Package {streetlamp}

streetlamp: Panel Econometrics for the Effect of Police Stop and Search on Recorded Crime

Description

streetlamp builds an area-by-month panel of recorded crime in England and Wales from the data.police.uk bulk archive, attaches stop and search intensity as a treatment variable, and provides estimators for evaluating policing interventions. It is about what stops achieve, not who is stopped: it never computes an ethnic disparity measure and reads stop and search files only to count stops by area and month.

Design principles

• Every panel carries a panel contract recording provenance, coverage, LSOA vintage and treatment definition. Estimators read it and refuse or warn when their assumptions are violated.

• "Force did not submit" is never a zero.

• Anti-social behaviour is not a crime: it is kept in a separate series and excluded from crime totals by default.

• Treatment effects are reported with the identifying assumption named in the output, and every estimator ships with a placebo or pre-trend diagnostic.

• No network access during ⁠R CMD check⁠, tests or examples.

Data attribution

Recorded crime, outcomes and stop and search counts derive from data.police.uk, and geography from the Office for National Statistics Open Geography Portal, both published under the Open Government Licence v3.0. Bank holidays derive from GOV.UK. See inst/NOTES/data_sources.md in the package sources for what was verified and when.

Author(s)

Maintainer: Mustapha Wasseja muswaseja@gmail.com

Authors:

• Souci Frissa

• Sarah Hamed

Other contributors:

• Black Thrive Global [copyright holder, funder]

See Also

Useful links:

• https://github.com/BlackThrive/streetlamp

• https://blackthrive.github.io/streetlamp/

• Report bugs at https://github.com/BlackThrive/streetlamp/issues

Spatial adjacency for a set of areas

Description

Builds a neighbour list and row-standardised spatial weights for area polygons with spdep, for the spillover estimators and for Moran's I diagnostics. Queen contiguity (any shared point) is the default; rook contiguity needs a shared edge; knn links each area to its k nearest neighbours by centroid and symmetrises the result. Areas with no neighbours (islands) are allowed and counted.

Usage

lamp_adjacency(boundaries, method = c("queen", "rook", "knn"), k = 6L)

Arguments

Value

A list of class lamp_adjacency with areas (identifiers in order), nb (an spdep neighbour list), listw (row-standardised weights, style = "W"), method, k, n_islands, boundary_vintage (taken from the vintage attribute of boundaries when present) and created.

See Also

Other panel: lamp_adjacency_from_nb(), lamp_boundaries(), lamp_changelog(), lamp_coverage(), lamp_panel(), lamp_population(), lamp_revintage()

Examples

squares <- sf::st_sf(
  area = c("a", "b", "c"),
  geometry = sf::st_sfc(
    sf::st_polygon(list(rbind(c(0, 0), c(1, 0), c(1, 1), c(0, 1), c(0, 0)))),
    sf::st_polygon(list(rbind(c(1, 0), c(2, 0), c(2, 1), c(1, 1), c(1, 0)))),
    sf::st_polygon(list(rbind(c(5, 5), c(6, 5), c(6, 6), c(5, 6), c(5, 5)))),
    crs = 27700
  )
)
adj <- lamp_adjacency(squares)
adj
adj$nb

Build an adjacency object from a neighbour list

Description

Wraps an existing spdep neighbour list, for example the lattice built by lamp_simulate(), in the object the spillover estimators expect.

Usage

lamp_adjacency_from_nb(nb, areas, method = "supplied")

Arguments

Value

A lamp_adjacency object.

See Also

Other panel: lamp_adjacency(), lamp_boundaries(), lamp_changelog(), lamp_coverage(), lamp_panel(), lamp_population(), lamp_revintage()

Examples

sim <- lamp_simulate(n_areas = 16, n_months = 12, design = "spillover", seed = 1)
truth <- attr(sim, "truth")
lamp_adjacency_from_nb(truth$nb, truth$lattice$area)

How stop and search activity follows recorded crime

Description

Regresses searches on the crime recorded in previous months, with area and month fixed effects. This is not an effect of crime on searching in any causal sense; it is a description of how activity is allocated, and it is here because without it the other estimators cannot be read honestly.

Usage

lamp_allocation(
  panel,
  stops = "stops",
  crime = "crime_total",
  crime_lags = 1:3,
  cluster = c("area", "force"),
  family = c("poisson", "ols_log")
)

Arguments

Value

A lamp_estimate of class lamp_allocation, with one coefficient per lag and diagnostics$total for their sum. The identifying assumption field states plainly that the result is descriptive.

Why this matters for every other estimate

If police search more where crime has just risen, then searching is a response to crime as well as a possible cause of it. A regression of crime on searches then mixes the two directions: the deterrent effect pushes the coefficient down, the allocation response pushes it up, and the estimate is the net of them. A positive coefficient here is the size of the channel that has to be argued away before a stop-crime elasticity or a difference-in-differences estimate can be called causal. A coefficient near zero does not establish exogeneity or rule out other allocation channels.

See Also

Other estimators: lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(n_areas = 24, n_months = 30, design = "continuous", seed = 1)
lamp_allocation(sim, crime_lags = 1:2)

Download force-month files from the data.police.uk archive

Description

Fetches the street, outcomes and stop-and-search CSV files for chosen months and forces from one or more archive snapshots, without downloading the whole zip: the zip's table of contents is read with an HTTP byte-range request and each wanted file is then fetched by range and inflated locally. Files already present in the cache with a matching checksum are skipped. Each fetched file is recorded with its CRC32 from the archive, a SHA-256 of its content and the time of download, which is what the panel contract reports as provenance.

Usage

lamp_archive_download(
  archives = "latest",
  months = NULL,
  forces = NULL,
  file_types = c("street", "outcomes", "stop-and-search"),
  dir = lamp_cache_dir(),
  index = NULL,
  relist = FALSE
)

Arguments

Value

A lamp_snapshot (see lamp_archive_snapshot()) for dir, invisibly.

See Also

Other ingest: lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

# Needs network access. Fetches the latest month for one force into a
# temporary directory, and prints a message instead of failing when offline.
cache <- tempfile("streetlamp-")
snap <- tryCatch(
  {
    idx <- lamp_archive_index(dir = cache)
    lamp_archive_download(
      months = format(max(idx$to), "%Y-%m"), forces = "dyfed-powys",
      dir = cache, index = idx
    )
  },
  streetlamp_error_network = function(e) message(conditionMessage(e))
)
snap

Index of the data.police.uk archive

Description

Lists the monthly archive snapshots published at https://data.police.uk/data/archive/: one zip per month, each holding the street, outcomes and stop-and-search CSV files for every force and, since the May 2017 snapshot, a rolling window of the latest 36 months (earlier snapshots hold every month back to December 2010). The index is fetched once and cached; pass refresh = TRUE to fetch it again. Without network access a cached index is returned with a warning, and an error of class streetlamp_error_network is raised if there is no cache.

Usage

lamp_archive_index(refresh = FALSE, dir = lamp_cache_dir())

Arguments

Value

A tibble of class lamp_archive_index with one row per snapshot: archive (identifier such as "2026-07"), label, url, from and to (first and last data month in the zip), n_months, size_gb (as shown on the page) and md5 (the published checksum of the zip). The attribute fetched_at records when the page was read.

See Also

Other ingest: lamp_archive_download(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

# Needs network access. The index is cached in a temporary directory, and
# the example prints a message instead of failing when offline.
idx <- tryCatch(
  lamp_archive_index(dir = tempfile("streetlamp-")),
  streetlamp_error_network = function(e) message(conditionMessage(e))
)
if (!is.null(idx)) tail(idx)

Register a locally downloaded archive zip

Description

Makes a complete archive zip that was downloaded by other means (for example in a browser) available to lamp_archive_snapshot() without copying it: the zip is listed and every member becomes readable in place.

Usage

lamp_archive_register(path, dir = lamp_cache_dir(), md5 = NA_character_)

Arguments

Value

The archive identifier, invisibly.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zip <- system.file("extdata", "archive", "2026-07.zip", package = "streetlamp")
lamp_archive_register(zip, dir = cache)
lamp_archive_snapshot(cache)

Snapshot of the archive files held locally

Description

Describes what the cache directory holds: which archive snapshots have been listed, and for every force-month file in them whether its content is available locally, either because it was fetched by lamp_archive_download() or because the whole zip was registered with lamp_archive_register(). The snapshot is the starting point for lamp_list_versions() and the readers.

Usage

lamp_archive_snapshot(dir = lamp_cache_dir())

Arguments

Value

A list of class lamp_snapshot with elements dir, archives (one row per archive: archive, source, kind, size, etag, last_modified, md5, n_members, listed_at) and members (one row per file: archive, member, month, force_id, file_type, crc32, csize, usize, available, path, sha256, fetched_at).

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zips <- list.files(
  system.file("extdata", "archive", package = "streetlamp"),
  pattern = "zip$", full.names = TRUE
)
for (z in zips) lamp_archive_register(z, dir = cache)
snap <- lamp_archive_snapshot(cache)
snap
table(snap$members$force_id, snap$members$file_type)

Area hierarchy for 2021 LSOAs

Description

One row per 2021 Lower layer Super Output Area in England and Wales with its 2021 Middle layer Super Output Area, 2022 local authority district, police force area and the police.uk force identifier. This is how lamp_panel() reaches the msoa21, lad and pfa levels and how it knows which areas belong to a force.

Usage

lamp_area_lookup()

Details

Source: Office for National Statistics lookups licensed under the Open Government Licence v3.0; see inst/NOTES/data_sources.md for the products and retrieval dates.

Value

A tibble with columns lsoa21, lsoa21_name, msoa21, msoa21_name, lad22, lad22_name, pfa, pfa_name and force_id.

See Also

Other bundled data: lamp_bank_holidays(), lamp_crime_types(), lamp_deprivation(), lamp_forces(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_sample_panel(), lamp_shocks()

Examples

lk <- lamp_area_lookup()
table(lk$force_id[lk$pfa_name == "West Yorkshire"])
length(unique(lk$msoa21))

Bank holidays in England and Wales

Description

Bank and public holidays in England and Wales from 2010 onwards, for seasonality controls. Rows from 2012 onwards are taken verbatim from the GOV.UK bank holidays feed (the current feed for 2019 onwards and archived copies of the same feed for 2012 to 2018); 2010 and 2011 are transcribed from the archived Directgov bank holidays page. Apostrophes are normalised to ASCII.

Usage

lamp_bank_holidays()

Value

A tibble with columns date (Date), title, notes (for example ⁠Substitute day⁠), bunting (logical, as published; NA for the transcribed 2010 and 2011 rows), source and source_url.

See Also

Other bundled data: lamp_area_lookup(), lamp_crime_types(), lamp_deprivation(), lamp_forces(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_sample_panel(), lamp_shocks()

Examples

bh <- lamp_bank_holidays()
range(bh$date)
bh[format(bh$date, "%Y") == "2023", c("date", "title")]

ONS digital boundaries as an sf layer

Description

Fetches the generalised, clipped (BGC, 20 m) boundaries for a geography from the ONS Open Geography Portal, in British National Grid (EPSG:27700), and caches them. Products: 2021 LSOAs, 2011 LSOAs, 2021 MSOAs, December 2022 local authority districts (England and Wales only) and December 2023 police force areas. The generalised boundaries suit adjacency and mapping; a search made within a few metres of a boundary can be assigned to the neighbouring area, so supply full-resolution boundaries yourself for point-in-polygon work that needs that precision.

Usage

lamp_boundaries(
  type = c("lsoa21", "lsoa11", "msoa21", "lad", "pfa"),
  dir = lamp_cache_dir(),
  refresh = FALSE
)

Arguments

Details

Without network access a cached copy is returned, and an error of class streetlamp_error_network is raised if there is none. A first fetch of the 2021 LSOA layer reads about 50 MB in 18 pages.

Source: Office for National Statistics licensed under the Open Government Licence v3.0; contains OS data, Crown copyright and database right 2024.

Value

An sf object with columns area (code), name and geometry (multipolygons, EPSG:27700), sorted by area, with attributes vintage (for example "lsoa21-bgc-v5"), product (the ONS title and item identifier) and retrieved.

See Also

Other panel: lamp_adjacency(), lamp_adjacency_from_nb(), lamp_changelog(), lamp_coverage(), lamp_panel(), lamp_population(), lamp_revintage()

Examples

# Needs network access. The boundaries are cached in a temporary directory,
# and the example prints a message instead of failing when offline.
pfa <- tryCatch(
  lamp_boundaries("pfa", dir = tempfile("streetlamp-")),
  streetlamp_error_network = function(e) message(conditionMessage(e))
)
if (!is.null(pfa)) {
  plot(sf::st_geometry(pfa))
  lamp_adjacency(pfa)
}

Location of the streetlamp cache

Description

Downloaded archive files, boundaries and lookups are cached on disk so that each is fetched once. By default the cache is a directory inside the session's temporary directory: it is deleted when R closes, and nothing is written to your file space.

Usage

lamp_cache_dir(create = TRUE)

lamp_cache_clear(subdir = NULL)

Arguments

Details

To keep downloads between sessions, which is worth doing before building a national panel of several gigabytes, choose a directory and set it with the option streetlamp.cache_dir or the environment variable STREETLAMP_CACHE_DIR (the option wins). The conventional place is the user cache directory, for example in .Rprofile:

options(streetlamp.cache_dir = tools::R_user_dir("streetlamp", "cache"))

lamp_cache_clear() deletes the cache, or one sub-directory of it.

Value

The cache directory path, invisibly for lamp_cache_clear().

Examples

# Point the cache at a temporary directory for the duration of the example
old <- options(streetlamp.cache_dir = tempfile("streetlamp-cache-"))
lamp_cache_dir()
lamp_cache_clear()
options(old)

The data.police.uk changelog as a table of refresh and gap notes

Description

Fetches and parses the publisher's changelog, which records when a force re-supplied months ("data refresh") or failed to supply a month ("not provided"). lamp_coverage() and lamp_panel() use it to mark force-months as refreshed and to annotate gaps. The page is fetched once and cached; pass refresh = TRUE to fetch it again.

Usage

lamp_changelog(refresh = FALSE, dir = lamp_cache_dir())

Arguments

Value

A tibble with columns entry_month (the changelog heading, a Date), force_name, force_id, file_type (street, outcomes, stop-and-search, a semicolon-separated combination, or all), action (refresh or not_provided), from, to (Dates, NA when the entry covers all months) and text (the entry as published). Entries that do not follow the refresh or gap patterns are kept with action = NA.

See Also

Other panel: lamp_adjacency(), lamp_adjacency_from_nb(), lamp_boundaries(), lamp_coverage(), lamp_panel(), lamp_population(), lamp_revintage()

Examples

# Needs network access. The page is cached in a temporary directory, and
# the example prints a message instead of failing when offline.
cl <- tryCatch(
  lamp_changelog(dir = tempfile("streetlamp-")),
  streetlamp_error_network = function(e) message(conditionMessage(e))
)
if (!is.null(cl)) cl[cl$force_id == "west-yorkshire", ]

The colours streetlamp draws with

Description

A named vector of the colours every plot() method in the package uses, so that figures drawn outside the package can match them. The series colours were chosen and checked as a set for separation under the common forms of colour vision deficiency; the coverage statuses are drawn from the same set so that a status never impersonates a series.

Usage

lamp_colours()

Value

A named character vector of hex colours. series is the main data colour, contrast the second series where a plot has two, accent the colour for the one mark the reader is meant to compare against (the actual estimate in a placebo distribution), and neutral the fill for marks that carry no identity. The ink, secondary, muted, grid, baseline, surface and shade entries are the text and chrome colours. The remaining entries are named for coverage statuses.

See Also

Other presentation: lamp_map(), lamp_table(), lamp_theme()

Examples

lamp_colours()[c("series", "contrast", "accent")]

The panel contract

Description

Every table streetlamp derives from the archive carries a contract as an attribute: the archive snapshots and versions it was built from, the coverage status of every force-month, the LSOA vintage, the crime scope, the stop-count definition and the treatment definition. lamp_contract() returns it; the contract survives subsetting with [ and dplyr verbs.

Usage

lamp_contract(x)

Arguments

Value

A list of class lamp_contract with elements source, snapshots, versions, coverage, geography, population, crime_scope, stops, treatment and created. Elements not yet applicable are NULL.

Examples

cache <- tempfile("streetlamp-cache-")
zip <- system.file("extdata", "archive", "2026-07.zip", package = "streetlamp")
lamp_archive_register(zip, dir = cache)
crime <- lamp_read_crime(cache, forces = "dyfed-powys")
contract <- lamp_contract(crime)
contract
contract$coverage

Coverage audit: which force-months exist, and can be compared

Description

lamp_coverage() returns the force by month by file-type grid behind a panel or a records table, with one status per cell (submitted, missing, not_read, refreshed, partial_suspected), the record count, the archive version used, any changelog note, and a mismatch flag for months where a stop-and-search file exists without a street file or the reverse. Its plot() method draws the grid as a heat map.

Usage

lamp_coverage(x, changelog = NULL)

lamp_coverage_compare(coverage, period_a, period_b, file_type = "street")

Arguments

Details

lamp_coverage_compare() summarises comparability per force between two periods: the share of months with usable files, and whether every month in both periods was submitted without refreshes, partial submissions or mismatches.

Value

lamp_coverage() returns a tibble of class lamp_coverage with columns force_id, month, file_type, status, n_records, archive, n_versions, versions_differ, note and mismatch.

lamp_coverage_compare() returns a tibble with one row per force: n_months_a, n_usable_a, n_months_b, n_usable_b (months with status submitted or refreshed), n_partial_a, n_partial_b, n_refreshed_a, n_refreshed_b, n_mismatch_a, n_mismatch_b and comparable (TRUE when every month in both periods is submitted, with no mismatch).

See Also

Other panel: lamp_adjacency(), lamp_adjacency_from_nb(), lamp_boundaries(), lamp_changelog(), lamp_panel(), lamp_population(), lamp_revintage()

Examples

panel <- lamp_sample_panel()
cov <- lamp_coverage(panel)
cov
table(cov$file_type, cov$status)
# Dyfed-Powys stopped submitting stop-and-search files in December 2025
cov[cov$mismatch %in% TRUE, c("force_id", "month", "file_type", "status")]
lamp_coverage_compare(cov, c("2024-08", "2025-07"), c("2025-08", "2026-07"),
  file_type = "stop-and-search"
)

Crime type classification used by data.police.uk

Description

data.police.uk assigns every record in the street-level files to one of fourteen categories: anti-social behaviour plus thirteen crime types. This table gives the labels exactly as they appear in the ⁠Crime type⁠ column of the archive CSV files, a snake_case key used for panel columns, the slug used by the police.uk API, and two package-defined groupings.

Usage

lamp_crime_types()

Details

Anti-social behaviour (ASB) is not a crime: ASB incidents carry no Crime ID and no outcome, so is_asb is TRUE and in_crime_total is FALSE for that row and lamp_panel() keeps ASB in its own series.

The category set was verified against the police.uk API and archive files on 2026-09-16. It has been stable since the May 2013 data, when data.police.uk introduced bicycle theft and theft from the person (previously within other theft), split public disorder and weapons into possession of weapons and public order, and renamed violent crime to violence and sexual offences. Files for September 2011 to April 2013 use eleven categories and files for December 2010 to August 2011 only six, with a very broad other crime. since gives the first data month of each current category; lamp_read_crime() maps legacy labels where the mapping is one to one and keeps ⁠Public disorder and weapons⁠ as its own level.

group follows the Home Office offence groups with the six theft categories combined; broad_group collapses further into violent, acquisitive, drugs, damage, other and asb. Both groupings are the package's own and are documented here so that results by offence group can be reproduced.

Value

A tibble with one row per category and columns crime_type, key, api_slug, is_asb, in_crime_total, group and broad_group.

See Also

lamp_outcome_types()

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_deprivation(), lamp_forces(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_sample_panel(), lamp_shocks()

Examples

lamp_crime_types()

Convert an estimate into crimes prevented per thousand searches

Description

Takes an elasticity or a treatment effect and expresses it as the number of crimes prevented per thousand searches at the sample mean, which is the form in which the question is usually asked. The conversion is arithmetic, not evidence: it inherits every assumption of the estimate it is given, and adds more of its own. The assumption chain is returned with the number and printed with it.

Usage

lamp_crimes_prevented(
  estimate,
  panel,
  per_stops = 1000,
  stops_added = NULL,
  outcome_column = NULL,
  n_boot = 2000,
  seed = NULL
)

Arguments

Details

For an elasticity e, a proportional change in searches dS/S changes crime by e * dS/S, so at mean monthly crime C and mean searches S the crimes prevented by an extra per_stops searches are -e * C * per_stops / S. The elasticity estimators use log(1 + stops), so their derivative has denominator S + 1. The count-mean model uses C in the numerator; the legacy log-one-plus outcome approximation uses C + 1. For a treatment effect in log points b, the effect on crime is C * (exp(b) - 1), and it is divided by the searches the intervention actually added, which you must supply as stops_added because a binary treatment does not say how much searching it involved.

Value

A list of class lamp_crimes_prevented with crimes_prevented, conf_low, conf_high, the inputs used, and assumptions, a character vector naming every step from the estimate to the number.

What this number does not include

It counts recorded crime only, so crimes not reported to the police are invisible to it, and any change in recording practice is counted as a change in crime. It assumes the effect is linear in searches over the range considered, and that the average effect applies at the margin, which is the opposite of what diminishing returns implies. It says nothing about the costs of searching, which fall on the people searched.

See Also

Other estimators: lamp_allocation(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(n_areas = 30, n_months = 30, design = "continuous", seed = 1)
el <- lamp_elasticity(sim, "crime_total", lags = 0:1)
lamp_crimes_prevented(el, sim)

Small-area deprivation indices for England and Wales

Description

The English Indices of Deprivation 2025 (IoD2025, 33,755 LSOAs) and the Welsh Index of Multiple Deprivation 2025 (WIMD2025, 1,917 LSOAs), both on 2021 LSOAs, as a compact table for covariates and heterogeneity analysis. The two indices are separate national rankings built from different indicators, so ranks and deciles are comparable only within a country; rank / n_lsoas gives a within-country percentile. Domain deciles cover the seven domains the two indices share; Wales's community safety domain is held in decile_crime and its physical environment domain in decile_environment. England's domain deciles are as published; Wales's are formed within Wales from the published domain ranks, with the same thresholds as its published overall decile.

Usage

lamp_deprivation()

Details

Sources: MHCLG, English indices of deprivation 2025, File 7 (published 30 October 2025, corrected 19 November 2025); Welsh Government, WIMD 2025 index and domain ranks by small area (published 27 November 2025). Both Open Government Licence v3.0; retrieved 2026-09-16.

Value

A tibble with columns lsoa21, country, index, score (IMD score, England only), rank, decile (1 is most deprived), decile_income, decile_employment, decile_education, decile_health, decile_crime, decile_housing, decile_environment and n_lsoas.

See Also

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_crime_types(), lamp_forces(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_sample_panel(), lamp_shocks()

Examples

dep <- lamp_deprivation()
table(dep$country, dep$decile)[, 1:3]

Audit the support and assumptions of an intervention design

Description

Reports missing submissions, pre/post support and detection-sensitive outcomes. A clean data audit does not establish causal identification; intervention assignment, contemporaneous changes and spillovers still require substantive evidence.

Usage

lamp_design_audit(
  panel,
  treatment = NULL,
  outcomes = "crime_total",
  min_pre = 6L,
  min_post = 6L
)

Arguments

Value

A list with issues (check, status, detail), area_support, and summary. Statuses describe data support or required arguments, not certification of causality.

See Also

Other diagnostics: lamp_effect_summary(), lamp_elasticity_robustness(), lamp_placebo(), lamp_pretrends(), lamp_trend_sensitivity()

Examples

sim <- lamp_simulate(design = "event", seed = 1)
lamp_design_audit(sim)

Detect surges in stop and search activity

Description

Flags area-months in which searches rise far above the area's recent level, and returns them as a candidate staggered-adoption table. Two methods are offered: a threshold of k standard deviations above the area's trailing twelve-month mean, and a structural break test from strucchange.

Usage

lamp_detect_surges(
  panel,
  method = c("threshold", "breakpoints"),
  k = 2,
  min_months = 12L,
  min_stops = 3L
)

Arguments

Value

A tibble of class lamp_surges with one row per area: area, adoption_month (the first flagged month, NA if none), n_flagged, trailing_mean, trailing_sd, peak_stops and peak_ratio (peak over trailing mean). Pass it to lamp_treatment() as adoption for a staggered design. The attribute flags holds every flagged area-month.

Endogeneity

A surge detected from the stop series is not an experiment. Police deploy searches where crime has risen, so an area's first surge month is correlated with its recent crime history and with anything else that prompted the deployment. Treating a detected surge as exogenous will mistake the response for the cause. Use this to explore where activity changed and to pick candidate cases for a design with an external source of variation, not as the primary design; and read lamp_allocation() alongside any estimate that uses it.

See Also

Other treatment: lamp_treatment()

Examples

sim <- lamp_simulate(n_areas = 16, n_months = 30, design = "event", seed = 3)
surges <- lamp_detect_surges(sim, min_stops = 5)
head(surges)

Difference-in-differences with staggered adoption

Description

Estimates the effect of an intervention that different areas adopt at different dates, using an estimator that compares each adopting cohort with areas that have not yet adopted rather than with areas already treated. Plain two-way fixed effects does the latter and can return an estimate outside the range of every area's true effect; this function exists so that staggered designs are not analysed that way.

Usage

lamp_did_staggered(
  panel,
  outcome = "crime_total",
  treatment,
  estimator = c("callaway_santanna", "sun_abraham", "imputation"),
  control_group = c("never_treated", "not_yet_treated"),
  cluster = c("area", "force"),
  family = c("ols_log", "ols_ihs", "identity"),
  window = c(-12, 12)
)

Arguments

Value

A lamp_estimate of class lamp_did_staggered. coefficients holds the dynamic (event-study) aggregation with a rel_time column; diagnostics$overall is the single summary effect with its standard error; diagnostics$group_time holds the group-time effects for the Callaway and Sant'Anna backend; and diagnostics$pretrend_p is that backend's pre-test.

Identifying assumption

Parallel trends holds for each adopting cohort against the chosen control group, from the period before that cohort adopts onward, and there is no anticipation: the outcome does not move before adoption in response to it. With control_group = "never_treated" the comparison is against areas that never adopt, which must exist; with "not_yet_treated" it is against areas that adopt later, which uses more data but assumes their later adoption carries no information about their current path.

Backends

• callaway_santanna (default) calls did::att_gt() and did::aggte(), giving one effect per cohort and period and the aggregations of them.

• sun_abraham uses fixest::sunab(), an interaction-weighted estimator fitted in one regression.

• imputation uses the didimputation package, which fits the untreated potential outcome and imputes: efficient, but it needs that package, which is in Suggests.

Counts are modelled on a transformed scale by all three backends, which is why the function warns when a count outcome meets ols_log or ols_ihs: the coefficients are effects on that scale.

References

Callaway, B. and Sant'Anna, P. H. C. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics 225(2), 200-230.

Sun, L. and Abraham, S. (2021). Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics 225(2), 175-199.

Borusyak, K., Jaravel, X. and Spiess, J. (2024). Revisiting event-study designs: robust and efficient estimation. Review of Economic Studies 91(6).

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

# small: most of this example's cost is loading the did package, which
# whichever example reaches it first has to pay
sim <- lamp_simulate(n_areas = 24, n_months = 18, design = "staggered", effect = -0.3, seed = 1)
ad <- attr(sim, "truth")$adoption
tr <- lamp_treatment(sim, "staggered", adoption = ad[!is.na(ad$adoption_month), ])
fit <- lamp_did_staggered(sim, "crime_total", tr)
fit
fit$diagnostics$overall

Weighted displacement quotient

Description

Measures whether crime prevented in a treated area reappeared in the areas around it. The quotient compares the change in a buffer of surrounding areas with the change in control areas, relative to the change in the treated areas, following Bowers and Johnson (2003).

Usage

lamp_displacement_quotient(
  panel,
  outcome = "crime_total",
  treated_areas,
  buffer_areas,
  control_areas,
  pre,
  post,
  n_boot = 1000,
  seed = NULL
)

Arguments

Details

The statistic is WDQ = ((B1/C1) - (B0/C0)) / ((A1/C1) - (A0/C0)), where A, B and C are total crime in the treated, buffer and control areas and the subscripts are the pre and post periods. Read it as:

• negative: crime moved into the buffer, so the gain in the treated area is partly or wholly displacement;

• between 0 and 1: the buffer improved too, a diffusion of benefit smaller than the treated area's gain;

• above 1: the buffer improved more than the treated area.

The quotient is a descriptive ratio, not a causal estimate: it assumes the control areas show what would have happened everywhere, and it is unstable when the treated-area change is close to zero, which the bootstrap interval will show as a very wide range.

Value

A list of class lamp_wdq with wdq, conf_low, conf_high, the underlying totals, success (the treated area's change net of controls) and interpretation.

References

Bowers, K. J. and Johnson, S. D. (2003). Measuring the geographical displacement and diffusion of benefit effects of crime prevention activity. Journal of Quantitative Criminology 19(3), 275-301.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(
  n_areas = 36, n_months = 20, design = "spillover",
  effect = -0.4, spillover_effect = 0.2, seed = 2
)
truth <- attr(sim, "truth")
areas <- truth$lattice$area
lamp_displacement_quotient(
  sim, "crime_total",
  treated_areas = truth$treated_areas[1:6],
  buffer_areas = truth$treated_areas[7:12],
  control_areas = setdiff(areas, truth$treated_areas),
  pre = c("2018-01", "2018-06"), post = c("2018-12", "2019-05"), n_boot = 50
)

Summarise effects using their joint covariance

Description

Computes a linear combination with the full covariance matrix. For staggered estimates, the default is the backend's overall ATT and its own weighting, rather than an equally weighted mean of dynamic estimates.

Usage

lamp_effect_summary(estimate, terms = NULL, weights = NULL, level = 0.95)

Arguments

Value

A one-row tibble with estimate, standard error, interval, confidence level and the aggregation definition. Unsupported joint covariance is an error, rather than an independence approximation.

See Also

Other diagnostics: lamp_design_audit(), lamp_elasticity_robustness(), lamp_placebo(), lamp_pretrends(), lamp_trend_sensitivity()

Examples

sim <- lamp_simulate(design = "continuous", seed = 1)
lamp_effect_summary(lamp_elasticity(sim, lags = 0:1))

Elasticity of crime with respect to stop and search

Description

Estimates how recorded crime responds to the intensity of searching, as a distributed lag of log-one-plus stops with area and month fixed effects. Three methods are offered, differing in how they handle the fact that areas move together.

Usage

lamp_elasticity(
  panel,
  outcome = "crime_total",
  stops = "stops",
  lags = 0:3,
  method = c("fe", "cce_mg", "cce_pooled"),
  cluster = c("area", "force"),
  min_months = 12L,
  family = NULL
)

Arguments

Value

A lamp_estimate of class lamp_elasticity. Coefficients are the lag-by-lag elasticities; diagnostics$long_run is their sum with a standard error, diagnostics$cd_test is Pesaran's test, and diagnostics$area_coefficients holds the per-area slopes for the mean group estimator. The CD test sums over every pair of areas, so above two thousand areas it is computed on a random sample of them, drawn with a fixed seed: cd_test$sampled says whether that happened and cd_test$n_areas_used how many areas went in.

What an elasticity here is and is not

With family = "poisson", a coefficient describes the log conditional mean of the count per unit of log(1 + stops). Its elasticity with respect to stops at intensity S is coefficient * S / (1 + S). With family = "ols_log", it describes ⁠log(1 + observed crime)⁠ instead; this is a different estimand, particularly when counts are small. Neither is automatically the effect of a decision to search more. Police send officers where crime is rising, so the association runs in both directions; lamp_allocation() measures that reverse channel and should be reported alongside. Read the elasticity as a description of the joint movement unless the variation in searching has an argued external source.

Methods

• fe: area and month fixed effects only.

• cce_mg: the common correlated effects mean group estimator. The cross-sectional averages of the outcome and the regressors are added to each area's own time-series regression, and the area coefficients are averaged; the standard error is the spread of those coefficients, which makes no assumption that areas share a slope.

• cce_pooled: one pooled regression with the same averages added, which is more precise if the slope really is common.

Pesaran's CD test is an auxiliary descriptive diagnostic in every case. It uses OLS residuals of log-one-plus counts with area effects only, including when the primary estimator is Poisson, because month dummies would remove the common factor by construction and make the statistic negative whatever the data looked like. A large statistic flags shared movements; it does not by itself prove that clustered intervals are invalid.

References

Pesaran, M. H. (2006). Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 74(4), 967-1012.

Chudik, A. and Pesaran, M. H. (2015). Common correlated effects estimation of heterogeneous dynamic panel data models with weakly exogenous regressors. Journal of Econometrics 188(2), 393-420.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_event_study(), lamp_spillover(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(n_areas = 30, n_months = 36, design = "continuous", seed = 1)
fit <- lamp_elasticity(sim, "crime_total", lags = 0:1)
fit
fit$diagnostics$cd_test$statistic

Evaluate an explicit elasticity specification grid

Description

Runs every requested outcome, lag window and family, retaining failures and optional leave-one-area-out estimates. Register the primary outcome and specification before examining the grid. These are dependent, exploratory checks; selecting a favourable row does not establish a cause. The OLS and Poisson rows target different quantities and should not be interpreted as estimates of one common elasticity.

Usage

lamp_elasticity_robustness(
  panel,
  outcomes = "crime_total",
  lags = list(current = 0L, short = 0:1, distributed = 0:3),
  families = c("poisson", "ols_log"),
  cluster = c("area", "force"),
  leave_one_out = FALSE
)

Arguments

Value

A tibble with every requested specification, omitted area, status, error message, lag-sum estimate and interval, and observed sample sizes.

See Also

Other diagnostics: lamp_design_audit(), lamp_effect_summary(), lamp_placebo(), lamp_pretrends(), lamp_trend_sensitivity()

Examples

sim <- lamp_simulate(design = "continuous", n_areas = 12, seed = 1)
lamp_elasticity_robustness(sim, lags = list(current = 0, short = 0:1))

Event study around a dated intervention

Description

Estimates one coefficient per month relative to the intervention, so that the path of the outcome before and after the event is visible rather than summarised in a single number. The months before the event are the pre-trend test: if the outcome was already moving, the parallel-trends assumption behind any difference-in-differences estimate is in doubt.

Usage

lamp_event_study(
  panel,
  outcome = "crime_total",
  event,
  window = c(-12, 12),
  reference = -1,
  cluster = c("area", "force"),
  family = c("poisson", "negbin", "ols_log", "ols_ihs"),
  controls = NULL,
  staggered_ok = FALSE
)

Arguments

Value

A lamp_estimate of class lamp_event_study. Its coefficients carry a rel_time column, and diagnostics$pretrend_p holds the joint test that every pre-period coefficient is zero.

Identifying assumption

In the absence of the event, treated areas would have followed the same path as controls, net of area and month fixed effects. The pre-period coefficients test a necessary consequence of this, not the assumption itself: passing the test does not establish parallel trends, and with few areas the test has little power to detect a trend that matters. See lamp_pretrends().

Staggered adoption

When areas adopt at different dates, relative-time dummies in a two-way fixed effects regression use already-treated areas as controls and are biased. This function never does that: with more than one adoption date it stops and explains, or, with staggered_ok = TRUE, hands the work to lamp_did_staggered() and says so. The result is then a lamp_did_staggered estimate, not an event study.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_spillover(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(n_areas = 24, n_months = 24, design = "event", effect = -0.25, seed = 4)
truth <- attr(sim, "truth")
# half the areas are treated; the rest are the comparison group
tr <- lamp_treatment(
  sim, "event",
  date = truth$event_date, scope = truth$treated_areas, window = c(-6, 6)
)
es <- lamp_event_study(sim, "crime_total", tr, window = c(-6, 6))
es
plot(es)

Police forces in the data.police.uk archive

Description

The 45 forces whose files appear in the archive: the 43 territorial forces of England and Wales, the Police Service of Northern Ireland and British Transport Police, with the identifier used in archive file names, the force name as published by the police.uk API, and the ONS police force area code (December 2025) where one exists.

Usage

lamp_forces()

Details

British Transport Police has no police force area; its records carry ⁠British Transport Police⁠ in the ⁠Falls within⁠ column rather than the territorial force in whose area the offence occurred. Greater Manchester Police has submitted no data since 2019 and is absent from recent archive snapshots.

Sources: https://data.police.uk/api/forces and the ONS Police Force Areas (December 2025) Names and Codes in the United Kingdom, both retrieved 2026-09-16; Open Government Licence v3.0.

Value

A tibble with columns force_id, name, pfa_code, pfa_name, country (england, wales, northern_ireland or NA) and notes.

See Also

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_crime_types(), lamp_deprivation(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_sample_panel(), lamp_shocks()

Examples

lamp_forces()

List the archive versions of every force-month file

Description

List the archive versions of every force-month file

Usage

lamp_list_versions(snapshot, file_types = NULL)

Arguments

Value

A tibble of class lamp_versions with one row per force, month and file type: n_versions, archives (list-column of archive identifiers, oldest first), crc32 (list-column, aligned with archives), differs (TRUE when the versions do not all share one checksum), usize_min, usize_max, and n_available (versions readable locally).

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zips <- list.files(
  system.file("extdata", "archive", package = "streetlamp"),
  pattern = "zip$", full.names = TRUE
)
for (z in zips) lamp_archive_register(z, dir = cache)
v <- lamp_list_versions(lamp_archive_snapshot(cache))
v[v$n_versions > 1, c("force_id", "month", "file_type", "n_versions", "differs")]

ONS LSOA code lists and the 2011 to 2021 lookup

Description

The Office for National Statistics lookup between 2011 and 2021 Lower layer Super Output Areas (LSOAs) in England and Wales, with the ONS change indicator, the 2022 local authority district, and a flag marking the single 2021 LSOA that the ONS best-fit lookup assigns to each 2011 LSOA. lamp_lsoa_codes() returns the complete code list for either vintage, which is what lamp_lsoa_vintage() matches against.

Usage

lamp_lsoa_lookup()

lamp_lsoa_codes(vintage = c("lsoa21", "lsoa11"))

Arguments

Details

lamp_lsoa_lookup() is the ONS exact-fit lookup (version 3): one row per 2011 to 2021 pair, so a 2011 LSOA that was split appears on several rows and a merged 2021 LSOA appears on several rows. change is the ONS change indicator: U unchanged, S split, M merged, X complex. best_fit is TRUE on the row chosen by the ONS best-fit lookup (version 2), which assigns every 2011 LSOA to exactly one 2021 LSOA; re-vintaging counts with the best_fit rows is deterministic but leaves the 2021 LSOAs that no 2011 LSOA best-fits to without data.

Source: Office for National Statistics licensed under the Open Government Licence v3.0. Retrieved 2026-09-16; see inst/NOTES/data_sources.md.

Value

lamp_lsoa_lookup() returns a tibble with columns lsoa11, lsoa11_name, lsoa21, lsoa21_name, change (factor), lad22, lad22_name and best_fit. lamp_lsoa_codes() returns a sorted character vector of codes.

See Also

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_crime_types(), lamp_deprivation(), lamp_forces(), lamp_outcome_types(), lamp_sample_panel(), lamp_shocks()

Examples

length(lamp_lsoa_codes("lsoa21"))
length(lamp_lsoa_codes("lsoa11"))
table(lamp_lsoa_lookup()$change)

Detect whether LSOA codes are 2011 or 2021 vintage

Description

Matches codes against the bundled ONS code lists (lamp_lsoa_codes()). Codes that exist in only one vintage decide the answer; when a file has none of those (possible for very small forces), the publication month decides, because data.police.uk moved to 2021 codes with the June 2023 data.

Usage

lamp_lsoa_vintage(x, month = NULL)

Arguments

Value

A tibble with one row per file (or one row for a vector) and columns vintage ("lsoa11", "lsoa21", "mixed" or "ambiguous"), basis ("codes" or "month"), n_codes, n_only_2011, n_only_2021, n_both and n_unknown, preceded by archive, file, force_id and month for records.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

lamp_lsoa_vintage(c("E01000001", "E01000002", "E01035000"))
lamp_lsoa_vintage(c("E01000001", "E01000002"), month = "2024-01")

Map a panel column, an area-level estimate or a synthetic control's donors

Description

Draws a choropleth of the areas in a panel. The geometry comes from lamp_boundaries() or lamp_sample_boundaries(); the fill comes from a panel column, from the per-area slopes of a mean-group elasticity, or from the donor weights of a synthetic control.

Usage

lamp_map(x, ...)

## S3 method for class 'lamp_panel'
lamp_map(
  x,
  column = "crime_total",
  boundaries,
  month = NULL,
  statistic = c("mean", "total", "rate"),
  per = 1000,
  border = NA,
  extent = c("panel", "all"),
  ...
)

## S3 method for class 'lamp_elasticity'
lamp_map(x, boundaries, border = NA, extent = c("panel", "all"), ...)

## S3 method for class 'lamp_synth'
lamp_map(x, boundaries, border = NA, extent = c("panel", "all"), ...)

## S3 method for class 'lamp_estimate'
lamp_map(x, ...)

Arguments

Value

A ggplot.

Missing is not zero

An area whose value cannot be computed, because every month of it was a force-month with no file or because it is absent from the panel, is filled with the missing colour from lamp_colours() and given its own legend entry. It is never shaded as a low value. The caption counts those areas, and counts areas present in one of the panel and the boundaries but not the other, so that a map with a hole in it says why.

See Also

Other presentation: lamp_colours(), lamp_table(), lamp_theme()

Examples

panel <- lamp_sample_panel()
boundaries <- lamp_sample_boundaries()
# one local authority, so that the example draws quickly; a whole force is
# a few seconds and a national panel a few minutes
leeds <- boundaries$area[grepl("^Leeds", boundaries$name)]
lamp_map(panel[panel$area %in% leeds, ], "stop_rate", boundaries)

Outcome classification used by data.police.uk

Description

The outcome categories that appear in the ⁠Outcome type⁠ column of the outcomes files and the ⁠Last outcome category⁠ column of the street files, with the police.uk API code and a package-defined grouping into six classes.

Usage

lamp_outcome_types()

Details

The grouping is the package's own:

• charged_or_summonsed: the suspect was charged or summonsed, including every subsequent court outcome supplied by the Ministry of Justice and offences taken into consideration (⁠Suspect charged as part of another case⁠).

• out_of_court: cautions, drugs possession warnings, penalty notices and community (local) resolutions.

• no_suspect: investigation complete with no suspect identified.

• evidential_difficulties: a suspect was identified but the police were unable to prosecute.

• other: action not in the public interest or taken by another body.

• unknown: still under investigation or no status update available.

is_court_outcome marks the categories that come from court records rather than from the police. data.police.uk reports that court outcomes from June 2019 onwards are unavailable, so these categories are sparse after that date and charged_or_summonsed should be read as the charge stage.

The category list was verified against the police.uk API documentation and against archive files on 2026-09-16. outcome_type is the string written in the archive CSV files; where the API documentation names a category differently (⁠Offender given penalty notice⁠ appears there as ⁠Offender given a penalty notice⁠) the API name is kept in api_name and lamp_read_outcomes() accepts either.

Value

A tibble with columns outcome_type, api_name, code, group (factor with the six levels above) and is_court_outcome.

See Also

lamp_crime_types()

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_crime_types(), lamp_deprivation(), lamp_forces(), lamp_lsoa_lookup(), lamp_sample_panel(), lamp_shocks()

Examples

lamp_outcome_types()

Build an area-by-month panel of crime and stops

Description

Turns crime records into a balanced area-by-month panel with one column per crime type, a crime total, an anti-social behaviour series, stop counts, population and a stop rate, and attaches a full panel contract. The panel covers every area in the police force areas of the forces present in crime (or the areas you name) and every month in the records.

Usage

lamp_panel(
  crime,
  outcomes = NULL,
  stops = NULL,
  area = c("lsoa21", "lsoa11", "msoa21", "lad", "pfa"),
  population = NULL,
  covariates = NULL,
  include_asb = FALSE,
  attribution = c("falls_within", "reported_by"),
  areas = NULL,
  adjacency = NULL,
  changelog = NULL
)

Arguments

Details

Records are placed by their LSOA code. Files published with 2011 codes are re-vintaged to 2021 with the bundled ONS lookup (or 2021 to 2011 when area = "lsoa11"), and the contract's geography$revintage element records how many records and areas were unchanged, split, merged or complex. Coarser areas (msoa21, lad, pfa) are reached through lamp_area_lookup().

"Force did not submit" is never a zero: counts are zero-filled only in force-months whose street file was submitted; force-months with no file are NA and carry coverage_status = "missing". Force-months whose record count is suspiciously low are kept but flagged partial_suspected; see lamp_coverage().

Anti-social behaviour is not a crime: it goes to the asb column and is excluded from crime_total unless include_asb = TRUE. The crime type columns are the snake_case keys of lamp_crime_types(); the legacy public_disorder_and_weapons column appears only when records from before May 2013 are present.

Value

A tibble of class lamp_panel with columns area, month, force_id, one column per crime type, crime_total, asb, stops, stops_s60, population, stop_rate, coverage_status, stops_status, any outcome and covariate columns, and a contract retrievable with lamp_contract().

See Also

Other panel: lamp_adjacency(), lamp_adjacency_from_nb(), lamp_boundaries(), lamp_changelog(), lamp_coverage(), lamp_population(), lamp_revintage()

Examples

cache <- tempfile("streetlamp-cache-")
zip <- system.file("extdata", "archive", "2026-07.zip", package = "streetlamp")
lamp_archive_register(zip, dir = cache)
crime <- lamp_read_crime(cache, forces = "dyfed-powys", months = "2026-07")
panel <- lamp_panel(crime, area = "lad")
panel
lamp_contract(panel)$coverage

# The bundled sample is a full two-force LSOA panel over 24 months
lamp_sample_panel()

Placebo tests for a streetlamp estimate

Description

Re-estimates the same specification under conditions in which the true effect is zero, and compares the real estimate with that distribution. An estimate that looks the same when the treatment is moved to the wrong date, the wrong areas, or an outcome the intervention cannot plausibly affect, is measuring something other than the intervention.

Usage

lamp_placebo(
  estimate,
  type = c("time", "space", "outcome"),
  n = 200,
  outcome = "bicycle_theft",
  seed = NULL
)

Arguments

Details

Types:

• time: the event is moved back to each month that leaves a full pre-period, and only pre-period data is used, so nothing real happens at the fake date.

• space: the same number of areas is drawn at random, keeping the real dates, and the treatment is reassigned to them.

• outcome: the same specification is run on a crime type the intervention is not expected to move (bicycle theft by default), which tests for anything that shifts recorded crime generally, such as a recording change.

The reported p value is the share of placebo estimates at least as large in absolute value as the real one. It is a randomisation p value, not a test of a specific null, and with few areas it is coarse.

Value

A list of class lamp_placebo with type, actual (the real estimate), distribution (a tibble of placebo estimates), p_value, n_valid and interpretation.

See Also

Other diagnostics: lamp_design_audit(), lamp_effect_summary(), lamp_elasticity_robustness(), lamp_pretrends(), lamp_trend_sensitivity()

Examples

sim <- lamp_simulate(n_areas = 20, n_months = 24, design = "event", effect = -0.3, seed = 6)
truth <- attr(sim, "truth")
tr <- lamp_treatment(sim, "event", date = truth$event_date, scope = truth$treated_areas)
fit <- lamp_twfe(sim, "crime_total", tr)
lamp_placebo(fit, type = "space", n = 20, seed = 1)

Census 2021 usual resident population by area

Description

Fetches the Census 2021 usual resident population of every 2021 LSOA in England and Wales from the NOMIS API (table TS001, dataset NM_2021_1), caches it, and returns it at the requested area level. Coarser levels are sums of LSOA values through lamp_area_lookup(); because ONS applies cell-key perturbation to small-area counts, these sums differ from the published MSOA and district totals by a few persons. 2011 LSOAs receive the population of the 2021 LSOAs that map to them: unchanged and split LSOAs pass their value to their single parent, and a 2021 LSOA formed by a merger shares its value equally among its parents, so lsoa11 values are approximate.

Usage

lamp_population(
  area = c("lsoa21", "msoa21", "lad", "pfa", "lsoa11"),
  dir = lamp_cache_dir(),
  refresh = FALSE
)

Arguments

Details

Without network access a cached download is returned, and an error of class streetlamp_error_network is raised if there is none.

Value

A tibble with columns area, name (NA for aggregated levels) and population, with attributes area (the level) and source (a list naming the NOMIS dataset, the census, the geography type, the retrieval time and the aggregation rule) that lamp_panel() copies into the contract.

See Also

Other panel: lamp_adjacency(), lamp_adjacency_from_nb(), lamp_boundaries(), lamp_changelog(), lamp_coverage(), lamp_panel(), lamp_revintage()

Examples

# Needs network access. The download is cached in a temporary directory,
# and the example prints a message instead of failing when offline.
pop <- tryCatch(
  lamp_population("pfa", dir = tempfile("streetlamp-")),
  streetlamp_error_network = function(e) message(conditionMessage(e))
)
if (!is.null(pop)) head(pop)

Test and interpret pre-trends

Description

Reports the joint test that every pre-period coefficient of an event study is zero, and answers the question the test cannot: how large a pre-trend could have been present without this test detecting it, and how much bias that undetected pre-trend would put into the post-period estimates.

Usage

lamp_pretrends(estimate, power = c(0.5, 0.8), level = 0.05)

Arguments

Details

The power calculation follows the argument in Roth (2022) and is implemented here rather than taken from another package. Under a linear violation of parallel trends with slope s per month, the expected pre-period coefficients are s times their relative time (measured from the reference month), so the Wald statistic is non-central chi-square with non-centrality ⁠s^2 * t' V^-1 t⁠, where t is that vector of relative times and V the clustered covariance of the pre-period coefficients. Inverting this gives the slope the test detects with a given probability. The bias column shows what that slope would add to each post-period coefficient if it continued, which is the quantity that matters: a test that passes while remaining blind to a trend big enough to explain the result is not reassurance.

Value

A list of class lamp_pretrends with test (statistic, degrees of freedom and p value), coefficients (the pre-period coefficients), power (a tibble of power, slope in log points per month, and bias_mean_post, the bias such a slope would add to the mean post-period coefficient) and interpretation.

References

Roth, J. (2022). Pretest with caution: event-study estimates after testing for parallel trends. American Economic Review: Insights 4(3), 305-322.

See Also

Other diagnostics: lamp_design_audit(), lamp_effect_summary(), lamp_elasticity_robustness(), lamp_placebo(), lamp_trend_sensitivity()

Examples

sim <- lamp_simulate(n_areas = 24, n_months = 24, design = "event", effect = -0.25, seed = 5)
truth <- attr(sim, "truth")
tr <- lamp_treatment(
  sim, "event",
  date = truth$event_date, scope = truth$treated_areas, window = c(-6, 6)
)
es <- lamp_event_study(sim, "crime_total", tr, window = c(-6, 6))
pt <- lamp_pretrends(es)
pt

Read street-level crime records from the archive

Description

Reads the selected street files into one table with a fixed snake_case schema, harmonises crime type labels across the three category sets that data.police.uk has used since December 2010, flags anti-social behaviour, detects the LSOA vintage of every file, and attaches a panel contract.

Usage

lamp_read_crime(
  dir = lamp_cache_dir(),
  versions = NULL,
  months = NULL,
  forces = NULL,
  fetch = FALSE
)

Arguments

Details

Crime types are returned as a factor whose levels are the fourteen current categories plus ⁠Public disorder and weapons⁠, the legacy category that was split in May 2013 and cannot be mapped forward. ⁠Violent crime⁠, the pre-2013 name of ⁠Violence and sexual offences⁠, is renamed. The raw label is kept in crime_type_raw, and the contract's crime_scope$category_sets records which set each file uses (six for December 2010 to August 2011, eleven to April 2013, fourteen since May 2013).

Anti-social behaviour rows have no Crime ID and no outcome; they are kept with is_asb = TRUE so that lamp_panel() can hold them in a separate series. Rows without coordinates or LSOA code are kept as NA.

Value

A tibble of class lamp_records with columns crime_id, month (Date, first of month), reported_by, falls_within, force_id (from the file name), longitude, latitude, location, lsoa_code, lsoa_name, lsoa_vintage, crime_type (factor), crime_type_raw, is_asb, last_outcome_category, context, archive and file, and a contract retrievable with lamp_contract() whose coverage element marks force-months with no file as missing.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zip <- system.file("extdata", "archive", "2026-07.zip", package = "streetlamp")
lamp_archive_register(zip, dir = cache)
crime <- lamp_read_crime(cache, forces = "dyfed-powys")
crime
table(crime$crime_type)
lamp_contract(crime)$coverage

Read outcome records from the archive

Description

Reads the selected outcomes files into one table with a fixed schema, classifies outcomes with lamp_outcome_types(), detects the LSOA vintage of every file, and attaches a panel contract. An outcomes file for a month lists the outcomes recorded in that month, whose crimes may date from earlier months, and a crime can have several outcomes.

Usage

lamp_read_outcomes(
  dir = lamp_cache_dir(),
  versions = NULL,
  months = NULL,
  forces = NULL,
  fetch = FALSE
)

Arguments

Value

A tibble of class lamp_records with columns crime_id, month, reported_by, falls_within, force_id, longitude, latitude, location, lsoa_code, lsoa_name, lsoa_vintage, outcome_type (factor), outcome_type_raw, outcome_group (factor with the six groups of lamp_outcome_types()), archive and file, plus a contract.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zip <- system.file("extdata", "archive", "2026-07.zip", package = "streetlamp")
lamp_archive_register(zip, dir = cache)
outcomes <- lamp_read_outcomes(cache, forces = "dyfed-powys", months = "2026-07")
table(outcomes$outcome_group)

Count stops by area and month

Description

Reads the selected stop-and-search files and counts searches by area and month, with a Section 60 flag. Only the Date, Latitude, Longitude and Legislation columns are read; the package never parses the ethnicity, object, outcome or other fields. Area assignment is by point-in-polygon in British National Grid (EPSG:27700) against boundaries you supply; without boundaries, counts are per force (the force that submitted the file). Stops are dated by the month of the file they were published in: the publisher stores local time as UTC, so a handful of searches made after 23:00 on the first night of a summer-time month carry the previous month's date, and the count of such rows is reported per file as n_outside_month.

Usage

lamp_read_stop_counts(
  dir = lamp_cache_dir(),
  versions = NULL,
  area = "force",
  boundaries = NULL,
  months = NULL,
  forces = NULL,
  fetch = FALSE
)

Arguments

Value

A tibble of class lamp_records with one row per area and month: area, month, force_id (the submitting force), stops, stops_s60 and archive; for area = "force" also stops_no_location (searches without coordinates, which cannot be placed in an area) and stops_no_legislation. The contract's stops element records the origin (streetlamp), the definition (⁠raw count⁠), the area type and per-file diagnostics, and its coverage element marks force-months with no stop-and-search file as missing.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_s60_legislation(), lamp_select_version(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zip <- system.file("extdata", "archive", "2026-07.zip", package = "streetlamp")
lamp_archive_register(zip, dir = cache)
stops <- lamp_read_stop_counts(cache, forces = c("west-yorkshire", "dyfed-powys"))
stops
lamp_contract(stops)$coverage

Build a report on a panel and the estimates made from it

Description

Renders a document that puts a set of estimates next to the things that decide whether they mean anything: the coverage of the panel they came from, the identifying assumption of each estimator, the pre-trend and placebo diagnostics, and the allocation model. The report is deliberately hard to read as a simple verdict, because the underlying question does not have one.

Usage

lamp_report(
  panel,
  estimates,
  file = NULL,
  title = "streetlamp report",
  diagnostics = TRUE,
  figures = NULL,
  quiet = FALSE
)

Arguments

Value

The path to the written file, invisibly.

Examples

sim <- lamp_simulate(n_areas = 20, n_months = 20, design = "event", seed = 1)
truth <- attr(sim, "truth")
tr <- lamp_treatment(sim, "event", date = truth$event_date, scope = truth$treated_areas)
out <- lamp_report(
  sim,
  list("Two-way fixed effects" = lamp_twfe(sim, "crime_total", tr)),
  file = tempfile(fileext = ".md"), quiet = TRUE
)
file.exists(out)

Re-vintage LSOA codes between 2011 and 2021

Description

Converts a vector of LSOA codes from one vintage to the other with the bundled ONS lookup (lamp_lsoa_lookup()) and reports how many areas were unchanged, split, merged or complex. From 2011 to 2021 each code maps to its best-fit 2021 LSOA. From 2021 to 2011 unchanged and split LSOAs map to their single parent; a 2021 LSOA formed by a merger maps to the first of its parents in code order, so lsoa21 panels are the recommended target.

Usage

lamp_revintage(codes, from = c("lsoa11", "lsoa21"), to = c("lsoa21", "lsoa11"))

Arguments

Value

A list with codes (the mapped vector, NA where a code is not in the lookup) and summary (counts of records and distinct areas by change type and the share of areas that were split or merged).

See Also

Other panel: lamp_adjacency(), lamp_adjacency_from_nb(), lamp_boundaries(), lamp_changelog(), lamp_coverage(), lamp_panel(), lamp_population()

Examples

r <- lamp_revintage(c("E01000001", "E01000002", "E01033768"), from = "lsoa11", to = "lsoa21")
r$codes
r$summary$share_split_or_merged

Legislation label for Section 60 searches

Description

The exact string that data.police.uk uses in the Legislation column of stop-and-search files for searches under section 60 of the Criminal Justice and Public Order Act 1994 (verified 2026-09-16).

Usage

lamp_s60_legislation()

Value

A string.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_select_version(), lamp_version_diff()

Examples

lamp_s60_legislation()

The bundled sample panel and boundaries

Description

A ready-made lamp_panel() for West Yorkshire and Dyfed-Powys: every 2021 LSOA in the two police force areas over the 24 most recent complete months in the July 2026 archive snapshot (August 2024 to July 2026), with crime by type, anti-social behaviour, stops and Section 60 stops assigned by location, Census 2021 population, stop rate, and deprivation deciles as covariates. Its contract records the archive versions, coverage, geography and adjacency it was built with. lamp_sample_boundaries() returns the matching ONS generalised boundaries, simplified to 50 m for bundling, as an sf layer in British National Grid.

Usage

lamp_sample_panel()

lamp_sample_boundaries()

Details

The panel is aggregated counts, not records, and is meant for examples, tests and the vignettes. Data: data.police.uk, Office for National Statistics and NOMIS, Open Government Licence v3.0; see inst/extdata/README.md.

Value

lamp_sample_panel() returns a lamp_panel; lamp_sample_boundaries() an sf object with columns area, name and geometry.

See Also

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_crime_types(), lamp_deprivation(), lamp_forces(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_shocks()

Examples

panel <- lamp_sample_panel()
panel
lamp_coverage(panel)
bnd <- lamp_sample_boundaries()
nrow(bnd)

Select one archive version per force-month file

Description

Select one archive version per force-month file

Usage

lamp_select_version(versions, rule = c("latest", "earliest"), prefer = NULL)

Arguments

Value

A tibble of class lamp_selection with one row per force, month and file type: the chosen archive, its crc32 and usize, n_versions, differs, alternatives (list-column of tibbles naming the other archives with their checksums and sizes) and rule.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_version_diff()

Examples

cache <- tempfile("streetlamp-cache-")
zips <- list.files(
  system.file("extdata", "archive", package = "streetlamp"),
  pattern = "zip$", full.names = TRUE
)
for (z in zips) lamp_archive_register(z, dir = cache)
v <- lamp_list_versions(lamp_archive_snapshot(cache))
sel <- lamp_select_version(v, rule = "latest")
sel[, c("force_id", "month", "file_type", "archive", "n_versions")]

National policy shocks affecting stop and search or recorded crime

Description

A curated table of dated national events for use as default intervention definitions in lamp_treatment(): the Best Use of Stop and Search Scheme, the 2019 and later changes to Section 60 authorisation conditions, COVID-19 lockdown periods, changes to crime recording practice, and other verified policy changes. Every row carries the source that establishes its dates; see inst/NOTES/data_sources.md in the package sources for the verification record.

Usage

lamp_shocks()

Value

A tibble with columns name (identifier), label, kind (stop_search_policy, crime_policy, covid, recording_practice or disorder), start and end (Date; end is NA for point events or regimes still in force), scope (national, england, wales or forces), forces (semicolon-separated police.uk force identifiers when scope is forces, otherwise NA), source_url, notes and verified (Date the source was checked).

See Also

Other bundled data: lamp_area_lookup(), lamp_bank_holidays(), lamp_crime_types(), lamp_deprivation(), lamp_forces(), lamp_lsoa_lookup(), lamp_outcome_types(), lamp_sample_panel()

Examples

shocks <- lamp_shocks()
shocks[shocks$kind == "stop_search_policy", c("name", "start", "end")]

Simulate a panel with a known treatment effect

Description

Generates an area-by-month panel from a known data-generating process, for testing estimators and for the simulation evidence in the vignettes. Crime counts are Poisson draws around area and month fixed effects, and the treatment multiplies the mean of the affected crime types by exp(effect), so effect is a log point effect: -0.1 is a 9.5 percent reduction.

Usage

lamp_simulate(
  n_areas = 36L,
  n_months = 24L,
  design = c("staggered", "event", "continuous", "spillover"),
  effect = -0.15,
  spillover_effect = 0.05,
  adoption = NULL,
  missing = 0,
  base_rate = 8,
  population = 1500,
  seed = NULL,
  cohort_effects = NULL,
  dynamic_effect = NULL,
  trend_violation = 0,
  dispersion = Inf,
  missing_mechanism = c("random", "high_outcome"),
  effect_scale = c("log", "additive")
)

Arguments

Details

Designs:

• event: one date; the areas in the treated half are treated from that month onward. Suits lamp_event_study().

• staggered: areas adopt in cohorts spread across the middle of the period, with never-treated areas kept as controls. Plain two-way fixed effects are biased here, which is what lamp_did_staggered() is for.

• continuous: every area has a stop intensity that varies over time, and the outcome responds to log(stops + 1) with elasticity effect.

• spillover: as event, and in addition the mean treatment of an area's first-order lattice neighbours shifts its outcome by spillover_effect, so an estimator that omits the neighbour term is biased.

Areas sit on a square lattice and belong to two forces, so that force-month missingness, clustering and adjacency are all exercisable. A share missing of force-months has no street file: those rows carry NA counts and coverage_status = "missing", exactly as a real panel does. Bicycle theft is never affected by treatment, which makes it the default placebo outcome in lamp_placebo().

Value

A lamp_panel with the usual columns plus treat (the treatment variable) and, for the event and staggered designs, rel_time and cohort. Its contract carries the treatment definition, and attr(x, "truth") holds the true effect, the implied effect on the crime total (effect_crime_total, smaller in size than effect because bicycle theft is unaffected), spillover_effect, design, the affected crime types, the treated areas, the event date, the adoption table and the neighbour list.

See Also

Other simulation: lamp_simulation_truth()

Examples

sim <- lamp_simulate(n_areas = 36, n_months = 24, design = "event", effect = -0.2)
sim
attr(sim, "truth")$effect
table(sim$coverage_status)

Recover exact expected outcomes from a simulated panel

Description

Keeps count-level effects and log mean ratios distinct. This function deliberately does not treat a log mean as the expectation of log counts.

Usage

lamp_simulation_truth(
  panel,
  outcome = c("affected_crime", "crime_total"),
  scale = c("identity", "log_mean")
)

Arguments

Value

A tibble of area-month counterfactual means and exact effects, including cohort, relative time and observed coverage status. For spillover designs the effect includes the simulated neighbour exposure.

See Also

Other simulation: lamp_simulate()

Examples

sim <- lamp_simulate(seed = 1)
head(lamp_simulation_truth(sim))

Own-area and neighbour effects estimated together

Description

Adds the treatment of an area's neighbours to the specification, so that the effect on the area itself and the effect on its neighbours are estimated in one model. This is what separates a reduction in crime from its displacement next door: if searching an area pushes offending into the surrounding areas, the own-area coefficient is negative and the neighbour coefficient positive, and the net effect is their sum.

Usage

lamp_spillover(
  panel,
  outcome = "crime_total",
  treatment,
  adjacency = NULL,
  rings = 1:2,
  cluster = c("area", "force"),
  family = c("poisson", "negbin", "ols_log", "ols_ihs"),
  controls = NULL
)

Arguments

Value

A lamp_estimate of class lamp_spillover, whose coefficients hold the own-area effect and one neighbour effect per ring, and whose diagnostics$net gives the implied net effect (own plus neighbours) with a standard error from the full covariance matrix.

Identifying assumption

Parallel trends as in two-way fixed effects, and in addition that the treatment reaches an area only through its own treatment and that of the rings included. If spillovers extend beyond the last ring, the areas used as controls are themselves affected, and every coefficient is biased toward zero.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_synth(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(
  n_areas = 49, n_months = 20, design = "spillover",
  effect = -0.3, spillover_effect = 0.15, seed = 1
)
adj <- lamp_adjacency_from_nb(attr(sim, "truth")$nb, attr(sim, "truth")$lattice$area)
truth <- attr(sim, "truth")
tr <- lamp_treatment(sim, "event", date = truth$event_date, scope = truth$treated_areas)
fit <- lamp_spillover(sim, "crime_total", tr, adjacency = adj, rings = 1)
fit

Synthetic control for a single treated area or force

Description

Builds a weighted average of untreated donor units whose pre-period path matches the treated unit's, and reads the gap between them afterwards as the effect. Inference is by permutation: the same procedure is run pretending each donor was treated, and the treated unit's post-period fit is compared with that distribution.

Usage

lamp_synth(
  panel,
  outcome = "crime_total",
  treated_unit,
  treatment_start,
  donors = NULL,
  method = c("internal", "synth", "gsynth"),
  placebo = TRUE,
  min_pre = 6L
)

Arguments

Value

A list of class lamp_synth with weights, path (a tibble of month, observed, synthetic and gap), effect (mean post-period gap), rmspe_pre, rmspe_post, rmspe_ratio, p_value (the share of donors with a ratio at least as large), placebos and assumption.

Identifying assumption

The weighted donors reproduce what the treated unit would have done without the intervention. This is credible only when the pre-period fit is close over a long window, no donor was itself affected by the intervention, and nothing else happened to the treated unit at the same time. A poor pre-period fit invalidates the comparison, which is why the pre-period root mean squared error is reported first.

References

Abadie, A., Diamond, A. and Hainmueller, J. (2010). Synthetic control methods for comparative case studies. Journal of the American Statistical Association 105(490), 493-505.

Xu, Y. (2017). Generalized synthetic control method. Political Analysis 25(1), 57-76.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_twfe(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(n_areas = 20, n_months = 30, design = "event", effect = -0.4, seed = 1)
truth <- attr(sim, "truth")
sc <- lamp_synth(
  sim, "crime_total",
  treated_unit = truth$treated_areas[1],
  treatment_start = truth$event_date,
  donors = setdiff(unique(sim$area), truth$treated_areas),
  placebo = FALSE
)
sc

A presentation-ready table from a streetlamp object

Description

Formats an estimate, a diagnostic or a coverage audit as a tibble of character columns with readable headers, rounded figures, confidence intervals in brackets and p values as ⁠<0.001⁠ where they are that small. The result is meant to be passed to knitr::kable() or a similar table renderer; the unrounded numbers remain available from tidy() and from the object itself.

Usage

lamp_table(x, ...)

## S3 method for class 'lamp_estimate'
lamp_table(x, digits = 3, ...)

## S3 method for class 'lamp_event_study'
lamp_table(x, digits = 3, ...)

## S3 method for class 'lamp_did_staggered'
lamp_table(x, digits = 3, ...)

## S3 method for class 'lamp_pretrends'
lamp_table(x, digits = 3, ...)

## S3 method for class 'lamp_synth'
lamp_table(x, digits = 3, ...)

## S3 method for class 'lamp_coverage'
lamp_table(x, ...)

Arguments

Value

A tibble of character columns. For estimates: the term, the estimate, its standard error, the 95 percent confidence interval and the p value. Event studies and staggered fits show months relative to the event in place of the term. For a pre-trend diagnostic: the power table. For a synthetic control: donor weights. For a coverage audit: counts of force-months by file type and status.

See Also

Other presentation: lamp_colours(), lamp_map(), lamp_theme()

Examples

sim <- lamp_simulate(n_areas = 24, n_months = 24, design = "event", effect = -0.2, seed = 2)
truth <- attr(sim, "truth")
tr <- lamp_treatment(sim, "event", date = truth$event_date, scope = truth$treated_areas)
lamp_table(lamp_twfe(sim, "crime_total", tr))
lamp_table(lamp_coverage(sim))

The ggplot2 theme streetlamp plots use

Description

A quiet theme built on ggplot2::theme_minimal(): recessive hairline grid, no minor grid or tick marks, left-aligned titles, muted axis text, and the legend above the plot. Every plot() method in the package applies it, and it can be added to any other ggplot.

Usage

lamp_theme(base_size = 11)

Arguments

Value

A ggplot2 theme object.

See Also

Other presentation: lamp_colours(), lamp_map(), lamp_table()

Examples

library(ggplot2)
ggplot(mtcars, aes(wt, mpg)) +
  geom_point(colour = lamp_colours()[["series"]]) +
  labs(title = "Weight and fuel economy") +
  lamp_theme()

Define the treatment

Description

Turns a panel into a treatment definition that the estimators can use: a continuous stop intensity, a binary intervention in named areas and months, staggered adoption dates by area, or a dated event applied to every area. The result records how the treatment was built, is stored in the contract of every estimate made with it, and is printed by the estimators alongside their identifying assumption.

Usage

lamp_treatment(
  panel,
  type = c("continuous", "binary", "staggered", "event"),
  measure = c("stop_rate", "stops", "stops_per_100_crimes"),
  transform = c("identity", "log", "ihs"),
  areas = NULL,
  window = NULL,
  adoption = NULL,
  event = NULL,
  date = NULL,
  scope = NULL
)

Arguments

Details

Types:

• continuous: measure is "stops", "stop_rate" (per 1,000 residents, the default) or "stops_per_100_crimes" (searches per 100 crimes recorded in the area's previous twelve months, which removes the mechanical dependence on population). transform is "identity", "log" (log of one plus the measure) or "ihs" (inverse hyperbolic sine, which handles zeros).

• binary: areas are treated during window, a pair of months, and untreated outside it; every other area is a control.

• staggered: adoption is a table of area and adoption_month. Areas absent from it are never treated. rel_time (months since adoption) and cohort are added.

• event: a dated shock applied to every area in scope, from lamp_shocks() by event name or from date, with an event window in months around it. rel_time is months since the event.

A continuous treatment is a dose, not an experiment: the allocation of searches to areas responds to crime, so estimates from it describe an association unless the variation is argued to be exogenous. See lamp_allocation().

Value

A lamp_treatment object: a list with type, column (the treatment variable's name), data (area, month, the treatment column and, where applicable, rel_time and cohort), details, n_treated_areas, n_control_areas and created.

See Also

Other treatment: lamp_detect_surges()

Examples

sim <- lamp_simulate(n_areas = 16, n_months = 18, design = "event", seed = 1)

# a dated event, twelve months either side
lamp_treatment(sim, "event", date = "2018-10", window = c(-6, 6))

# stop intensity as a continuous dose
lamp_treatment(sim, "continuous", measure = "stop_rate", transform = "ihs")

# a surge in named areas
lamp_treatment(sim, "binary", areas = unique(sim$area)[1:4], window = c("2018-06", "2018-12"))

Sensitivity to a specified differential linear trend

Description

Adjusts an event-study average for user-specified untreated trend slopes. Each slope is an assumption in the model's outcome units per month, not a trend estimated and treated as known. Intervals reflect sampling uncertainty conditional on each slope. This is a transparent linear bias calculation, not a general robust confidence procedure or a test of parallel trends.

Usage

lamp_trend_sensitivity(
  estimate,
  slopes,
  periods = NULL,
  reference = -1L,
  level = 0.95
)

Arguments

Value

A tibble with the assumed slope, implied bias and adjusted estimate and interval for each scenario.

See Also

Other diagnostics: lamp_design_audit(), lamp_effect_summary(), lamp_elasticity_robustness(), lamp_placebo(), lamp_pretrends()

Examples

sim <- lamp_simulate(design = "event", seed = 1)
truth <- attr(sim, "truth")
tr <- lamp_treatment(sim, "event", date = truth$event_date, scope = truth$treated_areas)
fit <- lamp_event_study(sim, "crime_total", tr, window = c(-4, 4))
lamp_trend_sensitivity(fit, c(-0.01, 0, 0.01))

Two-way fixed effects estimate of the effect of stops on crime

Description

Fits the outcome on the treatment with area and month fixed effects, so that the effect is identified from changes within an area over time, net of anything common to all areas in a month. Standard errors are clustered at the level given in cluster.

Usage

lamp_twfe(
  panel,
  outcome = "crime_total",
  treatment,
  controls = NULL,
  cluster = c("area", "force"),
  family = c("poisson", "negbin", "ols_log", "ols_ihs"),
  force_month = FALSE
)

Arguments

Value

A lamp_estimate of class lamp_twfe: coefficients with clustered standard errors and 95 percent intervals, the identifying assumption, the exclusion table and a diagnostics list holding the dispersion statistic and, when the panel carries adjacency, Moran's I of the residuals.

Identifying assumption

Treated and control areas would have followed parallel paths in the outcome had the treatment not changed, once area and month fixed effects are removed. With a continuous treatment this also requires that the intensity of searching is unrelated to what else was changing in the area, which police allocation makes doubtful: see lamp_allocation().

When this estimator is the wrong one

With staggered adoption, two-way fixed effects compares later-treated areas against already-treated ones, and the estimate can lie outside the range of every area's true effect. Use lamp_did_staggered() instead; this function warns when the treatment has more than one adoption date.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_synth(), tidy.lamp_estimate()

Examples

sim <- lamp_simulate(n_areas = 24, n_months = 24, design = "event", effect = -0.2, seed = 2)
truth <- attr(sim, "truth")
tr <- lamp_treatment(sim, "event", date = truth$event_date, scope = truth$treated_areas)
fit <- lamp_twfe(sim, "crime_total", tr)
fit
tidy(fit)

Compare the archive versions of one force-month file

Description

Reads every available version of a file and reports what differs: the number of rows, and for street and outcomes files the crime identifiers present in one version but not another. Street files routinely differ between snapshots only in their ⁠Last outcome category⁠ column, which is why row and identifier counts, not checksums, decide whether a force re-supplied a month.

Usage

lamp_version_diff(
  snapshot,
  force,
  month,
  file_type = c("street", "outcomes", "stop-and-search"),
  fetch = FALSE
)

Arguments

Value

A tibble with one row per version: archive, n_rows, n_ids (distinct crime identifiers; NA for stop-and-search files), ids_not_in_others (identifiers absent from every other version) and crc32. The attribute outcome_changes gives, for street files, the number of shared crimes whose last outcome category differs between the oldest and newest versions.

See Also

Other ingest: lamp_archive_download(), lamp_archive_index(), lamp_archive_register(), lamp_archive_snapshot(), lamp_list_versions(), lamp_lsoa_vintage(), lamp_read_crime(), lamp_read_outcomes(), lamp_read_stop_counts(), lamp_s60_legislation(), lamp_select_version()

Examples

cache <- tempfile("streetlamp-cache-")
zips <- list.files(
  system.file("extdata", "archive", package = "streetlamp"),
  pattern = "zip$", full.names = TRUE
)
for (z in zips) lamp_archive_register(z, dir = cache)
snap <- lamp_archive_snapshot(cache)
lamp_version_diff(snap, "west-yorkshire", "2026-05", "outcomes")

Objects exported from other packages

Description

These objects are imported from other packages. Follow the links below to see their documentation.

generics

tidy

Tidy a streetlamp estimate

Description

Tidy a streetlamp estimate

Usage

## S3 method for class 'lamp_estimate'
tidy(x, ...)

Arguments

Value

A tibble with columns term, estimate, std_error, statistic, p_value, conf_low and conf_high.

See Also

Other estimators: lamp_allocation(), lamp_crimes_prevented(), lamp_did_staggered(), lamp_displacement_quotient(), lamp_elasticity(), lamp_event_study(), lamp_spillover(), lamp_synth(), lamp_twfe()