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

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

panel

A lamp_panel from lamp_panel().

outcome

The outcome column, for example "crime_total" or a crime type key from lamp_crime_types().

treatment

A lamp_treatment() object, or the name of a panel column.

controls

Optional panel columns to include as covariates.

cluster

"area" (default) or "force".

family

"poisson" (default, a Poisson pseudo-likelihood suited to counts), "negbin", "ols_log" or "ols_ihs".

force_month

Add force-by-month fixed effects, which absorb anything that moved a whole force in a month (a recording change, a force-wide operation) at the cost of identifying only from within-force variation.

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
#> 
#> ── streetlamp estimate: lamp_twfe 
#> Outcome: crime_total; family: Poisson pseudo-likelihood on counts
#> Treatment: event; clustered by area
#> Identifying assumption: Parallel trends: treated and control areas would have
#> moved together in the outcome, net of area and month fixed effects.
#> Sample: 576 area-months in 24 areas; 0 rows dropped for coverage.
#> Dispersion: 0.972
#>   Term             Estimate  Std. error            95% CI       p
#>   ───────────────  ────────  ──────────  ────────────────  ──────
#>   Treated × after    −0.276      0.0572  [−0.388, −0.164]  <0.001
tidy(fit)
#> # A tibble: 1 × 7
#>   term  estimate std_error statistic    p_value conf_low conf_high
#>   <chr>    <dbl>     <dbl>     <dbl>      <dbl>    <dbl>     <dbl>
#> 1 treat   -0.276    0.0572     -4.82 0.00000143   -0.388    -0.164