% Auto-generated
\begin{table}[htbp]
  \centering
  \begin{threeparttable}
    \caption{Robustness: Blend Performance Across Subsamples}
    \label{tab:robustness}
    \begin{tabular}{@{}lcccc@{}}
      \toprule
      Subsample & $N$ & $R^2_{\text{OOS}}$ & Median AE & Within $\pm$15\% \\
      \midrule
      \textit{Full sample} & 19{,}123 & 0.779 & 14.0\% & 53\% \\
      \quad Price tercile 1 (low) & 6{,}375 & n/a$^{\dagger}$ & 15.6\% & 49\% \\
      \quad Price tercile 2 (mid) & 6{,}412 & n/a$^{\dagger}$ & 11.0\% & 64\% \\
      \quad Price tercile 3 (high) & 6{,}336 & n/a$^{\dagger}$ & 16.4\% & 47\% \\
      \quad Realtor.ca & 15{,}356 & 0.802 & 14.0\% & 53\% \\
      \quad DuProprio & 3{,}767 & 0.579 & 14.0\% & 54\% \\
      \quad Measured area & 9{,}446 & 0.773 & 13.5\% & 55\% \\
      \quad Imputed area & 9{,}677 & 0.783 & 14.4\% & 52\% \\
      \bottomrule
    \end{tabular}
    \begin{tablenotes}[flushleft]
      \footnotesize
      \item \textit{Notes:} Out-of-sample performance of the text\,+\,vision blend
      (Table~\ref{tab:models}, Panel~B) computed on subsamples of the same 5-fold
      cross-validated predictions. Price terciles are defined on the actual list price.
      ``Measured area'' are listings with a reported or room-reconstructed living area
      (priced by Model~A); ``Imputed area'' listings are priced by the area-free model.
      $^{\dagger}$Within a single price tercile the range of $\ln P$ is heavily
      restricted, so $R^2$---which normalizes by within-group variance---is
      uninformative; the median absolute error is the comparable accuracy metric across
      terciles.
    \end{tablenotes}
  \end{threeparttable}
\end{table}