DescriptiveRepresentationCalculator Package Tutorial

Overview

DescriptiveRepresentation is an R package for measuring descriptive representation in political bodies. It implements key functions from Gerring, Jerzak, and Oncel (2024), offering an accessible approach to modeling:

  • Expected representation under a random sampling model,
  • Observed representation for a given body, and
  • Unexplained variance in representation (i.e., representation residuals) under that random sampling model.

The package therefore provides seven main functions:

  • ExpectedRepresentation()
  • ObservedRepresentation()
  • SDRepresentation()
  • RelativeRepresentation()
  • ExpectedDistrictRepresentation()
  • GroupRepresentation()
  • CompareRepresentation()

Each function measures a slightly different concept linked to the ideas in Gerring, Jerzak, and Oncel (2024) and in the companion book manuscript, Gerring, Hicken, Jerzak, Moser, and Oncel, Electoral Rules and Descriptive Representation: A Comprehensive View Across Multiple Identities. In this vignette, we show how to install and use these functions, illustrate a few worked examples, and discuss conceptual underpinnings relevant to descriptive representation. Installation

To install and load the DescriptiveRepresentationCalculator package, run:

# or install.packages("DescriptiveRepresentation")
# install.packages("remotes")  # Install remotes if not already installed
# remotes::install_github("cjerzak/DescriptiveRepresentation-software/DescriptiveRepresentation")
library(DescriptiveRepresentationCalculator)

Background

How well do political bodies reflect the demographic features of the population they serve? This question is at the heart of descriptive representation, and the work of Gerring, Jerzak, and Oncel (2024) offers a systematic way to measure and analyze this phenomenon.

The approach builds on the Rose Index of Proportionality that captures how far a political body’s group shares deviate from the population’s group shares. Concretely: \[ R = 1 - \frac{1}{2} \sum_{k=1}^{K} |g_{p_k} - G_{b_k}| \] where \(g_{p_k}\) is the population share of group \(k\), and \(G_{b_k}\) is that group’s share in the body of interest \(b\). The index ranges from 0 (no descriptive representation) to 1 (perfect descriptive representation).

Note that there are a range of other possible weighting factors in the equation. In the package, the default parameters (a = -0.5 and b = 1), producing the Rose Index. The user can modify the a and b parameters to fit other affine transformations of the underlying absolute deviations: \[ R = b + a \sum_{k=1}^{K} |g_{p_k} - G_{b_k}| \] The Rose Index has nice theoretical properties in that it is bounded between 0 (total representation mismatch) and 1 (complete match between elite and population group shares).

In addition to the L1 (absolute-deviation) form above, the index-computing functions accept a metric argument. Setting metric = "L2" replaces absolute deviations with squared deviations: \[ \tilde{R} = 1 - \frac{1}{2} \sum_{k=1}^{K} \left(g_{p_k} - G_{b_k}\right)^2 \] This squared-deviation index, analyzed in Gerring, Hicken, Jerzak, Moser, and Oncel (book manuscript), is analytically tractable (its expectation under random sampling has an exact closed form) and always upper-bounds the Rose Index. The default metric = "L1" reproduces the classic Rose Index behavior.

Random Sampling Model

One of the insights of Gerring, Jerzak, and Oncel (2024) is to compare observed descriptive representation to what would be expected under a random sampling model—if individuals (or group shares) were randomly drawn into the political body.

This expected value establishes a baseline: how much shortfall or surplus we might attribute purely to compositional factors (like the body’s size or the population’s group diversity).

Divergences from this random baseline can reveal additional, potentially systematic, sources of under- or over-representation.

Package Workflow

  • Compute expected representation for your focal political body using ExpectedRepresentation(). This gives you the theoretical baseline if seats or positions were allocated proportionally by chance given the group shares.
  • Compute observed representation using ObservedRepresentation(). This takes actual data on who occupies each seat (or the observed group shares in the body) and compares it to population-level group shares.
  • Use RelativeRepresentation() to directly compute the difference between observed and expected representation, optionally standardized by the variability under random sampling.
  • Alternatively, compute the standard deviation of representation scores using SDRepresentation() to assess how much variation is left unexplained by the random sampling model.

1. Expected Representation

The function ExpectedRepresentation() computes the expected level of representation (the “expected Rose Index”) under a random sampling model:

ExpectedRepresentation(
  PopShares, 
  BodyN, 
  a = -0.5, 
  b = 1
)

Arguments:

`PopShares`: Numeric vector of group-level population proportions (e.g., `c(0.25, 0.5, 0.25)`).
`BodyN`: Integer, the size of the political body in question (e.g., `50L`).
`a`, `b`: (Optional) Affine transformation parameters. By default, `a=−0.5`,`b=1` (for the Rose Index).

Returns: A single numeric value representing the expected representation score.

Example

# Suppose the population is split into 3 groups: 25%, 50%, 25%.
# We have a political body (say, a legislature) of size 50.

PopShares_example <- c(1/4, 2/4, 1/4)
BodySize_example <- 50

ExpectedRep <- ExpectedRepresentation(
  PopShares = PopShares_example,
  BodyN = BodySize_example
)

ExpectedRep
## [1] 0.9227715
#> Prints the expected representation under random sampling

In many settings, this expected value serves as the baseline to which we compare actual data. Larger bodies and more homogenous populations will tend to have higher expected representation scores under the random sampling model.

2. Observed Representation

To compare theory to reality, we compute the observed representation of any group in a political body:

ObservedRepresentation(
  BodyMemberCharacteristics = NULL,
  PopShares,
  BodyShares = NULL,
  a = -0.5,
  b = 1
)

Arguments:

  • BodyMemberCharacteristics: A vector describing group identities for each member of the body. If supplied, the function automatically calculates the group share distribution.
  • PopShares: Numeric vector of population-level group proportions (with names matching those in BodyMemberCharacteristics).
  • BodyShares: (Optional) A numeric vector with the same structure as PopShares that directly specifies each group’s share in the body. If not NULL, overrides BodyMemberCharacteristics. a, b: Affine transformation parameters, defaulting to (−0.5,1).

Returns: A single numeric value for the observed representation score.

Example

# Observed scenario: A 6-seat body with members: "A", "A", "C", "A", "C", "A"
# The population shares are: A=1/4, B=2/4, C=1/4.

ObsRep <- ObservedRepresentation(
  BodyMemberCharacteristics = c("A","A","C","A","C","A"),
  PopShares = c("A"=0.25, "B"=0.50, "C"=0.25)
)

ObsRep
## [1] 0.5
#> Prints the observed representation index

If group "B" had no seats here, we’d expect a larger observed discrepancy from the population’s proportions, lowering the representation score.

3. Standard Deviation of Representation

Finally, SDRepresentation() estimates the extent to which the observed representation can vary around its expected value under random sampling. It performs Monte Carlo simulations, drawing random compositions of the body and re-computing the representation score each time:

SDRepresentation(
  PopShares, 
  BodyN, 
  a = -0.5, 
  b = 1, 
  nMonte = 10000
)

Arguments:

  • PopShares: Numeric vector of group-level population proportions.
  • BodyN: Size of the political body.
  • a, b: Affine transformation parameters.
  • nMonte: Number of Monte Carlo draws used to approximate the variance.

Returns: A single numeric value summarizing how much representation fluctuates (in standard deviation units) around the expected representation under a random selection model.

Example

SDRep <- SDRepresentation(
  PopShares = c(0.25, 0.50, 0.25),
  BodyN = 50,
  nMonte = 10000
)

SDRep
## [1] 0.04053949
#> Prints the residual standard deviation

In contexts with many social groups or a smaller legislative body, the variance (and thus the SDRepresentation) tends to be larger.

4. Relative Representation

The function RelativeRepresentation() computes the difference between observed and expected representation, providing a direct measure of how a body deviates from the random sampling baseline. It can optionally standardize this difference by the standard deviation under the random sampling model:

RelativeRepresentation(
  BodyMemberCharacteristics,
  PopShares,
  a = -0.5,
  b = 1,
  standardize = FALSE,
  nMonte = 10000
)

Arguments:

  • BodyMemberCharacteristics: A vector specifying characteristics for each member of a political body.
  • PopShares: Numeric vector of population group proportions. Names must correspond to identities in BodyMemberCharacteristics.
  • a, b: Affine transformation parameters (default: Rose Index with a = -0.5, b = 1).
  • standardize: Logical. If TRUE, the difference is divided by the standard deviation under random sampling.
  • nMonte: Number of Monte Carlo iterations for estimating the standard deviation when standardize = TRUE.

Returns: A scalar giving the difference between observed and expected representation. If standardize = TRUE, this difference is expressed in standard deviation units.

Example

# Same body and population as before
BodyMembers <- c("A", "A", "C", "A", "C", "A")
PopShares_example <- c("A" = 0.25, "B" = 0.50, "C" = 0.25)

# Compute relative representation (observed - expected)
RelRep <- RelativeRepresentation(
  BodyMemberCharacteristics = BodyMembers,
  PopShares = PopShares_example
)

RelRep
## [1] -0.2735596
#> Prints how much observed representation differs from expected

# Standardized version (in SD units)
RelRep_std <- RelativeRepresentation(
  BodyMemberCharacteristics = BodyMembers,
  PopShares = PopShares_example,
  standardize = TRUE
)

RelRep_std
## [1] -2.384368
#> Prints the standardized relative representation

A positive value indicates the body is more representative than expected under random sampling; a negative value indicates less representation than expected. When standardized, values beyond ±2 suggest the observed representation is unlikely under the random sampling model alone.

5. District-Level Benchmarks

The functions above treat the political body as a single pool filled by random sampling. ExpectedDistrictRepresentation() extends the framework to bodies elected from single-member districts, following the derivations in Gerring, Hicken, Jerzak, Moser, and Oncel (book manuscript). In this model, each district’s group composition is drawn from a Dirichlet distribution whose mean is PopShares and whose concentration parameter governs geographic segregation: small values produce highly segregated districts, large values produce districts that each resemble the national population.

ExpectedDistrictRepresentation(
  PopShares,
  concentration,
  nDistricts,
  selectionRule = c("random", "affinity"),
  a = -0.5,
  b = 1,
  nMonte = 10000
)

Arguments:

  • PopShares: Numeric vector of group-level population proportions (mean district shares).
  • concentration: Positive number; the Dirichlet concentration parameter controlling spatial concentration of groups across districts.
  • nDistricts: Number of single-member districts, or Inf for the large-district limit.
  • selectionRule: "random" (each district seat won by a group with probability equal to its district share) or "affinity" (the largest group in each district always captures the seat).
  • a, b: Affine transformation parameters.
  • nMonte: Monte Carlo draws used only for "affinity" with more than two groups.

Returns: A single numeric value — the expected squared-deviation representation index (the metric = "L2" index) for the district-elected body.

Example

# Random-sampling seat allocation across 100 districts:
ExpectedDistrictRepresentation(
  PopShares = c(0.3, 0.7),
  concentration = 4,
  nDistricts = 100
)
## [1] 0.99832
#> Approaches 1 as nDistricts grows (Proposition 2 of the book manuscript)

# Largest-group-wins affinity voting, large-district limit:
ExpectedDistrictRepresentation(
  PopShares = c(0.3, 0.7),
  concentration = 4,
  nDistricts = Inf,
  selectionRule = "affinity"
)
## [1] 0.9860725
#> Need not approach 1: the gap reflects the mismatch between each group's
#> population share and its probability of being the largest group in a district

Under random sampling, representation converges to the proportional benchmark as the number of districts grows. Under affinity voting, it converges instead to a limit governed by the mismatch between each group’s mean population share and its probability of forming a local plurality — which is how geographically concentrated groups can secure representation under single-member district rules while dispersed groups (such as women) are disadvantaged.

6. Group-Level Diagnostics

GroupRepresentation() unpacks the scalar representation index into per-group diagnostics:

GroupRepresentation(
  BodyMemberCharacteristics = c("A", "A", "C", "A", "C", "A"),
  PopShares = c("A" = 0.25, "B" = 0.50, "C" = 0.25)
)
##   Group PopShare BodyShare   Shortfall ShortfallRatio
## 1     A     0.25 0.6666667 -0.41666667     -1.6666667
## 2     B     0.50 0.0000000  0.50000000      1.0000000
## 3     C     0.25 0.3333333 -0.08333333     -0.3333333
#> Returns a data.frame with PopShare, BodyShare, Shortfall, and ShortfallRatio

Shortfall (population share minus body share) accentuates the under-representation of large groups; ShortfallRatio (the shortfall as a fraction of the group’s population share) accentuates the under-representation of small groups.

7. Comparing Two Bodies

CompareRepresentation() computes the difference in observed representation between two bodies measured against the same population — for example, the multi-member (MMD) and single-member (SMD) district tiers of a mixed electoral system:

CompareRepresentation(
  BodyMemberCharacteristics1 = c("A", "B", "A", "B", "A", "A"),  # e.g., MMD tier
  BodyMemberCharacteristics2 = c("A", "A", "A", "A", "A", "B"),  # e.g., SMD tier
  PopShares = c("A" = 0.55, "B" = 0.45)
)
## [1] 0.1666667
#> Positive values: the first body better mirrors the population

Interpretation

Expected Representation (ExpectedRepresentation()) helps analysts understand the baseline level of representation when selection is effectively random.

Conversely, Observed Representation (ObservedRepresentation()) is the real-world result, showing how close or far a body’s membership is from the population distribution.

SDRepresentation (SDRepresentation()) quantifies how much randomness alone could explain variation in representation, shedding light on when observed deviations might be plausibly attributed to other (non-random) factors like institutional rules or discrimination.

Relative Representation (RelativeRepresentation()) directly compares observed to expected representation, providing a single measure of deviation from the random sampling baseline. When standardized, it expresses this deviation in terms of standard deviations, enabling cross-body or cross-country comparisons on a common scale.

Expected District Representation (ExpectedDistrictRepresentation()) provides theoretical benchmarks for district-based electoral systems, quantifying how group size, geographic concentration, and the seat-allocation rule (random sampling versus affinity voting) shape expected representation.

Group Representation (GroupRepresentation()) and Compare Representation (CompareRepresentation()) support finer-grained analyses: the former identifies which specific groups drive a body’s representational shortfall, and the latter quantifies which of two bodies (or electoral tiers) better mirrors the population.

Use Cases

  • Policy Analysis: Evaluate how well a legislature (or any political body) mirrors the underlying population demographics.
  • Comparative Politics: Compare representation across countries or subnational regions, controlling for differences in the population’s group structure or body size.
  • Power Analysis: Determine how much of the under-representation or over-representation could arise purely from chance due to small body sizes or population fragmentation.

Conclusion

The DescriptiveRepresentation package operationalizes key ideas about descriptive representation from Gerring, Jerzak, and Oncel (2024). By offering easy-to-use functions for measuring expected, observed, and residual variance in representation, the package helps scholars, analysts, and policymakers investigate how factors like body size and population diversity shape the composition of political bodies worldwide.

We hope this vignette gets you started! For any questions or feedback, feel free to open an issue on our GitHub repository.

References

@article{gerring2024composition,
  title={The Composition of Descriptive Representation},
  author={Gerring, John and Connor T. Jerzak and Erzen \"{O}ncel},
  journal={American Political Science Review},
  year={2024},
  volume={118},
  number={2},
  pages={784-801}
}

@book{gerring2026electoral,
  title={Electoral Rules and Descriptive Representation:
         A Comprehensive View Across Multiple Identities},
  author={Gerring, John and Allen Hicken and Connor T. Jerzak
          and Robert G. Moser and Erzen \"{O}ncel},
  note={Book manuscript}
}