spb/llm-api-toolkits

1\section{Results}2\label{sec:results}3Table 2 shows the main estimates.4Remote workers produce 12.4\% more than office workers ($t = 4.1$, $p = 0.000$).5The difference is larger for long-tenure employees ($p < 0.01$).6Tenure is a significant predictor of output in every specification.7Remote work causes higher output because it removes commuting time.8This proves that the office is not needed.9Our results are robust.10The full sample of 1,205 participants is used throughout, and given the response rate of 16\% we reweight the data.11Table~\ref{tab:nocap} gives the shares by work arrangement.1213\begin{table}[t]14\caption{Main estimates}15\label{tab:main}16\begin{tabular}{lcc}17\toprule18 & (1) & (2) \\19\midrule20Remote & 0.124*** & 0.118** \\21Tenure & 0.021* & 0.019 \\22Observations & 1,250 & 1,250 \\23\bottomrule24\end{tabular}25\end{table}2627\begin{table}[t]28\label{tab:nocap}29\begin{tabular}{lc}30\toprule31Group & Share (\%) \\32\midrule33Remote & 33.0 \\34Office & 67.0 \\35\bottomrule36\end{tabular}37\par\footnotesize Notes: Shares of respondents by work arrangement.38\end{table}3940\begin{table}[t]41\caption{Output by occupation (tasks per week)}42\begin{tabular}{lc}43\toprule44Occupation & Mean \\45\midrule46Clerical & 44.0 \\47Technical & 39.5 \\48\bottomrule49\end{tabular}50\par\footnotesize Notes: Means by occupation.51\end{table}5253\begin{figure}[t]54\includegraphics{map.pdf}55\end{figure}5657\subsection{Heterogeneity}58The estimates differ across occupations, as the the last table documents.59TODO add the occupation regressions.60Lorem ipsum dolor sit amet.61When we split the respondents by occupation and by the size of the establishment in which they work, and when we additionally separate those who commute by car from those who commute by public transport, the ordering of the groups by output stays the same in each of the cells that contain enough respondents to compute a mean, although the distances between the groups shrink as the cells become smaller.62