network.survival.estimator: network survival estimator

In networkreporting: Tools for using Network Reporting Estimators

Description

use an aggregate multiplicity estimator and the respondents' own network size estimates to estimate hidden population sizes

Usage

network.survival.estimator_(resp.data, attribute.data, attribute.names,
  known.populations, total.kp.size = 1, weights, attribute.weights,
  dropmiss = NULL, verbose = TRUE)

network.survival.estimator(resp.data, attribute.data, attribute.names,
  known.populations, total.kp.size = 1, weights, attribute.weights,
  verbose = TRUE)

Arguments

resp.data	the dataframe that has a row for each respondent, with reported connections to the groups of known size, as well as the attributes. Note that the column names of the attributes should match their names in attribute.data
attribute.data	A dataframe with the reported attributes of hidden population members reported by survey respondents. There should be one row for each time a respondent reports a hidden population member. For example, to estimate death rates, there should be one row for each report of a death.
attribute.names	the names of the columns of attribute.data and resp.data that contain the attribute information.
known.populations	the names of the columns in resp.data that have responses to the known population questions
total.kp.size	the size of the probe alters, i.e., the sum of the known population sizes
weights	the weights or weights column for the respondent data
attribute.weights	the weights or weights column for the alter data
dropmiss	see report.aggregator
verbose	if TRUE, print information to screen

Details

This function takes two sources of data as input: first, it requires a long-form dataframe with the attributes of the reported members of the hidden population. For example, if we are asking about emigres and we collect the age and sex of each reported emigrant, then the long form dataset might look like:

age	sex	weight
15	m	2.10
58	f	1.15
33	m	3.67

The second source of data we need is the known population responses for the respondents, along with the *same* attributes for each respondent. For example, in the situation above, we would also require a dataset like this to be passed in

age	sex	weight	hm.teachers	hm.nurses	...
20	f	2.10	4	0	...
44	m	1.65	0	2	...
60	m	2.75	1	1	...

Value

the network reporting estimate of the hidden population's size (as a prevalence) broken down by the categories defined by all combinations of attribute.names.

Technical note

This function assumes that the sampling weights are standard analysis weights and *not* relative weights. Standard analysis weights should provide an estimate for the size of the frame population when added up; relative weights, on the other hand, will sum to the number of respondents in the sample. Demographic and Health surveys typically have relative weights, which must be converted into standard sampling weights before using this function.

TODO

• handle missing values

• think about whether or not this is the best way to handle N.F

• write more general agg mult est fn and call that

• make unit tests

networkreporting documentation built on May 2, 2019, 1:52 p.m.