\begin{table}[!ht]
\caption{Covariate balance before and after matching (normalised differences)}\label{table:referee_matchbal}
\centering
\begin{tabular}{lcc}\hline\hline
Baseline covariate & nd, full sample & nd, matched sample \\ \hline
Population (log) &  0.099 &  0.020 \\
Public-sector employment (log) &  0.072 &  0.005 \\
Private-sector employment (log) &  0.128 &  0.015 \\
Number of firms (log) &  0.115 &  0.013 \\
\hline Matched neighbours (T / C) & & 780 / 780 \\ \hline\hline
\end{tabular}\par
\begin{tablenotes}\footnotesize
\item Imbens--Rubin normalised differences $(\bar x_T-\bar x_C)/\sqrt{(s_T^2+s_C^2)/2}$ on the pre-event, calendar-adjusted baseline covariates. \emph{Full sample}: all treated and comparison neighbours. \emph{Matched sample}: treated neighbours and their 1:1 nearest-neighbour propensity-score match (logit pscore on the four covariates, no replacement, common support, caliper $0.10$). Matching is on the baseline only; cohort timing and the not-yet-treated comparison logic of the Callaway--Sant'Anna estimator are unchanged.
\end{tablenotes}\end{table}