Si però mi permetto di insistere. Perdonami per questo.
Io ho messo così:
\small
\begin{longtable}[]{lll}
\caption{\normalsize Variables descriptions and sources}\label{datasources}\\
\toprule
\textbf{Variable}&\textbf{Description}&\textbf{Source}\\
\toprule
\endfirsthead
\caption[]{\normalsize Variables descriptions and sources (continued)}\\
\toprule
\textbf{Variable}&\textbf{Description}&\textbf{Source}\\
\toprule
\endhead
\bottomrule
\multicolumn{3}{l}{\strut\emph{Continues on next page}} \\
\endfoot
\bottomrule
\endlastfoot
\begin{tabularx}{\columnwidth}{XXX}
nimp of $i$ from $j$&nominal CIF imports of country $i$ from country $j$, in current millions of US\$&\textcite{olympic}, originally from \emph{DOTS}-IMF\\
\midrule
nexp from $i$ to $j$&nominal FOB exports from country $i$ to country $j$, in current millions of US\$&\textcite{olympic}, originally from \emph{DOTS}-IMF\\
\midrule
defl&US-GDP deflator, 2005=1000&\emph{IFS}-IMF(2011), accessed via beyond 20/20 WDS on courtesy of University of Cambridge\\
\midrule
ln of real imports&$ln(nimp/defl)\times100$&own calculations\\
\midrule
ln of real exports&$ln(nexp/defl)\times100$&own calculations\\
\midrule
Polity IV index($polity$)*&recoded “polity2” indicator, $1=most autocratic$, $21=most democratic$&\textcite{qog}, constructed by \textcite{polity4}\\
\midrule
Freedom House index($fh$)*&average of recoded “Political Rights” and “Civil Liberties”, $1=least free$, $21=most free$&\textcite{qog}, constructed by \textcite{fh}\\
\midrule
Chga index ($chga$)*&dummy variable which is unity for democratic states&\textcite{qog}, originally constructed by \textcite{prew}, updated by \textcite{cheibub}\\
\midrule
Freedom House/Polity IV ($fh/polity$)*&ratio of the recoded Freedom House index and Polity IV index, $0=least democratic$, $21=most democratic$&\textcite{qog}\\
\midrule
Freedom House and Polity IV ($fhpolity$)*& average of recoded $polity$ and $fh$, $1=autocracy$, $21=democracy$&own construction\\
\midrule
institutional similarity&difference in absolute value between \inst of country $i$ and \inst of country $j$, lower values=higher similarity&own construction\\
\midrule
ln GDP*& natural logarithm of real GDP (constant US\$)&\textcite{olympic}, originally from Penn World Tables Mark 6.2 and World Development Indicators\\
\midrule
landlocked&dummy variable taking the value of 1 if at least one trading partner is landlocked&\textcite{Cepii}\\
\midrule
ln distw& natural logarithm of the distance between the most populated cities of the trading partners&\textcite{Cepii}\\
\midrule
common language&dummy variable for both trading partners sharing an official language&\textcite{Cepii}\\
\midrule
common border&dummy variable for trading partners sharing a land border&\textcite{Cepii}\\
\midrule
colony&dummy variable for trading partners ever in a colonial relationship&\textcite{Cepii}\\
\midrule
common colonizer&dummy variable for common colonizer&\textcite{Cepii}\\
\midrule
colony post 1945&dummy variable for pairs in colonial relationship post 1945&\textcite{Cepii}\\
\midrule
same country&dummy variable for countries that are or were the same nation&\textcite{Cepii}\\
\midrule
common legal origin&dummy variable for pairs sharing a common legal origin&\textcite{Cepii}\\
\midrule
regional trade agreement&dummy variable for pairs that are member of the same regional trade agreement at time $t$&\textcite{olympic}, originally from WTO website \url{http://www.wto.org/english/tratop_e/region_e/eif_e.xls}\\
\midrule
common currency&dummy variable for pairs that use the same currency at time $t$&\textcite{rosecurrency}, updated by \textcite{jose}\\
\midrule
GATT/WTO membership&dummy variable for GATT/WTO members at time $t$&\textcite{rosecurrency}, originally from WTO(2007)\\
\midrule
ln GDP per capita*&ln (GDP/population), constant US\$&\textcite{rosecurrency}, originally from Penn World Tables Mark 6.2 and World Development Indicators\\
\midrule
ln population&natural logarithm of total population&\textcite{rosecurrency}, originally from Penn World Tables Mark 6.2 and World Development Indicators\\
\addlinespace
\multicolumn{3}{@{}l@{}}{\parbox{\columnwidth}{
* for these variables $i$ and $j$ indicate the values for the
importing and exporting country, respectively}}
\end{tabularx}
\end{longtable}
ed il risultato è che longtable non mi spezza la tabella, e \textbf{Variable}&\textbf{Description}&\textbf{Source}
non sono in corrispondenza delle colonne! Come mai, secondo te?