Did Proposition 99 cut California's cigarette consumption?
In January 1989, California's Proposition 99 raised the state cigarette tax by 25 cents per pack and funded a comprehensive anti-smoking education campaign. The honest counterfactual question — what would California's cigarette sales have been without the policy? — cannot be answered by rewinding history. The synthetic control method (SCM) sidesteps this by building a weighted recipe of other US states whose pre-1989 smoking trajectories looked indistinguishable from California's, and using that synthetic stand-in as the missing counterfactual.
This app reproduces the post's three headline findings interactively. In four tabs you will see how a sparse five-donor recipe (Utah, Nevada, Montana, Colorado, Connecticut) is built, watch the actual–synthetic gap open after 1989 and peak at −26.4 packs per capita in 1999, and rank California against placebo runs on the other 38 states.
The headline finding in one picture
The synthetic California is built almost entirely from five of thirty-eight donor states. The remaining 33 states receive exactly zero weight. This is what "sparse, convex combination" looks like when you plot the weights.
Utah, Nevada, Montana, Colorado, and Connecticut are the five states whose pre-1989 smoking trajectories and demographics most closely matched California's. A reader new to US tobacco-control history gets a one-line summary: Synthetic California ≈ 33% Utah + 24% Nevada + 20% Montana + 16% Colorado + 7% Connecticut.
Donor Recipe
The full 38-state donor pool. See which states were considered, and which got non-zero weight. Compare predictor balance side-by-side and inspect the V-weights.
The Gap
Actual vs synthetic California, 1970–2000. Toggle the gap-only view to isolate the post-1989 divergence from the levels.
Placebo Ranking
The full in-space placebo distribution. Trim by pre-treatment fit and see California rank 1 of 20 (pseudo p = 0.05).
Glossary (open a card if a term is unfamiliar)
Synthetic control method (SCM)
Donor pool
Unit weights W
Predictor weights V
Pre-treatment fit (RMSE, R²)
ATT (treatment effect)
In-space placebo
MSPE ratio (post / pre)
The donor recipe — who builds synthetic California?
The optimization searches the 38-state donor pool for the convex combination whose pre-treatment predictors look most like California. The result is a sparse five-state solution. Below: the optimised weights for all 38 donor states (sorted), plus the predictor balance that explains why these five were picked.
Optimal donor weights
Steel blue = dominant donor (Utah). Orange = secondary donors (Nevada, Montana, Colorado, Connecticut). All other 33 states are dimmed because their optimised weights are exactly zero.
Predictor balance — why these five states win
For each of the 7 matched predictors, this table compares the treated California value, the synthetic California value (from the optimised W), and the simple average across the full donor pool.
Look at cigsale(1988): treated 90.1 vs synthetic 91.7
(sample mean: 113.8). California's pre-policy cigarette sales were
already well below the national average — synthetic California reproduces
that with only 1.7% bias, while the simple donor-pool average is off by
26.3%. Look at cigsale(1975): treated 127.1 vs synthetic
127.1 (sample mean: 136.9). The match on a key pre-treatment outcome
year is essentially perfect — that is why the post-1989 divergence is
informative.
What to look for
- The five-donor sparsity is a feature, not a bug. When only a few donors closely resemble the treated unit, the optimiser correctly concentrates weight on them. The other 33 get exactly zero — no contribution. This is structurally different from a difference-in-differences design that gives equal weight to every control.
- The match is excellent on outcome-relevant predictors (all seven predictor biases ≤ 2.2%, five of them ≤ 0.3%). Compare this to the simple donor-pool average, which shows bias of 26.3% on cigsale(1988) and 14.9% on cigsale(1980).
- The V-weights tell you which predictors matter. Age 15–24 (V = 0.546) and cigsale(1975) (V = 0.422) together carry 97% of the predictor-weight mass. Log income receives essentially zero weight — the outer loop discovered that lnincome was not informative for distinguishing California from the donor pool on the outcome.
Actual California vs synthetic California, 1970–2000
These are the real numbers from the post's analysis.do run.
Pre-1989 (grey shaded), the two paths should be indistinguishable — that
is the point of the construction. Post-1989, any divergence is the
estimated causal effect of Proposition 99, expressed in packs per capita.
View
What to look for
- Pre-1989 the two lines are nearly indistinguishable. Both fall from about 123 to 90 packs across the 19 pre-treatment years. RMSE = 1.76 packs and R² = 0.974 — the synthetic is doing what it was constructed to do.
- Starting at 1989 the lines diverge. The gap opens at −7.6 packs in 1989 and deepens monotonically through the 1990s. By 1999 the deficit reaches its largest single-year value of −26.4 packs per capita. This corresponds to roughly a 38% reduction in cigarette sales relative to the no-policy counterfactual.
- The effect compounds rather than fades. Unlike short-run shocks, the divergence widens over the entire post-period — consistent with cumulative behavioural change, anti-smoking campaigns reinforcing each year, and declining social acceptability of smoking.
In-space placebo distribution — is California's gap unusual?
The same algorithm is re-run with each of the other 38 states as the "treated" unit. If Proposition 99 had no causal effect, then California's post/pre MSPE ratio should be unremarkable in this distribution. The chart below shows where it actually sits.
Filter
How to read this
- Toggle to "Full" view to see the entire 39-state distribution. California's MSPE ratio of 123.5 is the highest of any state — its post/pre ratio is 54% larger than Georgia (80.0), the runner-up. The probability of obtaining a ratio this extreme by chance, drawing one state from the full pool, is 1/39 ≈ 0.026.
- Switch back to "Trimmed." The
cut(2)filter excludes 19 states whose pre-treatment MSPE is more than twice California's. This removes states like Maine and New Hampshire, where high pre-fit error means even a small post-period gap divides into a large ratio (mechanical, not informative). Among the 20 states with comparable pre-fit, California again ranks 1st, giving pseudo p = 1/20 = 0.050. - The honest takeaway from this tab: the two p-values (0.026 unfiltered, 0.050 trimmed) bracket the conventional 5% threshold. The unfiltered p is below 5%; the trimmed p sits exactly at it. Both, together with the visual evidence in Tab 3, support the conclusion that Proposition 99 had a real, statistically detectable causal effect on California's cigarette consumption.
Connecting back to the previous tabs
The placebo evidence (Tab 4) and the donor-weight structure (Tab 2) should be read together. The five states that compose synthetic California (Utah, Nevada, Montana, Colorado, Connecticut) are not the placebos that rank near the top — Georgia, Virginia, Missouri, Texas, and Oklahoma are. Those high-ranking placebos are states whose own synthetic-control fits happen to have large post-1988 gaps for reasons unrelated to tobacco policy (cross-border shopping, smuggling, demographic shifts). California's gap stands out relative to this null distribution — which is exactly the hypothesis test the in-space placebo is designed to run.
The post also runs an in-time placebo (fake treatment at 1985) and leave-one-out robustness (excluding each weighted donor one at a time). Both confirm the baseline: the in-time placebo finds small effects of −3 to −9 packs in the fake-treatment window 1985–1988, growing sharply to −14 to −26 packs only after the real 1989 onset; the LOO analysis shows the year-2000 effect ranges from −23.5 to −28.4 packs across the five LOO specifications, never approaching zero.