ham (version 1.1.0)

group: Group level confidence intervals and between-group variation

Description

Group level confidence intervals and between-group variation

Usage

group(
  x,
  y,
  z = NULL,
  dataf,
  dist = "t",
  conf.int = 0.95,
  increment = 1,
  rolling = NULL,
  quarts = FALSE,
  cluster = FALSE
)

Value

list of confidence intervals for outcomes by groups, over time, and clustering measures. Some values returned in alphabetical and numerical order based on the group.

Arguments

x

group predictor variable name.

y

outcome variable name.

z

time period variable name.

dataf

name of data frame object.

dist

indicate the distribution used for confidence intervals. Options for the t, binomial, and exact Poisson distributions. Options are 't', 'b', and 'p'. Default is the 't'.

conf.int

select the confidence interval level. Default is 0.95.

increment

specify the increment in time periods. Selecting 3 if data uses the month as the unit of time will give confidence intervals, each based on 3 months. Default is 1.

rolling

indicate the number of time periods for the 'rolling average'. The rolling average consists of >1 time periods but subsequent point estimate increase by a unit of 1. For example, the common 12-month rolling average is based on months 1-12 of data, followed by the next estimate using months 2-13, 3-14, and so on until the last month in the data has been reached. Default is NULL.

quarts

logical TRUE or FALSE that indicates whether to convert continuous x into 4 groups based on quartiles of x. Default is FALSE.

cluster

logical TRUE or FALSE to generate measures of between-group variation such as the Intra-Class Correlation, Median Odds Ratio, or Design Effect. Default is FALSE. Uses binary outcome formula (between-group variance/(between-group variance + (3.14^2/3)) for ICC in Rabe-Hesketh which may be more appropriate for multilevel models. ICC, MOR, DE may be less reliable for binomial and Poisson distributions, use caution.

References

Merlo, J. (2006). A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. Journal of Epidemiological Health, 60, 4, 290-297. https://doi.org/10.1136/jech.2004.029454.

Muthen, B. & Satorra, A. (1995). Complex Sample Data in Structural Equation Modeling. Sociological Methodology, 25, 267-316. https://doi.org/10.2307/271070.

Rabe-Hesketh, S. & Skrondal, A. (2008). Multilevel and Longitudinal Modeling Using Stata, Second Edition. ISBN: 978-1-59718-040-5.

Examples

#default t distribution results
group(x="program", y="los", dataf=hosprog)
#Rounding LOS to integers
hp2 <- hosprog; hp2$los2 <- round(hp2$los, 0)
#Exact Poisson confidence intervals
group(x="program", y="los2", dataf=hp2, dist="p")
#Rolling 6-months of data
group(x="program", y="los", z="month", dataf=hosprog, dist="t", rolling=6)
#Data returned separately for rolling 6-months of data and 3-month increments (e.g., quarters)
group(x="program", y="los", z="month", dataf=hosprog, dist="t", increment=3, rolling=6)
#Quartile groups for continuous risk score and returned clustering info
group(x="risk", y="los", dataf=hosprog, quarts=TRUE, cluster=TRUE)
#Binomial distribution with less conservative 90% confidence intervals
group(x="risk", y="rdm30", dataf=hosprog, quarts=TRUE, dist="b", conf.int=0.90)