vignettes/tidybayes.Rmd
tidybayes.Rmd
This vignette introduces the tidybayes
package, which facilitates the use of tidy data (one observation per row) with Bayesian models in R. This vignette is geared towards working with tidy data in general-purpose modeling functions like JAGS or Stan. For a similar introduction to the use of tidybayes
with high-level modeling functions such as those in brms
or rstanarm
, see vignette("tidy-brms")
or vignette("tidy-rstanarm")
. This vignette also describes how to use ggdist
(the sister package to tidybayes
) for visualizing model output.
The default output (and sometimes input) data formats of popular modeling functions like JAGS and Stan often don’t quite conform to the ideal of tidy data. For example, input formats might expect a list instead of a data frame, and for all variables to be encoded as numeric values (requiring translation of factors to numeric values and the creation of index variables to store the number of levels per factor or the number of observations in a data frame). Output formats will often be in matrix form (requiring conversion for use with libraries like ggplot), and will use numeric indices (requiring conversion back into factor level names if the you wish to make meaningfully-labeled plots or tables). tidybayes
automates all of these sorts of tasks.
There are a few core ideas that run through the tidybayes
API that should (hopefully) make it easy to use:
Tidy data does not always mean all parameter names as values. In contrast to the ggmcmc
library (which translates model results into a data frame with a Parameter
and value
column), the spread_draws
function in tidybayes
produces data frames where the columns are named after parameters and (in some cases) indices of those parameters, as automatically as possible and using a syntax as close to the same way you would refer to those variables in the model’s language as possible. A similar function to ggmcmc
’s approach is also provided in gather_draws
, since sometimes you do want variable names as values in a column. The goal is for tidybayes
to do the tedious work of figuring out how to make a data frame look the way you need it to, including turning parameters with indices like "b[1,2]"
and the like into tidy data for you.
Fit into the tidyverse. tidybayes
methods fit into a workflow familiar to users of the tidyverse
(dplyr
, tidyr
, ggplot2
, etc), which means fitting into the pipe (%>%
) workflow, using and respecting grouped data frames (thus spread_draws
and gather_draws
return results already grouped by variable indices, and methods like median_qi
calculate point summaries and intervals for variables and groups simultaneously), and not reinventing too much of the wheel if it is already made easy by functions provided by existing tidyverse
packages (unless it makes for much clearer code for a common idiom). For compatibility with other package column names (such as broom::tidy
), tidybayes
provides transformation functions like to_broom_names
that can be dropped directly into data transformation pipelines.
Focus on composable operations and plotting primitives, not monolithic plots and operations. Several other packages (notably bayesplot
and ggmcmc
) already provide an excellent variety of pre-made methods for plotting Bayesian results. tidybayes
shies away from duplicating this functionality. Instead, it focuses on providing composable operations for generating and manipulating Bayesian samples in a tidy data format, and graphical primitives for ggplot
that allow you to build custom plots easily. Most simply, where bayesplot
and ggmcmc
tend to have functions with many options that return a full ggplot object, tidybayes
tends towards providing primitives (like geom
s) that you can compose and combine into your own custom plots. I believe both approaches have their place: pre-made functions are especially useful for common, quick operations that don’t need customization (like many diagnostic plots), while composable operations tend to be useful for more complex custom plots (in my opinion).
Sensible defaults make life easy. But options (and the data being tidy in the first place) make it easy to go your own way when you need to.
Variable names in models should be descriptive, not cryptic. This principle implies avoiding cryptic (and short) subscripts in favor of longer (but descriptive) ones. This is a matter of readability and accessibility of models to others. For example, a common pattern among Stan users (and in the Stan manual) is to use variables like J
to refer to the number of elements in a group (e.g., number of participants) and a corresponding index like j
to refer to specific elements in that group. I believe this sacrifices too much readability for the sake of concision; I prefer a pattern like n_participant
for the size of the group and participant
(or a mnemonic short form like p
) for specific elements. In functions where names are auto-generated (like compose_data
), tidybayes
will (by default) assume you want these sorts of more descriptive names; however, you can always override the default naming scheme.
tidybayes
aims to support a variety of models with a uniform interface. Currently supported models include rstan, cmdstanr, brms, rstanarm, runjags, rjags, jagsUI, coda::mcmc and coda::mcmc.list, posterior::draws, MCMCglmm, and anything with its own as.mcmc.list
implementation. If you install the tidybayes.rethinking package, models from the rethinking package are also supported.
For an up-to-date list of supported models, see ?"tidybayes-models"
.
The following libraries are required to run this vignette:
library(magrittr)
library(dplyr)
library(forcats)
library(modelr)
library(ggdist)
library(tidybayes)
library(ggplot2)
library(cowplot)
library(emmeans)
library(broom)
library(rstan)
library(rstanarm)
library(brms)
library(bayesplot)
library(MCMCglmm)
library(RColorBrewer)
theme_set(theme_tidybayes() + panel_border())
These options help Stan run faster:
rstan_options(auto_write = TRUE)
options(mc.cores = parallel::detectCores())
To demonstrate tidybayes
, we will use a simple dataset with 10 observations from 5 conditions each:
set.seed(5)
n = 10
n_condition = 5
ABC =
tibble(
condition = rep(c("A","B","C","D","E"), n),
response = rnorm(n * 5, c(0,1,2,1,-1), 0.5)
)
A snapshot of the data looks like this:
head(ABC, 10)
## # A tibble: 10 x 2
## condition response
## <chr> <dbl>
## 1 A -0.420
## 2 B 1.69
## 3 C 1.37
## 4 D 1.04
## 5 E -0.144
## 6 A -0.301
## 7 B 0.764
## 8 C 1.68
## 9 D 0.857
## 10 E -0.931
This is a typical tidy format data frame: one observation per row. Graphically:
compose_data
to prepare a data frame for the model
Shunting data from a data frame into a format usable in samplers like JAGS or Stan can involve a tedious set of operations, like generating index variables storing the number of operations or the number of levels in a factor. compose_data
automates these operations.
A hierarchical model of our example data might fit an overall mean across the conditions (overall_mean
), the standard deviation of the condition means (condition_mean_sd
), the mean within each condition (condition_mean[condition]
) and the standard deviation of the responses given a condition mean (response_sd
):
data {
int<lower=1> n;
int<lower=1> n_condition;
int<lower=1, upper=n_condition> condition[n];
real response[n];
}
parameters {
real overall_mean;
vector[n_condition] condition_zoffset;
real<lower=0> response_sd;
real<lower=0> condition_mean_sd;
}
transformed parameters {
vector[n_condition] condition_mean;
condition_mean = overall_mean + condition_zoffset * condition_mean_sd;
}
model {
response_sd ~ cauchy(0, 1); // => half-cauchy(0, 1)
condition_mean_sd ~ cauchy(0, 1); // => half-cauchy(0, 1)
overall_mean ~ normal(0, 5);
condition_zoffset ~ normal(0, 1); // => condition_mean ~ normal(overall_mean, condition_mean_sd)
for (i in 1:n) {
response[i] ~ normal(condition_mean[condition[i]], response_sd);
}
}
We have compiled and loaded this model into the variable ABC_stan
.
This model expects these variables as input:
n
: number of observationsn_condition
: number of conditionscondition
: a vector of integers indicating the condition of each observationresponse
: a vector of observationsOur data frame (ABC
) only has response
and condition
, and condition
is in the wrong format (it is a factor instead of numeric). However, compose_data
can generate a list containing the above variables in the correct format automatically. It recognizes that condition
is a factor and converts it to a numeric, adds the n_condition
variable automatically containing the number of levels in condition
, and adds the n
column containing the number of observations (number of rows in the data frame):
compose_data(ABC)
## $condition
## [1] 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
##
## $n_condition
## [1] 5
##
## $response
## [1] -0.42042774 1.69217967 1.37225407 1.03507138 -0.14427956 -0.30145399 0.76391681 1.68231434 0.85711318
## [10] -0.93094589 0.61381517 0.59911027 1.45980370 0.92123282 -1.53588002 -0.06949307 0.70134345 0.90801662
## [19] 1.12040863 -1.12967770 0.45025597 1.47093470 2.73398095 1.35338054 -0.59049553 -0.14674092 1.70929454
## [28] 2.74938691 0.67145895 -1.42639772 0.15795752 1.55484708 3.10773029 1.60855182 -0.26038911 0.47578692
## [37] 0.49523368 0.99976363 0.11890706 -1.07130406 0.77503018 0.59878841 1.96271054 1.94783398 -1.22828447
## [46] 0.28111168 0.55649574 1.76987771 0.63783576 -1.03460558
##
## $n
## [1] 50
This makes it easy to skip right to running the model without munging the data yourself:
m = sampling(ABC_stan, data = compose_data(ABC), control = list(adapt_delta = 0.99))
The results look like this:
m
## Inference for Stan model: a079ffbff39f9d3c47919d14be5398a6.
## 4 chains, each with iter=2000; warmup=1000; thin=1;
## post-warmup draws per chain=1000, total post-warmup draws=4000.
##
## mean se_mean sd 2.5% 25% 50% 75% 97.5% n_eff Rhat
## overall_mean 0.63 0.02 0.62 -0.55 0.28 0.63 0.95 1.87 915 1
## condition_zoffset[1] -0.39 0.02 0.49 -1.41 -0.71 -0.39 -0.05 0.58 973 1
## condition_zoffset[2] 0.35 0.02 0.49 -0.62 0.04 0.33 0.66 1.31 902 1
## condition_zoffset[3] 1.11 0.02 0.59 -0.02 0.70 1.11 1.50 2.30 868 1
## condition_zoffset[4] 0.37 0.02 0.49 -0.62 0.04 0.36 0.69 1.31 905 1
## condition_zoffset[5] -1.39 0.02 0.65 -2.74 -1.81 -1.36 -0.94 -0.23 1031 1
## response_sd 0.56 0.00 0.06 0.46 0.52 0.56 0.60 0.70 1793 1
## condition_mean_sd 1.23 0.02 0.51 0.62 0.90 1.12 1.43 2.50 997 1
## condition_mean[1] 0.20 0.00 0.17 -0.14 0.09 0.20 0.32 0.55 4885 1
## condition_mean[2] 1.00 0.00 0.18 0.65 0.89 1.01 1.12 1.34 4606 1
## condition_mean[3] 1.84 0.00 0.18 1.48 1.72 1.84 1.95 2.19 4893 1
## condition_mean[4] 1.02 0.00 0.18 0.68 0.90 1.02 1.14 1.37 4274 1
## condition_mean[5] -0.89 0.00 0.18 -1.23 -1.01 -0.89 -0.77 -0.53 4725 1
## lp__ 0.30 0.08 2.34 -5.02 -1.08 0.63 2.06 3.79 915 1
##
## Samples were drawn using NUTS(diag_e) at Thu Dec 30 18:15:48 2021.
## For each parameter, n_eff is a crude measure of effective sample size,
## and Rhat is the potential scale reduction factor on split chains (at
## convergence, Rhat=1).
spread_draws
Now that we have our results, the fun begins: getting the draws out in a tidy format! The default methods in Stan for extracting draws from the model do so in a nested format:
## List of 6
## $ overall_mean : num [1:4000(1d)] 1.113 0.818 0.432 0.794 0.528 ...
## ..- attr(*, "dimnames")=List of 1
## .. ..$ iterations: NULL
## $ condition_zoffset: num [1:4000, 1:5] -0.9652 -0.4328 -0.0276 -0.4277 -0.0597 ...
## ..- attr(*, "dimnames")=List of 2
## .. ..$ iterations: NULL
## .. ..$ : NULL
## $ response_sd : num [1:4000(1d)] 0.696 0.587 0.651 0.571 0.566 ...
## ..- attr(*, "dimnames")=List of 1
## .. ..$ iterations: NULL
## $ condition_mean_sd: num [1:4000(1d)] 0.974 1.068 0.53 1.456 0.951 ...
## ..- attr(*, "dimnames")=List of 1
## .. ..$ iterations: NULL
## $ condition_mean : num [1:4000, 1:5] 0.173 0.356 0.417 0.171 0.472 ...
## ..- attr(*, "dimnames")=List of 2
## .. ..$ iterations: NULL
## .. ..$ : NULL
## $ lp__ : num [1:4000(1d)] -1.598 0.887 -5.352 3.171 0.319 ...
## ..- attr(*, "dimnames")=List of 1
## .. ..$ iterations: NULL
There are also methods for extracting draws as matrices or data frames in Stan (and other model types, such as JAGS and MCMCglmm, have their own formats).
The spread_draws
method yields a common format for all model types supported by tidybayes
. It lets us instead extract draws into a data frame in tidy format, with a .chain
and .iteration
column storing the chain and iteration for each row (if available), a .draw
column that uniquely indexes each draw, and the remaining columns corresponding to model variables or variable indices. The spread_draws
method accepts any number of column specifications, which can include names for variables and names for variable indices. For example, we can extract the condition_mean
variable as a tidy data frame, and put the value of its first (and only) index into the condition
column, using a syntax that directly echoes how we would specify indices of the condition_mean
variable in the model itself:
m %>%
spread_draws(condition_mean[condition]) %>%
head(10)
## # A tibble: 10 x 5
## # Groups: condition [1]
## condition condition_mean .chain .iteration .draw
## <int> <dbl> <int> <int> <int>
## 1 1 0.00544 1 1 1
## 2 1 -0.0836 1 2 2
## 3 1 0.0324 1 3 3
## 4 1 0.113 1 4 4
## 5 1 0.157 1 5 5
## 6 1 0.218 1 6 6
## 7 1 0.276 1 7 7
## 8 1 0.0130 1 8 8
## 9 1 0.152 1 9 9
## 10 1 0.192 1 10 10
As-is, the resulting variables don’t know anything about where their indices came from. The index of the condition_mean
variable was originally derived from the condition
factor in the ABC
data frame. But Stan doesn’t know this: it is just a numeric index to Stan, so the condition
column just contains numbers (1, 2, 3, 4, 5
) instead of the factor levels these numbers correspond to ("A", "B", "C", "D", "E"
).
We can recover this missing type information by passing the model through recover_types
before using spread_draws
. In itself recover_types
just returns a copy of the model, with some additional attributes that store the type information from the data frame (or other objects) that you pass to it. This doesn’t have any useful effect by itself, but functions like spread_draws
use this information to convert any column or index back into the data type of the column with the same name in the original data frame. In this example, spread_draws
recognizes that the condition
column was a factor with five levels ("A", "B", "C", "D", "E"
) in the original data frame, and automatically converts it back into a factor:
m %>%
recover_types(ABC) %>%
spread_draws(condition_mean[condition]) %>%
head(10)
## # A tibble: 10 x 5
## # Groups: condition [1]
## condition condition_mean .chain .iteration .draw
## <chr> <dbl> <int> <int> <int>
## 1 A 0.00544 1 1 1
## 2 A -0.0836 1 2 2
## 3 A 0.0324 1 3 3
## 4 A 0.113 1 4 4
## 5 A 0.157 1 5 5
## 6 A 0.218 1 6 6
## 7 A 0.276 1 7 7
## 8 A 0.0130 1 8 8
## 9 A 0.152 1 9 9
## 10 A 0.192 1 10 10
Because we often want to make multiple separate calls to spread_draws
, it is often convenient to decorate the original model using recover_types
immediately after it has been fit, so we only have to call it once:
m %<>% recover_types(ABC)
Now we can omit the recover_types
call before subsequent calls to spread_draws
.
point_interval
functions: [median|mean|mode]_[qi|hdi]
tidybayes
provides a family of functions for generating point summaries and intervals from draws in a tidy format. These functions follow the naming scheme [median|mean|mode]_[qi|hdi]
, for example, median_qi
, mean_qi
, mode_hdi
, and so on. The first name (before the _
) indicates the type of point summary, and the second name indicates the type of interval. qi
yields a quantile interval (a.k.a. equi-tailed interval, central interval, or percentile interval) and hdi
yields a highest density interval. Custom point or interval functions can also be applied using the point_interval
function.
For example, we might extract the draws corresponding to the overall mean and standard deviation of observations:
m %>%
spread_draws(overall_mean, response_sd) %>%
head(10)
## # A tibble: 10 x 5
## .chain .iteration .draw overall_mean response_sd
## <int> <int> <int> <dbl> <dbl>
## 1 1 1 1 0.0672 0.576
## 2 1 2 2 0.0361 0.576
## 3 1 3 3 1.17 0.551
## 4 1 4 4 0.378 0.576
## 5 1 5 5 0.359 0.583
## 6 1 6 6 0.332 0.621
## 7 1 7 7 0.226 0.641
## 8 1 8 8 0.107 0.637
## 9 1 9 9 0.225 0.609
## 10 1 10 10 -0.0395 0.521
Like with condition_mean[condition]
, this gives us a tidy data frame. If we want the median and 95% quantile interval of the variables, we can apply median_qi
:
m %>%
spread_draws(overall_mean, response_sd) %>%
median_qi(overall_mean, response_sd)
## # A tibble: 1 x 9
## overall_mean overall_mean.lower overall_mean.up~ response_sd response_sd.low~ response_sd.upp~ .width .point .interval
## <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 0.633 -0.546 1.87 0.558 0.456 0.698 0.95 median qi
median_qi
summarizes each input column using its median. If there are multiple columns to summarize, each gets its own x.lower
and x.upper
column (for each column x
) corresponding to the bounds of the .width
% interval. If there is only one column, the names .lower
and .upper
are used for the interval bounds.
We can specify the columns we want to get medians and intervals from, as above, or if we omit the list of columns, median_qi
will use every column that is not a grouping column or a special column (like .chain
, .iteration
, or .draw
). Thus in the above example, overall_mean
and response_sd
are redundant arguments to median_qi
because they are also the only columns we gathered from the model. So we can simplify the previous code to the following:
m %>%
spread_draws(overall_mean, response_sd) %>%
median_qi()
## # A tibble: 1 x 9
## overall_mean overall_mean.lower overall_mean.up~ response_sd response_sd.low~ response_sd.upp~ .width .point .interval
## <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 0.633 -0.546 1.87 0.558 0.456 0.698 0.95 median qi
When we have a variable with one or more indices, such as condition_mean
, we can apply median_qi
(or other functions in the point_interval
family) as we did before:
m %>%
spread_draws(condition_mean[condition]) %>%
median_qi()
## # A tibble: 5 x 7
## condition condition_mean .lower .upper .width .point .interval
## <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A 0.199 -0.142 0.549 0.95 median qi
## 2 B 1.01 0.651 1.34 0.95 median qi
## 3 C 1.84 1.48 2.19 0.95 median qi
## 4 D 1.02 0.681 1.37 0.95 median qi
## 5 E -0.890 -1.23 -0.529 0.95 median qi
How did median_qi
know what to aggregate? Data frames returned by spread_draws
are automatically grouped by all index variables you pass to it; in this case, that means it groups by condition
. median_qi
respects groups, and calculates the point summaries and intervals within all groups. Then, because no columns were passed to median_qi
, it acts on the only non-special (.
-prefixed) and non-group column, condition_mean
. So the above shortened syntax is equivalent to this more verbose call:
m %>%
spread_draws(condition_mean[condition]) %>%
group_by(condition) %>% # this line not necessary (done automatically by spread_draws)
median_qi(condition_mean)
## # A tibble: 5 x 7
## condition condition_mean .lower .upper .width .point .interval
## <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A 0.199 -0.142 0.549 0.95 median qi
## 2 B 1.01 0.651 1.34 0.95 median qi
## 3 C 1.84 1.48 2.19 0.95 median qi
## 4 D 1.02 0.681 1.37 0.95 median qi
## 5 E -0.890 -1.23 -0.529 0.95 median qi
When given only a single column, median_qi
will use the names .lower
and .upper
for the lower and upper ends of the intervals.
tidybayes
also provides an implementation of posterior::summarise_draws()
for grouped data frames (tidybayes::summaries_draws.grouped_df()
), which you can use to quickly get convergence diagnostics:
m %>%
spread_draws(condition_mean[condition]) %>%
summarise_draws()
## # A tibble: 5 x 11
## # Groups: condition [5]
## condition variable mean median sd mad q5 q95 rhat ess_bulk ess_tail
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 A condition_mean 0.201 0.199 0.172 0.171 -0.0847 0.481 1.00 4779. 3172.
## 2 B condition_mean 1.00 1.01 0.176 0.173 0.711 1.28 1.00 4660. 3492.
## 3 C condition_mean 1.84 1.84 0.178 0.174 1.54 2.12 1.00 4794. 3601.
## 4 D condition_mean 1.02 1.02 0.177 0.178 0.738 1.31 1.00 4303. 3060.
## 5 E condition_mean -0.888 -0.890 0.180 0.177 -1.18 -0.593 1.00 4758. 3248.
geom_pointinterval
Plotting medians and intervals is straightforward using ggdist::geom_pointinterval()
or ggdist::stat_pointinterval()
, which are similar to ggplot2::geom_pointrange()
but with sensible defaults for multiple intervals. For example:
m %>%
spread_draws(condition_mean[condition]) %>%
ggplot(aes(y = fct_rev(condition), x = condition_mean)) +
stat_pointinterval()
These functions have .width = c(.66, .95)
by default (showing 66% and 95% intervals), but this can be changed by passing a .width
argument to ggdist::stat_pointinterval()
.
stat_eye
The ggdist::stat_halfeye()
geom provides a shortcut to generating “half-eye plots” (combinations of intervals and densities). This example also demonstrates how to change the interval probability (here, to 90% and 50% intervals):
m %>%
spread_draws(condition_mean[condition]) %>%
ggplot(aes(y = fct_rev(condition), x = condition_mean)) +
stat_halfeye(.width = c(.90, .5))
Or say you want to annotate portions of the densities in color; the fill
aesthetic can vary within a slab in all geoms and stats in the ggdist::geom_slabinterval()
family, including ggdist::stat_halfeye()
. For example, if you want to annotate a domain-specific region of practical equivalence (ROPE), you could do something like this:
m %>%
spread_draws(condition_mean[condition]) %>%
ggplot(aes(y = fct_rev(condition), x = condition_mean, fill = stat(abs(x) < .8))) +
stat_halfeye() +
geom_vline(xintercept = c(-.8, .8), linetype = "dashed") +
scale_fill_manual(values = c("gray80", "skyblue"))
stat_slabinterval
There are a variety of additional stats for visualizing distributions in the ggdist::geom_slabinterval()
family of stats and geoms:
See vignette("slabinterval", package = "ggdist")
for an overview.
.width =
argument
If you wish to summarise the data before plotting (sometimes useful for large samples), median_qi()
and its sister functions can also produce an arbitrary number of probability intervals by setting the .width =
argument:
m %>%
spread_draws(condition_mean[condition]) %>%
median_qi(.width = c(.95, .8, .5))
## # A tibble: 15 x 7
## condition condition_mean .lower .upper .width .point .interval
## <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A 0.199 -0.142 0.549 0.95 median qi
## 2 B 1.01 0.651 1.34 0.95 median qi
## 3 C 1.84 1.48 2.19 0.95 median qi
## 4 D 1.02 0.681 1.37 0.95 median qi
## 5 E -0.890 -1.23 -0.529 0.95 median qi
## 6 A 0.199 -0.0145 0.421 0.8 median qi
## 7 B 1.01 0.777 1.22 0.8 median qi
## 8 C 1.84 1.61 2.06 0.8 median qi
## 9 D 1.02 0.797 1.25 0.8 median qi
## 10 E -0.890 -1.11 -0.658 0.8 median qi
## 11 A 0.199 0.0870 0.317 0.5 median qi
## 12 B 1.01 0.887 1.12 0.5 median qi
## 13 C 1.84 1.72 1.95 0.5 median qi
## 14 D 1.02 0.904 1.14 0.5 median qi
## 15 E -0.890 -1.01 -0.765 0.5 median qi
The results are in a tidy format: one row per index (condition
) and probability level (.width
). This facilitates plotting. For example, assigning -.width
to the size
aesthetic will show all intervals, making thicker lines correspond to smaller intervals:
m %>%
spread_draws(condition_mean[condition]) %>%
median_qi(.width = c(.95, .66)) %>%
ggplot(aes(
y = fct_rev(condition), x = condition_mean, xmin = .lower, xmax = .upper,
# size = -.width means smaller probability interval => thicker line
# this can be omitted, geom_pointinterval includes it automatically
# if a .width column is in the input data.
size = -.width
)) +
geom_pointinterval()
ggdist::geom_pointinterval()
includes size = -.width
as a default aesthetic mapping to facilitate exactly this usage.
Intervals are nice if the alpha level happens to line up with whatever decision you are trying to make, but getting a shape of the posterior is better (hence eye plots, above). On the other hand, making inferences from density plots is imprecise (estimating the area of one shape as a proportion of another is a hard perceptual task). Reasoning about probability in frequency formats is easier, motivating quantile dotplots (Kay et al. 2016, Fernandes et al. 2018), which also allow precise estimation of arbitrary intervals (down to the dot resolution of the plot, 100 in the example below).
Within the slabinterval family of geoms in tidybayes is the dots
and dotsinterval
family, which automatically determine appropriate bin sizes for dotplots and can calculate quantiles from samples to construct quantile dotplots. ggdist::stat_dots()
is the variant designed for use on samples:
m %>%
spread_draws(condition_mean[condition]) %>%
ggplot(aes(x = condition_mean, y = fct_rev(condition))) +
stat_dotsinterval(quantiles = 100)
The idea is to get away from thinking about the posterior as indicating one canonical point or interval, but instead to represent it as (say) 100 approximately equally likely points.
The point_interval()
family of functions follow the naming scheme [median|mean|mode]_[qi|hdi|hdci]
, and all work in the same way as median_qi()
: they take a series of names (or expressions calculated on columns) and summarize those columns with the corresponding point summary function (median, mean, or mode) and interval (qi, hdi, or hdci). qi
yields a quantile interval (a.k.a. equi-tailed interval, central interval, or percentile interval), hdi
yields one or more highest (posterior) density interval(s), and hdci
yields a single (possibly) highest-density continuous interval. These can be used in any combination desired.
The *_hdi
functions have an additional difference: In the case of multimodal distributions, they may return multiple intervals for each probability level. Here are some draws from a multimodal normal mixture:
Passed through mode_hdi()
, we get multiple intervals at the 80% probability level:
## # A tibble: 2 x 6
## x .lower .upper .width .point .interval
## <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 -0.0153 -1.52 1.60 0.8 mode hdi
## 2 -0.0153 3.04 5.00 0.8 mode hdi
This is easier to see when plotted:
multimodal_draws %>%
ggplot(aes(x = x)) +
stat_slab(aes(y = 0)) +
stat_pointinterval(aes(y = -0.5), point_interval = median_qi, .width = c(.95, .80)) +
annotate("text", label = "median, 80% and 95% quantile intervals", x = 6, y = -0.5, hjust = 0, vjust = 0.3) +
stat_pointinterval(aes(y = -0.25), point_interval = mode_hdi, .width = c(.95, .80)) +
annotate("text", label = "mode, 80% and 95% highest-density intervals", x = 6, y = -0.25, hjust = 0, vjust = 0.3) +
xlim(-3.25, 18) +
scale_y_continuous(breaks = NULL)
spread_draws()
supports extracting variables that have different indices. It automatically matches up indices with the same name, and duplicates values as necessary to produce one row per all combination of levels of all indices. For example, we might want to calculate the difference between each condition mean and the overall mean. To do that, we can extract draws from the overall mean and all condition means:
m %>%
spread_draws(overall_mean, condition_mean[condition]) %>%
head(10)
## # A tibble: 10 x 6
## # Groups: condition [5]
## .chain .iteration .draw overall_mean condition condition_mean
## <int> <int> <int> <dbl> <chr> <dbl>
## 1 1 1 1 0.0672 A 0.00544
## 2 1 1 1 0.0672 B 1.03
## 3 1 1 1 0.0672 C 1.84
## 4 1 1 1 0.0672 D 1.25
## 5 1 1 1 0.0672 E -0.724
## 6 1 2 2 0.0361 A -0.0836
## 7 1 2 2 0.0361 B 0.873
## 8 1 2 2 0.0361 C 1.78
## 9 1 2 2 0.0361 D 1.12
## 10 1 2 2 0.0361 E -0.884
Within each draw, overall_mean
is repeated as necessary to correspond to every index of condition_mean
. Thus, the dplyr::mutate()
function can be used to take the differences over all rows, then we can summarize with median_qi()
:
m %>%
spread_draws(overall_mean, condition_mean[condition]) %>%
mutate(condition_offset = condition_mean - overall_mean) %>%
median_qi(condition_offset)
## # A tibble: 5 x 7
## condition condition_offset .lower .upper .width .point .interval
## <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A -0.420 -1.69 0.754 0.95 median qi
## 2 B 0.363 -0.863 1.59 0.95 median qi
## 3 C 1.20 -0.0333 2.44 0.95 median qi
## 4 D 0.397 -0.885 1.61 0.95 median qi
## 5 E -1.50 -2.85 -0.340 0.95 median qi
We can use combinations of variables with difference indices to generate predictions from the model. In this case, we can combine the condition means with the residual standard deviation to generate predictive distributions from the model, then show the distributions using ggdist::stat_interval()
and compare them to the data:
m %>%
spread_draws(condition_mean[condition], response_sd) %>%
mutate(y_rep = rnorm(n(), condition_mean, response_sd)) %>%
median_qi(y_rep, .width = c(.95, .8, .5)) %>%
ggplot(aes(y = fct_rev(condition), x = y_rep)) +
geom_interval(aes(xmin = .lower, xmax = .upper)) + #auto-sets aes(color = fct_rev(ordered(.width)))
geom_point(aes(x = response), data = ABC) +
scale_color_brewer()
If this model is well-calibrated, about 95% of the data should be within the outer intervals, 80% in the next-smallest intervals, and 50% in the smallest intervals.
Altogether, data, posterior predictions, and posterior distributions of the means:
draws = m %>%
spread_draws(condition_mean[condition], response_sd)
reps = draws %>%
mutate(y_rep = rnorm(n(), condition_mean, response_sd))
ABC %>%
ggplot(aes(y = condition)) +
stat_interval(aes(x = y_rep), .width = c(.95, .8, .5), data = reps) +
stat_pointinterval(aes(x = condition_mean), .width = c(.95, .66), position = position_nudge(y = -0.3), data = draws) +
geom_point(aes(x = response)) +
scale_color_brewer()
compare_levels()
allows us to compare the value of some variable across levels of some factor. By default it computes all pairwise differences, though this can be changed using the comparison =
argument:
m %>%
spread_draws(condition_mean[condition]) %>%
compare_levels(condition_mean, by = condition) %>%
ggplot(aes(y = condition, x = condition_mean)) +
stat_halfeye()
gather_draws
and gather_variables
We might also prefer all model variable names to be in a single column (long-format) instead of as column names. There are two methods for obtaining long-format data frames with tidybayes
, whose use depends on where and how in the data processing chain you might want to transform into long-format: gather_draws()
and gather_variables()
. There are also two methods for wide (or semi-wide) format data frame, spread_draws()
(described previously) and tidy_draws()
.
gather_draws()
is the counterpart to spread_draws()
, except it puts all variable names in a .variable
column and all draws in a .value
column:
m %>%
gather_draws(overall_mean, condition_mean[condition]) %>%
median_qi()
## # A tibble: 6 x 8
## .variable condition .value .lower .upper .width .point .interval
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 condition_mean A 0.199 -0.142 0.549 0.95 median qi
## 2 condition_mean B 1.01 0.651 1.34 0.95 median qi
## 3 condition_mean C 1.84 1.48 2.19 0.95 median qi
## 4 condition_mean D 1.02 0.681 1.37 0.95 median qi
## 5 condition_mean E -0.890 -1.23 -0.529 0.95 median qi
## 6 overall_mean NA 0.633 -0.546 1.87 0.95 median qi
Note that condition = NA
for the overall_mean
row, because it does not have an index with that name in the specification passed to gather_draws()
.
While this works well if we do not need to perform computations that involve multiple columns, the semi-wide format returned by spread_draws()
is very useful for computations that involve multiple columns names, such as the calculation of the condition_offset
above. If we want to make intermediate computations on the format returned by spread_draws
and then gather variables into one column, we can use gather_variables()
, which will gather all non-grouped variables that are not special columns (like .chain
, .iteration
, or .draw
):
m %>%
spread_draws(overall_mean, condition_mean[condition]) %>%
mutate(condition_offset = condition_mean - overall_mean) %>%
gather_variables() %>%
median_qi()
## # A tibble: 15 x 8
## condition .variable .value .lower .upper .width .point .interval
## <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A condition_mean 0.199 -0.142 0.549 0.95 median qi
## 2 A condition_offset -0.420 -1.69 0.754 0.95 median qi
## 3 A overall_mean 0.633 -0.546 1.87 0.95 median qi
## 4 B condition_mean 1.01 0.651 1.34 0.95 median qi
## 5 B condition_offset 0.363 -0.863 1.59 0.95 median qi
## 6 B overall_mean 0.633 -0.546 1.87 0.95 median qi
## 7 C condition_mean 1.84 1.48 2.19 0.95 median qi
## 8 C condition_offset 1.20 -0.0333 2.44 0.95 median qi
## 9 C overall_mean 0.633 -0.546 1.87 0.95 median qi
## 10 D condition_mean 1.02 0.681 1.37 0.95 median qi
## 11 D condition_offset 0.397 -0.885 1.61 0.95 median qi
## 12 D overall_mean 0.633 -0.546 1.87 0.95 median qi
## 13 E condition_mean -0.890 -1.23 -0.529 0.95 median qi
## 14 E condition_offset -1.50 -2.85 -0.340 0.95 median qi
## 15 E overall_mean 0.633 -0.546 1.87 0.95 median qi
Note how overall_mean
is now repeated here for each condition, because we have performed the gather after spreading model variables across columns.
Finally, if we want raw model variable names as columns names instead of having indices split out as their own column names, we can use tidy_draws()
. Generally speaking spread_draws()
and gather_draws()
are typically more useful than tidy_draws()
, but it is provided as a common method for generating data frames from many types of Bayesian models, and is used internally by gather_draws()
and spread_draws()
:
m %>%
tidy_draws() %>%
head(10)
## # A tibble: 10 x 23
## .chain .iteration .draw overall_mean `condition_zoffset[1]` `condition_zoffset[2]` `condition_zoffs~ `condition_zoff~
## <int> <int> <int> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 1 1 1 0.0672 -0.0237 0.369 0.682 0.455
## 2 1 2 2 0.0361 -0.0468 0.327 0.681 0.424
## 3 1 3 3 1.17 -0.827 -0.156 0.454 -0.00124
## 4 1 4 4 0.378 -0.260 0.603 1.66 0.646
## 5 1 5 5 0.359 -0.207 0.483 1.78 0.716
## 6 1 6 6 0.332 -0.132 0.563 1.89 0.672
## 7 1 7 7 0.226 0.0642 1.11 1.54 1.03
## 8 1 8 8 0.107 -0.0823 0.737 1.32 0.794
## 9 1 9 9 0.225 -0.0611 0.571 1.44 0.755
## 10 1 10 10 -0.0395 0.200 0.885 1.69 0.919
## # ... with 15 more variables: condition_zoffset[5] <dbl>, response_sd <dbl>, condition_mean_sd <dbl>,
## # condition_mean[1] <dbl>, condition_mean[2] <dbl>, condition_mean[3] <dbl>, condition_mean[4] <dbl>,
## # condition_mean[5] <dbl>, lp__ <dbl>, accept_stat__ <dbl>, stepsize__ <dbl>, treedepth__ <dbl>, n_leapfrog__ <dbl>,
## # divergent__ <dbl>, energy__ <dbl>
Combining tidy_draws()
with gather_variables()
also allows us to derive similar output to ggmcmc::ggs()
, if desired:
m %>%
tidy_draws() %>%
gather_variables() %>%
head(10)
## # A tibble: 10 x 5
## # Groups: .variable [1]
## .chain .iteration .draw .variable .value
## <int> <int> <int> <chr> <dbl>
## 1 1 1 1 overall_mean 0.0672
## 2 1 2 2 overall_mean 0.0361
## 3 1 3 3 overall_mean 1.17
## 4 1 4 4 overall_mean 0.378
## 5 1 5 5 overall_mean 0.359
## 6 1 6 6 overall_mean 0.332
## 7 1 7 7 overall_mean 0.226
## 8 1 8 8 overall_mean 0.107
## 9 1 9 9 overall_mean 0.225
## 10 1 10 10 overall_mean -0.0395
But again, this approach does not handle variable indices for us automatically, so using spread_draws()
and gather_draws()
is generally recommended unless you do not have variable indices to worry about.
You can use regular expressions in the specifications passed to spread_draws()
and gather_draws()
to match multiple columns by passing regex = TRUE
. Our example fit contains variables named condition_mean[i]
and condition_zoffset[i]
. We could extract both using a single regular expression:
m %>%
spread_draws(`condition_.*`[condition], regex = TRUE) %>%
head(10)
## # A tibble: 10 x 6
## # Groups: condition [1]
## condition condition_mean condition_zoffset .chain .iteration .draw
## <chr> <dbl> <dbl> <int> <int> <int>
## 1 A 0.00544 -0.0237 1 1 1
## 2 A -0.0836 -0.0468 1 2 2
## 3 A 0.0324 -0.827 1 3 3
## 4 A 0.113 -0.260 1 4 4
## 5 A 0.157 -0.207 1 5 5
## 6 A 0.218 -0.132 1 6 6
## 7 A 0.276 0.0642 1 7 7
## 8 A 0.0130 -0.0823 1 8 8
## 9 A 0.152 -0.0611 1 9 9
## 10 A 0.192 0.200 1 10 10
This result is equivalent in this case to spread_draws(c(condition_mean, condition_zoffset)[condition])
, but does not require us to list each variable explicitly—this can be useful, for example, in models with naming schemes like b_[some name]
for coefficients.
To demonstrate drawing fit curves with uncertainty, let’s fit a slightly naive model to part of the mtcars
dataset using brms::brm()
:
m_mpg = brm(
mpg ~ hp * cyl,
data = mtcars,
file = "models/tidybayes_m_mpg.rds" # cache model (can be removed)
)
We can draw fit curves with probability bands using add_epred_draws()
and ggdist::stat_lineribbon()
:
mtcars %>%
group_by(cyl) %>%
data_grid(hp = seq_range(hp, n = 51)) %>%
add_epred_draws(m_mpg) %>%
ggplot(aes(x = hp, y = mpg, color = ordered(cyl))) +
stat_lineribbon(aes(y = .epred)) +
geom_point(data = mtcars) +
scale_fill_brewer(palette = "Greys") +
scale_color_brewer(palette = "Set2")
Or we can sample a reasonable number of fit lines (say 100) and overplot them:
mtcars %>%
group_by(cyl) %>%
data_grid(hp = seq_range(hp, n = 101)) %>%
# NOTE: this shows the use of ndraws to subsample within add_epred_draws()
# ONLY do this IF you are planning to make spaghetti plots, etc.
# NEVER subsample to a small sample to plot intervals, densities, etc.
add_epred_draws(m_mpg, ndraws = 100) %>%
ggplot(aes(x = hp, y = mpg, color = ordered(cyl))) +
geom_line(aes(y = .epred, group = paste(cyl, .draw)), alpha = .1) +
geom_point(data = mtcars) +
scale_color_brewer(palette = "Dark2")
For more examples of fit line uncertainty, see the corresponding sections in vignette("tidy-brms")
or vignette("tidy-rstanarm")
.
point_interval
with broom::tidy
: A model comparison example
Combining to_broom_names()
with median_qi()
(or more generally, the point_interval()
family of functions) makes it easy to compare results against models supported by broom::tidy()
. For example, let’s compare our model’s fits for conditional means against an ordinary least squares (OLS) regression:
m_linear = lm(response ~ condition, data = ABC)
Combining emmeans::emmeans
with broom::tidy
, we can generate tidy-format summaries of conditional means from the above model:
linear_results = m_linear %>%
emmeans(~ condition) %>%
tidy(conf.int = TRUE) %>%
mutate(model = "OLS")
linear_results
## # A tibble: 5 x 9
## condition estimate std.error df conf.low conf.high statistic p.value model
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <chr>
## 1 A 0.182 0.173 45 -0.167 0.530 1.05 3.00e- 1 OLS
## 2 B 1.01 0.173 45 0.665 1.36 5.85 5.13e- 7 OLS
## 3 C 1.87 0.173 45 1.53 2.22 10.8 4.15e-14 OLS
## 4 D 1.03 0.173 45 0.678 1.38 5.93 3.97e- 7 OLS
## 5 E -0.935 0.173 45 -1.28 -0.586 -5.40 2.41e- 6 OLS
We can derive corresponding fits from our model:
bayes_results = m %>%
spread_draws(condition_mean[condition]) %>%
median_qi(estimate = condition_mean) %>%
to_broom_names() %>%
mutate(model = "Bayes")
bayes_results
## # A tibble: 5 x 8
## condition estimate conf.low conf.high .width .point .interval model
## <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr> <chr>
## 1 A 0.199 -0.142 0.549 0.95 median qi Bayes
## 2 B 1.01 0.651 1.34 0.95 median qi Bayes
## 3 C 1.84 1.48 2.19 0.95 median qi Bayes
## 4 D 1.02 0.681 1.37 0.95 median qi Bayes
## 5 E -0.890 -1.23 -0.529 0.95 median qi Bayes
Here, to_broom_names()
will convert .lower
and .upper
into conf.low
and conf.high
so the names of the columns we need to make the comparison (condition
, estimate
, conf.low
, and conf.high
) all line up easily. This makes it simple to combine the two tidy data frames together using bind_rows
, and plot them:
bind_rows(linear_results, bayes_results) %>%
mutate(condition = fct_rev(condition)) %>%
ggplot(aes(y = condition, x = estimate, xmin = conf.low, xmax = conf.high, color = model)) +
geom_pointinterval(position = position_dodge(width = .3))
Observe the shrinkage towards the overall mean in the Bayesian model compared to the OLS model.
bayesplot
using unspread_draws
and ungather_draws
Functions from other packages might expect draws in the form of a data frame or matrix with variables as columns and draws as rows. That is the format returned by tidy_draws()
, but not by gather_draws()
or spread_draws()
, which split indices from variables out into columns.
It may be desirable to use the spread_draws()
or gather_draws()
functions to transform your draws in some way, and then convert them back into the draw \(\times\) variable format to pass them into functions from other packages, like bayesplot
. The unspread_draws()
and ungather_draws()
functions invert spread_draws()
and gather_draws()
to return a data frame with variable column names that include indices in them and draws as rows.
As an example, let’s re-do the previous example of compare_levels()
, but use bayesplot::mcmc_areas()
to plot the results instead of ggdist::stat_eye()
. First, the result of compare_levels()
looks like this:
m %>%
spread_draws(condition_mean[condition]) %>%
compare_levels(condition_mean, by = condition) %>%
head(10)
## # A tibble: 10 x 5
## # Groups: condition [1]
## .chain .iteration .draw condition condition_mean
## <int> <int> <int> <chr> <dbl>
## 1 1 1 1 B - A 1.02
## 2 1 2 2 B - A 0.957
## 3 1 3 3 B - A 0.921
## 4 1 4 4 B - A 0.881
## 5 1 5 5 B - A 0.676
## 6 1 6 6 B - A 0.599
## 7 1 7 7 B - A 0.813
## 8 1 8 8 B - A 0.940
## 9 1 9 9 B - A 0.750
## 10 1 10 10 B - A 0.794
To get a version we can pass to bayesplot::mcmc_areas()
, all we need to do is invert the spread_draws()
call we started with:
m %>%
spread_draws(condition_mean[condition]) %>%
compare_levels(condition_mean, by = condition) %>%
unspread_draws(condition_mean[condition]) %>%
head(10)
## # A tibble: 10 x 13
## .chain .iteration .draw `condition_mean[B - A]` `condition_mean[~ `condition_mean[~ `condition_mean~ `condition_mean~
## <int> <int> <int> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 1 1 1 1.02 1.84 0.814 1.25 0.224
## 2 1 2 2 0.957 1.86 0.906 1.20 0.248
## 3 1 3 3 0.921 1.76 0.838 1.13 0.213
## 4 1 4 4 0.881 1.96 1.08 0.925 0.0446
## 5 1 5 5 0.676 1.95 1.27 0.904 0.228
## 6 1 6 6 0.599 1.75 1.15 0.693 0.0941
## 7 1 7 7 0.813 1.14 0.331 0.754 -0.0594
## 8 1 8 8 0.940 1.61 0.673 1.01 0.0653
## 9 1 9 9 0.750 1.78 1.03 0.968 0.218
## 10 1 10 10 0.794 1.72 0.927 0.834 0.0393
## # ... with 5 more variables: condition_mean[D - C] <dbl>, condition_mean[E - A] <dbl>, condition_mean[E - B] <dbl>,
## # condition_mean[E - C] <dbl>, condition_mean[E - D] <dbl>
We can pass that into bayesplot::mcmc_areas()
directly. The drop_indices = TRUE
argument to unspread_draws()
indicates that .chain
, .iteration
, and .draw
should not be included in the output:
m %>%
spread_draws(condition_mean[condition]) %>%
compare_levels(condition_mean, by = condition) %>%
unspread_draws(condition_mean[condition], drop_indices = TRUE) %>%
bayesplot::mcmc_areas()
If you are instead working with tidy draws generated by gather_draws()
or gather_variables()
, the ungather_draws()
function will transform those draws into the draw \(\times\) variable format. It has the same syntax as unspread_draws()
.
emmeans
(formerly lsmeans
)
The emmeans::emmeans()
function provides a convenient syntax for generating marginal estimates from a model, including numerous types of contrasts. It also supports some Bayesian modeling packages, like MCMCglmm
, rstanarm
, and brms
. However, it does not provide draws in a tidy format. The gather_emmeans_draws()
function converts output from emmeans
into a tidy format, keeping the emmeans
reference grid and adding a .value
column with long-format draws.
(Another approach to using emmeans
contrast methods is to use emmeans_comparison()
to convert emmeans contrast methods into comparison functions that can be used with tidybayes::compare_levels()
. See the documentation of emmeans_comparison()
for more information).
rstanarm
or brms
Both rstanarm
and brms
behave similarly when used with emmeans::emmeans()
. The example below uses rstanarm
, but should work similarly for brms
.
Given this rstanarm
model:
m_rst = stan_glm(response ~ condition, data = ABC)
We can use emmeans::emmeans()
to get conditional means with uncertainty:
## # A tibble: 5 x 7
## condition .value .lower .upper .width .point .interval
## <fct> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A 0.177 -0.170 0.537 0.95 median qi
## 2 B 1.02 0.668 1.37 0.95 median qi
## 3 C 1.87 1.53 2.21 0.95 median qi
## 4 D 1.03 0.673 1.38 0.95 median qi
## 5 E -0.935 -1.29 -0.581 0.95 median qi
Or emmeans::emmeans()
with emmeans::contrast()
to do all pairwise comparisons:
m_rst %>%
emmeans( ~ condition) %>%
contrast(method = "pairwise") %>%
gather_emmeans_draws() %>%
median_qi()
## # A tibble: 10 x 7
## contrast .value .lower .upper .width .point .interval
## <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A - B -0.842 -1.34 -0.335 0.95 median qi
## 2 A - C -1.70 -2.20 -1.21 0.95 median qi
## 3 A - D -0.848 -1.34 -0.344 0.95 median qi
## 4 A - E 1.11 0.624 1.64 0.95 median qi
## 5 B - C -0.859 -1.33 -0.362 0.95 median qi
## 6 B - D -0.0144 -0.502 0.491 0.95 median qi
## 7 B - E 1.95 1.47 2.47 0.95 median qi
## 8 C - D 0.845 0.359 1.32 0.95 median qi
## 9 C - E 2.81 2.31 3.31 0.95 median qi
## 10 D - E 1.95 1.45 2.49 0.95 median qi
See the documentation for emmeans::pairwise.emmc()
for a list of the numerous contrast types supported by emmeans::emmeans()
.
As before, we can plot the results instead of using a table:
m_rst %>%
emmeans( ~ condition) %>%
contrast(method = "pairwise") %>%
gather_emmeans_draws() %>%
ggplot(aes(x = .value, y = contrast)) +
stat_halfeye()
gather_emmeans_draws()
also supports emm_list
objects, which contain estimates from different reference grids (see emmeans::ref_grid()
for more information on reference grids). An additional column with the default name of .grid
is added to indicate the reference grid for each row in the output:
## # A tibble: 15 x 9
## condition contrast .grid .value .lower .upper .width .point .interval
## <chr> <chr> <chr> <dbl> <dbl> <dbl> <dbl> <chr> <chr>
## 1 A NA emmeans 0.177 -0.170 0.537 0.95 median qi
## 2 B NA emmeans 1.02 0.668 1.37 0.95 median qi
## 3 C NA emmeans 1.87 1.53 2.21 0.95 median qi
## 4 D NA emmeans 1.03 0.673 1.38 0.95 median qi
## 5 E NA emmeans -0.935 -1.29 -0.581 0.95 median qi
## 6 NA A - B contrasts -0.842 -1.34 -0.335 0.95 median qi
## 7 NA A - C contrasts -1.70 -2.20 -1.21 0.95 median qi
## 8 NA A - D contrasts -0.848 -1.34 -0.344 0.95 median qi
## 9 NA A - E contrasts 1.11 0.624 1.64 0.95 median qi
## 10 NA B - C contrasts -0.859 -1.33 -0.362 0.95 median qi
## 11 NA B - D contrasts -0.0144 -0.502 0.491 0.95 median qi
## 12 NA B - E contrasts 1.95 1.47 2.47 0.95 median qi
## 13 NA C - D contrasts 0.845 0.359 1.32 0.95 median qi
## 14 NA C - E contrasts 2.81 2.31 3.31 0.95 median qi
## 15 NA D - E contrasts 1.95 1.45 2.49 0.95 median qi
MCMCglmm
Let’s do the same example as above again, this time using MCMCglmm::MCMCglmm()
instead of rstanarm
. The only different when using MCMCglmm()
is that to use MCMCglmm()
with emmeans()
you must also pass the original data used to fit the model to the emmeans()
call (see vignette("models", package = "emmeans"))
for more information).
First, we’ll fit the model:
# MCMCglmm does not support tibbles directly,
# so we convert ABC to a data.frame on the way in
m_glmm = MCMCglmm(response ~ condition, data = as.data.frame(ABC))
Now we can use emmeans()
and gather_emmeans_draws()
exactly as we did with rstanarm
, but we need to include a data
argument in the emmeans()
call:
m_glmm %>%
emmeans( ~ condition, data = ABC) %>%
contrast(method = "pairwise") %>%
gather_emmeans_draws() %>%
ggplot(aes(x = .value, y = contrast)) +
stat_halfeye()