Skip to contents

Estimate death rates from network reporting data

network_survival_estimator.Rd

The generic network survival estimator. Given reports about alters connected to survey respondents by some tie, it produces age-specific rates using both the individual-visibility and the aggregate-visibility estimators.

Usage

network_survival_estimator(
  rel.dat,
  ego.id,
  alter.id,
  frame.indicator,
  alter.sex = "sex",
  cell.config,
  weights,
  boot.weights = NULL,
  return.boot = FALSE,
  visibility = vis_from_clique(),
  tie,
  discretize.exp = FALSE,
  .arg.labels = character(0),
  .data.label = "rel.dat"
)

Arguments

rel.dat

The long-form ego X alter dataset: one row per reported alter, per ego

ego.id

String naming the column of rel.dat with the survey respondent's id

alter.id

String naming the column of rel.dat with the alter's id

frame.indicator

String naming the 0/1 column of rel.dat saying whether each alter is in the frame population

alter.sex

String naming the alter attribute that enters the estimation cells alongside age and time period. Called alter.sex because sex is what it is in every current application; carrying several such attributes, rather than one plus cell.config$covars, is future work

cell.config

An object from cell_config() configuring the cells

weights

String naming the column of rel.dat with the sampling weight

boot.weights

Optional dataframe of bootstrap resampled weights; see Details

return.boot

If TRUE, and boot.weights is given, return every bootstrap estimate rather than only their summaries

visibility

A visibility_rule saying how each alter's visibility is derived. Defaults to vis_from_clique(), which is exact for a clique tie and refuses any other structure

tie

A tie_config() saying what kind of tie these reports are about. Required; see above

discretize.exp

Boolean for whether exposure should be discretized. Not yet implemented

.arg.labels

Internal. Named character vector letting a wrapper phrase the up-front column-check message in its own argument names

.data.label

Internal. Name to use for the data argument in that message

Value

a list with asdr.ind (individual-visibility estimates), asdr.agg (aggregate-visibility estimates), ec.dat, esc.dat, and a vis_provenance object saying how visibility was arrived at

Details

siblingsurvival::sibling_estimator() is this function with the sibling names and the clique tie filled in; if you are working with sibling histories, use that.

The tie is required

There is no default tie, and that is the point. Which kind of tie a set of reports is about cannot be read off the data: on a tie that is not a clique, the default vis_from_clique() rule still returns a finite, plausible number, and it can be wrong. Measured against socsim ground truth on a roster that pools maternal and paternal cousins — which is not a clique, even though each line separately is — it overstates visibility by 1.089x for off-frame alters against 1.061x for on-frame ones. Because a death is always off-frame while exposure is a mixture, that differential biases the rate rather than cancelling out of it. See tie_config().

Details

If you want estimated sampling variances, pass a data frame boot.weights. It is assumed to have a column named whatever ego.id is, and then columns boot_weight_1, ..., boot_weight_M.

See also

siblingsurvival::sibling_estimator(), tie_config(), vis_from_clique(), vis_from_donor()