Character State Choice - We Have a Long Way to Go

[Mickey Mortimer <Mickey_Mortimer111@msn.com>]

As I reported last time, I'm in the process of checking every published data
matrix associated with a computer-run parsimony analysis of coelurosaurs.
I'm recording each character in my list, and noting who else has used it.
This is so that once I'm done with this phase of my analysis, I can claim to
have tested all characters previously used, and can give them the correct
attribution.  While doing this, several problems with standard choices for
character states have become apparent.

There are far too many instances where characters are defined as a feature
being smaller or larger than some other feature.  Some examples are "naris
larger than antorbital fossa" and "manual phalanx II-2 shorter than II-1".
In actuality, of course, taxa show a continuously varying range of
proportions for any such character, and gaps made of proportions not filled
by any taxon will fill up as we discover new specimens.  Unfortunately,
given the limitations of PAUP, we need to break up such a range of
proportions into a few states.  This is problematic for a few reasons.  If
we use too few states, we'll lose potentially useful information.  This
happens a lot when people use only two states.  Like "proximodorsal ischial
process - absent/present".  We all recall how much flack ABSRDists get for
placing confuciusornithids and enantiornithines together based on this, when
Unenlagia, Sinornithosaurus and other more basal paravians have it too.  But
those of confuciusornithids and enantiornithines are much larger, and by
keeping the state limited to absent vs. present, we're losing that
information.  If we use too many states, minor factors like distortion,
measuring errors and individual variation will play an important role.  The
Theropod Working Group claimed to exclude such characters because of these
reasons, and said "because the sample sizes of these taxa are far too small
for statistical analysis to identify gap between states, using these
characters would require arbitrarily drawn distinctions among states that
would severely compromise the results".  One still finds examples in their
character list though, such as "scapula longer than humerus" or "ischium
two-thirds or less pubis length".  Such worries about the objectivity of
states are valid, however.  We all know the classic divisions, but why
should we divide ischiopubic ratios at the 2/3 or 3/4 mark?  "Simple" points
of division like ">50%" or "<33%" look better to us than ones like ">57%" or
"<138%", but that's just a consequence of our number system, it should have
no bearing on phylogenetic considerations.  And we must be aware of the
features we use as a size reference.  Xu et al. (1999) had a character
relating ulnar to metatarsal length, but what's the rationale?  The two
elements have little relation to each other, and a difference in their
ratios can be as easily explained by a change in ulnar size OR a change in
metatarsal size.  Taxa with elongate metatarsi could clade with those that
have shortened forearms.  I realized this while measuring the classic bird
analysis character comparing naris and antorbital fenestra length.
Proceratosaurus has a huge naris, but a large antorbital fenestra as well,
so got a low ratio.  Sinornithomimus has a small naris, but a very short
antorbital fenestra, so got a large ratio.  I've chosen orbital+jugal height
for cranial ratios and metacarpal II length for manual ones, but there's
still bias here.  The highly reduced metacarpal II of mononykines should
show that much.  A related problem is analyzing the length of a feature with
multiple components, whose change in size could be due to any or all of its
parts.  Is the first digit short because its metacarpal is short, its
phalanx, or both?  And why is a digit shortened by a reduced metacarpal
homologously transformed compared to one shortened by a reduced phalanx?
And if you include both "metacarpal length" and "digit length" as characters
in your analysis, they will be correlated and weight that manual variation
compared to other characters.  Then there's those characters whose states
are defined with gaps between them, like "radius 30-50% of humeral length
(0); 70-100% (1)".  What about those taxa who fall in between the two states
(which usually end up being a lot of them).  Choosing character states is
really a difficult job, and it's best to have all your measurements
available before you do it.  This is one reason it's taking me so long.
Making an analysis like mine is more than just plugging in everyone elses'
characters and checking each taxon for them.  I can't claim my methods are
completely objective, as I don't see any way to make them such currently,
but they're hopefully as good as any other analysis.

As an example of what I've described above, here's my measurements for the
metacarpal I/II ratio in coelurosaurs, and all the equivalent characters
from analyses done from 1986-1999.  How would YOU define the states?

Gauthier 1986 character 61 - manus gracile and elongate, especially digits
two and three and their metacarpals, and metacarpal I only one-third the
length of metacarpal II.
= Holtz 1994 character 124 - metacarpal I one third the length of metacarpal
II or shorter.
= Perez-Moreno et al., 1994 character 19 - length of mcI greater than half
the length of mcII.
= Russell and Dong 1994 character metacarpal I at least half length of
metacarpal II (0); only one-third lengtrh of metacarpal II (1).
= Novas 1996 character 13 - digit I proportions - transverse dimension
subequal to digit II, and longitudinally shorter than digit II (0); digit I
larger than the remaining digits of the hands (1).
= Novas 1996 character 47 - digit I proportions - digit I ends at level of
mid-length of phalanx 2, digit II (0); digit I ends at level of mid-length
of phalanx I, digit II (1).
Martin 1997 character 24 - metacarpal I greatly enlarged.
= Chiappe et al. 1998 character 56 - Alular metacarpal (I)/major metacarpal
(II) length ratio smaller than or equivalent to 0.30: absent (0), present
(1).
= Chiappe et al. 1998 character 58 - Alular digit (I) long, exceeding the
distal end of the major metacarpal (0), or short, not surpassing this
metacarpal (1).
= Chiappe et al. 1998 character 59 - Alular digit (I) large, robust, and
dorsoventrally compressed: absent (0), present (1).
= Forster et al. 1998 character 62 - metacarpal I greater than (0), or less
than (1) one-third the length of MC II. (Tarsitano and Hecht, 1980;
Gauthier, 1986)
= Harris 1998 character 98 - ratio of metacarpal I length : metacarpal II
length - >1/3 (0); <1/3 (1). (Holtz, 1994)
= Sereno 1999 character 136 - Metacarpal I, length: less than 40% (0),
approximately 50% (1), or 60% or more, of metacarpal II length.
= Sereno 1999 character 80 - Manual digit I, length: 50% (0), or 30% (1), of
length of digit II.
= Xu et al. 1999 character 64 - Metacarpal I at least half length of
metacarpal II (0) or only one-third length of metacarpal II (1) (Russell and
Dong, 1993).

Allosaurus 59-64
Acrocanthosaurus 53 (Currie and Carpenter, 2000)
Gorgosaurus 49-67 (Lambe, 1917; Matthew and Brown, 1923)
?Albertosaurus ~50 (Holtz, 2001)
Tyrannosaurus 63 (Brochu, 2003)
Nqwebasaurus 63 (de Klerk et al., 2000)
Sinosauropteryx 41-49 (Currie and Chen, 2001)
Huaxiagnathus 48 (Hwang et al., 2004) 19 mm / 40 mm
NGMC 2124 ~83 (pers. obs.)
Compsognathus 42 (Ostrom, 1972; Gauthier and Gishlick, 2000)
Scipionyx 37 (del Sasso and Signore, 1998)
Pelecanimimus 81 (Perez-Moreno et al., 1994)
Harpymimus 60 (Barsbold and Osmolska, 1990)
Sinornithomimus 75 (Kobayashi and Lu, 2003)
Archaeornithomimus 86 (Smith and Galton, 1990)
Ornithomimidae 85-107 (Parks, 1933; Sternberg, 1933; Osmolska et al., 1972;
Nicholls and Russell, 1981)
Anserimimus 96 (Barsbold, 1988)
Deinocheirus 93 (Osmolska and Roniewicz, 1969)
Mononykinae 125-137 (Perle et al., 1994)
Beipiaosaurus ~40 (Xu et al., 1999)
Alxasaurus 50 (Russell and Dong, 1993)
Erliansaurus 49 (Xu et al., 2002)
Therizinosaurus 51 (Barsbold, 1979)
Caudipteryx 39 (pers. obs.) 39 (Zhou et al., 2000)
Oviraptoridae 44-73 (Barsbold et al., 1990; Dong and Currie, 1996)
Heyuannia ~40 (Lu, 2002)
Protarchaeopteryx 42 (pers. obs.)
Yixianosaurus 40 (Xu and Wang, 2003)
?Sinovenator <50 (Clark et al., 2002)
IGM 100/44 38 (Barsbold et al., 1987)
Sinornithoides 30 (Currie and Dong, 1993)
Deinonychus 46-52 (Ostrom, 1969)
Velociraptor 38 (Norell and Makovicky, 1999)
Bambiraptor 35-36 (Burnham, 2004)
Graciliraptor 32 (Xu and Wang, 2004)
Sinornithosaurus 34 (Xu et al., 1999)
NGMC 91 32 (pers. obs.)
Cryptovolans ~31 (Czerkas et al., 2002)
Microraptor <25 (Hwang et al., 2002)
Microraptor gui 28 (Xu et al., 2003)
Scansoriopterygidae 38 (Czerkas and Yuan, 2002)
Archaeopteryx 28-31 (Wellnhofer, 1974; 1993)
Wellnhoferia ~27 (Wellnhofer, 1992)
Jeholornis ~25 (Zhang and Zhou, 2002)
Shenzhouraptor 22 (Ji et al., 2003)
Jixiangornis 25 (Ji et al., 2002)
Sapeornis 21-25 (Zhang and Zhou, 2003)
Omnivoropteryx ~20 (Czerkas and Ji, 2002)
Confuciusornis ~43 (Chiappe et al., 1999)
Changchengornis 45 (Chiappe et al., 1999)
"Proornis" ~50 (pers. obs.)
Protopteryx ~23 (Zhang and Zhou, 2000)
Longipteryx 26 (Zhang and Zhou, 2001)
Eocathayornis ~24 (Zhou, 2002)
Jibeinia 22 (Hou, 1997)
Spanish nestling ~14 (SVP, 2002)
Concornis ~19 (Sanz et al., 1995)
Eoalulavis <15 (Sanz et al., 2002)
Noguerornis ~19 (Chiappe and Lacasa-Ruiz, 2002)
Liaoxiornis ~16 (Chiappe, 2002)
GMV-2158 25 (Chiappe, 2002)
GMV-2159 <27 (Chiappe, 2002)
Eoenantiornis ~32 (Hou et al., 1999)
Sinornis 28 (Sereno et al., 2002)
Longchengornis ~17 (Hou, 1997)
Songlingornis <27 (Hou, 1997)
Enantiornis ~15 (Chiappe and Walker, 2002)
Neuquenornis ~17 (Chiappe and Calvo, 1994)
Yanornis 26 (Zhang and Zhou, 2001)
Yixianornis 24 (Zhang and Zhou, 2001)
Ambiortus ~17 (Kurochkin, 1999)
Apsaravis <36 (Clarke and Norell, 2002)
Ornithurae ~23 (Clarke, 2002)

Mickey Mortimer