Case Study: Measuring Group Alignment in the CES

Case Study: Measuring Group Alignment in the Cooperative Election Study (CES)

The Cooperative Election Study (CES) is a large-scale collaborative survey fielded annually around U.S. elections, with tens of thousands of respondents. It includes a rich set of policy preference items, making it well-suited for measuring intra-group alignment across political, demographic, and geographic subgroups.

data(ces)

Exploring the Data

Before measuring alignment, it is worth inspecting which items are available across survey waves and how support is distributed across groups.

First, plot_coverage() shows which items were fielded in each wave, which is necessary for the selection of issue basket prior to any temporal analyses.

ces |>
    plot_coverage(wave_col="year", ques_stem="immig")

ces |>
    plot_coverage(wave_col="year", ques_stem="spend")

Meanwhile, plot_support() summarizes the share of respondents supporting each item (i.e. selecting the top-coded category for that column), which allows for a helpful birds-eye view of individual issue majorities.

# Cross-sectional support 
ces |>
  filter(year == 2024) |>
  plot_support(weight_col="weight", group_col="pid3", ques_stem="immig") +
  ggtitle("Support for CES Immigration Items (2024)")

Similarly, plot_pairwise_support() gives a sense of how correlated responses are across items within a group — a useful diagnostic for whether items are tapping a coherent underlying dimension.

We can use this to assess, for instance, Democrats’ pairwise support for different spending proposals:

ces |>
  filter(pid3 == "Democrat", year == 2024) |>
  plot_pairwise_support(ques_stem="spend")

Measuring Alignment

The core items used in the alignment analysis below cover immigration, environmental policy, gun control, military spending, and trade — areas where partisan divides are typically most pronounced in the CES.

CES Core Policy Preference Items (click to expand)

abortion_20weeks: Ban abortions after the 20th week of pregnancy
abortion_always: Always allow a woman to obtain an abortion as a matter of choice
abortion_conditional: Permit abortion only in cases of rape, incest, or danger to woman’s life
abortion_coverage: Allow employers to decline coverage of abortions in insurance plans
abortion_expenditures: Prohibit the expenditure of federal funds for any abortion
abortion_prohibition: Make abortions illegal in all circumstances
enviro_airwateracts: Strengthen EPA enforcement of the Clean Air Act and Clean Water Act even if it costs jobs
enviro_carbon: Give the Environmental Protection Agency power to regulate carbon dioxide emissions
enviro_mpg_raise: Raise required fuel efficiency standards for automobiles
enviro_renewable: Require states to use a minimum amount of renewable fuels for electricity generation
guns_assaultban: Ban assault rifles
guns_bgchecks: Require criminal background checks for all gun sales
guns_names: Prohibit state and local governments from publishing names and addresses of gun owners
guns_permits: Make it easier for people to obtain concealed-carry permits
immig_border: Increase border security along the U.S.–Mexico border
immig_deport: Deport undocumented immigrants
immig_employer: Penalize employers for hiring undocumented immigrants
immig_legalize: Grant legal status to undocumented immigrants meeting work and tax requirements
immig_police: Allow police to question individuals about immigration status
immig_reduce: Reduce overall levels of immigration
immig_report: Require local police to report undocumented immigrants to federal authorities
immig_services: Provide public services to undocumented immigrants
immig_wall: Build a wall along the U.S.–Mexico border
military_democracy: Use military force to promote democracy abroad
military_genocide: Use military force to stop genocide
military_helpun: Use military force to help the United Nations enforce international law
military_oil: Use military force to ensure access to foreign oil supplies
military_protectallies: Use military force to protect U.S. allies under attack
military_terroristcamp: Use military force to destroy terrorist training camps
trade_canmex_except: Exclude Canada and Mexico from U.S. trade agreements
trade_canmex_include: Include Canada and Mexico in U.S. trade agreements
trade_china: Support or oppose free trade agreements with China

We begin with a simpler case: measuring alignment on spending preference items alone. plot_cumulative_majority() shows the distribution of item-level agreement, ordered by support level — a useful first look at where consensus is strong or weak within a group.

ces24_spend_align <- ces |>
  filter(year == 2024) |>
  measure_alignment(
    ques_stem="spend",
    group_col="pid3",
    id_col="id"
  )

## 
## [0/5] Pivoting...
## [1/5] Computing group-wise plurality opinions
## [2/5] Measuring question-level respondent group alignment
## [3/5] Sorting questions by alignment within each group
## [4/5] Measuring respondent's overall group alignment
## [5/5] Computing cumulative pluralities

ces24_spend_align |>
  plot_cumulative_majority()

Now we turn to the broader core policy battery. plot_individual_alignments() shows the full distribution of pairwise alignment scores across all respondent pairs within each group. The further right the distribution is shifted, the more internally cohesive the group.

ces24_core_align <- ces |>
  filter(year == 2024) |>
  measure_alignment(
    ques_stem="(abort|immig|enviro|guns|military|trade)",
    group_col="pid3",
    id_col="id"
  )

## 
## [0/5] Pivoting...
## [1/5] Computing group-wise plurality opinions
## [2/5] Measuring question-level respondent group alignment
## [3/5] Sorting questions by alignment within each group
## [4/5] Measuring respondent's overall group alignment
## [5/5] Computing cumulative pluralities

ces24_core_align |>
  plot_individual_alignments(show_dists_as="histogram")

The alignment curve summarizes how alignment accumulates as the threshold for “agreement” increases. A group whose curve stays high across thresholds exhibits robust consensus; a curve that drops off quickly suggests only shallow or item-specific agreement.

ces24_core_align |>
  plot_alignment_curve(group_colors=pid_colors, interactive=TRUE)

Additionally, we can output a single dataframe with all of the metrics in one, using summary():

summary(ces24_core_align, format="long")

pid3	statistic	value
Democrat	Cumulative Perfect Alignment	0.05
Democrat	Cumulative Weak Alignment	0.97
Democrat	Alignment Mean	0.79
Democrat	Alignment SE	0.00
Democrat	N Respondents	22982.00
Democrat	Mean Plurality	0.79
Democrat	Cumulative Issue Alignment N	8.00
Democrat	Cumulative Issue Alignment Prop	0.47
Republican	Cumulative Perfect Alignment	0.00
Republican	Cumulative Weak Alignment	0.88
Republican	Alignment Mean	0.67
Republican	Alignment SE	0.00
Republican	N Respondents	18663.00
Republican	Mean Plurality	0.67
Republican	Cumulative Issue Alignment N	4.00
Republican	Cumulative Issue Alignment Prop	0.24
Independent/Other	Cumulative Perfect Alignment	0.03
Independent/Other	Cumulative Weak Alignment	0.93
Independent/Other	Alignment Mean	0.74
Independent/Other	Alignment SE	0.00
Independent/Other	N Respondents	15913.00
Independent/Other	Mean Plurality	0.74
Independent/Other	Cumulative Issue Alignment N	5.00
Independent/Other	Cumulative Issue Alignment Prop	0.29

Alignment with Different Missing Value Assumptions

By default, measure_alignment() excludes missing responses from the denominator. An alternative is to treat NA responses as explicit non-support — a more conservative assumption that increases the penalty for non-response. Under this assumption, alignment tends to fall noticeably for all groups, reflecting the substantive uncertainty that unanswered items introduce.

ces24_core_align_na <- ces |>
  filter(year == 2024) |>
  measure_alignment(
    ques_stem="(abort|immig|enviro|guns|military|trade)",
    group_col="pid3",
    id_col="id",
    treat_na="unaligned"
  )

## 
## [0/5] Pivoting...
## [1/5] Computing group-wise plurality opinions
## [2/5] Measuring question-level respondent group alignment
## [3/5] Sorting questions by alignment within each group
## [4/5] Measuring respondent's overall group alignment
## [5/5] Computing cumulative pluralities

ces24_core_align_na |>
    plot_alignment_curve(group_colors = pid_colors)

Alignment Over Time

Policy preferences are not static, and neither is within-group consensus. measure_alignment_waves() computes alignment separately for each survey wave, allowing us to track whether groups are becoming more or less internally cohesive over time.

To begin, let’s actually check which issues questions are consistently asked over time.

# Temporal support
ces |>
  plot_support(wave_col="year", weight_col="weight", 
               group_col="pid3", ques_stem="immig") +
  ggtitle("Support for CES Immigration Items")

# Measure alignment over time
ces_pid_align <- ces |>
  measure_alignment_waves(
    ques_stem="(abort|immig|enviro|guns|military|trade)",
    group_col  = "pid3",
    wave_col   = "year",
    id_col     = "id",
    weight_col = "weight",
    verbose    = FALSE
  )

The alignment mean summarizes, on average, what share of a group’s members agree on each policy item. Higher values reflect greater internal consensus. Examining this metric over time reveals whether partisan sorting on core policy dimensions is intensifying, stabilizing, or reversing.

ces_pid_align |>
  plot_group_stat_over_time(
    metric       = "alignment_mean",
    wave_col     = "year",
    group_col    = "pid3",
    group_label  = "Party ID",
    group_colors = pid_colors
  )

Cumulative weak alignment is a more lenient threshold: it counts the share of items on which at least a bare majority of group members agree. This metric is less sensitive to the handful of items where near-unanimous consensus exists, and captures whether a group agrees directionally across most issues — even if not overwhelmingly so.

ces_pid_align |>
  plot_group_stat_over_time(
    metric       = "cumulative_weak_alignment",
    wave_col     = "year",
    group_col    = "pid3",
    group_label  = "Party ID",
    group_colors = pid_colors
  )

Alignment on Different Issue Baskets

Alignment scores can be sensitive to the choice of issue baskets. The core policy items above (immigration, guns, environment, etc.) may capture different dimensions of consensus than, say, spending preferences. The spending battery asks respondents to trade off government spending levels across policy domains, which may produce different alignment patterns than direct policy support questions.

CES National Spending Items (click to expand)

spending_education: Increase government spending on education
spending_healthcare: Increase government spending on health care
spending_infrastructure: Increase government spending on infrastructure
spending_police: Increase government spending on police
spending_welfare: Increase government spending on welfare

One advantage of using the spending issue basket is that it is consistently asked in the same format across multiple waves othe CES.

# Plot support
ces |>
  plot_support(wave_col="year", weight_col="weight", 
               group_col="pid3", ques_stem="spend") +
  ggtitle("Support for CES Spending Items")

We can see that the results are largely the same as with using the core issues basket.

# Measure alignment
ces_spend_align <- ces |>
  measure_alignment_waves(
    ques_stem  = "spend",
    group_col  = "pid3",
    wave_col   = "year",
    id_col     = "id",
    weight_col = "weight",
    verbose    = FALSE
  )

# Plot alignment means
ces_spend_align |>
  plot_group_stat_over_time(
    metric       = "alignment_mean",
    wave_col     = "year",
    group_col    = "pid3",
    group_label  = "Party ID",
    group_colors = pid_colors
  )

# Plot cumulative weak alignment
ces_spend_align |>
  plot_group_stat_over_time(
    metric       = "cumulative_weak_alignment",
    wave_col     = "year",
    group_col    = "pid3",
    group_label  = "Party ID",
    group_colors = pid_colors
  )

Alignment Within Different Groups

Party ID is a natural grouping variable for studying policy alignment, but survalign works with any categorical grouping.

Racial Alignment

ces24_race_align <- ces |>
  filter(year == 2024) |>
  measure_alignment(
    ques_stem="(abort|immig|enviro|guns|military|trade)",
    group_col="race5",
    id_col="id"
  )

## 
## [0/5] Pivoting...
## [1/5] Computing group-wise plurality opinions
## [2/5] Measuring question-level respondent group alignment
## [3/5] Sorting questions by alignment within each group
## [4/5] Measuring respondent's overall group alignment
## [5/5] Computing cumulative pluralities

ces24_race_align |>
  plot_alignment_curve(interactive = TRUE)

Educational Alignment

ces24_educ_align <- ces |>
  filter(year == 2024) |>
  measure_alignment(
    ques_stem="(abort|immig|enviro|guns|military|trade)",
    group_col="educ5",
    id_col="id"
  )

## 
## [0/5] Pivoting...
## [1/5] Computing group-wise plurality opinions
## [2/5] Measuring question-level respondent group alignment
## [3/5] Sorting questions by alignment within each group
## [4/5] Measuring respondent's overall group alignment
## [5/5] Computing cumulative pluralities

ces24_educ_align |>
  plot_alignment_curve(interactive = TRUE)

A technical tutorial of survalign with the Cooperative Election Study