Visualizing sdmTMB conditional effects using visreg

If the code in this vignette has not been evaluated, a rendered version is available on the documentation site under ‘Articles’.

We can use the visreg package to visually inspect the conditional effects of explanatory variables in sdmTMB models.

We will use the built-in Pacific cod survey data for this example. We will fit a model with scaled depth and a factor effect of year using a Tweedie distribution.

pcod_2011$fyear <- as.factor(pcod_2011$year)
mesh <- make_mesh(pcod_2011, c("X", "Y"), cutoff = 20)
#> as(<dgCMatrix>, "dgTMatrix") is deprecated since Matrix 1.5-0; do as(., "TsparseMatrix") instead
fit <- sdmTMB(
  density ~ 0 + s(depth_scaled) + fyear,
  data = pcod_2011,
  mesh = mesh,
  spatial = "off", # off for vignette building speed
  family = tweedie(link = "log")
)
#> Warning in checkMatrixPackageVersion(): Package version inconsistency detected.
#> TMB was built with Matrix version 1.5.3
#> Current Matrix version is 1.5.1
#> Please re-install 'TMB' from source using install.packages('TMB', type = 'source') or ask CRAN for a binary version of 'TMB' matching CRAN's 'Matrix' package

We can then plot the effect of depth in link space. This shows the value of depth on the x-axis and the change in response on the y-axis, holding all other variables constant.

visreg(fit, xvar = "depth_scaled")

Note that by default, randomized quantile residuals are included. These should be normally distributed in link space if the model is consistent with the data. These can be turned off with partial = FALSE

The default, a visreg plot uses base graphics. We can set gg = TRUE to get a ggplot version:

g <- visreg(fit, xvar = "depth_scaled", gg = TRUE)
g + xlab("Depth scaled")

We can also grab the underlying data it creates and make our own plot:

d <- visreg(fit, xvar = "depth_scaled", plot = FALSE)
head(d$fit)
#>   depth_scaled fyear density visregFit visregLwr visregUpr
#> 1    -2.933490  2011       0  2.629109  1.122037  4.136180
#> 2    -2.877239  2011       0  2.524428  1.158454  3.890403
#> 3    -2.820988  2011       0  2.420117  1.185479  3.654754
#> 4    -2.764737  2011       0  2.316777  1.200200  3.433353
#> 5    -2.708486  2011       0  2.215281  1.199393  3.231169
#> 6    -2.652235  2011       0  2.116732  1.180156  3.053308
g <- ggplot(d$fit, aes(x = depth_scaled, y = visregFit)) +
  geom_line() +
  geom_ribbon(aes(ymin = visregLwr, ymax = visregUpr), alpha = 0.5)

# residuals are in d$res
head(d$res)
#>   depth_scaled fyear density  visregRes visregPos
#> 1    0.7408840  2011       0   5.598588      TRUE
#> 2   -2.7062455  2011       0   1.929910     FALSE
#> 3   -2.9334899  2011       0   1.758294     FALSE
#> 4   -0.2224156  2011       0   3.533598     FALSE
#> 5    1.8685071  2011       0  -3.572073      TRUE
#> 6    2.3175211  2011       0 -10.493536     FALSE

g + geom_point(aes(y = visregRes), data = d$res, size = 1, alpha = 0.4)

We can look at the effect of depth in each year separately, either in separate plots or overlaid on the same plot.

visreg(fit, xvar = "depth_scaled", by = "fyear", gg = TRUE)

visreg(fit, xvar = "depth_scaled", by = "fyear", overlay = TRUE, gg = TRUE)

And at the effect of depth on the response scale, rather than link (log) scale.

visreg(fit, xvar = "depth_scaled", scale = "response")

In this case, visreg adds “rug” lines indicating the where the data were observed along the x-axis variable.

visreg() can also be used to visualize just the categorical effect of year.

visreg(fit, xvar = "fyear")

We can also simultaneously compare both depth and year with a two-dimensional contour plot.

visreg2d(fit, xvar = "fyear", yvar = "depth_scaled")

visreg2d() is different from visreg() in that it doesn’t take a gg argument. Instead, it takes plot.type = c("image", "persp", "rgl", "gg"). For example:

visreg2d(fit, xvar = "fyear", yvar = "depth_scaled", plot.type = "gg")

visreg2d(fit, xvar = "fyear", yvar = "depth_scaled", plot.type = "persp")

Delta models

visreg() can also be used to plot the output of delta models from sdmTMB() by using similar code for the previous plots, but using the sdmTMB wrapper function visreg_delta() and specifying model = 1 for the encounter (0 vs. non-zero) model or model = 2 for the positive component model (e.g., Gamma, lognormal). For example:

fit_dg <- sdmTMB(
  density ~ s(depth_scaled, year, k = 8),
  data = pcod_2011,
  mesh = pcod_mesh_2011,
  spatial = "off", # for vignette speed
  family = delta_gamma()
)

visreg_delta(fit_dg, xvar = "depth_scaled", model = 1, gg = TRUE)

visreg_delta(fit_dg, xvar = "depth_scaled", model = 2, gg = TRUE)

Note that for plotting with visreg_delta(), categorical variables like year need to be designated as a factor in the data frame, as in the example above with fyear, rather than in the model formula (e.g., + as.factor(year)).

Julia Indivero, Sean Anderson, Lewis Barnett, Philina English, Eric Ward

2022-11-24

Delta models