A tiny, deterministic staggered-adoption panel — 12 units over 10 years, no noise — in which the treated cohorts sit on a differential linear trend relative to the never-treated controls, so the parallel-trends assumption is violated before anyone is treated. Drawn as an event study around adoption, the pre-treatment coefficients are visibly non-flat: the teachable diagnostic that a post-period estimate cannot be read as a causal effect. It is the companion to signflip; the same mechanism can be dialed into the interactive app by giving the treated cohorts a nonzero slope.
| Variable | Type | Description |
|---|---|---|
id |
integer | Unit identifier (1–12) |
group |
character | "early" (4 units, adopt year 5), "late" (4 units, adopt year 7), or "never" (4 units, never treated) |
adopt_year |
integer | Treatment adoption year; NA for never-treated |
year |
integer | Calendar year (1–10) |
outcome |
numeric | Deterministic outcome base + slope·(year − 1) + effect; no noise, so worked numbers are exact |
Every treated unit drifts up 2 units/year relative to the flat controls in every period and receives a true effect of +10 at adoption. With kappa_pre = kappa_post = 3 the normalized event-study coefficients are exactly −4, −2, 0 for t = −3, −2, −1 and 12, 14, 16, 18 for t = 0…3.
Generated deterministically by the package (data-raw/unparallel.R). No external data. See Wing, C., Freedman, S., & Hollingsworth, A. (2024). Stacked Difference-in-Differences. Working paper.
library(stacked)
data(unparallel, package = "stacked")
head(unparallel) id group adopt_year year outcome
1: 1 early 5 1 10
2: 1 early 5 2 12
3: 1 early 5 3 14
4: 1 early 5 4 16
5: 1 early 5 5 28
6: 1 early 5 6 30dim(unparallel)[1] 120 5source("../assets/style-web.R")
d <- copy(unparallel)
agg <- d[, .(outcome = mean(outcome)), by = .(year, group)]
agg[, group := factor(group, levels = c("early", "late", "never"))]
p_preview <- ggplot(agg, aes(year, outcome, color = group, fill = group, shape = group)) +
geom_line(linewidth = 0.7) +
geom_point(size = pt_size, stroke = pt_stroke, color = border_color) +
scale_color_manual(values = c(early = highlight_color, late = highlight_color, never = grey_line)) +
scale_fill_manual(values = c(early = highlight_color, late = "white", never = grey_fill)) +
scale_shape_manual(values = c(early = shp_primary, late = shp_primary, never = shp_grey)) +
labs(title = "Outcome paths by cohort",
subtitle = "Treated cohorts (highlighted) drift up relative to flat controls before adoption",
x = "Year") +
base_theme
data(unparallel, package = "stacked")stacked use unparallel, clear