Consider the following example and its output with the standard plain
style that applies sentence casing to titles of @article
s, but not to @book
s.
\documentclass[british]{article}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{babel}
\usepackage{filecontents}
\begin{filecontents}{\jobname.bib}
@article{uthor:one,
author = {Anne Uthor},
title = {The Alligator Population in {T}allahassee},
year = {1981},
}
@article{uthor:two,
author = {Anne Uthor},
title = {The Alligator Population in {Tallahassee}},
year = {1982},
}
@article{uthor:three,
author = {Anne Uthor},
title = {{The Alligator Population in Tallahassee}},
year = {1983},
}
@book{elk:one,
author = {Anne Elk},
title = {The Brontosaurus Population in {T}admarton},
year = {1971},
}
@book{elk:two,
author = {Anne Elk},
title = {The Brontosaurus Population in {Tadmarton}},
year = {1972},
}
@book{elk:three,
author = {Anne Elk},
title = {{The Brontosaurus Population in Tadmarton}},
year = {1973},
}
\end{filecontents}
\begin{document}
\nocite{*}
\bibliographystyle{plain}
\bibliography{\jobname}
\end{document}

If you look really closely, you'll see that the {T}a
approach (...:one
) leaves a slightly larger space between "T" and "a" than the other two. That is because in pdfLaTeX the group destroys the kerning between letters (it doesn't destroy kerning in LuaTeX, though). Hence, I advise against method 2.
Furthermore, the completely protected title field (...:three
) is never converted to sentence case. Now one might argue that this WYSIWYG approach is a useful way to suppress sentence casing if it is not desired, but I think that makes the entire point of case changing a bit of a farce. If you don't want sentence casing at all, you are better off picking a style that does not apply it rather than suppressing it so cruelly once and for all. I find completely protected title fields particuarly terrible if they are already in sentence case: title = {{A title with words in it}},
, one sometimes sees that in auto-generated .bib
files and it makes me shudder (see also Software-generated bibliographic entries: common errors and other mistakes to check before use for more things that make me shudder).
That leaves the option of protecting only those words that need protection in their entirety.
title = {Rise and Fall of {Word}}
I would recommend to input titles in title casing and to protect those words that need protection. Everything else can be taken care of by the style.
See also What is the proper casing to use when storing titles in the bibliography database? and Implementation of "Title Case" in Bibtex.