\begin{table}[!ht]
\caption{Spillover ATTs: full vs matched sample (Callaway--Sant'Anna)}\label{table:referee_matched}
\centering
\begin{tabular}{lcccc}\hline\hline
Neighbour outcome (log) & ATT full (\%) & N full & ATT matched (\%) & N matched \\ \hline
Population &   0.45 ( 0.31) &       23,585 &   0.45 ( 0.33) &       21,032 \\
Current expenditure &   1.08 ( 0.66) &       23,581 &   1.30 ( 0.67) &       21,028 \\
Education expenditure &   3.71 ( 1.01) &       23,465 &   4.15 ( 1.03) &       20,928 \\
Public-sector employment &   2.67 ( 1.29) &       23,108 &   2.11 ( 1.25) &       20,671 \\
Private-sector employment &  -1.05 ( 1.53) &       23,541 &  -0.24 ( 1.55) &       20,997 \\
Number of firms &  -1.56 ( 1.00) &       23,541 &  -0.85 ( 1.03) &       20,997 \\
Firms (services) &  -1.50 ( 1.05) &       23,541 &  -0.87 ( 1.07) &       20,997 \\
Federal transfers (ex-FPM) &   4.45 ( 2.98) &       23,522 &   4.25 ( 3.03) &       20,982 \\
\hline\hline
\end{tabular}\par
\begin{tablenotes}\footnotesize
\item Overall (simple-weighted) Callaway--Sant'Anna ATTs in percent (log coefficients $\times 100$), influence-function standard errors clustered at the neighbour level in parentheses. \emph{Full}: all treated and comparison neighbours (the headline sample). \emph{Matched}: treated neighbours and their 1:1 nearest-neighbour propensity-score match on the four baseline covariates of Table~\ref{table:referee_matchbal} ($780$ of $886$ treated neighbours find a match on common support). Because matching changes the cross-section of neighbours, it also changes the realized cohort composition and the not-yet-treated comparison units; the matched column is therefore a matched-sample estimand rather than the full-sample estimand re-weighted, and I report it as a check on whether baseline differences between treated and comparison neighbours drive the results. The concordance of signs and the broadly similar magnitudes across the two columns indicate that they do not.
\end{tablenotes}\end{table}