Package 'surveygraph'

Title: Network Representations of Attitudes
Description: A tool for computing network representations of attitudes, extracted from tabular data such as sociological surveys. Development of surveygraph software and training materials was initially funded by the European Union under the ERC Proof-of-concept programme (ERC, Attitude-Maps-4-All, project number: 101069264). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.
Authors: Samuel Unicomb [aut, cre], Ana Jovancevic [aut], Caoimhe O'Reilly [aut], Alejandro Dinkelberg [aut], Pádraig MacCarron [aut], David O'Sullivan [aut], Paul Maher [aut], Mike Quayle [aut]
Maintainer: Samuel Unicomb <[email protected]>
License: MIT + file LICENSE
Version: 0.1.3
Built: 2025-02-16 05:32:14 UTC
Source: https://github.com/surveygraph/surveygraphr

Help Index

Outputs the survey projection onto the agent or symbolic layer

Description

make_projection() outputs the agent or symbolic network corresponding to a survey, i.e. the row or column projection.

Usage

make_projection(
  data,
  layer = NULL,
  threshold_method = NULL,
  method_value = NULL,
  centre = NULL,
  dummycode = NULL,
  likert = NULL,
  mincomps = NULL,
  similarity_metric = NULL,
  verbose = NULL
)

Arguments

data

A data frame corresponding to a survey
layer

A string flag specifying which layer to project

  • "agent" produces the network corresponding to the agents, which we assume to be rows in data

  • "symbolic" produces the network corresponding to the symbols, or items, which we assume to be columns in data

threshold_method

A string flag specifying how edges are selected in the network representation.

  • "raw_similarity" means we remove all edges whose weight, meaning node similarity, is below a specified threshold.

  • "target_lcc" finds the value of the threshold that results in the network whose largest connected component is as close as possible to a specified value. In general a range of thresholds will satisfy this condition, and we choose the upper limit of this range.

  • "target_ad" finds the value of the threshold that results in the network whose average degree is as close as possible to a specified value.

method_value

A utility variable that we interpret according to the threshold_method chosen.

  • If threshold_method = "raw_similarity", then method_value is interpreted as the similarity threshold, and thus is in the range ⁠[-1, 1]⁠. A value of -1 means no edges are removed, and a value of 1 means all edges are removed.

  • If threshold_method = "target_lcc", then method_value is interpreted as the desired fractional size of the largest connected component, in the range ⁠[0, 1]⁠. E.g., when set to 0, no nodes are connected, and if set to 1, the network is as sparse as possible while remaining fully connected.

  • If threshold_method = "target_ad", then method_value is interpreted as the desired average degree. We assume that method_value is normalised to the range ⁠[0, 1]⁠ When method_value = 0, then no nodes are connected, and if method_value = 1, the network is complete, meaning it contains every possible edge.

centre

If TRUE, we shift edge weights from ⁠[0, 1]⁠ to ⁠[-1, 1]⁠. Defaults to FALSE, as most network analysis applications require positive edge weights.
dummycode

flag that indicates whether we dummycode data.
likert

Specifies the range of the Likert scale contained in data.
mincomps

The minimum number of valid comparisons that must be made when computing the similarity between rows or columns in the data. If at least one of the entries in the fields being compared is NA, then the comparison is invalid.
similarity_metric

This currently has just one allowed value, namely the Manhattan distance, which is the default.
verbose

This is a debugging flag that prints out survey data after a pre-processing step, but before being supplied to the C++ routines that compute the network representation.

Value

A data frame corresponding to the edge list of the specified network. It contains three columns named

  • u, the first node adjacent to the edge

  • v, the second node adjacent to the edge, and

  • weight, the similarity between nodes u and v

Examples

S <- make_synthetic_data(20, 5)

Outputs a synthetic survey using a simple model

Description

make_synthetic_data() outputs a synthetic survey, generated using a simple, stochastic model of polarisation.

Usage

make_synthetic_data(
  nrow,
  ncol,
  minority = 0.5,
  polarisation = 0,
  correlation = 0.85,
  scale = 10
)

Arguments

nrow

The number of rows in the survey
ncol

The number of columns in the survey
minority

The fraction of nodes in the smaller of the two polarised groups
polarisation

The degree of polarisation among the system's agents
correlation

Probability that group item corresponds to polarisation
scale

Range of the Likert scale

Value

A data frame corresponding to a survey.

Examples

S <- make_synthetic_data(200, 8)

Illustrates how network properties vary with the similarity threshold

Description

make_threshold_profile() outputs properties of the agent or symbolic network as a function of similarity threshold.

Usage

make_threshold_profile(data, layer)

Arguments

data

A data frame corresponding to the attitudes held by agents with respect to a number of items
layer

A string flag specifying the type of network to be extracted,

  • "agent" produces the network corresponding to the agents, which we assume to be rows in data

  • "symbolic" produces the network corresponding to the symbols, or items, which we assume to be columns in data

Details

Note that this routine is expensive on large graphs. We study networks over the full range of similarity thresholds ⁠[-1, 1]⁠, and as a result, produce networks that are complete at the lower limit of that range. Note that by default we will subsample the provided survey with the C++ implementation in order to avoid memory issues. We could then allow a flag that turns off the subsampling step, at the user's peril.

Value

A data frame containing properties of the agent or symbolic network as a function of the similarity threshold. In particular, it contains three columns named

  • threshold, the value of the similarity threshold

  • ad, the average degree resulting from threshold, and

  • lcc, the size of the largest connected component resulting from threshold

Examples

S <- make_synthetic_data(20, 5)

surveygraph: network representations of attitudes

Description

This page is a work in progress.

Details

The surveygraph package functions can be summarised as follows.

reading functions

The reading functions import survey datasets to R so they can be passed to C++ routines. A bunch of file formats need to be accounted for.

network generating functions

The network functions are implemented in C++.

Author(s)

Maintainer: Samuel Unicomb [email protected]

Authors:

  • Ana Jovancevic

  • Caoimhe O'Reilly

  • Alejandro Dinkelberg

  • Pádraig MacCarron

  • David O'Sullivan

  • Paul Maher

  • Mike Quayle

See Also

Useful links: