inferences

(EXPERIMENTAL) Bootstrap and Simulation-Based Inference

Description

Warning: This function is experimental. It may be renamed, the user interface may change, or the functionality may migrate to arguments in other marginaleffects functions.

Apply this function to a marginaleffects object to change the inferential method used to compute uncertainty estimates.

Usage

inferences(x, method, R = 1000, conf_type = "perc", ...)

Arguments

x	Object produced by one of the core `marginaleffects` functions.
method	String "delta": delta method standard errors "boot" package "fwb": fractional weighted bootstrap "rsample" package "simulation" from a multivariate normal distribution (Krinsky & Robb, 1986) "mi" multiple imputation for missing data
R	Number of resamples or simulations.
conf_type	String: type of bootstrap interval to construct. `boot`: "perc", "norm", "basic", or "bca" `fwb`: "perc", "norm", "basic", "bc", or "bca" `rsample`: "perc" or "bca" `simulation`: argument ignored.
...	If `method=“boot”`, additional arguments are passed to `boot::boot()`. If `method=“fwb”`, additional arguments are passed to `fwb::fwb()`. If `method=“rsample”`, additional arguments are passed to `rsample::bootstraps()`. If `method=“simulation”`, additional arguments are ignored.

Details

When method=“simulation”, we conduct simulation-based inference following the method discussed in Krinsky & Robb (1986):

Draw R sets of simulated coefficients from a multivariate normal distribution with mean equal to the original model’s estimated coefficients and variance equal to the model’s variance-covariance matrix (classical, "HC3", or other).
Use the R sets of coefficients to compute R sets of estimands: predictions, comparisons, or slopes.
Take quantiles of the resulting distribution of estimands to obtain a confidence interval and the standard deviation of simulated estimates to estimate the standard error.

When method=“fwb”, drawn weights are supplied to the model fitting function’s weights argument; if the model doesn’t accept non-integer weights, this method should not be used. If weights were included in the original model fit, they are extracted by weights() and multiplied by the drawn weights. These weights are supplied to the wts argument of the estimation function (e.g., comparisons()).

Value

A marginaleffects object with simulation or bootstrap resamples and objects attached.

References

Krinsky, I., and A. L. Robb. 1986. “On Approximating the Statistical Properties of Elasticities.” Review of Economics and Statistics 68 (4): 715–9.

King, Gary, Michael Tomz, and Jason Wittenberg. "Making the most of statistical analyses: Improving interpretation and presentation." American journal of political science (2000): 347-361

Dowd, Bryan E., William H. Greene, and Edward C. Norton. "Computation of standard errors." Health services research 49.2 (2014): 731-750.

Examples

library(marginaleffects)

library(marginaleffects)
library(magrittr)
set.seed(1024)
mod <- lm(Sepal.Length ~ Sepal.Width * Species, data = iris)

# bootstrap
avg_predictions(mod, by = "Species") %>%
  inferences(method = "boot")


    Species Estimate Std. Error 2.5 % 97.5 %
 setosa         5.01     0.0344  4.93   5.07
 versicolor     5.94     0.0619  5.82   6.07
 virginica      6.59     0.0788  6.44   6.75

Columns: Species, estimate, std.error, conf.low, conf.high

avg_predictions(mod, by = "Species") %>%
  inferences(method = "rsample")


    Species Estimate 2.5 % 97.5 %
 setosa         5.01  4.94   5.07
 versicolor     5.94  5.82   6.06
 virginica      6.59  6.44   6.75

Columns: Species, estimate, conf.low, conf.high

# Fractional (bayesian) bootstrap
avg_slopes(mod, by = "Species") %>%
  inferences(method = "fwb") %>%
  posterior_draws("rvar") %>%
  data.frame()

         term                        contrast    Species  estimate predicted
1 Sepal.Width                     mean(dY/dX)     setosa 0.6904897  5.055715
2 Sepal.Width                     mean(dY/dX) versicolor 0.8650777  6.307983
3 Sepal.Width                     mean(dY/dX)  virginica 0.9015345  6.881900
4     Species mean(versicolor) - mean(setosa)     setosa 1.4992211  5.055715
5     Species mean(versicolor) - mean(setosa) versicolor 1.3843422  6.307983
6     Species mean(versicolor) - mean(setosa)  virginica 1.4199582  6.881900
7     Species  mean(virginica) - mean(setosa)     setosa 1.9912967  5.055715
8     Species  mean(virginica) - mean(setosa) versicolor 1.8524292  6.307983
9     Species  mean(virginica) - mean(setosa)  virginica 1.8954823  6.881900
  predicted_hi predicted_lo  std.error  conf.low conf.high         rvar
1     5.055881     5.055715 0.08102709 0.5487008 0.8583514 0.69 ± 0.081
2     6.308191     6.307983 0.20712193 0.4808519 1.2809997 0.88 ± 0.207
3     6.882117     6.881900 0.23704875 0.3867165 1.3338278 0.89 ± 0.237
4     6.567507     5.055715 0.15422309 1.2037516 1.7935150 1.50 ± 0.154
5     6.307983     4.848568 0.08693716 1.2216215 1.5524924 1.38 ± 0.087
6     6.394491     4.917617 0.08908202 1.2437332 1.5875035 1.42 ± 0.089
7     7.062207     5.055715 0.12038738 1.7300974 2.2013722 1.99 ± 0.120
8     6.791747     4.848568 0.12456281 1.6287253 2.1275295 1.86 ± 0.125
9     6.881900     4.917617 0.09656870 1.7196001 2.0992510 1.90 ± 0.097

# Simulation-based inference
slopes(mod) %>%
  inferences(method = "simulation") %>%
  head()


        Term Contrast Estimate Std. Error 2.5 % 97.5 %
 Sepal.Width    dY/dX    0.696      0.167 0.366   1.01
 Sepal.Width    dY/dX    0.696      0.167 0.366   1.01
 Sepal.Width    dY/dX    0.696      0.167 0.366   1.01
 Sepal.Width    dY/dX    0.696      0.167 0.366   1.01
 Sepal.Width    dY/dX    0.696      0.167 0.366   1.01
 Sepal.Width    dY/dX    0.696      0.167 0.366   1.01

Columns: rowid, term, contrast, estimate, std.error, conf.low, conf.high, predicted, predicted_hi, predicted_lo, tmp_idx, Sepal.Length, Sepal.Width, Species