Confidence intervals for wisp model parameters

Returns confidence intervals for the fitted wisp model parameters. These were already computed once, at whatever significance level (alpha) was in effect when sample_stats last ran on this model (the default, both for sample_stats() and for the statistical analysis wisp runs internally, is alpha = 0.05, i.e. a 95\ columns of object$stats$parameters.

Usage

# S3 method for class 'wisp'
confint(object, parm = NULL, level = 0.95, ...)

Arguments

object: List, output of the wisp function (an object of class "wisp").
parm: Character vector, or NULL (default). Names of parameters to include (matching object$param.names); if NULL, confidence intervals for all parameters are returned. Follows the same parm semantics as confint.
level: Numeric, the confidence level required. Defaults to 0.95, matching the confidence level wisp/sample_stats compute by default; see Details above for what happens when a different level is requested.
...: Additional arguments passed on to sample_stats when level != 0.95 (e.g. Bonferroni, conv.resamples.only, verbose).

Value

A matrix with one row per (selected) parameter and two columns, CI.low and CI.high, row-named by parameter.

Details

If level = 0.95 (the default, matching sample_stats()'s own default alpha = 0.05), the already-stored CI.low/CI.high columns are returned directly – exactly as they were computed during the original fit (respecting whatever Bonferroni/conv.resamples.only choices were made then, even if those differ from sample_stats()'s own defaults). If a different level is requested, sample_stats is transparently re-run (with alpha = 1 - level, and otherwise its own defaults for Bonferroni and conv.resamples.only, since this model does not record which choices were used for its originally-stored statistics) to recompute the intervals at the requested level; this is cheap, since it only re-quantiles the resampled parameters already stored on the model; it does not refit anything.

Note

As with coef.wisp, the row names mix several different kinds of quantities. "Rt" rows are confidence bounds on the fitted rate on the log(sum + 1) scale of the summed count within a bin, not the original per-instance rate – to recover the latter, exponentiate each bound, subtract one, and divide by the mean number of data rows per bin (see coef.wisp, the wisp examples, or the "Poisson processes and sigmoids" tutorial, for the full back-transformation). "tslope" and "tpoint" rows are already on their natural, directly interpretable scales and need no such transformation. Whenever the random-effect grouping variable (ran) has more than one level, the matrix additionally contains "wfactor_rate", "wfactor_slope", and "wfactor_point" rows (one per random-effect level) – these are warping factors feeding into the warp function described in the "Random effects and warping" tutorial, a different kind of quantity altogether from the like-named base parameters; neither the "Rt" back-transformation above nor "use as-is" applies to them.

Examples

set.seed(1)
grid <- expand.grid(bin = 1:20, age = c(12, 18), ran = c("r1", "r2"), rep = 1:6)
base_rate <- ifelse(grid$bin <= 10, 3, 9)
age_effect <- ifelse(grid$age == 18, 1.5, 1)
grid$count <- rpois(nrow(grid), base_rate * age_effect)
countdata <- data.frame(
  bin = grid$bin, count = grid$count, age = grid$age, ran = grid$ran,
  species = "demoGene", context = "demoContext"
)
model <- wisp(
  countdata,
  variables      = list(timeseries = "age"),
  model.settings = list(LRO_cost = "none"),
  plot.settings  = list(print.plots = FALSE),
  MCMC.settings  = list(MCMC.steps = 100, MCMC.burnin = 20),
  bootstraps.num = 10,
  max.fork       = 1,
  verbose        = FALSE
)
confint(model)
#>                                                      CI.low     CI.high
#> baseline_demoContext_Rt_demoGene_Tns/Blk1       3.016476952  3.12833343
#> beta_Rt_demoContext_demoGene_18_X_Tns/Blk1      0.213825918  0.41373890
#> baseline_demoContext_Rt_demoGene_Tns/Blk2       4.075587853  4.16161032
#> beta_Rt_demoContext_demoGene_18_X_Tns/Blk2      0.314764126  0.39523704
#> baseline_demoContext_tslope_demoGene_Tns/Blk1   5.519569538 14.52583539
#> beta_tslope_demoContext_demoGene_18_X_Tns/Blk1  1.767844600  4.83561112
#> baseline_demoContext_tpoint_demoGene_Tns/Blk1  10.171114279 10.76825512
#> beta_tpoint_demoContext_demoGene_18_X_Tns/Blk1 -0.125822886  0.52492211
#> wfactor_point_r1_X_demoGene                    -0.042736430  0.30219300
#> wfactor_rate_r1_X_demoGene                     -0.020552776  0.11270932
#> wfactor_slope_r1_X_demoGene                     0.006024845  0.26510370
#> wfactor_point_r2_X_demoGene                    -0.233002271  0.02874054
#> wfactor_rate_r2_X_demoGene                     -0.114990011  0.01830464
#> wfactor_slope_r2_X_demoGene                     0.057493081  2.14208315
confint(model, level = 0.90)
#>                                                      CI.low     CI.high
#> baseline_demoContext_Rt_demoGene_Tns/Blk1       3.024515476  3.12277975
#> beta_Rt_demoContext_demoGene_18_X_Tns/Blk1      0.215707296  0.40962098
#> baseline_demoContext_Rt_demoGene_Tns/Blk2       4.078552713  4.15504201
#> beta_Rt_demoContext_demoGene_18_X_Tns/Blk2      0.316450735  0.39504406
#> baseline_demoContext_tslope_demoGene_Tns/Blk1   5.528391381 14.39380198
#> beta_tslope_demoContext_demoGene_18_X_Tns/Blk1  1.779599603  4.75791994
#> baseline_demoContext_tpoint_demoGene_Tns/Blk1  10.210434661 10.75089289
#> beta_tpoint_demoContext_demoGene_18_X_Tns/Blk1 -0.121094838  0.50167546
#> wfactor_point_r1_X_demoGene                    -0.031560061  0.28558263
#> wfactor_rate_r1_X_demoGene                     -0.005847111  0.11047191
#> wfactor_slope_r1_X_demoGene                     0.011613842  0.26299121
#> wfactor_point_r2_X_demoGene                    -0.226639689  0.01865384
#> wfactor_rate_r2_X_demoGene                     -0.109331077  0.01157228
#> wfactor_slope_r2_X_demoGene                     0.069949611  1.88225845