It has to do with the way people absorb printed information or, more accurately, how readers of Western texts are acculturated into this.
Tables are textual devices and are, to some or other approximate degree, structured to be read in the same left->right, top->bottom order (*). To this degree, they therefore do not differ in important information processing ways from the body text. Normal text (not, e.g., poetry or other intentionally disruptive forms) is ordinarily structured in a "pyramidal" form: general and introductory concepts are best placed at the beginning, with increasingly fleshed-out details placed later on. The reading eye that saccades across text and the brain that controls it has learnt beginning at the age of two or whatever to most efficiently gather (Western) textual information in this way. By our age, we cannot not process text this way.
Graphics, on the other hand, are iconic. The eye does not scan pictures for information in remotely the same way (in fact entirely different cerebral systems are in charge of gathering and interpreting textual and graphical information - you can knock out one part of the brain and impair one system without noticeably affecting the performance of the other; see, e.g., almost any popular writing by Oliver Sacks). It would be almost impossible to prevent the normal human brain from scanning and interpreting, no matter how briefly, a picture on the page before forcing it to reading the caption (no matter where the caption is placed because by that time the attentional mechanisms in the part of the brain that is now switched on is directing the eye to saccade in non-linear sequences).
So, to answer your question, good typesetting sets out to provide readers the most productive (rate of information processing for expended effort) reading experience that it can. For text: the caption summary is placed above the details to fit in with linear eye saccades and the pyramid principle of text interpretation; for graphics: the graphic is placed first to fit in with quite different attentional control mechanisms, non-linear eye saccading and non-linear information processing.
(*) Factor in acculturation. Good typesetting for R->L or bottom->top reading orders will follow these general principles (for biologically human species (**)), leading however to different outcomes.
(**) The point is that it's our brain biology that has ultimately led to the development of these rules. Here's something to think about: what are the optimal typesetting rules for non-human species, e.g., maybe sapient machines, or echo-locating information gatherers like bats or porpoises. Where would they like to "see" their captions? :))
I just ran into the same issue. (It's really annoying when trying to produce print-quality preprint versions of the paper.) Poking around in pnastwo.cls, I found the following fix:
Find the first line in pnastwo.cls which says
\ifx\@captype\xtable
Change this to
\ifx\@captype\table
Alternatively, if you don't want to touch the class file, you can
put the following into the preamble of your document:
% Fix wierd behavior which prevents table captions from appearing for
% tables in the body of the article
\makeatletter
\long\def\@makecaption#1#2{%
\ifx\@captype\table
\let\currtabcaption\relax
\gdef\currtabcaption{
\tabnumfont\relax #1. \tabtextfont\relax#2\par
\vskip\belowcaptionskip
}
\else
\vskip\abovecaptionskip
\sbox\@tempboxa{\fignumfont#1.\figtextfont\hskip.5em\relax #2}%
\ifdim \wd\@tempboxa >\hsize
\fignumfont\relax #1.\figtextfont\hskip.5em\relax#2\par
\else
\global \@minipagefalse
\hb@xt@\hsize{\hfil\box\@tempboxa\hfil}%
\fi
\fi
}
\makeatother
One additional comment: once this bug is fixed, you'll almost certainly run into another issue which you'd want fixed if you use the PNAS two-column class to produce production-quality output: Unless you define floats with the h "here" placement, it loses the labels, so all cross-referencing of figures and tables breaks. Here is the fix, for direct pasting into the document preamble. You can also patch up the macro \DonormalEndcol pnastwo.cls, it's the same mistake 6 times over...
% And another fix. PNAS class loses the label of floats unless they
% were defined with the [h] option (so not really floats at all). It
% all comes down to wrong scope in the following routine which pushes
% out the floats onto the page. This is the fixed version:
\makeatletter
\def\DonormalEndcol{%
%% top float ==>
\ifx\toporbotfloat\xtopfloat%
%% figure ==>
\ifcaptypefig%
\expandafter\gdef\csname topfloat\the\figandtabnumber\endcsname{%
\vbox{\vskip\PushOneColTopFig%
\unvbox\csname figandtabbox\the\loopnum\endcsname%
\vskip\abovefigcaptionskip%
\csname caption\the\loopnum\endcsname%
\csname letteredcaption\the\loopnum\endcsname%
\csname continuedcaption\the\loopnum\endcsname%
\csname letteredcontcaption\the\loopnum\endcsname
\ifredefining%
\csname label\the\loopnum\endcsname%
\expandafter\gdef\csname topfloat\the\loopnum\endcsname{}\fi}%
\vskip\intextfloatskip%%
\vskip-4pt %% probably an artifact of topskip??
}%
\else%
%% plate ==>
\ifcaptypeplate%
\expandafter\gdef\csname topfloat\the\figandtabnumber\endcsname{%
\vbox{\vskip\PushOneColTopFig%
\unvbox\csname figandtabbox\the\loopnum\endcsname
\vskip\abovefigcaptionskip
\csname caption\the\loopnum\endcsname
\csname letteredcaption\the\loopnum\endcsname
\csname continuedcaption\the\loopnum\endcsname
\csname letteredcontcaption\the\loopnum\endcsname
\ifredefining
\csname label\the\loopnum\endcsname
\expandafter\gdef\csname topfloat\the\loopnum\endcsname{}\fi}
\vskip\intextfloatskip %%
\vskip-4pt %% probably an artifact of topskip??
}%
\else% table ==>
\expandafter\gdef\csname topfloat\the\figandtabnumber\endcsname{%
\vbox{\vskip\PushOneColTopTab %%
\csname caption\the\loopnum\endcsname
\csname letteredcaption\the\loopnum\endcsname
\csname continuedcaption\the\loopnum\endcsname
\csname letteredcontcaption\the\loopnum\endcsname
\vskip\captionskip
\unvbox\csname figandtabbox\the\loopnum\endcsname
\ifredefining
\csname label\the\loopnum\endcsname
\expandafter\gdef\csname topfloat\the\loopnum\endcsname{}\fi
}\vskip\intextfloatskip %% why don't we need this?
\vskip-10pt}
\fi\fi%
%
\else% bottom float
%
\ifcaptypefig
\expandafter\gdef\csname botfloat\the\figandtabnumber\endcsname{%
\vskip\intextfloatskip
\vbox{\unvbox\csname figandtabbox\the\loopnum\endcsname
\vskip\abovefigcaptionskip
\csname caption\the\loopnum\endcsname
\csname letteredcaption\the\loopnum\endcsname%
\csname continuedcaption\the\loopnum\endcsname%
\csname letteredcontcaption\the\loopnum\endcsname%
\vskip\PushOneColBotFig%%
\ifredefining%
\csname label\the\loopnum\endcsname
\expandafter\gdef\csname botfloat\the\loopnum\endcsname{}\fi}}%
\else
\ifcaptypeplate
\expandafter\gdef\csname botfloat\the\figandtabnumber\endcsname{%
\vskip\intextfloatskip
\vbox{\unvbox\csname figandtabbox\the\loopnum\endcsname
\vskip\abovefigcaptionskip
\csname caption\the\loopnum\endcsname
\csname letteredcaption\the\loopnum\endcsname%
\csname continuedcaption\the\loopnum\endcsname%
\csname letteredcontcaption\the\loopnum\endcsname%
\vskip\PushOneColBotFig%%
\ifredefining%
\csname label\the\loopnum\endcsname
\expandafter\gdef\csname botfloat\the\loopnum\endcsname{}\fi}}%
\else% TABLE
\expandafter\gdef\csname botfloat\the\figandtabnumber\endcsname{%
\vskip\intextfloatskip
\vbox{\csname caption\the\loopnum\endcsname
\csname letteredcaption\the\loopnum\endcsname
\csname continuedcaption\the\loopnum\endcsname
\csname letteredcontcaption\the\loopnum\endcsname%
\vskip.5\intextfloatskip
\unvbox\csname figandtabbox\the\loopnum\endcsname%
\vskip\PushOneColBotTab
\ifredefining%
\csname label\the\loopnum\endcsname
\expandafter\gdef\csname botfloat\the\loopnum\endcsname{}\fi}}%
\fi\fi\fi}
\makeatother
Best Answer
table
is a floating environment. TeX has special rules for placing floats so that they cause minimal disruption of the surrounding text.You can give LaTeX hints of how to place your floats. For example, to make LaTeX try to place it where you inserted it, specify the option
h
(for “here”):Note that this won’t always work because LaTeX still tries to make the text look nice. Your hint is only that: a hint, not a definite command.
You can force LateX to obey your command to place a floating environment HERE by loading the package
float
and specifying the optionH
(capital letter!):But notice that this can lead to an ugly layout.