Package 'merTools'

Description

Extract a data frame of a single row that represents the average observation in a merMod object. This function also allows the user to pass a series of conditioning argument to calculate the average observation conditional on other characteristics.

Usage

averageObs(merMod, varList = NULL, origData = NULL, ...)

Arguments

Details

Each character and factor variable in the data.frame is assigned to the modal category and each numeric variable is collapsed to the mean. Currently if mode is a tie, returns a "." Uses the collapseFrame function.

For models with a scalar left-hand side (e.g. lmer(y ~ ...)), the response column is included in the output and is set to the mean of the observed response. For models with a matrix-valued left-hand side – most commonly two-column cbind() specifications in binomial GLMMs such as glmer(cbind(successes, failures) ~ ..., family = binomial) – the response column is omitted from the output. A matrix response cannot be meaningfully collapsed to a single "average" value, and averageObs() is primarily intended to produce newdata for predict / predictInterval, both of which ignore the response column in newdata. Callers that iterate over names(averageObs(merMod)) or compare against merMod@frame should not assume column parity for matrix-LHS models.

Value

a data frame with a single row for the average observation, but with full factor levels. See details for more.

Description

Take an entire dataframe and summarize it in one row by using the mean and mode.

Usage

collapseFrame(data)

Arguments

Details

Each character and factor variable in the data.frame is assigned to the modal category and each numeric variable is collapsed to the mean. Currently if mode is a tie, returns a "."

Value

a data frame with a single row

Description

Draw is used to select a single observation out of an R object. Additional parameters allow the user to control how that observation is chosen in order to manipulate that observation later. This is a generic function with methods for a number of objects.

Usage

draw(object, type = c("random", "average"), varList = NULL, seed = NULL, ...)

## S3 method for class 'merMod'
draw(object, type = c("random", "average"), varList = NULL, seed = NULL, ...)

Arguments

Details

In cases of tie, ".", may be substituted for factors.

Value

a data.frame with a single row representing the desired observation

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
# Random case
draw(fm1, type = "random")
# Average
draw(fm1, type = "average")
# Subset
draw(fm1, type = "average", varList = list("Subject" = "308"))

Description

expectedRank calculates the expected rank and the percentile expected rank of any random term in a merMod object. A simple ranking of the estimated random effects (as produced by ranef) is not satisfactory because it ignores any amount of uncertainty.

Usage

expectedRank(merMod, groupFctr = NULL, term = NULL)

Arguments

Details

Inspired by Lingsma et al. (2010, see also Laird and Louis 1989), expectedRank sums the probability that each level of the grouping factor is greater than every other level of the grouping factor, similar to a two-sample t-test.

The formula for the expected rank is:

$ExpectedRank_i = 1 + \sum \phi((\theta_i - \theta_k) / \sqrt(var(\theta_i)+var(\theta_k))$

where $\phi$ is the standard normal distribution function, $\theta$ is the estimated random effect and $var(\theta)$ is the posterior variance of the estimated random effect. We add one to the sum so that the minimum rank is one instead of zero so that in the case where there is no overlap between the variances of the random effects (or if the variances are zero), the expected rank equals the actual rank. The ranks are ordered such that the winners have ranks that are greater than the losers.

The formula for the percentile expected rank is:

$100 * (ExpectedRank_i - 0.5) / N_grps$

where $N_grps$ is the number of grouping factor levels. The percentile expected rank can be interpreted as the fraction of levels that score at or below the given level.

NOTE: expectedRank will only work under conditions that lme4::ranef will work. One current example of when this is not the case is for models when there are multiple terms specified per factor (e.g. uncorrelated random coefficients for the same term, e.g. lmer(Reaction ~ Days + (1 | Subject) + (0 + Days | Subject), data = sleepstudy))

Value

A data.frame with the following five columns:

groupFctr

a character representing name of the grouping factor

groupLevel

a character representing the level of the grouping factor

term

a character representing the formula term for the group

estimate

effect estimate from lme4::ranef(, condVar=TRUE)).

std.error

the posterior variance of the estimate random effect (from lme4::ranef(, condVar=TRUE)); named "term"_var.

ER

The expected rank.

pctER

The percentile expected rank.

References

Laird NM and Louis TA. Empirical Bayes Ranking Methods. Journal of Education Statistics. 1989;14(1)29-46. Available at doi:10.2307/1164724.

Lingsma HF, Steyerberg EW, Eijkemans MJC, et al. Comparing and ranking hospitals based on outcome: results from The Netherlands Stroke Survey. QJM: An International Journal of Medicine. 2010;103(2):99-108. doi:10.1093/qjmed/hcp169

Examples

#For a one-level random intercept model
m1 <- lmer(Reaction ~ Days + (1 | Subject), sleepstudy)
(m1.er <- expectedRank(m1))

#For a one-level random intercept model with multiple random terms
m2 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
#ranked by the random slope on Days
(m2.er1 <- expectedRank(m2, term="Days"))
#ranked by the random intercept
(m2.er2 <- expectedRank(m2, term="int"))

#For a two-level model with random intercepts
m3 <- lmer(y ~ service * dept + (1|s) + (1|d), InstEval)
#Ranked by the random intercept on 's'
(m3.er1 <- expectedRank(m3, groupFctr="s", term="Intercept"))

Description

Find link function family

Usage

famlink(object, resp = object@resp)

Arguments

Value

the link function and family

Description

Display model fit summary of x or x like objects, fast

Usage

fastdisp(x, ...)

## S3 method for class 'merMod'
fastdisp(x, ...)

## S3 method for class 'merModList'
fastdisp(x, ...)

Arguments

Details

Faster than the implementation in the arm package because it avoids refitting

The time saving is only noticeable for large, time-consuming (g)lmer fits.

Value

A list with model summary components (call, coef, se, ngrps, AIC, n, and fit statistics), returned invisibly. The summary is also printed to the console.

See Also

display

Examples

#Compare the time for displaying this modest model
require(arm)
m1 <- lmer(y ~ lectage + studage + (1|d) + (1|s), data=InstEval)
system.time(display(m1))
system.time(fastdisp(m1))

Description

Simulate fixed effects from merMod FEsim simulates fixed effects from merMod object posterior distributions

Usage

FEsim(merMod, n.sims = 200, oddsRatio = FALSE, seed = NULL)

Arguments

Details

Use the Gelman sim technique to build fixed effect estimates and confidence intervals. Uses the sim function in the arm package

Value

a data frame with the following columns

term

Name of fixed term (intercept/coefficient)

mean

Mean of the simulations

median

Median of the simulations

sd

Standard deviation of the simulations, NA if oddsRatio=TRUE

Examples

require(lme4)
m2 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
fe2 <- FEsim(m2, 25)
head(fe2)

Description

Extract all warning msgs from a merMod object

Usage

fetch.merMod.msgs(x)

Arguments

findFormFuns used by averageObs to calculate proper averages

Description

The purpose is to properly derive data for the average observation in the data by being 'aware' of formulas that contain interactions and/or function calls. For example, in the old behavior, if the formula contained a square term specified as I(x^2), we were returning the mean of x(^2) not the square of mean(x).

Usage

findFormFuns(merMod, origData = NULL)

Arguments

Details

Matrix-valued response columns (e.g. the cbind(successes, failures) left-hand side of a binomial GLMM) are detected and dropped from the working frame before averaging, since they cannot be collapsed to a single scalar. The returned frame therefore has no response column for matrix-LHS models; see averageObs for rationale.

Value

a data frame with a single row for the average observation, but with full factor levels. See details for more.

Description

Extract fixed-effects estimates for a merModList

Usage

## S3 method for class 'merModList'
fixef(object, add.dropped = FALSE, ...)

Arguments

Details

Extract the estimates of the fixed-effects parameters from a list of fitted merMod models. Takes the mean of the individual fixef objects for each of the component models in the merModList.

Value

a named, numeric vector of fixed-effects estimates.

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
fixef(mod)

Description

Identify if a merMod has weights

Usage

hasWeights(merMod)

Arguments

Value

TRUE or FALSE for whether the model has weights

Description

A key example dataset used for examples in the HLM software manual. Included here for use in replicating HLM analyses in R.

Usage

hsb

Format

A data frame with 7,185 observations on the following 8 variables.

schid

a numeric vector, 160 unique values

mathach

a numeric vector for the performance on a standardized math assessment

female

a numeric vector coded 0 for male and 1 for female

ses

a numeric measure of student socio-economic status

minority

a numeric vector coded 0 for white and 1 for non-white students

schtype

a numeric vector coded 0 for public and 1 for private schools

meanses

a numeric, the average SES for each school in the data set

size

a numeric for the number of students in the school

Details

The data file used for this presentation is a subsample from the 1982 High School and Beyond Survey and is used extensively in Hierarchical Linear Models by Raudenbush and Bryk. It consists of 7,185 students nested in 160 schools.

Source

Data made available by UCLA Institute for Digital Research and Education (IDRE) online: https://stats.oarc.ucla.edu/other/hlm/hlm-mlm/introduction-to-multilevel-modeling-using-hlm/

References

Stephen W. Raudenbush and Anthony S. Bryk (2002). Hierarchical Linear Models: Applications and Data Analysis Methods (2nd ed.). SAGE.

Examples

data(hsb)
head(hsb)

Description

Calculate the intraclass correlation using mixed effect models

Usage

ICC(outcome, group, data, subset = NULL)

Arguments

Value

a numeric for the intraclass correlation

Examples

data(sleepstudy)
ICC(outcome = "Reaction", group = "Subject", data = sleepstudy)

Description

Apply a multilevel model to a list of data frames

Apply a Bayesian multilevel model to a list of data frames

Apply a generalized linear multilevel model to a list of data frames

Apply a Bayesian generalized linear multilevel model to a list of data frames

Usage

lmerModList(formula, data, parallel = FALSE, ...)

blmerModList(formula, data, parallel = FALSE, ...)

glmerModList(formula, data, parallel = FALSE, ...)

bglmerModList(formula, data, parallel = FALSE, ...)

Arguments

Details

Parallel computing is provided by the 'futures' package, and its extension the 'future.apply' package to provide the 'future_lapply' function for easy parallel computations on lists. To use this package, simply register a parallel backend using the 'plan()' function from 'futures' - an example is to use 'plan(multisession)'

Value

a list of fitted merMod objects of class merModList

a merModList

a merModList

a merModList

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
summary(mod)

Description

A list of fitted mixed-effects models from lme4

Description

Extract averaged fixed effect parameters across a list of merMod objects

Usage

modelFixedEff(modList, ...)

Arguments

Details

The Rubin correction for combining estimates and standard errors from Rubin (1987) is applied to adjust for the within and between imputation variances.

Value

a data.frame of the averaged fixed effect parameters

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
modelFixedEff(mod)

Description

Extract model information from a merMod

Usage

modelInfo(object)

Arguments

Value

Simple summary information about the object, number of observations, number of grouping terms, AIC, and residual standard deviation

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
modelInfo(mod[[1]])
lapply(mod, modelInfo)

Description

Extract data.frame of random effect statistics from merMod List

Usage

modelRandEffStats(modList)

Arguments

Value

a data.frame

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
modelRandEffStats(mod)

Description

Extract all warning msgs from a merMod object

Usage

plot_sim_error_chks(
  type = c("FE", "RE"),
  level = 0.95,
  stat = c("mean", "median"),
  sd = TRUE,
  sigmaScale = NULL,
  oddsRatio = FALSE,
  labs = FALSE,
  facet = TRUE
)

Arguments

Description

Plot the simulated fixed effects on a ggplot2 chart

Usage

plotFEsim(
  data,
  level = 0.95,
  stat = "median",
  sd = TRUE,
  intercept = FALSE,
  sigmaScale = NULL,
  oddsRatio = FALSE
)

Arguments

Value

a ggplot2 plot of the coefficient effects

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
 (p1 <- plotFEsim(FEsim(fm1)))

Description

Plot the output of one or more REimpact calls on a single chart. Each point is the weighted-average fitted value for a bin of the expected-rank distribution of a grouping factor, with a confidence interval derived from the weighted standard error. Supplying a named list of REimpact results overlays them on the same axes – for example to compare the influence of two different grouping factors, or the same factor across two models – which previously required hand-assembling the data frames (#84).

Usage

plotREimpact(data, level = 0.95, facet = TRUE, point_size = 2.5)

Arguments

Value

a ggplot2 object.

See Also

REimpact

Examples

m1 <- lmer(Reaction ~ Days + (1 | Subject), sleepstudy)
 imp <- REimpact(m1, newdata = sleepstudy[1, ], groupFctr = "Subject",
                 breaks = 4)
 plotREimpact(imp)

 # Compare two grouping factors on the same chart
 g1 <- lmer(y ~ lectage + studage + (1 | d) + (1 | s), data = InstEval)
 d_eff <- REimpact(g1, newdata = InstEval[1, ], groupFctr = "d", breaks = 4)
 s_eff <- REimpact(g1, newdata = InstEval[1, ], groupFctr = "s", breaks = 4)
 plotREimpact(list("Instructor (d)" = d_eff, "Student (s)" = s_eff))

Description

Plot the simulated random effects on a ggplot2 chart. Points that are distinguishable from zero (i.e. the confidence band based on level does not cross the red line) are highlighted. Currently, the plots are ordered according to the grouping factor.

Usage

plotREsim(
  data,
  level = 0.95,
  stat = "median",
  sd = TRUE,
  sigmaScale = NULL,
  oddsRatio = FALSE,
  labs = FALSE,
  facet = TRUE
)

Arguments

Value

a ggplot2 plot of the coefficient effects

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
 (p1 <- plotREsim(REsim(fm1)))
 #Plot just the random effects for the Days slope
 (p2 <- plotREsim(REsim(fm1), facet= list(groupFctr= "Subject", term= "Days")))

Description

This function provides a way to capture model uncertainty in predictions from multi-level models fit with lme4. By drawing a sampling distribution for the random and the fixed effects and then estimating the fitted value across that distribution, it is possible to generate a prediction interval for fitted values that includes all variation in the model except for variation in the covariance parameters, theta. This is a much faster alternative than bootstrapping for models fit to medium to large datasets.

Usage

predictInterval(
  merMod,
  newdata = NULL,
  which = c("full", "fixed", "random", "all"),
  level = 0.8,
  n.sims = 1000,
  stat = c("median", "mean"),
  type = c("linear.prediction", "probability"),
  include.resid.var = TRUE,
  returnSims = FALSE,
  seed = NULL,
  .parallel = FALSE,
  fix.intercept.variance = FALSE,
  ignore.fixed.terms = NULL,
  new.levels = c("zero", "draw")
)

Arguments

Value

a data.frame with three columns: fit, lwr and upr. If 'returnSims' is TRUE the attribute 'sim.results' contains the full simulation array.

Description

Summarize a merMod list

Usage

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

Arguments

Value

a summary object of model information

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
summary(mod)

Description

Print the summary of a merMod list

Usage

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

Arguments

Value

summary content printed to console

Description

Select a random observation from the model frame of a merMod

Usage

randomObs(merMod, varList, seed = NULL)

Arguments

Details

Each factor variable in the data frame has all factor levels from the full model.frame stored so that the new data is compatible with predict.merMod

Value

a data frame with a single row for a random observation, but with full factor levels. See details for more.

Description

Extract random-effects estimates for a merModList

Usage

## S3 method for class 'merModList'
ranef(object, ...)

Arguments

Details

Extract the estimates of the random-effects parameters from a list of fitted merMod models. Takes the mean of the individual ranef objects for each of the component models in the merModList.

Value

a named, numeric vector of random-effects estimates.

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
ranef(mod)

Description

Extract the correlations between the slopes and the intercepts from a model

Usage

REcorrExtract(model)

Arguments

Value

a numeric vector of the correlations among the effects

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
REcorrExtract(fm1)

Description

Extracts random effect terms from an lme4 model

Usage

REextract(merMod)

Arguments

Value

a data frame with the following columns

groupFctr

The name of the grouping factor associated with the random effects

groupID

The level of the grouping factor associated with the random effects

'term'

One column per random effect, the name is derived from the merMod

'term'_se

One column per random effect, the name is derived from the merMod

Examples

m2 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
rfx <- REextract(m2)
#Note the column names
head(rfx)

Description

REimpact calculates the average predicted value for each row of a new data frame across the distribution of expectedRank for a merMod object. This allows the user to make meaningful comparisons about the influence of random effect terms on the scale of the response variable, for user-defined inputs, and accounting for the variability in grouping terms.

Usage

REimpact(merMod, newdata, groupFctr = NULL, term = NULL, breaks = 3, ...)

Arguments

Details

The function predicts the response at every level in the random effect term specified by the user. Then, the expected rank of each group level is binned to the number of bins specified by the user. Finally, a weighted mean of the fitted value for all observations in each bin of the expected ranks is calculated using the inverse of the variance as the weight – so that less precise estimates are downweighted in the calculation of the mean for the bin. Finally, a standard error for the bin mean is calculated.

This function uses the formula for variance of a weighted mean recommended by Cochran (1977).

Value

A data.frame with all unique combinations of the number of cases, rows in the newdata element, and number of bins:

case

The row number of the observation from newdata.

bin

The ranking bin for the expected rank, the higher the bin number, the greater the expected rank of the groups in that bin.

AvgFitWght

The weighted mean of the fitted values for case i in bin k

AvgFitWghtSE

The standard deviation of the mean of the fitted values for case i in bin k.

nobs

The number of group effects contained in that bin.

References

Gatz, DF and Smith, L. The Standard Error of a Weighted Mean Concentration. I. Bootstrapping vs other methods. Atmospheric Environment. 1995;11(2)1185-1193. doi:10.1016/1352-2310(94)00210-C

Cochran, WG. 1977. Sampling Techniques (3rd Edition). Wiley, New York.

See Also

expectedRank, predictInterval

Examples

#For a one-level random intercept model
m1 <- lmer(Reaction ~ Days + (1 | Subject), sleepstudy)
m1.er <- REimpact(m1, newdata = sleepstudy[1, ], breaks = 2)
#For a one-level random intercept model with multiple random terms
m2 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
#ranked by the random slope on Days
m2.er1 <- REimpact(m2,  newdata = sleepstudy[1, ],
           groupFctr = "Subject", term="Days")
#ranked by the random intercept
m2.er2 <- REimpact(m2, newdata = sleepstudy[1, ],
             groupFctr = "Subject", term="int")

# You can also pass additional arguments to predictInterval through REimpact
g1 <- lmer(y ~ lectage + studage + (1|d) + (1|s), data=InstEval)
zed <- REimpact(g1, newdata = InstEval[9:12, ], groupFctr = "d", n.sims = 50,
                include.resid.var = TRUE)
zed2 <- REimpact(g1, newdata = InstEval[9:12, ], groupFctr = "s", n.sims = 50,
                 include.resid.var = TRUE)
zed3 <- REimpact(g1, newdata = InstEval[9:12, ], groupFctr = "d", breaks = 5,
                n.sims = 50, include.resid.var = TRUE)

Description

REmargins calculates the average predicted value for each row of a new data frame across the distribution of expectedRank for a merMod object. This allows the user to make meaningful comparisons about the influence of random effect terms on the scale of the response variable, for user-defined inputs, and accounting for the variability in grouping terms.

Usage

REmargins(
  merMod,
  newdata = NULL,
  groupFctr = NULL,
  term = NULL,
  breaks = 4,
  .parallel = FALSE,
  ...
)

Arguments

Details

The function simulates the

The function predicts the response at every level in the random effect term specified by the user. Then, the expected rank of each group level is binned to the number of bins specified by the user. Finally, a weighted mean of the fitted value for all observations in each bin of the expected ranks is calculated using the inverse of the variance as the weight – so that less precise estimates are downweighted in the calculation of the mean for the bin. Finally, a standard error for the bin mean is calculated.

Value

A data.frame with all unique combinations of the number of cases, rows in the newdata element:

...

The columns of the original data taken from newdata

case

The row number of the observation from newdata. Each row in newdata will be repeated for all unique levels of the grouping_var, term, and breaks.

grouping_var

The grouping variable the random effect is being marginalized over.

term

The term for the grouping variable the random effect is being marginalized over.

breaks

The ntile of the effect size for grouping_var and term

original_group_level

The original grouping value for this case

fit_combined

The predicted value from predictInterval for this case simulated at the Nth ntile of the expected rank distribution of grouping_var and term

upr_combined

The upper bound of the predicted value.

lwr_combined

The lower bound of the predicted value.

fit_XX

For each grouping term in newdata the predicted value is decomposed into its fit components via predictInterval and these are all returned here

upr_XX

The upper bound for the effect of each grouping term

lwr_XX

The lower bound for the effect of each grouping term

fit_fixed

The predicted fit with all the grouping terms set to 0 (average)

upr_fixed

The upper bound fit with all the grouping terms set to 0 (average)

lwr_fixed

The lower bound fit with all the grouping terms set to 0 (average)

References

Gatz, DF and Smith, L. The Standard Error of a Weighted Mean Concentration. I. Bootstrapping vs other methods. Atmospheric Environment. 1995;11(2)1185-1193. doi:10.1016/1352-2310(94)00210-C

Cochran, WG. 1977. Sampling Techniques (3rd Edition). Wiley, New York.

See Also

expectedRank, predictInterval

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
mfx <- REmargins(merMod = fm1, newdata = sleepstudy[1:10,])

# You can also pass additional arguments to predictInterval through REimpact
 g1 <- lmer(y ~ lectage + studage + (1|d) + (1|s), data=InstEval)
 margin_df <- REmargins(g1, newdata = InstEval[20:25, ], groupFctr = c("s"),
                        breaks = 4)
 margin_df <- REmargins(g1, newdata = InstEval[20:25, ], groupFctr = c("d"),
                         breaks = 3)

Description

For a user specified quantile (or quantiles) of the random effect terms in a merMod object. This allows the user to easily identify the observation associated with the nth percentile effect.

Usage

REquantile(merMod, quantile, groupFctr, term = "(Intercept)")

Arguments

Value

a vector of the level of the random effect grouping term that corresponds to each quantile

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
REquantile(fm1, quantile = 0.25, groupFctr = "Subject")
REquantile(fm1, quantile = 0.25, groupFctr = "Subject", term = "Days")

Description

Extract the standard deviation of the random effects from a merMod object

Usage

REsdExtract(model)

Arguments

Value

a numeric vector for standard deviations of the random effects

Examples

fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
REsdExtract(fm1)

Description

Simulate random effects from merMod REsim simulates random effects from merMod object posterior distributions

Usage

REsim(merMod, n.sims = 200, oddsRatio = FALSE, seed = NULL)

Arguments

Details

Use the Gelman sim technique to build empirical Bayes estimates. Uses the sim function in the arm package

Value

a data frame with the following columns

groupFctr

Name of the grouping factor

groupID

Level of the grouping factor

term

Name of random term (intercept/coefficient)

mean

Mean of the simulations

median

Median of the simulations

sd

Standard deviation of the simulations, NA if oddsRatio=TRUE

Examples

require(lme4)
m2 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
re2 <- REsim(m2, 25)
head(re2)

Description

Extract the Root Mean Squared Error for a lmerMod object

Usage

RMSE.merMod(merMod, scale = FALSE)

Arguments

Value

a numeric which represents the RMSE

Examples

require(lme4)
m2 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
RMSE.merMod(m2)

Description

Strips out transformations from variable names in data frames

Usage

sanitizeNames(data)

Arguments

Value

a data frame with variable names cleaned to remove factor() construction

Description

Set up parallel environment

Usage

setup_parallel()

Value

Nothing

Description

shinyMer launches a shiny app that allows you to interactively explore an estimated merMod using functions from merTools.

Usage

shinyMer(merMod, simData = NULL, pos = 1)

Arguments

Value

A shiny app

Description

Randomly reorder a dataframe by row

Usage

shuffle(data)

Arguments

Value

a data frame of the same dimensions with the rows reordered randomly

Description

For each random effect term the function draws 'n.sims' samples from the conditional multivariate normal distribution (BLUPs + post‑var). The result is a named list where each element is an array of dimensions 'nrow(newdata)' × 'n.sims'.

Usage

simulate_random_effects(
  merMod,
  newdata,
  n.sims,
  .parallel = FALSE,
  seed = NULL,
  new.levels = c("zero", "draw")
)

Arguments

Value

List of matrices (rows = observations, cols = simulations).

Description

Strips attributes off of a data frame that come with a merMod model.frame

Usage

stripAttributes(data)

Arguments

Value

a data frame with variable names cleaned to remove all attributes except for names, row.names, and class

Description

Bootstrap a subset of an lme4 model

Usage

subBoot(merMod, n = NULL, FUN, R = 100, seed = NULL, warn = FALSE)

Arguments

Details

This function allows users to estimate parameters of a large merMod object using bootstraps on a subset of the data.

Value

a data.frame of parameters extracted from each of the R replications. The original values are appended to the top of the matrix.

Examples

(fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy))
resultMatrix <- subBoot(fm1, n = 160, FUN = thetaExtract, R = 20)

Description

Split a data.frame by elements in a list

Usage

subsetList(data, list)

Arguments

Value

a data frame with values that match the conditions in the list

Description

Print the results of a merMod list

Usage

## S3 method for class 'merModList'
summary(object, ...)

Arguments

Value

summary content printed to console

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
print(mod)

Description

Create a factor variable and include unobserved levels for compatibility with model prediction functions

Usage

superFactor(x, fullLev)

Arguments

Value

a factor variable with all observed levels of x and all levels of x in fullLev

Examples

regularFactor <- c("A", "B", "C")
regularFactor <- factor(regularFactor)
levels(regularFactor)
# Now make it super
newLevs <- c("D", "E", "F")
regularFactor <- superFactor(regularFactor, fullLev = newLevs)
levels(regularFactor) # now super

Description

A convenience function that returns the theta parameters for a merMod object.

Usage

thetaExtract(merMod)

Arguments

Value

a vector of the covariance, theta, parameters from a merMod

See Also

merMod

Examples

(fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy))
thetaExtract(fm1) #(a numeric vector of the covariance parameters)

Description

Extract the variances and correlations for random effects from a merMod list

Usage

## S3 method for class 'merModList'
VarCorr(x, sigma = 1, rdig = 3L, ...)

Arguments

Value

a list with two elements "stddev" and "correlation" for the standard deviations and correlations averaged across models in the list

Examples

sim_list <- replicate(n = 10,
        expr = sleepstudy[sample(row.names(sleepstudy), 180),],
        simplify=FALSE)
fml <- "Reaction ~ Days + (Days | Subject)"
mod <- lmerModList(fml, data = sim_list)
VarCorr(mod)

Description

Creates a new data.frame with copies of the original observation, each assigned to a different user-specified value of a variable. Allows the user to look at the effect on predicted values of changing either a single variable or multiple variables.

Usage

wiggle(data, varlist, valueslist)

Arguments

Details

If the variable specified is a factor, then wiggle will return it as a character.

Value

a data.frame with each row assigned to the one of the new variable combinations. All variable combinations are returned, eg wiggling two variables with 3 and 4 variables respectively will return a new dataset with 3 * 4 = 12 observations.

Examples

data(iris)
wiggle(iris[3,], varlist = "Sepal.Width", valueslist = list(c(1, 2, 3, 5)))
wiggle(iris[3:5,], "Sepal.Width", valueslist = list(c(1, 2, 3, 5)))
wiggle(iris[3,], c("Sepal.Width", "Petal.Length"), list(c(1,2,3,5), c(3,5,6)))
wiggle(iris[3:5,], c("Sepal.Width", "Petal.Length"), list(c(1,2,3,5), c(3,5,6)))