Summary for a fitted mvgam models — summary.mvgam • mvgam

These functions take a fitted mvgam or jsdgam object and return various useful summaries

Usage

# S3 method for mvgam
summary(object, include_betas = TRUE, smooth_test = TRUE, digits = 2, ...)

# S3 method for mvgam_prefit
summary(object, ...)

# S3 method for mvgam
coef(object, summarise = TRUE, ...)

Arguments

object: list object returned from mvgam
include_betas: Logical. Print a summary that includes posterior summaries of all linear predictor beta coefficients (including spline coefficients)? Defaults to TRUE but use FALSE for a more concise summary
smooth_test: Logical. Compute estimated degrees of freedom and approximate p-values for smooth terms? Defaults to TRUE, but users may wish to set to FALSE for complex models with many smooth or random effect terms
digits: The number of significant digits for printing out the summary; defaults to 2.
...: Ignored
summarise: logical. Summaries of coefficients will be returned if TRUE. Otherwise the full posterior distribution will be returned

Value

For summary.mvgam, an object of class mvgam_summary containing:

model_spec: Model specification details (formulas, family, dimensions)
parameters: Parameter estimates and significance tests
diagnostics: MCMC convergence diagnostics
sampling_info: Sampling algorithm details

For summary.mvgam_prefit, a list is printed on-screen showing the model specifications

For coef.mvgam, either a matrix of posterior coefficient distributions (if summarise == FALSE or data.frame of coefficient summaries)

Details

summary.mvgam and summary.mvgam_prefit return brief summaries of the model's call, along with posterior intervals for some of the key parameters in the model. Note that some smooths have extra penalties on the null space, so summaries for the rho parameters may include more penalty terms than the number of smooths in the original model formula. Approximate p-values for smooth terms are also returned, with methods used for their calculation following those used for mgcv equivalents (see summary.gam for details). The Estimated Degrees of Freedom (edf) for smooth terms is computed using either edf.type = 1 for models with no trend component, or edf.type = 0 for models with trend components. These are described in the documentation for jagam. Experiments suggest these p-values tend to be more conservative than those that might be returned from an equivalent model fit with summary.gam using method = 'REML'

coef.mvgam returns either summaries or full posterior estimates for GAM component coefficients

Author

Nicholas J Clark

Examples

# \donttest{
simdat <- sim_mvgam(seasonality = "hierarchical")

mod <- mvgam(
  y ~ series +
    s(season, bs = "cc", k = 6) +
    s(season, series, bs = "fs", k = 4),
  data = simdat$data_train,
  chains = 2,
  silent = 2
)
#> Warning: model has repeated 1-d smooths of same variable.
#> Warning: model has repeated 1-d smooths of same variable.
#> Warning: model has repeated 1-d smooths of same variable.

mod_summary <- summary(mod)
mod_summary
#> GAM formula:
#> y ~ series + s(season, bs = "cc", k = 6) + s(season, series, 
#>     bs = "fs", k = 4)
#> <environment: 0x561c03abdd68>
#> 
#> Family:
#> poisson
#> 
#> Link function:
#> log
#> 
#> Trend model:
#> None
#> 
#> N series:
#> 3 
#> 
#> N timepoints:
#> 75 
#> 
#> Status:
#> Fitted using Stan 
#> 2 chains, each with iter = 1000; warmup = 500; thin = 1 
#> Total post-warmup draws = 1000
#> 
#> GAM coefficient (beta) estimates:
#>                       2.5%     50%  97.5% Rhat n_eff
#> (Intercept)         -1.900 -0.3800  2.000 1.00   254
#> seriesseries_2      -2.400  0.7600  2.600 1.00   324
#> seriesseries_3      -1.700  1.0000  2.900 1.00   367
#> s(season).1         -1.800 -1.3000 -0.720 1.00   282
#> s(season).2          0.330  0.7700  2.000 1.01   109
#> s(season).3          0.110  0.6000  1.800 1.00   121
#> s(season).4          0.079  0.4600  0.920 1.00   273
#> s(season,series).1  -0.200  0.0900  0.710 1.00   180
#> s(season,series).2  -0.860 -0.1900  0.037 1.00   101
#> s(season,series).3  -1.300  0.2900  2.700 1.00   263
#> s(season,series).4  -0.210  0.0083  0.350 1.00   322
#> s(season,series).5  -0.590 -0.0730  0.200 1.00   319
#> s(season,series).6  -0.640 -0.0720  0.130 1.01   112
#> s(season,series).7  -2.700 -0.0920  1.900 1.00   355
#> s(season,series).8  -0.220 -0.0110  0.200 1.01   345
#> s(season,series).9  -0.340  0.0082  0.410 1.00   266
#> s(season,series).10 -0.600 -0.0037  0.210 1.01   115
#> s(season,series).11 -2.500 -0.0070  2.300 1.00   241
#> s(season,series).12 -0.190  0.0130  0.290 1.02   214
#> 
#> Approximate significance of GAM smooths:
#>                    edf Ref.df Chi.sq  p-value    
#> s(season)        3.708      4 62.223 1.33e-05 ***
#> s(season,series) 5.160     12  9.786    0.998    
#> ---
#> Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
#> 
#> Stan MCMC diagnostics:
#> ✔ No issues with effective samples per iteration
#> ✔ Rhat looks good for all parameters
#> ✔ No issues with divergences
#> ✔ No issues with maximum tree depth
#> 
#> Samples were drawn using sampling(hmc). For each parameter, n_eff is a
#>   crude measure of effective sample size, and Rhat is the potential scale
#>   reduction factor on split MCMC chains (at convergence, Rhat = 1)
#> 
#> Use how_to_cite() to get started describing this model
# }