Raton FL, 2012.

Little R, An H. Robust likelihood-based analysis of multivariate data with missing values. Statistica Sinica 14:949-968, 2004.

van Buuren S, Brand JPL, Groothuis-Oudshoorn CGM, Rubin DB. Fully conditional specifications in multivariate imputation. J Stat Comp Sim 72:1049-1064, 2006.

de Groot JAH, Janssen KJM, Zwinderman AH, Moons KGM, Reitsma JB. Multiple imputation to correct for partial verification bias revisited. Stat Med 27:5880-5889, 2008.

Siddique J. Multiple imputation using an iterative hot-deck with distance-based donor selection. Stat Med 27:83-102, 2008.

White IR, Royston P, Wood AM. Multiple imputation using chained equations: Issues and guidance for practice. Stat Med 30:377-399, 2011.

Curnow E, Carpenter JR, Heron JE, et al: Multiple imputation of missing data under missing at random: compatible imputation models are not sufficient to avoid bias if they are mis-specified. J Clin Epi June 9, 2023. DOI:10.1016/j.jclinepi.2023.06.011.

See Also

fit.mult.impute, transcan, areg, naclus, naplot, mice, dotchart3, Ecdf, completer

Examples

# Check that aregImpute can almost exactly estimate missing values when
# there is a perfect nonlinear relationship between two variables
# Fit restricted cubic splines with 4 knots for x1 and x2, linear for x3
set.seed(3)
x1 <- rnorm(200)
x2 <- x1^2
x3 <- runif(200)
m <- 30
x2[1:m] <- NA
a <- aregImpute(~x1+x2+I(x3), n.impute=5, nk=4, match='closest')
a
matplot(x1[1:m]^2, a$imputed$x2)
abline(a=0, b=1, lty=2)

x1[1:m]^2
a$imputed$x2


# Multiple imputation and estimation of variances and covariances of
# regression coefficient estimates accounting for imputation
# Example 1: large sample size, much missing data, no overlap in
# NAs across variables
x1 <- factor(sample(c('a','b','c'),1000,TRUE))
x2 <- (x1=='b') + 3*(x1=='c') + rnorm(1000,0,2)
x3 <- rnorm(1000)
y  <- x2 + 1*(x1=='c') + .2*x3 + rnorm(1000,0,2)
orig.x1 <- x1[1:250]
orig.x2 <- x2[251:350]
x1[1:250] <- NA
x2[251:350] <- NA
d <- data.frame(x1,x2,x3,y, stringsAsFactors=TRUE)
# Find value of nk that yields best validating imputation models
# tlinear=FALSE means to not force the target variable to be linear
f <- aregImpute(~y + x1 + x2 + x3, nk=c(0,3:5), tlinear=FALSE,
                data=d, B=10) # normally B=75
f
# Try forcing target variable (x1, then x2) to be linear while allowing
# predictors to be nonlinear (could also say tlinear=TRUE)
f <- aregImpute(~y + x1 + x2 + x3, nk=c(0,3:5), data=d, B=10)
f

## Not run: 
# Use 100 imputations to better check against individual true values
f <- aregImpute(~y + x1 + x2 + x3, n.impute=100, data=d)
f
par(mfrow=c(2,1))
plot(f)
modecat <- function(u) {
 tab <- table(u)
 as.numeric(names(tab)[tab==max(tab)][1])
}
table(orig.x1,apply(f$imputed$x1, 1, modecat))
par(mfrow=c(1,1))
plot(orig.x2, apply(f$imputed$x2, 1, mean))
fmi <- fit.mult.impute(y ~ x1 + x2 + x3, lm, f, 
                       data=d)
sqrt(diag(vcov(fmi)))
fcc <- lm(y ~ x1 + x2 + x3)
summary(fcc)   # SEs are larger than from mult. imputation

## End(Not run)
## Not run: 
# Example 2: Very discriminating imputation models,
# x1 and x2 have some NAs on the same rows, smaller n
set.seed(5)
x1 <- factor(sample(c('a','b','c'),100,TRUE))
x2 <- (x1=='b') + 3*(x1=='c') + rnorm(100,0,.4)
x3 <- rnorm(100)
y  <- x2 + 1*(x1=='c') + .2*x3 + rnorm(100,0,.4)
orig.x1 <- x1[1:20]
orig.x2 <- x2[18:23]
x1[1:20] <- NA
x2[18:23] <- NA
#x2[21:25] <- NA
d <- data.frame(x1,x2,x3,y, stringsAsFactors=TRUE)
n <- naclus(d)
plot(n); naplot(n)  # Show patterns of NAs
# 100 imputations to study them; normally use 5 or 10
f  <- aregImpute(~y + x1 + x2 + x3, n.impute=100, nk=0, data=d)
par(mfrow=c(2,3))
plot(f, diagnostics=TRUE, maxn=2)
# Note: diagnostics=TRUE makes graphs similar to those made by:
# r <- range(f$imputed$x2, orig.x2)
# for(i in 1:6) {  # use 1:2 to mimic maxn=2
#   plot(1:100, f$imputed$x2[i,], ylim=r,
#        ylab=paste("Imputations for Obs.",i))
#   abline(h=orig.x2[i],lty=2)
# }

table(orig.x1,apply(f$imputed$x1, 1, modecat))
par(mfrow=c(1,1))
plot(orig.x2, apply(f$imputed$x2, 1, mean))


fmi <- fit.mult.impute(y ~ x1 + x2, lm, f, 
                       data=d)
sqrt(diag(vcov(fmi)))
fcc <- lm(y ~ x1 + x2)
summary(fcc)   # SEs are larger than from mult. imputation

## End(Not run)

## Not run: 
# Study relationship between smoothing parameter for weighting function
# (multiplier of mean absolute distance of transformed predicted
# values, used in tricube weighting function) and standard deviation
# of multiple imputations.  SDs are computed from average variances
# across subjects.  match="closest" same as match="weighted" with
# small value of fweighted.
# This example also shows problems with predicted mean
# matching almost always giving the same imputed values when there is
# only one predictor (regression coefficients change over multiple
# imputations but predicted values are virtually 1-1 functions of each
# other)

set.seed(23)
x <- runif(200)
y <- x + runif(200, -.05, .05)
r <- resid(lsfit(x,y))
rmse <- sqrt(sum(r^2)/(200-2))   # sqrt of residual MSE

y[1:20] <- NA
d <- data.frame(x,y)
f <- aregImpute(~ x + y, n.impute=10, match='closest', data=d)
# As an aside here is how to create a completed dataset for imputation
# number 3 as fit.mult.impute would do automatically.  In this degenerate
# case changing 3 to 1-2,4-10 will not alter the results.
imputed <- impute.transcan(f, imputation=3, data=d, list.out=TRUE,
                           pr=FALSE, check=FALSE)
sd <- sqrt(mean(apply(f$imputed$y, 1, var)))

ss <- c(0, .01, .02, seq(.05, 1, length=20))
sds <- ss; sds[1] <- sd

for(i in 2:length(ss)) {
  f <- aregImpute(~ x + y, n.impute=10, fweighted=ss[i])
  sds[i] <- sqrt(mean(apply(f$imputed$y, 1, var)))
}

plot(ss, sds, xlab='Smoothing Parameter', ylab='SD of Imputed Values',
     type='b')
abline(v=.2,  lty=2)  # default value of fweighted
abline(h=rmse, lty=2)  # root MSE of residuals from linear regression

## End(Not run)

## Not run: 
# Do a similar experiment for the Titanic dataset
getHdata(titanic3)
h <- lm(age ~ sex + pclass + survived, data=titanic3)
rmse <- summary(h)$sigma
set.seed(21)
f <- aregImpute(~ age + sex + pclass + survived, n.impute=10,
                data=titanic3, match='closest')
sd <- sqrt(mean(apply(f$imputed$age, 1, var)))

ss <- c(0, .01, .02, seq(.05, 1, length=20))
sds <- ss; sds[1] <- sd

for(i in 2:length(ss)) {
  f <- aregImpute(~ age + sex + pclass + survived, data=titanic3,
                  n.impute=10, fweighted=ss[i])
  sds[i] <- sqrt(mean(apply(f$imputed$age, 1, var)))
}

plot(ss, sds, xlab='Smoothing Parameter', ylab='SD of Imputed Values',
     type='b')
abline(v=.2,   lty=2)  # default value of fweighted
abline(h=rmse, lty=2)  # root MSE of residuals from linear regression

## End(Not run)


set.seed(2)
d <- data.frame(x1=runif(50), x2=c(rep(NA, 10), runif(40)),
                x3=c(runif(4), rep(NA, 11), runif(35)))
reformM(~ x1 + x2 + x3, data=d)
reformM(~ x1 + x2 + x3, data=d, nperm=2)
# Give result or one of the results as the first argument to aregImpute

# Constrain imputed values for two variables
# Require imputed values for x2 to be above 0.2
# Assume x1 is never missing and require imputed values for
# x3 to be less than the recipient's value of x1
a <- aregImpute(~ x1 + x2 + x3, data=d,
                constraint=list(x2 = expression(d$x2 > 0.2),
                                x3 = expression(d$x3 < r$x1)))
a

Description

Produces 1-alpha confidence intervals for binomial probabilities.

Usage

binconf(x, n, alpha=0.05,
        method=c("wilson","exact","asymptotic","all"),
        include.x=FALSE, include.n=FALSE, return.df=FALSE)

Arguments

Value

a matrix or data.frame containing the computed intervals and, optionally, x and n.

Author(s)

Rollin Brant, Modified by Frank Harrell and
Brad Biggerstaff
Centers for Disease Control and Prevention
National Center for Infectious Diseases
Division of Vector-Borne Infectious Diseases
P.O. Box 2087, Fort Collins, CO, 80522-2087, USA
[email protected]

References

A. Agresti and B.A. Coull, Approximate is better than "exact" for interval estimation of binomial proportions, American Statistician, 52:119–126, 1998.

R.G. Newcombe, Logit confidence intervals and the inverse sinh transformation, American Statistician, 55:200–202, 2001.

L.D. Brown, T.T. Cai and A. DasGupta, Interval estimation for a binomial proportion (with discussion), Statistical Science, 16:101–133, 2001.

Examples

binconf(0:10,10,include.x=TRUE,include.n=TRUE)
binconf(46,50,method="all")

Description

biVar is a generic function that accepts a formula and usual data, subset, and na.action parameters plus a list statinfo that specifies a function of two variables to compute along with information about labeling results for printing and plotting. The function is called separately with each right hand side variable and the same left hand variable. The result is a matrix of bivariate statistics and the statinfo list that drives printing and plotting. The plot method draws a dot plot with x-axis values by default sorted in order of one of the statistics computed by the function.

spearman2 computes the square of Spearman's rho rank correlation and a generalization of it in which x can relate non-monotonically to y. This is done by computing the Spearman multiple rho-squared between (rank(x), rank(x)^2) and y. When x is categorical, a different kind of Spearman correlation used in the Kruskal-Wallis test is computed (and spearman2 can do the Kruskal-Wallis test). This is done by computing the ordinary multiple R^2 between k-1 dummy variables and rank(y), where x has k categories. x can also be a formula, in which case each predictor is correlated separately with y, using non-missing observations for that predictor. biVar is used to do the looping and bookkeeping. By default the plot shows the adjusted rho^2, using the same formula used for the ordinary adjusted R^2. The F test uses the unadjusted R2.

spearman computes Spearman's rho on non-missing values of two variables. spearman.test is a simple version of spearman2.default.

chiSquare is set up like spearman2 except it is intended for a categorical response variable. Separate Pearson chi-square tests are done for each predictor, with optional collapsing of infrequent categories. Numeric predictors having more than g levels are categorized into g quantile groups. chiSquare uses biVar.

Usage

biVar(formula, statinfo, data=NULL, subset=NULL,
      na.action=na.retain, exclude.imputed=TRUE, ...)

## S3 method for class 'biVar'
print(x, ...)

## S3 method for class 'biVar'
plot(x, what=info$defaultwhat,
                       sort.=TRUE, main, xlab,
                       vnames=c('names','labels'), ...)

spearman2(x, ...)

## Default S3 method:
spearman2(x, y, p=1, minlev=0, na.rm=TRUE, exclude.imputed=na.rm, ...)

## S3 method for class 'formula'
spearman2(formula, data=NULL,
          subset, na.action=na.retain, exclude.imputed=TRUE, ...)

spearman(x, y)

spearman.test(x, y, p=1)

chiSquare(formula, data=NULL, subset=NULL, na.action=na.retain,
          exclude.imputed=TRUE, ...)

Arguments

Details

Uses midranks in case of ties, as described by Hollander and Wolfe. P-values for Spearman, Wilcoxon, or Kruskal-Wallis tests are approximated by using the t or F distributions.

Value

spearman2.default (the function that is called for a single x, i.e., when there is no formula) returns a vector of statistics for the variable. biVar, spearman2.formula, and chiSquare return a matrix with rows corresponding to predictors.

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[email protected]

References

Hollander M. and Wolfe D.A. (1973). Nonparametric Statistical Methods. New York: Wiley.

Press WH, Flannery BP, Teukolsky SA, Vetterling, WT (1988): Numerical Recipes in C. Cambridge: Cambridge University Press.

See Also

combine.levels, varclus, dotchart3, impute, chisq.test, cut2.

Examples

x <- c(-2, -1, 0, 1, 2)
y <- c(4,   1, 0, 1, 4)
z <- c(1,   2, 3, 4, NA)
v <- c(1,   2, 3, 4, 5)

spearman2(x, y)
plot(spearman2(z ~ x + y + v, p=2))

f <- chiSquare(z ~ x + y + v)
f

Description

Bootstraps Kaplan-Meier estimate of the probability of survival to at least a fixed time (times variable) or the estimate of the q quantile of the survival distribution (e.g., median survival time, the default).

Usage

bootkm(S, q=0.5, B=500, times, pr=TRUE)

Arguments

Details

bootkm uses Therneau's survfitKM function to efficiently compute Kaplan-Meier estimates.

Value

a vector containing B bootstrap estimates

Side Effects

updates .Random.seed, and, if pr=TRUE, prints progress of simulations

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University School of Medicine
[email protected]

References

Akritas MG (1986): Bootstrapping the Kaplan-Meier estimator. JASA 81:1032–1038.

See Also

survfit, Surv, Survival.cph, Quantile.cph

Examples

# Compute 0.95 nonparametric confidence interval for the difference in
# median survival time between females and males (two-sample problem)
set.seed(1)
library(survival)
S <- Surv(runif(200))      # no censoring
sex <- c(rep('female',100),rep('male',100))
med.female <- bootkm(S[sex=='female',], B=100) # normally B=500
med.male   <- bootkm(S[sex=='male',],   B=100)
describe(med.female-med.male)
quantile(med.female-med.male, c(.025,.975), na.rm=TRUE)
# na.rm needed because some bootstrap estimates of median survival
# time may be missing when a bootstrap sample did not include the
# longer survival times

Description

Uses method of Fleiss, Tytun, and Ury (but without the continuity correction) to estimate the power (or the sample size to achieve a given power) of a two-sided test for the difference in two proportions. The two sample sizes are allowed to be unequal, but for bsamsize you must specify the fraction of observations in group 1. For power calculations, one probability (p1) must be given, and either the other probability (p2), an odds.ratio, or a percent.reduction must be given. For bpower or bsamsize, any or all of the arguments may be vectors, in which case they return a vector of powers or sample sizes. All vector arguments must have the same length.

Given p1, p2, ballocation uses the method of Brittain and Schlesselman to compute the optimal fraction of observations to be placed in group 1 that either (1) minimize the variance of the difference in two proportions, (2) minimize the variance of the ratio of the two proportions, (3) minimize the variance of the log odds ratio, or (4) maximize the power of the 2-tailed test for differences. For (4) the total sample size must be given, or the fraction optimizing the power is not returned. The fraction for (3) is one minus the fraction for (1).

bpower.sim estimates power by simulations, in minimal time. By using bpower.sim you can see that the formulas without any continuity correction are quite accurate, and that the power of a continuity-corrected test is significantly lower. That's why no continuity corrections are implemented here.

Usage

bpower(p1, p2, odds.ratio, percent.reduction, 
       n, n1, n2, alpha=0.05)


bsamsize(p1, p2, fraction=.5, alpha=.05, power=.8)


ballocation(p1, p2, n, alpha=.05)


bpower.sim(p1, p2, odds.ratio, percent.reduction, 
           n, n1, n2, 
           alpha=0.05, nsim=10000)

Arguments

Details

For bpower.sim, all arguments must be of length one.

Value

for bpower, the power estimate; for bsamsize, a vector containing the sample sizes in the two groups; for ballocation, a vector with 4 fractions of observations allocated to group 1, optimizing the four criteria mentioned above. For bpower.sim, a vector with three elements is returned, corresponding to the simulated power and its lower and upper 0.95 confidence limits.

AUTHOR

Frank Harrell

Department of Biostatistics

Vanderbilt University

[email protected]

References

Fleiss JL, Tytun A, Ury HK (1980): A simple approximation for calculating sample sizes for comparing independent proportions. Biometrics 36:343–6.

Brittain E, Schlesselman JJ (1982): Optimal allocation for the comparison of proportions. Biometrics 38:1003–9.

Gordon I, Watson R (1996): The myth of continuity-corrected sample size formulae. Biometrics 52:71–6.

See Also

samplesize.bin, chisq.test, binconf

Examples

bpower(.1, odds.ratio=.9, n=1000, alpha=c(.01,.05))
bpower.sim(.1, odds.ratio=.9, n=1000)
bsamsize(.1, .05, power=.95)
ballocation(.1, .5, n=100)


# Plot power vs. n for various odds ratios  (base prob.=.1)
n  <- seq(10, 1000, by=10)
OR <- seq(.2,.9,by=.1)
plot(0, 0, xlim=range(n), ylim=c(0,1), xlab="n", ylab="Power", type="n")
for(or in OR) {
  lines(n, bpower(.1, odds.ratio=or, n=n))
  text(350, bpower(.1, odds.ratio=or, n=350)-.02, format(or))
}


# Another way to plot the same curves, but letting labcurve do the
# work, including labeling each curve at points of maximum separation
pow <- lapply(OR, function(or,n)list(x=n,y=bpower(p1=.1,odds.ratio=or,n=n)),
              n=n)
names(pow) <- format(OR)
labcurve(pow, pl=TRUE, xlab='n', ylab='Power')


# Contour graph for various probabilities of outcome in the control
# group, fixing the odds ratio at .8 ([p2/(1-p2) / p1/(1-p1)] = .8)
# n is varied also
p1 <- seq(.01,.99,by=.01)
n  <- seq(100,5000,by=250)
pow <- outer(p1, n, function(p1,n) bpower(p1, n=n, odds.ratio=.8))
# This forms a length(p1)*length(n) matrix of power estimates
contour(p1, n, pow)

Description

Producess side-by-side box-percentile plots from several vectors or a list of vectors.

Usage

bpplot(..., name=TRUE, main="Box-Percentile Plot", 
       xlab="", ylab="", srtx=0, plotopts=NULL)

Arguments

Value

There are no returned values

Side Effects

A plot is created on the current graphics device.

BACKGROUND

Box-percentile plots are similiar to boxplots, except box-percentile plots supply more information about the univariate distributions. At any height the width of the irregular "box" is proportional to the percentile of that height, up to the 50th percentile, and above the 50th percentile the width is proportional to 100 minus the percentile. Thus, the width at any given height is proportional to the percent of observations that are more extreme in that direction. As in boxplots, the median, 25th and 75th percentiles are marked with line segments across the box.

Author(s)

Jeffrey Banfield
[email protected]
Modified by F. Harrell 30Jun97

References

Esty WW, Banfield J: The box-percentile plot. J Statistical Software 8 No. 17, 2003.

See Also

panel.bpplot, boxplot, Ecdf, bwplot

Examples

set.seed(1)
x1 <- rnorm(500)
x2 <- runif(500, -2, 2)
x3 <- abs(rnorm(500))-2
bpplot(x1, x2, x3)
g <- sample(1:2, 500, replace=TRUE)
bpplot(split(x2, g), name=c('Group 1','Group 2'))
rm(x1,x2,x3,g)

Description

For any number of cross-classification variables, bystats returns a matrix with the sample size, number missing y, and fun(non-missing y), with the cross-classifications designated by rows. Uses Harrell's modification of the interaction function to produce cross-classifications. The default fun is mean, and if y is binary, the mean is labeled as Fraction. There is a print method as well as a latex method for objects created by bystats. bystats2 handles the special case in which there are 2 classifcation variables, and places the first one in rows and the second in columns. The print method for bystats2 uses the print.char.matrix function to organize statistics for cells into boxes.

Usage

bystats(y, ..., fun, nmiss, subset)
## S3 method for class 'bystats'
print(x, ...)
## S3 method for class 'bystats'
latex(object, title, caption, rowlabel, ...)
bystats2(y, v, h, fun, nmiss, subset)
## S3 method for class 'bystats2'
print(x, abbreviate.dimnames=FALSE,
   prefix.width=max(nchar(dimnames(x)[[1]])), ...)
## S3 method for class 'bystats2'
latex(object, title, caption, rowlabel, ...)

Arguments

Value

for bystats, a matrix with row names equal to the classification labels and column names N, Missing, funlab, where funlab is determined from fun. A row is added to the end with the summary statistics computed on all observations combined. The class of this matrix is bystats. For bystats, returns a 3-dimensional array with the last dimension corresponding to statistics being computed. The class of the array is bystats2.

Side Effects

latex produces a .tex file.

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[email protected]

See Also

interaction, cut, cut2, latex, print.char.matrix, translate

Examples

## Not run: 
bystats(sex==2, county, city)
bystats(death, race)
bystats(death, cut2(age,g=5), race)
bystats(cholesterol, cut2(age,g=4), sex, fun=median)
bystats(cholesterol, sex, fun=quantile)
bystats(cholesterol, sex, fun=function(x)c(Mean=mean(x),Median=median(x)))
latex(bystats(death,race,nmiss=FALSE,subset=sex=="female"), digits=2)
f <- function(y) c(Hazard=sum(y[,2])/sum(y[,1]))
# f() gets the hazard estimate for right-censored data from exponential dist.
bystats(cbind(d.time, death), race, sex, fun=f)
bystats(cbind(pressure, cholesterol), age.decile, 
        fun=function(y) c(Median.pressure   =median(y[,1]),
                          Median.cholesterol=median(y[,2])))
y <- cbind(pressure, cholesterol)
bystats(y, age.decile, 
        fun=function(y) apply(y, 2, median))   # same result as last one
bystats(y, age.decile, fun=function(y) apply(y, 2, quantile, c(.25,.75)))
# The last one computes separately the 0.25 and 0.75 quantiles of 2 vars.
latex(bystats2(death, race, sex, fun=table))

## End(Not run)

Description

Capitalizes the first letter of each element of the string vector.

Usage

capitalize(string)

Arguments

Value

Returns a vector of charaters with the first letter capitalized

Author(s)

Charles Dupont

Examples

capitalize(c("Hello", "bob", "daN"))

Description

Uses the method of Peterson and George to compute the power of an interaction test in a 2 x 2 setup in which all 4 distributions are exponential. This will be the same as the power of the Cox model test if assumptions hold. The test is 2-tailed. The duration of accrual is specified (constant accrual is assumed), as is the minimum follow-up time. The maximum follow-up time is then accrual + tmin. Treatment allocation is assumed to be 1:1.

Usage

ciapower(tref, n1, n2, m1c, m2c, r1, r2, accrual, tmin, 
         alpha=0.05, pr=TRUE)

Arguments

Value

power

Side Effects

prints

AUTHOR

Frank Harrell

Department of Biostatistics

Vanderbilt University

[email protected]

References

Peterson B, George SL: Controlled Clinical Trials 14:511–522; 1993.

See Also

cpower, spower

Examples

# Find the power of a race x treatment test.  25% of patients will
# be non-white and the total sample size is 14000.  
# Accrual is for 1.5 years and minimum follow-up is 5y.
# Reduction in 5-year mortality is 15% for whites, 0% or -5% for
# non-whites.  5-year mortality for control subjects if assumed to
# be 0.18 for whites, 0.23 for non-whites.
n <- 14000
for(nonwhite.reduction in c(0,-5)) {
  cat("\n\n\n% Reduction in 5-year mortality for non-whites:",
      nonwhite.reduction, "\n\n")
  pow <- ciapower(5,  .75*n, .25*n,  .18, .23,  15, nonwhite.reduction,  
                  1.5, 5)
  cat("\n\nPower:",format(pow),"\n")
}

Description

Takes a set of coordinates in any of the 5 coordinate systems (usr, plt, fig, dev, or tdev) and returns the same points in all 5 coordinate systems.

Usage

cnvrt.coords(x, y = NULL, input = c("usr", "plt", "fig", "dev","tdev"))

Arguments

Details

Every plot has 5 coordinate systems:

usr (User): the coordinate system of the data, this is shown by the tick marks and axis labels.

plt (Plot): Plot area, coordinates range from 0 to 1 with 0 corresponding to the x and y axes and 1 corresponding to the top and right of the plot area. Margins of the plot correspond to plot coordinates less than 0 or greater than 1.

fig (Figure): Figure area, coordinates range from 0 to 1 with 0 corresponding to the bottom and left edges of the figure (including margins, label areas) and 1 corresponds to the top and right edges. fig and dev coordinates will be identical if there is only 1 figure area on the device (layout, mfrow, or mfcol has not been used).

dev (Device): Device area, coordinates range from 0 to 1 with 0 corresponding to the bottom and left of the device region within the outer margins and 1 is the top and right of the region withing the outer margins. If the outer margins are all set to 0 then tdev and dev should be identical.

tdev (Total Device): Total Device area, coordinates range from 0 to 1 with 0 corresponding to the bottom and left edges of the device (piece of paper, window on screen) and 1 corresponds to the top and right edges.

Value

A list with 5 components, each component is a list with vectors named x and y. The 5 sublists are:

Note

You must provide both x and y, but one of them may be NA.

This function is becoming depricated with the new functions grconvertX and grconvertY in R version 2.7.0 and beyond. These new functions use the correct coordinate system names and have more coordinate systems available, you should start using them instead.

Author(s)

Greg Snow [email protected]

See Also

par specifically 'usr','plt', and 'fig'. Also 'xpd' for plotting outside of the plotting region and 'mfrow' and 'mfcol' for multi figure plotting. subplot, grconvertX and grconvertY in R2.7.0 and later

Examples

old.par <- par(no.readonly=TRUE)

par(mfrow=c(2,2),xpd=NA)

# generate some sample data
tmp.x <- rnorm(25, 10, 2)
tmp.y <- rnorm(25, 50, 10)
tmp.z <- rnorm(25, 0, 1)

plot( tmp.x, tmp.y)

# draw a diagonal line across the plot area
tmp1 <- cnvrt.coords( c(0,1), c(0,1), input='plt' )
lines(tmp1$usr, col='blue')

# draw a diagonal line accross figure region
tmp2 <- cnvrt.coords( c(0,1), c(1,0), input='fig')
lines(tmp2$usr, col='red')

# save coordinate of point 1 and y value near top of plot for future plots
tmp.point1 <- cnvrt.coords(tmp.x[1], tmp.y[1])
tmp.range1 <- cnvrt.coords(NA, 0.98, input='plt')

# make a second plot and draw a line linking point 1 in each plot
plot(tmp.y, tmp.z)

tmp.point2 <- cnvrt.coords( tmp.point1$dev, input='dev' )
arrows( tmp.y[1], tmp.z[1], tmp.point2$usr$x, tmp.point2$usr$y,
 col='green')

# draw another plot and add rectangle showing same range in 2 plots

plot(tmp.x, tmp.z)
tmp.range2 <- cnvrt.coords(NA, 0.02, input='plt')
tmp.range3 <- cnvrt.coords(NA, tmp.range1$dev$y, input='dev')
rect( 9, tmp.range2$usr$y, 11, tmp.range3$usr$y, border='yellow')

# put a label just to the right of the plot and
#  near the top of the figure region.
text( cnvrt.coords(1.05, NA, input='plt')$usr$x,
	cnvrt.coords(NA, 0.75, input='fig')$usr$y,
	"Label", adj=0)

par(mfrow=c(1,1))

## create a subplot within another plot (see also subplot)

plot(1:10, 1:10)

tmp <- cnvrt.coords( c( 1, 4, 6, 9), c(6, 9, 1, 4) )

par(plt = c(tmp$dev$x[1:2], tmp$dev$y[1:2]), new=TRUE)
hist(rnorm(100))

par(fig = c(tmp$dev$x[3:4], tmp$dev$y[3:4]), new=TRUE)
hist(rnorm(100))

par(old.par)

Description

These functions are used on ggplot2 objects or as layers when building a ggplot2 object, and to facilitate use of gridExtra. colorFacet colors the thin rectangles used to separate panels created by facet_grid (and probably by facet_wrap). A better approach may be found at https://stackoverflow.com/questions/28652284/. arrGrob is a front-end to gridExtra::arrangeGrob that allows for proper printing. See https://stackoverflow.com/questions/29062766/store-output-from-gridextragrid-arrange-into-an-object/. The arrGrob print method calls grid::grid.draw.

Usage

colorFacet(g, col = adjustcolor("blue", alpha.f = 0.3))

arrGrob(...)

## S3 method for class 'arrGrob'
print(x, ...)

Arguments

Author(s)

Sandy Muspratt

Examples

## Not run: 
require(ggplot2)
s <- summaryP(age + sex ~ region + treatment)
colorFacet(ggplot(s))   # prints directly
# arrGrob is called by rms::ggplot.Predict and others

## End(Not run)

Description

Combine Infrequent Levels of a Categorical Variable

Usage

combine.levels(
  x,
  minlev = 0.05,
  m,
  ord = is.ordered(x),
  plevels = FALSE,
  sep = ","
)

Arguments

Details

After turning 'x' into a 'factor' if it is not one already, combines levels of 'x' whose frequency falls below a specified relative frequency 'minlev' or absolute count 'm'. When 'x' is not treated as ordered, all of the small frequency levels are combined into '"OTHER"', unless 'plevels=TRUE'. When 'ord=TRUE' or 'x' is an ordered factor, only consecutive levels are combined. New levels are constructed by concatenating the levels with 'sep' as a separator. This is useful when comparing ordinal regression with polytomous (multinomial) regression and there are too many categories for polytomous regression. 'combine.levels' is also useful when assumptions of ordinal models are being checked empirically by computing exceedance probabilities for various cutoffs of the dependent variable.

Value

a factor variable, or if 'ord=TRUE' an ordered factor variable

Author(s)

Frank Harrell

Examples

x <- c(rep('A', 1), rep('B', 3), rep('C', 4), rep('D',1), rep('E',1))
combine.levels(x, m=3)
combine.levels(x, m=3, plevels=TRUE)
combine.levels(x, ord=TRUE, m=3)
x <- c(rep('A', 1), rep('B', 3), rep('C', 4), rep('D',1), rep('E',1),
       rep('F',1))
combine.levels(x, ord=TRUE, m=3)

Description

Generates a plotly attribute plot given a series of possibly overlapping binary variables

Usage

combplotp(
  formula,
  data = NULL,
  subset,
  na.action = na.retain,
  vnames = c("labels", "names"),
  includenone = FALSE,
  showno = FALSE,
  maxcomb = NULL,
  minfreq = NULL,
  N = NULL,
  pos = function(x) 1 * (tolower(x) %in% c("true", "yes", "y", "positive", "+",
    "present", "1")),
  obsname = "subjects",
  ptsize = 35,
  width = NULL,
  height = NULL,
  ...
)

Arguments

Details

Similar to the UpSetR package, draws a special dot chart sometimes called an attribute plot that depicts all possible combination of the binary variables. By default a positive value, indicating that a certain condition pertains for a subject, is any of logical TRUE, numeric 1, "yes", "y", "positive", "+" or "present" value, and all others are considered negative. The user can override this determination by specifying her own pos function. Case is ignored in the variable values.

The plot uses solid dots arranged in a vertical line to indicate which combination of conditions is being considered. Frequencies of all possible combinations are shown above the dot chart. Marginal frequencies of positive values for the input variables are shown to the left of the dot chart. More information for all three of these component symbols is provided in hover text.

Variables are sorted in descending order of marginal frqeuencies and likewise for combinations of variables.

Value

plotly object

Author(s)

Frank Harrell

Examples

if (requireNamespace("plotly")) {
  g <- function() sample(0:1, n, prob=c(1 - p, p), replace=TRUE)
  set.seed(2); n <- 100; p <- 0.5
  x1 <- g(); label(x1) <- 'A long label for x1 that describes it'
  x2 <- g()
  x3 <- g(); label(x3) <- 'This is<br>a label for x3'
  x4 <- g()
  combplotp(~ x1 + x2 + x3 + x4, showno=TRUE, includenone=TRUE)

  n <- 1500; p <- 0.05
  pain       <- g()
  anxiety    <- g()
  depression <- g()
  soreness   <- g()
  numbness   <- g()
  tiredness  <- g()
  sleepiness <- g()
  combplotp(~ pain + anxiety + depression + soreness + numbness +
            tiredness + sleepiness, showno=TRUE)
}

Description

Create imputed dataset(s) using transcan and aregImpute objects

Usage

completer(a, nimpute, oneimpute = FALSE, mydata)

Arguments

Details

Similar in function to mice::complete, this function uses transcan and aregImpute objects to impute missing data and returns the completed dataset(s) as a dataframe or a list. It assumes that transcan is used for single regression imputation.

Value

      A single or a list of completed dataset(s).

Author(s)

      Yong-Hao Pua, Singapore General Hospital

Examples

## Not run: 
mtcars$hp[1:5]    <- NA
mtcars$wt[1:10]   <- NA
myrform <- ~ wt + hp + I(carb)
mytranscan  <- transcan( myrform,  data = mtcars, imputed = TRUE,
  pl = FALSE, pr = FALSE, trantab = TRUE, long = TRUE)
myareg      <- aregImpute(myrform, data = mtcars, x=TRUE, n.impute = 5)
completer(mytranscan)                    # single completed dataset
completer(myareg, 3, oneimpute = TRUE)
# single completed dataset based on the `n.impute`th set of multiple imputation
completer(myareg, 3)
# list of completed datasets based on first `nimpute` sets of multiple imputation
completer(myareg)
# list of completed datasets based on all available sets of multiple imputation
# To get a stacked data frame of all completed datasets use
# do.call(rbind, completer(myareg, data=mydata))
# or use rbindlist in data.table

## End(Not run)

Description

Merges an object by the names of its elements. Inserting elements in value into x that do not exists in x and replacing elements in x that exists in value with value elements if protect is false.

Usage

consolidate(x, value, protect, ...)
## Default S3 method:
consolidate(x, value, protect=FALSE, ...)

consolidate(x, protect, ...) <- value

Arguments

Author(s)

Charles Dupont

See Also

names

Examples

x <- 1:5
names(x) <- LETTERS[x]

y <- 6:10
names(y) <- LETTERS[y-2]

x                  # c(A=1,B=2,C=3,D=4,E=5)
y                  # c(D=6,E=7,F=8,G=9,H=10)

consolidate(x, y)      # c(A=1,B=2,C=3,D=6,E=7,F=8,G=9,H=10)
consolidate(x, y, protect=TRUE)      # c(A=1,B=2,C=3,D=4,E=5,F=8,G=9,H=10)

Description

contents is a generic method for which contents.data.frame is currently the only method. contents.data.frame creates an object containing the following attributes of the variables from a data frame: names, labels (if any), units (if any), number of factor levels (if any), factor levels, class, storage mode, and number of NAs. print.contents.data.frame will print the results, with options for sorting the variables. html.contents.data.frame creates HTML code for displaying the results. This code has hyperlinks so that if the user clicks on the number of levels the browser jumps to the correct part of a table of factor levels for all the factor variables. If long labels are present ("longlabel" attributes on variables), these are printed at the bottom and the html method links to them through the regular labels. Variables having the same levels in the same order have the levels factored out for brevity.

contents.list prints a directory of datasets when sasxport.get imported more than one SAS dataset.

If options(prType='html') is in effect, calling print on an object that is the contents of a data frame will result in rendering the HTML version. If run from the console a browser window will open.

Usage

contents(object, ...)
## S3 method for class 'data.frame'
contents(object, sortlevels=FALSE, id=NULL,
  range=NULL, values=NULL, ...)
## S3 method for class 'contents.data.frame'
print(x,
    sort=c('none','names','labels','NAs'), prlevels=TRUE, maxlevels=Inf,
    number=FALSE, ...) 
## S3 method for class 'contents.data.frame'
html(object,
           sort=c('none','names','labels','NAs'), prlevels=TRUE, maxlevels=Inf,
           levelType=c('list','table'),
           number=FALSE, nshow=TRUE, ...)
## S3 method for class 'list'
contents(object, dslabels, ...)
## S3 method for class 'contents.list'
print(x,
    sort=c('none','names','labels','NAs','vars'), ...)

Arguments

Value

an object of class "contents.data.frame" or "contents.list". For the html method is an html character vector object.

Author(s)

Frank Harrell
Vanderbilt University
[email protected]

See Also

describe, html, upData, extractlabs, hlab

Examples

set.seed(1)
dfr <- data.frame(x=rnorm(400),y=sample(c('male','female'),400,TRUE),
                  stringsAsFactors=TRUE)
contents(dfr)
dfr <- upData(dfr, labels=c(x='Label for x', y='Label for y'))
attr(dfr$x, 'longlabel') <-
 'A very long label for x that can continue onto multiple long lines of text'

k <- contents(dfr)
print(k, sort='names', prlevels=FALSE)
## Not run: 
html(k)
html(contents(dfr))            # same result
latex(k$contents)              # latex.default just the main information

## End(Not run)

Description

Assumes exponential distributions for both treatment groups. Uses the George-Desu method along with formulas of Schoenfeld that allow estimation of the expected number of events in the two groups. To allow for drop-ins (noncompliance to control therapy, crossover to intervention) and noncompliance of the intervention, the method of Lachin and Foulkes is used.

Usage

cpower(tref, n, mc, r, accrual, tmin, noncomp.c=0, noncomp.i=0, 
       alpha=0.05, nc, ni, pr=TRUE)

Arguments

Details

For handling noncompliance, uses a modification of formula (5.4) of Lachin and Foulkes. Their method is based on a test for the difference in two hazard rates, whereas cpower is based on testing the difference in two log hazards. It is assumed here that the same correction factor can be approximately applied to the log hazard ratio as Lachin and Foulkes applied to the hazard difference.

Note that Schoenfeld approximates the variance of the log hazard ratio by 4/m, where m is the total number of events, whereas the George-Desu method uses the slightly better 1/m1 + 1/m2. Power from this function will thus differ slightly from that obtained with the SAS samsizc program.

Value

power

Side Effects

prints

Author(s)

Frank Harrell
Department of Biostatistics
Vanderbilt University
[email protected]

References

Peterson B, George SL: Controlled Clinical Trials 14:511–522; 1993.

Lachin JM, Foulkes MA: Biometrics 42:507–519; 1986.

Schoenfeld D: Biometrics 39:499–503; 1983.

See Also

spower, ciapower, bpower

Examples

#In this example, 4 plots are drawn on one page, one plot for each
#combination of noncompliance percentage.  Within a plot, the
#5-year mortality % in the control group is on the x-axis, and
#separate curves are drawn for several % reductions in mortality
#with the intervention.  The accrual period is 1.5y, with all
#patients followed at least 5y and some 6.5y.


par(mfrow=c(2,2),oma=c(3,0,3,0))


morts <- seq(10,25,length=50)
red <- c(10,15,20,25)


for(noncomp in c(0,10,15,-1)) {
  if(noncomp>=0) nc.i <- nc.c <- noncomp else {nc.i <- 25; nc.c <- 15}
  z <- paste("Drop-in ",nc.c,"%, Non-adherence ",nc.i,"%",sep="")
  plot(0,0,xlim=range(morts),ylim=c(0,1),
           xlab="5-year Mortality in Control Patients (%)",
           ylab="Power",type="n")
  title(z)
  cat(z,"\n")
  lty <- 0
  for(r in red) {
        lty <- lty+1
        power <- morts
        i <- 0
        for(m in morts) {
          i <- i+1
          power[i] <- cpower(5, 14000, m/100, r, 1.5, 5, nc.c, nc.i, pr=FALSE)
        }
        lines(morts, power, lty=lty)
  }
  if(noncomp==0)legend(18,.55,rev(paste(red,"% reduction",sep="")),
           lty=4:1,bty="n")
}
mtitle("Power vs Non-Adherence for Main Comparison",
           ll="alpha=.05, 2-tailed, Total N=14000",cex.l=.8)
#
# Point sample size requirement vs. mortality reduction
# Root finder (uniroot()) assumes needed sample size is between
# 1000 and 40000
#
nc.i <- 25; nc.c <- 15; mort <- .18
red <- seq(10,25,by=.25)
samsiz <- red


i <- 0
for(r in red) {
  i <- i+1
  samsiz[i] <- uniroot(function(x) cpower(5, x, mort, r, 1.5, 5,
                                          nc.c, nc.i, pr=FALSE) - .8,
                       c(1000,40000))$root
}


samsiz <- samsiz/1000
par(mfrow=c(1,1))
plot(red, samsiz, xlab='% Reduction in 5-Year Mortality',
	 ylab='Total Sample Size (Thousands)', type='n')
lines(red, samsiz, lwd=2)
title('Sample Size for Power=0.80\nDrop-in 15%, Non-adherence 25%')
title(sub='alpha=0.05, 2-tailed', adj=0)

Description

Cs makes a vector of character strings from a list of valid R names. .q is similar but also makes uses of names of arguments.

Usage

Cs(...)
.q(...)

Arguments

Value

character string vector. For .q there will be a names attribute to the vector if any names appeared in ....

See Also

sys.frame, deparse

Examples

Cs(a,cat,dog)
# subset.data.frame <- dataframe[,Cs(age,sex,race,bloodpressure,height)]
.q(a, b, c, 'this and that')
.q(dog=a, giraffe=b, cat=c)

Description

Read comma-separated text data files, allowing optional translation to lower case for variable names after making them valid S names. There is a facility for reading long variable labels as one of the rows. If labels are not specified and a final variable name is not the same as that in the header, the original variable name is saved as a variable label. Uses read.csv if the data.table package is not in effect, otherwise calls fread.

Usage

csv.get(file, lowernames=FALSE, datevars=NULL, datetimevars=NULL,
        dateformat='%F',
        fixdates=c('none','year'), comment.char="", autodate=TRUE,
        allow=NULL, charfactor=FALSE,
        sep=',', skip=0, vnames=NULL, labels=NULL, text=NULL, ...)

Arguments

Details

csv.get reads comma-separated text data files, allowing optional translation to lower case for variable names after making them valid S names. Original possibly non-legal names are taken to be variable labels if labels is not specified. Character or factor variables containing dates can be converted to date variables. cleanup.import is invoked to finish the job.

Value

a new data frame.

Author(s)

Frank Harrell, Vanderbilt University

See Also

sas.get, data.frame, cleanup.import, read.csv, strptime, POSIXct, Date, fread

Examples

## Not run: 
dat <- csv.get('myfile.csv')

# Read a csv file with junk in the first row, variable names in the
# second, long variable labels in the third, and junk in the 4th row
dat <- csv.get('myfile.csv', vnames=2, labels=3, skip=4)

## End(Not run)

Description

curveRep finds representative curves from a relatively large collection of curves. The curves usually represent time-response profiles as in serial (longitudinal or repeated) data with possibly unequal time points and greatly varying sample sizes per subject. After excluding records containing missing x or y, records are first stratified into kn groups having similar sample sizes per curve (subject). Within these strata, curves are next stratified according to the distribution of x points per curve (typically measurement times per subject). The clara clustering/partitioning function is used to do this, clustering on one, two, or three x characteristics depending on the minimum sample size in the current interval of sample size. If the interval has a minimum number of unique values of one, clustering is done on the single x values. If the minimum number of unique x values is two, clustering is done to create groups that are similar on both min(x) and max(x). For groups containing no fewer than three unique x values, clustering is done on the trio of values min(x), max(x), and the longest gap between any successive x. Then within sample size and x distribution strata, clustering of time-response profiles is based on p values of y all evaluated at the same p equally-spaced x's within the stratum. An option allows per-curve data to be smoothed with lowess before proceeding. Outer x values are taken as extremes of x across all curves within the stratum. Linear interpolation within curves is used to estimate y at the grid of x's. For curves within the stratum that do not extend to the most extreme x values in that stratum, extrapolation uses flat lines from the observed extremes in the curve unless extrap=TRUE. The p y values are clustered using clara.

print and plot methods show results. By specifying an auxiliary idcol variable to plot, other variables such as treatment may be depicted to allow the analyst to determine for example whether subjects on different treatments are assigned to different time-response profiles. To write the frequencies of a variable such as treatment in the upper left corner of each panel (instead of the grand total number of clusters in that panel), specify freq.

curveSmooth takes a set of curves and smooths them using lowess. If the number of unique x points in a curve is less than p, the smooth is evaluated at the unique x values. Otherwise it is evaluated at an equally spaced set of x points over the observed range. If fewer than 3 unique x values are in a curve, those points are used and smoothing is not done.

Usage

curveRep(x, y, id, kn = 5, kxdist = 5, k = 5, p = 5,
         force1 = TRUE, metric = c("euclidean", "manhattan"),
         smooth=FALSE, extrap=FALSE, pr=FALSE)

## S3 method for class 'curveRep'
print(x, ...)

## S3 method for class 'curveRep'
plot(x, which=1:length(res),
                        method=c('all','lattice','data'),
                        m=NULL, probs=c(.5, .25, .75), nx=NULL, fill=TRUE,
                        idcol=NULL, freq=NULL, plotfreq=FALSE,
                        xlim=range(x), ylim=range(y),
                        xlab='x', ylab='y', colorfreq=FALSE, ...)
curveSmooth(x, y, id, p=NULL, pr=TRUE)

Arguments