control.ergm.ego: Control parameters for `ergm.ego`.

Description

Constructs and checks the list of control parameters for estimation by ergm.ego.

Usage

control.ergm.ego(
  ppopsize = c("auto", "samp", "pop"),
  ppopsize.mul = 1,
  ppop.wt = c("round", "sample"),
  stats.wt = c("data", "ppop"),
  stats.est = c("survey", "asymptotic", "bootstrap", "jackknife", "naive"),
  boot.R = 10000,
  ignore.max.alters = TRUE,
  ergm = control.ergm(),
  ...
)

Value

A list with arguments as components.

Arguments

ppopsize, ppopsize.mul

Parameters to determine the size \(|N'|\) of the pseudopopulation network. ppopsize can be

"auto"

If the popsize (\(|N|\)) argument is specified and is different from 1, as if "pop"; otherwise, as "samp".

"samp"

set \(|N'|\) based on the sample size: \(|N'|=|S| \times \code{popsize.mul}\)

"pop"

set \(|N'|\) based on the population size: \(|N'|=|N| \times \code{popsize.mul}\)

a number

set \(|N'|\) directly (popsize.mul ignored)

a network object

use the specified network as the pseudo-population network directly; use at your own risk

a data frame

use the specified data frame as the pseudo-population; use at your own risk

The default is to use the same pseudopopulation size as the sample size, but, particularly if there are sampling weights in the data, it should be bigger.

Note that depending on ppop.wt, this may only be an approximate target specification, with the actual constructed pseudopopulation network being slightly bigger or smaller.

ppop.wt

Because each ego must be represented in the pseuodopopulation network an integral number of times, if the sample is weighted (or the target \(|N'|\) calculated from ppopsize and ppopsize.mul is not a multiple of the sample size), it may not be possible, for a finite \(|N'|\) to represent each ego exactly according to its relative weight, and ppop.wt controls how the fractional egos are allocated:

"round": (default) Rather than treating ppopsize as a hard setting, calculate \(|N'| w_i / w_\cdot\) for each ego \(i\) and round it to the nearest integer. Then, the \(|N'|\) actually used will be the sum of these rounded freqencies.

"sample"

Resample in proportion to \(w_i\).

stats.wt

Weight assigned to each ego's contribution to the ERGM's sufficient statistic:

"data": (default) Use weights \(|N'| w_i / w_\cdot\) for each ego \(i\) as in the data.
"ppop": Use weights ultimately used in the pseudopopulation network.

stats.est, boot.R

Method to be used to estimate the ERGM's sufficient statistics and their variance:

"survey": Variance estimator returned by survey::svymean(), appropriate to the design of the dataset.

"asymptotic"

Delta method, as derived by Krivitsky and Morris (2017), assuming the ego weights are sampled alongside the egos.

(default)

Delta method, as derived by Krivitsky and Morris (2017), assuming the ego weights are sampled alongside the egos.

"bootstrap"

Nonparametric bootstrap with bias correction, resampling egos, using R replications.

"jackknife"

Jackknife with bias correction.

"naive"

"Naive" estimator, assuming that weights are fixed.

ignore.max.alters

if TRUE, ignores any constraints on the number of nominations. Used to be FALSE, now TRUE in light of the findings of Krivitsky et. al (2020).

ergm

Control parameters for the ergm() call to fit the model, constructed by control.ergm().

...

Not used at this time.

Author

Pavel N. Krivitsky

References

Pavel N. Krivitsky and Martina Morris (2017). "Inference for social network models from egocentrically sampled data, with application to understanding persistent racial disparities in HIV prevalence in the US." Annals of Applied Statistics, 11(1): 427–455. tools:::Rd_expr_doi("10.1214/16-AOAS1010")

Pavel N. Krivitsky, Martina Morris, and Michał Bojanowski (2019). "Inference for Exponential-Family Random Graph Models from Egocentrically-Sampled Data with Alter–Alter Relations." NIASRA Working Paper 08-19. https://www.uow.edu.au/niasra/publications/

Pavel N. Krivitsky, Michał Bojanowski, and Martina Morris (2020). "Impact of survey design on estimation of exponential-family random graph models from egocentrically-sampled data." Social Networks, to appear. tools:::Rd_expr_doi("10.1016/j.socnet.2020.10.001")