optCL: Calculate the Optimal Visual Testing Confidence Level

In davidaarmstrong/psre: Presenting Statistical Results Effectively

Calculate the Optimal Visual Testing Confidence Level

Description

Calculates the Optimal Visual Testing (OVT) confidence level. The OVT level is a level you can use to make confidence intervals such that the overlapping (or non-overlapping) of confidence intervals preserves the pairwise testing results. That is, statistically different estimates have confidence intervals that do not overlap and statistically indistinguishable intervals have confidence intervals that do overlap. It does not always work perfectly, but it generally results in fewer inferential errors than the nominal level.

Usage

optCL(
  obj = NULL,
  varname = NULL,
  b = NULL,
  v = NULL,
  resdf = Inf,
  level = 0.95,
  quasi_vars = NULL,
  add_ref = TRUE,
  grid_range = c(0.75, 0.99),
  grid_length = 100,
  adjust = p.adjust.methods[c(8, 1:7)]
)

Arguments

obj	A model object, on which coef and vcov can be called. Either obj and varname or b and v must be specified.
varname	The name of a variable whose coefficients will be used.
b	Optional vector of coefficients to be passed into the function. it overrides the coefficients in obj. Either obj and varname or b and v must be specified.
v	Optional variance-covariance matrix. This can be specified even if obj and varname are specified. It replaces the variance-covaraince matrix from the model.
resdf	If only b and v are passed in, this gives the residual degrees of freedom for the t-statistics.
level	The confidence level to use for testing.
quasi_vars	An optional vector of quasi-variances that will be used to make the confidence intervals.
add_ref	If obj and varname are passed in, an optional 0 is added to the front of the vector of coefficients, along with a leading row and column of zeros on the variance-covariance matrix to represent the reference category.
grid_range	The range of values over which to do the grid search.
grid_length	The number of values in the grid.
adjust	String giving the method used to adjust the p-values for multiplicity. All methods allowed in p.adjust.methods are permitted. None is the default.

Value

A list with the following elements:

opt_levels

The optimal confidence levels that all have identical minimal error rates.

opt_diffs

The sum of differences between upper and lower bounds that characterize the appropriate visual tests. Larger numbers are better.

opt_errors

The proportion of errors across all simple contrasts.

lev_errors

The proportion of errors made at the nominal significance level.

tot_comps

The total number of comparisons

lev_dat

If there are inferential errors at the nominal level, this is a data frame that has all of the information about which comparisons are not appropriately represented by the overlaps in confidence intervals.

err_dat

If there are inferential errors at the optimal level, this is a data frame that has all of the information about which comparisons remain not appropriately represented by the overlaps in optimized confidence intervals.

Examples

data(wvs)
wvs$civ2 <- "Other"
wvs$civ2 <- ifelse(wvs$civ == 9, 
                   "Western", 
                   wvs$civ2)
wvs$civ2 <- ifelse(wvs$civ == 6, 
                   "Latin American", 
                   wvs$civ2)
wvs$civ2 <- as.factor(wvs$civ2)

intmod <- lm(resemaval ~ civ2 * pct_secondary, 
             data=wvs)

ss2 <- simple_slopes(intmod, 
                     "pct_secondary", 
                     "civ2")
o2 <- optCL(b=ss2$est$slope, v=ss2$v)

davidaarmstrong/psre documentation built on Aug. 28, 2024, 6:35 a.m.