estimate

estimate is a lightweight package helping with model-based estimations, used in the computation of marginal means, contrast analysis and predictions.

Installation

Run the following:

install.packages("devtools")
devtools::install_github("easystats/estimate")

library("estimate")

Documentation

Click on the buttons above to access the package documentation and the easystats blog, and check-out these vignettes:

• Data grids
• Marginal means
• Contrast analysis
• Use a model to make predictions
• Describe non-linear curves

Features

The package is built around 5 main functions:

• estimate_means(): Estimates the average values at each factor levels
• estimate_contrasts(): Estimates and tests contrasts between different factor levels
• estimate_slopes(): Estimates the slopes of numeric predictors at different factor levels
• estimate_response(): Predict the response variable using the model
• estimate_smooth(): Describes a non-linear term (e.g. in GAMs) by its linear parts

These functions are powered by the data_grid() function, a smart tool for guessing the appropriate reference grid.

The package currently only supports rstanarm models, but will be expanded to cover a large variety of frequentist and Bayesian models.

Examples

Estimate marginal means

library(rstanarm)

model <- stan_glm(Sepal.Width ~ Species, data = iris)

estimate_means(model)

## Species    | Median |       89% CI
## ----------------------------------
## setosa     |   3.43 | [3.35, 3.50]
## versicolor |   2.77 | [2.69, 2.85]
## virginica  |   2.98 | [2.90, 3.05]

Contrast analysis

estimate_contrasts(model)
## Level1     |     Level2 | Median |         89% CI |     pd | % in ROPE | Median (std.)
## --------------------------------------------------------------------------------------
## setosa     | versicolor |   0.65 |   [0.55, 0.76] |   100% |        0% |          1.50
## setosa     |  virginica |   0.45 |   [0.34, 0.56] |   100% |        0% |          1.03
## versicolor |  virginica |  -0.20 | [-0.31, -0.09] | 99.85% |     6.10% |         -0.47

library(see)

plot(estimate_contrasts(model), estimate_means(model))

Check the contrasts at different points of another linear predictor

model <- stan_glm(Sepal.Width ~ Species * Petal.Length, data = iris)

estimate_contrasts(model, modulate = "Petal.Length", length = 3)

## Level1     |     Level2 | Petal.Length | Median |        89% CI |     pd | % in ROPE | Median (std.)
## ----------------------------------------------------------------------------------------------------
## setosa     | versicolor |         1.00 |   1.53 |  [1.06, 2.02] |   100% |        0% |          3.52
## setosa     |  virginica |         1.00 |   1.22 |  [0.69, 1.78] | 99.98% |     0.02% |          2.81
## versicolor |  virginica |         1.00 |  -0.31 | [-0.98, 0.45] | 74.72% |    13.35% |         -0.71
## setosa     | versicolor |         3.95 |   1.78 |  [1.03, 2.57] |   100% |        0% |          4.07
## setosa     |  virginica |         3.95 |   1.81 |  [1.04, 2.66] |   100% |     0.02% |          4.14
## versicolor |  virginica |         3.95 |   0.03 | [-0.18, 0.25] | 59.85% |    52.33% |          0.07
## setosa     | versicolor |         6.90 |   2.00 |  [0.53, 3.83] | 97.42% |     1.03% |          4.58
## setosa     |  virginica |         6.90 |   2.40 |  [0.92, 4.17] | 99.00% |     0.78% |          5.50
## versicolor |  virginica |         6.90 |   0.38 | [-0.08, 0.78] | 91.67% |    10.82% |          0.87

Find a predictor’s slopes at each factor level

estimate_slopes(model)
## Species    | Median |       89% CI |     pd | % in ROPE | Median (std.)
## -----------------------------------------------------------------------
## setosa     |   0.41 | [0.13, 0.73] | 98.32% |     4.70% |          1.66
## versicolor |   0.33 | [0.19, 0.48] | 99.98% |     0.75% |          1.33
## virginica  |   0.21 | [0.09, 0.33] | 99.80% |     6.78% |          0.86

Generate predictions from your model to compare it with original data

estimate_response(model)

Species	Petal.Length	Median	CI_low	CI_high
setosa	1.4	3.40	2.91	3.92
setosa	1.4	3.40	2.83	3.87
setosa	1.3	3.36	2.84	3.83
setosa	1.5	3.44	2.92	3.94
setosa	1.4	3.40	2.88	3.91
setosa	1.7	3.53	3.03	4.07

Estimate the link between the response and a predictor

model <- stan_glm(Sepal.Width ~ poly(Petal.Length, 2), data=iris)

estimate_link(model)

Petal.Length	Median	CI_low	CI_high
1.00	3.62	3.52	3.73
1.98	3.18	3.11	3.24
2.97	2.90	2.83	2.97
3.95	2.78	2.71	2.86
4.93	2.83	2.78	2.89
5.92	3.05	2.96	3.14
6.90	3.44	3.25	3.63

Describe the smooth term by its linear parts

estimate_smooth(model)
## Part | Start |  End |   Size | Trend | Linearity
## ------------------------------------------------
## 1    |  1.00 | 4.11 | 53.50% | -0.01 |      0.94
## 2    |  4.11 | 6.90 | 47.00% |  0.01 |      0.94