Plot Dynamic Network Model Diagnostics

Plots dynamic network model diagnostics calculated in netdx().

Usage

# S3 method for class 'netdx'
plot(
  x,
  type = "formation",
  method = "l",
  sims = NULL,
  stats = NULL,
  duration.imputed = TRUE,
  sim.lines = FALSE,
  sim.col = NULL,
  sim.lwd = NULL,
  mean.line = TRUE,
  mean.smooth = TRUE,
  mean.col = NULL,
  mean.lwd = 2,
  mean.lty = 1,
  qnts = 0.5,
  qnts.col = NULL,
  qnts.alpha = 0.5,
  qnts.smooth = TRUE,
  targ.line = TRUE,
  targ.col = NULL,
  targ.lwd = 2,
  targ.lty = 2,
  plots.joined = NULL,
  legend = NULL,
  grid = FALSE,
  xlim = NULL,
  xlab = NULL,
  ylim = NULL,
  ylab = NULL,
  momentary = FALSE,
  window = NULL,
  maxdeg = NULL,
  unique.partners = TRUE,
  ...
)

Arguments

x

An EpiModel object of class netdx.

type

Plot type, with options of "formation" for network model formation statistics, "duration" for dissolution model statistics for average edge duration, "dissolution" for dissolution model statistics for proportion of ties dissolved per time step, or "cumldeg" for the cumulative degree distribution.

method

Plot method, with options of "l" for line plots and "b" for box plots. For type = "cumldeg", "b" gives a bar plot of the degree distribution.

sims

A vector of simulation numbers to plot.

stats

Statistics to plot. For type = "formation", stats are among those specified in the call to netdx(); for type = "duration", "dissolution", stats are among those of the dissolution model (without offset()). The default is to plot all statistics.

duration.imputed

If type = "duration", a logical indicating whether or not to impute starting times for relationships extant at the start of the simulation. Defaults to TRUE when type = "duration".

sim.lines

If TRUE, plot individual simulation lines. Default is to plot lines for one-group models but not for two-group models.

sim.col

Vector of any standard R color format for simulation lines.

sim.lwd

Line width for simulation lines.

mean.line

If TRUE, plot mean of simulations across time.

mean.smooth

If TRUE, use a loess smoother on the mean line.

mean.col

Vector of any standard R color format for mean lines.

mean.lwd

Line width for mean lines.

mean.lty

Line type for mean lines.

qnts

If numeric, plot polygon of simulation quantiles based on the range implied by the argument (see details). If FALSE, suppress polygon from plot.

qnts.col

Vector of any standard R color format for polygons.

qnts.alpha

Transparency level for quantile polygons, where 0 = transparent and 1 = opaque (see adjustcolor function).

qnts.smooth

If TRUE, use a loess smoother on quantile polygons.

targ.line

If TRUE, plot target or expected value line for the statistic of interest.

targ.col

Vector of standard R colors for target statistic lines, with default colors based on RColorBrewer color palettes.

targ.lwd

Line width for the line showing the target statistic values.

targ.lty

Line type for the line showing the target statistic values.

plots.joined

If TRUE, combine all statistics in one plot, versus one plot per statistic if FALSE.

legend

If TRUE, plot default legend.

grid

If TRUE, a grid is added to the background of plot (see grid() for details), with default of nx by ny.

xlim

the x limits (x1, x2) of the plot. Note that x1 > x2 is allowed and leads to a ‘reversed axis’.

The default value, NULL, indicates that the range of the finite values to be plotted should be used.

xlab

a label for the x axis, defaults to a description of x.

ylim

the y limits of the plot.

ylab

a label for the y axis, defaults to a description of y.

momentary

If TRUE, add the momentary degree distribution to the cumulative degree plot for comparison. Used only when type = "cumldeg".

window

A vector of length 2 giving the first and last time step over which degrees are counted, with the default of the full simulation. Used only when type = "cumldeg".

maxdeg

Highest degree to show on the x-axis, with degrees above it collapsed into a final "maxdeg+" category. Used only when type = "cumldeg".

unique.partners

If TRUE, a dyad that forms, dissolves, and then re-forms within the window counts as one cumulative partner rather than two. Used only when type = "cumldeg".

...

Additional arguments to pass.

Details

The plot function for netdx objects will generate plots of two types of model diagnostic statistics that run as part of the diagnostic tools within that function. The formation plot shows the summary statistics requested in nwstats.formula, where the default includes those statistics in the network model formation formula specified in the original call to netest().

The duration plot shows the average age of existing edges at each time step, up until the maximum time step requested. The age is used as an estimator of the average duration of edges in the equilibrium state. When duration.imputed = FALSE, edges that exist at the beginning of the simulation are assumed to start with an age of 1, yielding a burn-in period before the observed mean approaches its target. When duration.imputed = TRUE, expected ages prior to the start of the simulation are calculated from the dissolution model, typically eliminating the need for a burn-in period.

The dissolution plot shows the proportion of the extant ties that are dissolved at each time step, up until the maximum time step requested. Typically, the proportion of ties that are dissolved is the reciprocal of the mean relational duration. This plot thus contains similar information to that in the duration plot, but should reach its expected value more quickly, since it is not subject to censoring.

The plots.joined argument will control whether the statistics are joined in one plot or plotted separately, assuming there are multiple statistics in the model. The default is based on the number of network statistics requested. The layout of the separate plots within the larger plot window is also based on the number of statistics.

The cumldeg plot shows the cumulative degree distribution: the proportion of nodes with each count of partners accumulated over the simulation, rather than the count held at one point in time. It requires a netdx object run with keep.tedgelist = TRUE, since it is calculated from the timed edgelist with get_degree_dist(). Setting momentary = TRUE adds the momentary degree distribution over the same window, calculated from the same edgelist, which is the distribution summarized by the degree(k) terms of the formation plot. Two models with the same mean degree and mean partnership duration have the same mean cumulative degree even when their momentary degree distributions differ, as the example below shows for models with high and low concurrency. Legend entries report the mean of each distribution.

Because the cumulative degree grows with the length of the observation window, use window to restrict it to a substantively meaningful period when comparing across models. The lines or bars show the mean across simulations, with qnts giving the inter-simulation quantile band; the targ.line, mean.smooth, qnts.smooth, and plots.joined arguments do not apply to this plot type.

Examples

if (FALSE) { # \dontrun{
# Network initialization and model parameterization
nw <- network_initialize(n = 500)
nw <- set_vertex_attribute(nw, "sex", rbinom(500, 1, 0.5))
formation <- ~edges + nodematch("sex")
target.stats <- c(500, 300)
coef.diss <- dissolution_coefs(dissolution = ~offset(edges) +
                  offset(nodematch("sex")), duration = c(50, 40))

# Estimate the model
est <- netest(nw, formation, target.stats, coef.diss, verbose = FALSE)

# Static diagnostics
dx1 <- netdx(est, nsims = 1e4, dynamic = FALSE,
             nwstats.formula = ~edges + meandeg + concurrent +
                                nodefactor("sex", levels = NULL) +
                                nodematch("sex"))
dx1

# Plot diagnostics
plot(dx1)
plot(dx1, stats = c("edges", "concurrent"), mean.col = "black",
     sim.lines = TRUE, plots.joined = FALSE)
plot(dx1, stats = "edges", method = "b",
     col = "seagreen3", grid = TRUE)

# Dynamic diagnostics
dx2 <- netdx(est, nsims = 10, nsteps = 500,
             nwstats.formula = ~edges + meandeg + concurrent +
                                nodefactor("sex", levels = NULL) +
                                nodematch("sex"))
dx2

# Formation statistics plots, joined and separate
plot(dx2, grid = TRUE)
plot(dx2, type = "formation", plots.joined = TRUE)
plot(dx2, type = "formation", sims = 1, plots.joined = TRUE,
     qnts = FALSE, sim.lines = TRUE, mean.line = FALSE)
plot(dx2, type = "formation", plots.joined = FALSE,
     stats = c("edges", "concurrent"), grid = TRUE)

plot(dx2, method = "b", col = "bisque", grid = TRUE)
plot(dx2, method = "b", stats = "meandeg", col = "dodgerblue")

# Duration statistics plot
par(mfrow = c(1, 2))
# With duration imputed
plot(dx2, type = "duration", sim.line = TRUE, sim.lwd = 0.3,
     targ.lty = 1, targ.lwd = 0.5)
# Without duration imputed
plot(dx2, type = "duration", sim.line = TRUE, sim.lwd = 0.3,
     targ.lty = 1, targ.lwd = 0.5, duration.imputed = FALSE)

# Dissolution statistics plot
plot(dx2, type = "dissolution", qnts = 0.25, grid = TRUE)
plot(dx2, type = "dissolution", method = "b", col = "pink1")

# Cumulative degree distribution: two models with the same mean degree and
# the same mean partnership duration, differing only in how much
# concurrency (overlapping partnerships) they allow
nw <- network_initialize(n = 500)
coef.diss <- dissolution_coefs(dissolution = ~offset(edges), duration = 25)

est.hi <- netest(nw, formation = ~edges + concurrent,
                 target.stats = c(200, 150), coef.diss = coef.diss,
                 verbose = FALSE)
est.lo <- netest(nw, formation = ~edges + concurrent,
                 target.stats = c(200, 40), coef.diss = coef.diss,
                 verbose = FALSE)

dx.hi <- netdx(est.hi, nsims = 5, nsteps = 250, keep.tedgelist = TRUE,
               nwstats.formula = ~edges + concurrent + degree(0:3))
dx.lo <- netdx(est.lo, nsims = 5, nsteps = 250, keep.tedgelist = TRUE,
               nwstats.formula = ~edges + concurrent + degree(0:3))

# The momentary degree distributions are far apart: about half the nodes in
# the high-concurrency model have no partner at any given time and a quarter
# have two, while most nodes in the low-concurrency model have exactly one.
# Over 250 time steps both accumulate about 8.5 partners per node.
par(mfrow = c(1, 2))
plot(dx.hi, type = "cumldeg", momentary = TRUE, xlim = c(0, 25),
     main = "High concurrency")
plot(dx.lo, type = "cumldeg", momentary = TRUE, xlim = c(0, 25),
     main = "Low concurrency")

# As bars, with the tail of the distribution collapsed into one category
# and the inter-simulation range shown
par(mfrow = c(1, 1))
plot(dx.hi, type = "cumldeg", method = "b", maxdeg = 15, qnts = 0.95)

# Partners accumulated over the last 100 steps only
plot(dx.hi, type = "cumldeg", window = c(150, 250), momentary = TRUE)

# The plotted distributions are returned invisibly
dd <- plot(dx.hi, type = "cumldeg")
tapply(dd$degree * dd$prop, dd$sim, sum)
} # }

Usage

Arguments

Details

See also

Examples