_
COMPREHENIIVE
vvvvOfvvvv
[)ICTI()~Al?"
MATHEMATICS
Chief Editor & Compiler:
~ Roger Thompson
~
ABHISHEK
All rights reserved. No part of this book may be reproduced in any
form, electronically or otherwise, in print, photoprint, micro film or
by any other means without written permission from the publisher.
ISBN 9788182473416
Copyright Publisher
Revised Edition 2010
Published by
ABHISHEK PUBLICATIONS,
S.C.O. 5759, Sector 17C,
CHANDIGARH1600 17 (India)
Ph.2707562,FaxOI722704668
Email: abhpub@yahoo.com
Preface
.2by2 table .
1. this is a twoway table ; A~
where the numbers of levels
of the ro·.v and columnclassi
.
I
T 45'
abacus
~ a Japanese counting device and
: calculator.
~ • abelian group
I a group in which the binary
rIO;'..
f
.'1...
°',;'"
.._'1 ~.if..... . •. .if. ...... (~, '=)
Wi'Z" w;;g,
• 454590 triangle
an isoscoles right triangle
========!!!!!!!!*
!!!!!6=" sbseisss IIIJlinetnlnsj'tmnllritm II
?peratio~ is commutative, that ~ • additive identity property
IS,. ab= bafor all elements a abd ; the sum .pf any number and
b m the group. : zero is the original number·
• abscissa ~ zero is the identity element of
the xcoordinate of a point in a ~ addition.
2dimensional coordinate sys ; • adjacent angles
tern . . .g
; two ht and nonzero
nonstraJ.
• absolute value : angles that have a common side
the positive value for a real ~ in the interior of the angle
n~ber, disregarding the sign. ; f?rmed by the nons:ommon
Wntten 1x I. For example, : SIdes
131 =3, 141 =4, and 101 =0. ~ lnIeriorangl ....nlhe
....... side oflhe InirIlIversaI A IntedonI on the ......
~~~~~~~~~~==~~~M
llilleeimlie elJllllhon lilleeimlie numb:.. =========~7
Mfine transformations of the ~ was applied to landmark
plane take squares into paral ; based shape .
lelograms and take circles into
ellipses of the same shape. M I 'g
fine transformations of a 3di I :or
mensional space take cubes
into parallelopipeds (sheared '.  !r<7H:~,
bricks) and spheres into ellip I
Y 'II
L "8;
.1 ' . '
_:l____ ~
soids all of the same shape. ,, '
~.::~~ . : .t,. ....
Similar results are produced
in higher dimensional spaces . I x,. x"
Equivalent to "uniform trans
formation" . As far as form is ~  algebraic equation
concerned' (that is, ignoring I an equation of the form
translation and rotation), any ~ f( x) = 0 where f is a polyno
affine transformation can be : mial.
diagrammed as a pure strain I
I
Wq = q. eq
same but now the sum of the q = (ra  i) . ax + q.m
squares of the axes is un st
changing. Still ignoring scale I
tlq = !6f.
n
(that is, as far as shape is con and
cerned), any affine transfor
mation can be also dia I
6= (~)!
grammed as a pure shear tak  I
ing a square into a parallelo : _ algebraic number
gram of unchanged base seg ~ a number that is the root of
ment and height. This dia ; an algebraic polynomial. For
gram of shear came into : example, (sqrt 2) is an alge
morphometrics via an applica ~ braic number because it is a
tion to principal components ~ solution of the equation x2 =2.
analysis somewhat before it
.MiJt'nes~====== II.
,!!!!8~~~~~~~~~~. tdgorithm I RltemateinterWrangles II
~ terion level (often 0.05). The
{~J.. l. .'1· ... {~. Ti.~. rz}' 11,
1.19ttbrUcl'·Z11'
I the transversal.
 alpha
also known as size or typel er I  alternate interior angles
ror. This is the probability that, if two parallel lines are cut by a
according to some null hypoth
esis, a statistical test will gener I y
ate a falsepositive error : af
firming a nonnull pattern by
chance. Conventional method I
;"';';'z
ology for statistical testing is, in
advance of undertaking the
test, to set a nominal alpha cri I
9
*=================
transversal, alternate interior ~ or the line containing the oppo
angles are inside the parallel ; site side, or the length of the
lines and on opposite sides of : altitude segment.
the transversal. ~ • altitude of a conic solid
• alternative hypothesis ~ the length of a segment whose
in hypothesis testing, a null hy ; endpoints are the vertex and a
pothesis (typically that there is : point on the plane of the base
no effect) is compared with an ~ that is perpendicular to the
alternative hypothesis (typically ~ plane of the base
that there is an effect, or that I • altitude of a cylindric solid
there is an effect of a particular
~ the distance be,tween the planes
sign). For example, in evaluat
: of the bases
ing whether a new cancer rem
edy works, the null hypothesis ~ • altitude of a trapezoid
typically would be that the rem I the distance between the bases
edy does not work, while the ~ of a trapeziod
alternative hypothesis would be :I • altitude of a triangle.
that the remedy does work. : the perpendicular segment
When the data are sufficiently I from a vertex to the line con
improbable under the assump ; taining the opposite side of a
tion that the null hypothesis is : triangle
true, the null hypothesis is re I
jected in favour of the alterna : • ambiguous
tive hypothesis. (This does not ~ not stable; changing
imply that the data are probable I • amicable numbers
under the assumption that the ; two numbers are said to be
alternative hypothesis is true, : amicable if each is equal to the
nor that the null hypothesis is ~ sum of the proper divisors of
false, nor that the alternative ; the other.
hypothesis is true.
; • analyse
• altitude : to break down into parts and
(of a triangle) A line segment ~ explain or demonstrate the
drawn from a vertu that is per I logic of a situation or a process.
pendicular to the opposite side
10 angle I anisotropy II
• angle I tex at the viewer's eye, one side
1. (of a polygon) an angle hav I horizontal, and the viewer's line
ing its vertex at one of the of sight to the object as the
polygon's vertices, and having I other side.
two of the polygon's sides as its I • angle of rotation
sides.
I the angle between a point and
2. two noncollinear rays (the
its image under a rotation, with
sides of the angle) having a com I
its vertex at the center of the
mon endpoint (the angle's ver I rotation. and sides that go
tex).
through the point and its im
• angle bisector age. Also, the measure of the
a ray that has the vertex of the I angle.
angle as its endpoint, and that I • angle ruler
divides the angle into two con a hinged ruler with a protrac
gruent angles.
tor attached for reading the
LI!OH'''' ..\.ooe •• _ I measure of an angle in degrees.
I • angle side
one of the two rays forming an
angle
• angular velocity
I (of an object moving around a
circle) the rate of change, with
respect to time, of the measure
• angle of depression I of the central angle that inter
(of an object lower than the I cepts the arc between the ob
viewer) The angle with its ver : ject and a ftxed point.
tex at the viewer's eye, one side I
horizontal, and the viewer's line I • anisotropy
of sight to the object as the anisotropy is a descriptor of one
other side. I aspect of an affine transforma
tion. In two dimensions, this is
• angle of elevation the ratio of the axes of the el
(of an object highet than a I lipse into which a circle is trans
viewer) The angle with its ver formed by an affine transfor
IllInnulus I lire length 11
~  apothem
: (of a regular polygon) a line
~ segment between the center of
; the polygon's circumscribed
: circle to a side of the polygon
~ that is also perpendicular to
; that side. Also, the length of
: that line segment.
I
:  arc
_ ANSI ~ (of a circle) Two points on the
acronym for the American Na ; circle (the endpoints of the arc)
tiona! Standards Institute. This : and the points of the circle be
body publishes specifications ~ tween them. An angle inter
for a number of standard pro I cepts an arc if the sides of the
MII,.",.riu======= II
12
II = = = = = = = . M J J , . " , . , ; u
13
mean, but it can also denote the ~ ther lands heads or lands tails
median, the· mode, the geomet ; is the sum of the chance that the
ric mean, and weighted means, : coin lands heads and the chance
among other things. ~ that the coin lands tails, because
I both cannot occur in the same
: coin toss. All other mathemati
~ cal facts about probability can
I be derived from these three axi
~ oms. For example, it is true that
: the chance that an event does
~ not occur is (100% the chance
; that the event occurs). This is a
: consequence of the second and
• axioms of probability ~ third axioms.
there are three axioms of prob I • axis
Iability: Chances
Th are always at .I (ofa cYlinder, ) the line segment
east z~ro. e c~ance that : connecting the centers of the
something happen! IS 100%. If I bases
two events cannot both occur at : .
the same time (if they are dis ~ • base
joint or mutually exclusive), the ; I. the side of an isoscoles tri
chance that either one occurs is : angle whose endpoints are the
the sum of the chances that ~ vertices of the base angles
each occurs. For example, con I 2. a side of a polygon or face of
sider an experiment that con : a solid used for reference when
sists of tossing a coin once. The ~ drawing an altitude or other
first axiom says that the chance I feature.
~at the coin lands heads, for ~ 3. a face of a solid used for ref
mstance, must be at least zero. : erence when drawing an alti
The second axiom says that the I tude or other feature.
chance that the coin either lands ; 4. the congruent parallel poly
heads or lands tails or lands on : gons of a prism. If the faces are
its edge or doesn't land at all is ~ all rectangular, any parallel pair
100%. The third axiom says; can be considered the bases.
that the chance that the coin ei
14
~~~~~~~~.
M s t h _ t i c s = = = = = = II
16 bias I biMr,ynumber II
and upon further considerations J swer that differs from the truth.
including the strength of the J The bias is the average (ex
pattern in the data and the pected) difference between the
sample size. Interest is gener J measurement and the truth. For
ally in the relative power of dif J example, if you get on the scale
ferent tests rather than in an with clothes on, that biases the
absolute value. It is question : measurement to be larger than
able whether the concept of J your true weight (this would be
BETA error is properly appli : a positive bias). The design of
cable without considering the an experiment or of a survey can
concept of sampling from a J also lead to bias. Bias can be
population, which is separate deliberate, but it is not neces
from the concerns of this Glos sarily so.
sary. Applicability of this rea I • biconditional
soning is also closely bound up I Q conditional and its converse
with the choice of test statistic. ~ where the converse is also true; r
.
r.(.,.~
~
,_Wool
t:'41 6:1 rt.,
•
. ~~~t:¥
0..__
_&1Jtt
I ,
• bijection
1d"'{~U
IAJ pI"D ",""
~~c;I , •• m /~,.,~
 , (I·/b"SV.>'1t',~ I
I I a onetoone onto function.
I
{ lAl .p}... ,l p ....)
rd'fr.l
~~.. toueo.
~, ~
....,,'..
l '.' •• I
,.",... .... ""'' 'O''.I0(h3 t14tiIJ ,  ::=::.:r ......... : I • bilateral symmetry
.....
i=:=::';:;;;
a,t"";""IJ.Q:d.. ,.th.
..r~J••• ........._ _
(2AIOJ1J:"'=,,=~
::::1Ia.J
WltS) J·.~~_~~L_ t!ffllW~
I reflectional symmetry with only
one line of symmetry.
....... ,,...t[!p,.~.., ......)., ih:::::" 1 • bimodal
I~.
::::=
=
Id~'!!.~I .... uacD
rJwJ·Mi;i;4~1/
(J.l#t.,"IQt3"'I""~""'''
It<W1
,\I) 11~
I I having two modes.
binary number
r. . .  "~~;;:c!W~");'~~5~ =.1
~ r.1'IWJ'>(.~s!:i'" ttel!'" a:u.. I •
I a number written to base 2.
~." ...,m:.)a~
... 
1loCt...
_?:ial.fUIJ~.~_::/
IiaJtJalnn
I\~_'_"~~&IJx),,~~:~~_"+I""'''
1umi~
r.a.,os.' ~.Jt'"
:=
5..,=r.a
_.. ...... "'.add ..."""
I;' :. :. .... • tIII OIl
• bias
a measurement procedure or J
. _ . . .2
32
.... • .•• .. 2
.
estimator is said to be biased if, I
on the average, it gives an an
II hiM", operation I biseet 17
1I=======MAthemsties
II bootstrapestimateofstandarderror I: model 19
Msthemsties=========~ II
20
=~~~~~~=*
suppose we are trymg , to evalu I,
M
ate a cold remedy by giving it ;
or a placebo to a group of n in : R
dividuals, randomly choosing ~
half the individuals to receive I Q
p
the remedy and half to receive :
the placebo. Consider the me ~
dian time to recovery for all the I
individuals (we assume every ~
N
one recovers from the cold :
eventually; to simplify things, ~ • box plot
we also ~sume that no one ,re ; a representation of data above
covered m exactly the median: a numbered scale where the
time, and that n is' even). By ~ "box" encloses all data between
defmition, half the individuals ;~ the median of the lower half
got better in less than, the me : (quartile 1) and the median of
dian time, and half 10 more ~ the upper half (quartile 3), with
than the median time. The in I a vertical line inside the box to
dividuals who received the: indicate the median of the data;
treatment are a random sample ~ a dot represents each of the high
of size n/2 from the set of n I and low values of the data, and
subjects, half of whom got bet ~ a horizontal line called a whis
ter in less than median time, and : ker connects each dot to the
half in longer than median ~ box.
time, If the remedy is ineffec; b h d b d
.
tIve, the number 0 f sub'Jects : • ranc
, an  oun . ,
w h0 recelve' d the remedy and I exploratIon
" , 'of a randOlrusanon
' 1 th
w ho recovered mess an me I : distrlbunon
' ,
m such a way as to
' 'lik th
dian orne lS e e sum 0 , f nl2 . anOclpate "Ithe effect' of the next
th
draws Wlth ' replacement firom a I, rannOffilSatIOn d 're atIve
. to ale
This
box with two tickets in it: one ; present r~ OffilSanon. ,
with a "1" on it and one with a : lows selecove search of ~aro~
"0"" I lar zones of a randoffilsatIon
on It. .. distrl'b'
uoon; m. the context 0 f a
~ randomisation test such selec
II =======MsthtmUJms
II hreaItdownpoint I canonicalcorreUJ";"nalysis 21
• byte
the amount of memory needed
to represent one character on a
computer, typically 8 bits.
• calculator notation ~ • canonical correlation
the symbols used by a calcula i analysis
tor for scientific notation. : a multivariate method for as
• caliban puzzle ; sessing the associations be
a logic puzzle in which one is I tween two sets of variables
asked to infer one or more facts : within a data set. The analysis
from a set of given facts. ~ focuses on pairs of linear com
I binations of variables (one for
• canonical ~ each set) ordered by the mag
a canonical description of any
: nitude of their correlations with
statistical situation is a descrip
~ each other. The fIrst such pair
tion in terms of extracted vec
; is determined so as to have the
tors that have especially simple
: maximal correlation of any
ordered relationships. For in
~ such linear combinations. Sub
stance, a canonical correlations
; sequent pairs have maximal cor
analysis describes the relation
: relation subject to the constraint
between two lists of variables
~ of being orthogonal to those
in terms of two lists of linear
I previously determined.
combinations that show a re
markable pattern of zero cor I
Msthmulties======= II
22 canonical 11am; analysis I causation, causal relation II
• canonical variates analysis ~. capacity
a method of multivariate analy ; the amount of liquid that can fill
sis in which the variation among : an object.
groups is expressed relative to I • I
the pooled withingroup cova I • cartesian p ane .
. . C . I . a rectangular coordinate svstem
nance matrIX. anoruca varI . . . J'
al . finds lin I COnSIStIng of a honzontal num
ates an. YSIS ear tr~s ber line (xaxis) and a vertical
fiormatlons of the data which I b lin ( .).
maximise the among group ~um ethr e . ~axIS , mtersect
. . , I mg at e ongm (zero on each
varIatIOn relatIve to the pooled . be lin)
withingroup variation. The ca ; num r e.
nonica! variates then may be I • categorical variable
displayed as an ordination to : a variable whose value ranges
show the group centroids and ~ over categories, such as {red,
scatter within groups. This may ; green, blue}, {male, female},
be thought of as a "data reduc : {Delhi, Calcutta, Mumabai},
tion" method in the sense that ~ {short, tall}, {Asian, Mrican
one wants to describe among ; American, Caucasian, Hispanic,
group differences in few dimen : Native American, Polynesian},
sions. The canonical variates are ~ {straight, curly}, etc. Some cat'
uncorrelated, however the vec ; egorical variables are ordinal.
tors of coefficients are not or : The distinction between cat
thogonal as in Principal Com ~ egorical variables and qualita
ponent Analysis. The method is I tive variables is a bit blurry.
closely related to multivariate .I • catenary
analysis of variance : a curve whose equation is y =
(MANOVA), multiple discrimi ~ (a/2)(e +e X/ 3 A chain sus
X/ 3 ) .
nant analysis, and canonical cor I pended from two points forms
relation analysis. A critical as : this curve
sumption is that the within ~ •.
group variancecovariance;. causation, causal relation
structure is similar otherwise : two variables are causally re
the pooling of th~ data over ~ lated if changes in the value of
groups is not very sensible. i one cause the other to change.
For example, if one heats a rigid
II =======.M.thelulies
II ceilingfunction I centroid 23
*=================
container filled with a gas, that I • center of gravity
causes the pressure of the gas I the mean of the coordinates of
in the container to increase. points in a figure, whether one,
Two variables can be associated I two, or threedimensional
without having any causal rela I • central angle
tion, and even if two variables
I (of a chord or arc) An angle
have a causal relation, their cor : whose vertex is the center of a
relation can be small or zero. I circle and whose sides pass
• ceiling function I through the endpoints of a
MlJthemR.tiu======= II
",,24=========== ..centroid size I Chebychev'si1UlJUR1ity II
_ centroid size  cevian
centroid size is the square root I a line segment extending from
of the sum of squared distances a vertex of a triangle to the op
of a set of landmarks from their I posite side.
centroid, or, equivalently, the I
square r'?Ot of the sum of the ~
variances of the landmarks
about that centroid in xand y I
directions. Centroid Size is used
in geometric morphometries _ chance variation, chance
because it is approximately I error
uncorrelated with every shape a random variable can be de
variable when landmarks are composed into a sum of its ex
distributed around lTiean posi ~ pected value and chance varia
tions by independent noise of ; tion around its expected value.
the same small variance at ev The expected value of the
ery landmark and in every di I chance variation is zero; the
rection. Centroid Size is the size I standard error of the chance
measure used to scale a con variation is the same as the stan
figuration of landmarks so they dard error of the random vari
can be plotted as a point in I ablethe size of a "typical" dif
Kendall"s shape space. The de ference between the random
nominator of the formula for variable and its expected value
the Procrustes distance between I
 Chebychev's inequality
two sets of landmark configu
I for lists: For every number
rations is the product of their
k>O, the fraction of elements
Centroid Sizes. in a list that are k SD's or fur
 certain event I ther from the arithmetic mean
an event is certain if ItS prob of the list is at most Ijk2. For
ability is 100%. Eve~ if an event random variables: For every
is certain, it might not occur. I number k>O, the probability
However, by the complement I that a random variable X is k
rule, the chance that it does not SEs or further from its expected
occur is 0%. I value is at most l/k2.
II chisquare curve I chisquared d~tion 25
Mslthema.ties======= II
26 chiSlJ.tl4red statistic I circumfoYence II
~============*~===========
• chisquared statistic
this is a longestablished test
statistic for measuring the ex
tent to which a set of categori :
cal outcomes depart from a. I
I
D
Q Chord
c
"A
II
28 corJJicient I combinatitms II
structed in a hierarchical man ~ • coincidental lines
ner and shown in the form of a ; lines that are identical (one and
treelike diagram called a den : the same)
drogram. ~ • collinear
• coefficient ~ lying on the same line.
a coefficient, in general, is a ~ • combinations
number multiplying a function. the number of combinations of
In multivariate data analysis, n things taken k at a time is the
usually the "function" is a vari I
number of ways of picking a
able measured over the cases of : subset ofk of the n things, with
the analysis, and the coefficients I
out replacement, and without
multiply these variable values I
regard to the order in which the
before we add them up to form elements of the subset are
a score. A coefficient is not the
pickc:d. The number of such
same as a loading. combinations is nCk = n!j(k!(n
• coincide k)!), where k! (pronounced "k
lying exactly on top of each factorial") is kx(kl)x(k2)x
other. Line segments that coin ... x 1. The numbers nCk are
cide are identical; they have all I also called the Binomial coeffi
the same points. I cients. From a set that has n el
ements one can form a total of
I 2n subsets of all sizes. For ex
I ample, from the set {a, b, c},
: which has 3 elements, one can
I
: form the 23 = 8 subsets U, {a},
I {b}, {c}, {a,b}, {a,c}, {b,c},
{a,b,c}. Because the number of
subsets with k elements one can
I form from a set with n elements
is nck, and the total number of
subsets of a set is the sum of
~ the numbers of possible subsets
; of each size, it follows that
nco +nC1 +nC2 + ... +nCn = 2n.
II ~m#n~~I~~n~~~~~~~~~~~2==9
The calculator has a button ~  complementary angles
(ncm ) that lets you compute the ; two angles whose measures
number of combinations of m : have the sum 90°.
things chosen from a set of n I
things. To use the button, first I
type the value of n, then push
ArSfc.
the nCm button, then type the
value of m, then press the "=" I
button.
8
~ ()
mLABC + mLCBD =90"
LABC and LCBD are complementary angles.
 commutative properties I
properties about order of addi :  complex numbers
tion' a + b = b + a; of multi ~ complex numbers are an alge
plication, a x b = b x a" ; braic way of coding points in
 compass : the ordinary Euclidean plane
a drawing tool used to draw ~ so that translation (shift of
circles at different radii ; position) corresponds to the
:I addition of complex numbers .
 compatible numbers : and both rescaling (enlarge
numbers that can be easily ma I ment or shrinking) and rota
nipulated and operated on men ~ tion correspond to multiplica
tally. : tion of complex numbers. In
_ complement rule ~ this system of notation, in
the probability of the comple ; vented by Gauss, the xaxis is
ment of an event is 100% mi : identified with the "real num
nus the probability of the event: ~ bers" (ordinary decimals
P(Ac) = 100% peA). ; numbers) and the yaxis is
: identified with "imaginary
thcompIement
I f b f .I num b ers" (t h e square roots 0 f
e.comp e~ent 0 a su .set 0 ; negative numbers). When you
a gIven set IS the collecnon of" ul" I "
r thi" b
all elements of the set that are ; ~ tip pomts on " s axiS hY
not elements of the subset. : t emse ves accord mg to t e
I rules, you get negative points
: on the "real" axis just defined.
~ Many operations on data in
MJJ.th_tics========== II
~30~~~~~~~~~~~~~~mmW~l~i~p70b~i~ /I
II =======MRthem4ries
31
*=================
we update the probability of A ~ if we are restricting the out
to reflect this new knowledge? ; come space to B, we need to
This is what the conditional : divide by the probability of B
probability does: it says how the ~ to make the probability of this
additional knowledge that B ; new S be 100%. On this scale,
occurred should affect the prob : the probability that AB hap
ability that A occurred quanti ~ pened is P(AB)fP(B). This is
tatively. For example, suppose , the deftnition of the conditional
that A and B are mutually ex : probability of A given B, pro
clusive. Then if B occurred, A ~ vided P(B) is not zero (division
did not, so the conditional prob ~ by zero is undefined). Note
ability that A occurred given ; that the special cases AB = {}
that B occurred is zero. At the : (A and B are mutually exclu
other extreme, suppose that B ~ sive) and AB = B (B is a subset
is a subset of A, so that A must ; of A) agree with our intuition
occur whenever B does. Then : as described at the top of this
if we learn that B occurred, A ~ paragraph. Conditional prob
must have occurred too, so the ; abilities satisfy the axioms of
conditional probability that A : probability, just as ordinary
occurred given that B occurred ~ probabilities do.
is 100%. For inbetween cases,
~  conditional proof
where A and B intersect, but B
; a proof of a conditional state
is not a subset of A, the condi
tional probability of A given B ,
: ment.
is a number between zero and :  cone
100%. Basically, one "restricts" ~ a solid whose surface consists
the outcome space S to consider , of a circle and its interior, and
only the part of S that is in B, : all points on line segments that
because we know that B oc ~ connect points on the circle to a
curred. For A to have happened ~ single point (the cone's vertex)
given that B happened requires ; that is not coplanar with the
that AB happened, so we are : circle. The circle and its interior
interested in the event AB. To ~ form the base of the cone. The
have a legitimate probability ; radius of a cone is the radius of
requires that P(S) = 100%, so : the base. The altitude of a cone
Msthemtllti&s========== II
~32~~~~~~~~~~~Mlmn~m~M II
is the line segment from the ~ the family, and these likelihoods
vertex to the plane of the base ; may be then used to defme a
an~ perpendic~ar to it. The ~ contiguous set of values which
?elgh~ of a cone 18 t.he length of : occupy a certain proportion of
Its aln~de. If the line segment I the total unit weight of the like
c~nnecnng the vertex of a cone : lihoods integrated over all val
with th~ center of its base is ~ ues of the test statistic, the con
perpendi~ar ~o the base, then ~ fidence interval is defmed by
the cone 18 a nght cone; other ' the minimum and maximum
wise it is oblique. ~ values of the range of values so
• confidence interval ~ defined. The proportion of the
for a given rerandomisation ; total weight within the range of
distribution, a family of related : values is regarded as an alpha
distributions may be defined ~ probabili~ ~at, the v~ue of the
according to a range of hypo I test stanstic hes WIthin this
the~cal values of the pattern ~ range, Generall~ the defmition
whIch the test statistic mea : of a confidence mterval cannot
sures. For instance, for the pit ~ be unique wi~out imposing
man permutation test to test for ' further constramts. Approaches
a scale shift between .two ~ to providing suitable con
groups, a related distribution ~ ~traints, s~ch that a confidence
may be formed by shifting all ; mtet;al will be unique, include
the observations in one group : d~g the confidence interval
by a common amount where ~ : to mclude the whole of one tail
this common shift is r~garded ; of the ~istribution; or to be
as a continuous variable, With : centred m some sense upon the
fmite numbers of data the num ~ outcome value; or to be centred
ber of related distributions will ; bet,ween TAILS of equal
be fmite, and typically consid ~ weIght" In, the ,cas,e of re
erably smaller than the number : randomIsanon distributions,
of points of the randomisation I these are discrete distributions
distribution. The likelihood of : so there will generally be no
the outcome value may be cal ~ range of,values with weight cor
culated for each distribution in I responding exactly to an arbi
: trary nominal alpha criterion
II ===:========Mstht:'l'/Ul.tU:s
II con.fideme level I congruentfiDU1'eS. 33
level, and the problem of non ~ the treatment (if any). For ex
uniqueness is therefore not gen ; ample, prominent statisticians
eratly solvable. : questioned whether differences
~ between individuals that led
; some to smoke and others not
: to (rather than the act of smok
~ ing itself) were responsible for
I the observed difference in the
~ frequencies with which smok
: ers and nonsmokers contract
I various illnesses. If that were
MIIthemtlnes======= \I
~34~~~~~~~~~~C~ongruen=.tpolygOllS I constantofRnetpmtUm II
• congruent polygons • consecutive vertices
two or more polygons with the I (of a polygon or polyhedron),
exact same size and shape. two vertices that are connected
I by a side or edge.
• congruent segments
two or more segments that have I • consensus configuration
the same measure or length. I a single set of landmarks in
tended to represent the central
• conic section
tendency of an observed sample
the cross section of a right cir
I for the production of superimpo
cular cone cut by a plane. An el
sitions, of a weight matrix, or
lipse, parabola, and hyperbola I
some other morphometric pur
are conic sections.
I pose. Often a consensus configu
~

=:. :\ ,, .......
1\ ration is ·computed to optimize
/ \ !
 ............
,~
•......:
;
.......
 ... ...... ~ some measure of fit to the full
I I .. I sample : in particular, the
='= I \  ) '/
Procrustes mean shape is com
:.,l_tt..
I \
i
l '
1\ \
f,I .\\ ~
/ \\
puted to minimise the sum of
squared Procrustes distances
(
.'    /
"


'/
\
I
I from the the consensus land
• conic solid marks to those of the sample.
the set of points between a point I
(the venex) and a noncoplan;:lI • consequent
region (the base), including the I the second or "then" part of a
point and the region. I conditional statement.
II =====.===MsJthmuJnes
II ~consts~~n~tffl,~te~oJi~'ch~'Q;~~~e~Iconti~~·n~Uf,~·ty=~ 35
MMIIenIaeics======____.11
38 .. ~~~~WD~e,!!!!!·c~O'1¥I1i~m!!!!!~!!!!!mee~!!!!!1coo!!!!!~"~rJt~~na~te II.
tive results based on samples of ~ 2. if P and q are two logical
convenience will be. i propositions, then the converse
T_. of the proposition (p IMPLIES :
~~=.:=.: ~ q) is the proposition (q IM
_of
v.,..., 1>..
~ i __i PLIES p).
of _..... .:..
. c.:
a.
..... ."" .....
1.8UJIa!i1C9
!
COl
iconversion factor •
~,# > . . ,,' .,.,  'Ii>! 00. : relationship between two umts
I
39
M/lthemR,riu======== II
is used for Pearson's producr I • correlation
moment correlation which is I a measure of linear association
the covariance divided by the between two (ordered) lists.
product of the standard ~ Two variables can be strongly
deviations,rxy=Sxy / Sx.Sy This ; correlated without having any
correlation coefficient IS + 1 or causal relationship, and two
1 when all values fall on a variables can have a causal re
straight line, not parallel to ei I lations hip and yet be
ther axis. However, there are uncorrelated.
also Kendall, Spearman, • corresponding angles
tetrachoric, etc. correlations I if two parallel lines are cut by a
which measure other aspects of ~ transversal, corresponding
die relation between two vari angles are translations of each
ables. , other along the transversal .
• correlation coefficient • cosine
the correlation coefficif!nt r is I (of an acute angle) The ratio of
a measure of how nearly a the length of the adjacent side
scatterplot falls on a straight I to the length of the hypotenuse
line. The correlation coeffi I in any right triangle containing
cient is always between 1 and I the angle.
+ 1. To compute the correla .
tion coefficient of a list of ; • coterminal angles
pairs of measurements (X,Y), : two angles that have the same
first transform X and Y indi .~" terminal side
vidually into standard units. "I • countable set
Multiply correspopding ele ; a set is countable if its elements
ments of the transformed : can be,put in one~toone corre
pairs to get a single list of ~ spondence with a subset of the
numbers. The correlation co ; integers. For eXaniple the sets
efficient is the mean of. that {O~ 7 3} {red, gre~n, blue}
list of. products. This page .~. { ... , , "'2 :1" 0" 1 2 ... }',
contams a .too~ tha~ lets y?U I {straiWtt, curly}, and the set of
generate" bIVarIate data WIth : all fractions are countable. If a
· coeffiICIent
any correIanon . YQU .I set is not countable
'
. , it is un
want.
II counterclockwise I criticalveUue 41
~~~======~*==~===========
countable. The set of all real ~ given transformation has only
numbers is uncountable. lone direction of covariants, but
: a full plane (four landmarks) or
• counterclockwise
in orientation, the direction in ~ hyperplane (five or more land
which points are named when I marks) of invariants.
· travelling on the line, the in'
if C<:MI:I'iAntl"lft.lrlcTII'rtSf.r rr'..¢qUCOtr~!lI,tha._orreI4UM ....rflcllll'4 ..,
(aC'lt.. lllt.~ rl'(HTl ")tIJvJ1(1U7'PolnbLtIC.WJst.~J
A
terior of the figure is on the left
side.
• counterexample
a situation in a conditional for I
which the antecedent is true, but
the conditional is false; aka con I
tradiction ~ • cover
• counting techniques ; a confidence interval is said to
a variety of methods used to : cover if the interval contains the
determine the total possible ~ true value ofl the parameter.
outcomes, typically in a prob I Before the data are collected,
ability situation, including the : the chance that the confidence
multiplication principle, trees ~ interval will contain the param
and lists. I eter value is the coverage prob
~ ability, which equals the confi
• covariant : dence level after the data are
a covariant of a particular shape I collected and the confidence in
change is a shape variable ~ terval is actually computed.
whose gradient vector as a func
tion of changes in any complete ; • coverage probability
set of shape coordinates lies : the coverage probability of a
precisely along the change in ~ procedure for making confi
question. For transformations ; dence intervals is the chance
of triangles, the relation be : that the procedure produces an
tween invariants and covariants ~ interval that covers the truth.
is a rotation by 90 degrees in ~ • critical value
the shapecoordinate plane. For ; the critical value in an hypoth
more than three landmarks, a : esis test is the value of the test
I
Msthernades,=======II
42 crossseaional study I Ciunulati; Probability Distribution Pmu:tUm... II
statistic beyond which we ~ has some or all of its digits re
would reject the null hypoth ; placed by letters or symbols and
esis. The critical value is set so : where the restoration of the
that the probability that the test ~ original digits is required. Each
statistic is beyond the critical ; letter represents a unique digit.
value is at most equal to the sig ~ • cube
nificance level if the null hy : a solid figure bounded by 6 con
pothesis be true. I
: gruent squares.
• crosssectional study I ub'IC equa0'on
r
. • • C
a crosssectlonal study com . I 'al ti f d
pares different individuals to ~ a po oml equa on 0 e
each other at the same time ; gree .
it looks at a crosssection of a ;
population. The differences be
tween those individuals can cQn ~
found with the effect being ex ;
plored. For example, in trying :
to determine the effect of age ~
on sexual promiscuity, a cross ;
sectional study would be likely :
to confound the effect of age ~
with the effect of the mores the ;
subjects were taught as children::
theoIder m
.
. di 'duals
ed·th VI
ably r~s WI a very ~ler ~
b I.
were...1:.a:'
pro .
C u1 u' 'D. b bili'ty
um a ve .ero a
D'IS m'bU u'on F unc0'on
. (CDF)
~ntthanatt1tuthde towards proIDlS . cu ; the cumulative distribution
Ity e younger sub~ects .. fun' f d . bl
Thus It · uld be . rud ; etlon 0 a ran om varia e
'b wo ...1:.a:
Imp• ent .
to: .IS the ch ance that the rand om
attn ute wuerences m proIDlS I . bl . 1 than ual
. th' C f. varia e IS ess or eq to
CUlty to e agmg process. ..: fun' f In bols
1 .rudinal tud I X, as a etlon 0 x, sym ,
ong! s y. : if F is the cdf of the random
• cryptarithm ~ variable X, then F(x) = P( X < =
a number puzzle in which an ; x). The cumulative distribution
indicated arithmetical operation : function must tend to zero as x
I
II :........... 1~tur,fue *=========!!!!!!4!!!!!!3
approaches minus inftnity, and ~  cyclic polygon . .
must tend to unity as x ap ; a polygon whose vernces lie on
proaches inftnity. It is a positive ~ a circle.
function, and increases mono : _ cyclic quadrilateral .
tonically: ify > x, thenF(y) >= ~ a quadrilateral that can be m
F(x). The cumulative distribu ; scribed in a circle.
tion function completely
characterises the probability dis ;  cylinder .
tribution of a random variable. : a solid whose surface COnsISts of
. ~ all points on two circl~ in ~o
 curved ~pace. I parallel planes, along WIth pomts
a space wlth coordinates and : in their interiors (the bases of the
a distance f~ction such that ~ cylinder), and all points o~ line
the area of Circles, volume of I segments joining the two arcles.
spheres, etc. are not pro~or : The axis of the cylinder is the line
tional to the ap.propnate ~ segment that joins the centers of
power of the radIUs, e. g., I the bases. The radius of the cyl
Kendall's shape space. In : inder is the radius of a base. An
curved spaces, the usual iI?tui  ~ altitude of a cylinder is a line seg
tions about what "straight I ment between. and perpendicu
lines" can be expected .
to do :I lar to the planes of the bases. The
will be fa~lty. For mstaI?ce, : height of a cylinder is ~e ~ength
corresponding to every trlan ~ of an altitude. If the axIS IS per
gular shape i? Kendall's sha~e ; pendicular to the basc:s, then the
space, there IS. another t?at I! : cylinder is a right cylinder; oth
"as far fronl ~t as 1?osslble, ~ erwise, it's oblique.
just like there IS a pomt on the I •• •
surface of the earth as far as :  cylindric s?lid
possible from where you now ~ the set o~ pomts ~~n a re
sit. ; gion ~d I~ trans~nondm.s~ce,
: including llle region an Its .lffi
 cycle ~ age
(of a periodic curve) One se~
tion of a curve that, when laid ;  cylindric surface
out repeatedly endtoend, :I the union of the bases and the
forms the entire curve. : lateral surface
~lhemsti&s======= II
=================*
44 dartI dejieientcotmlintJte I
.dart ~ The~e are 52! (52 factorial)
a concave kite. ~ possible orderings.
~ data . ; • decompose numbers
inform~tlon
used as a basis for ; break up numbers into addends
reasonmg. . or factors.
• decagon ~ • deductive reasoning
a tensided polygon ~ reasoning accepted as logical
• decimal number ; from agreedupon assumptions
a number written to the base 10. : and proven facts.
I
• decision rule :  deductive system
a rule for comparing the out ~ an arrangement of premises
come value of alpha with a ; and theorems, in which each
nominal alpha criterion level: theorem can be proved by de
(such as 0.05). An outcome ~ ducti~e reasoning using only the
value smaller (more extreme) ; premIses ~nd ~revious theo
than the nominal alpha criterion : rems, and m which each defini
level leads to a decision of sta ~ tion uses only terms that have
tistical significance of the find ; been defmed previously in the
ing that the test statistic has a : system.
value other than its (null) ~ • deficient coordinate
hypothesised value. ~ in addition to landmark loca
• deck of cards ; tions, a digitiser can be used to
a standard deck of playing cards : supply information of other
contains 52 cards, 13 each of ~ sorts. For example, a point can
four suits: spades, hearts, dia ~ ?e used .to encode part of the
monds, and clubs. The thirteen ; inf~rma~o~ about a curving arc
cards of each suit are {ace 2 3 : by Identifying the spot at which
4? 5,6,7,8,9, 10, jack, q~n: ~ the ar~ lies farthest from some
king}. The face cards are {jack, I other unage structure (perhaps
queen, king}. It is typically as : another such curving arc). The
sumed that if a deck of cards is ~ null model of independent
shuflled well, it is equally likely ; Gauss~ noise does not apply
to be in each possible ordering. ; to posItion along the tangent
II========~~Mstherntmes
II MJkimtnumbwl derWed~~~~~~~~~~
45·
MMhemsnes=========== II
46
=========. describe I dijfermtUJte II
ments such as speed in miles per ~ • diagonal
hour. I. a line segment connecting two
• digimetic ~  direction
a cryptarithm in which digits ; the way a number goes positive
represent other digits. : or negative
I
• digit :  direction of a translation
in the decimal system, one of ~ the compass direction in which
the numbers 0, I, 2, 3, 4, 5, 6, ; a translation goes.
7, 8, 9. ~ _ disc
 dihedral angle : a circle together with its inte
the angle formed by two planes ~ rior.
meeting in space.
~  dis.crete distribution
; a probability distribution of
: some statistic, based upon an
~ algebraic formula or upon re
I randomisation or upon actual
• discriminant analysis
i. into the fluid .
a broad class of methods con idissection
cemed with the development of : the result of dividing a figure
rules for assigning unclassified ~ into pieces.
objects/specimens to previously I • distance
II ~=====~Msthemstir;s
II euclitleandistancema;t'l'ixanalysis I ;"",t==========5=5
which particular sample is ob ~  Euler segment
tained, which is random. A ca ; the line segment containing the
nonical example of an estimator : centroid of a triangle, whose
is the sample mean, which is an ~ endpoints are the orthocenter
estimator of the population ; and the circumcenter of the tri
mean. : angle.
I
 euclidean distance matrix :  Euler's formula for poly·
analysis ~ hedrons octahedron
EDMA. A method for the sta I an eightsided polyhedron. The
tistical analysis of full matrices of : regular octahedron is one of the
all interlandmark distances, av ~ Platonic solids.
eraging elementwise within ~ _ Euler's constant
samples, and then comparing I the limit of the series 1/1 + 1/
those averages between samples ~ 2+1/3+ ... +1/nln n as n goes
by computing the ratios of cor : to infInity. Its value is approxi
responding mean distances. ~ mately 0.577216 .
• euclidean space I _ even function
a space where distances be
; a function f(x) is called an even
tween two points are defmed as
: function if f(x) =f( x) for all x.
Euclidean distances in some I
system of coordinates. :  even node
~ a node that has an even num
• euler line
; ber of arcs
the line through a triangle'S cir
cumcenter, orthocenter, and ;  even nwnber
centroid. Named after Swiss : an integer that is divisible by 2.
I
mathematician and physicist : _ event
Leonhard Euler. ~ an event is a subset of outcome
A Euler or nine.point circIc of ABC. ; space. An event determined by
: a random variable is an event
~ of the form A=(X is in A).
; When the random variable X is
: observed, that determines
~ whether or not A occurs: if the
56
~~~~~~~=.
value of X happens to be in A, ~ group for the development and
A occurs; if not, A does not oc ; promulgation of the ideas of re
cur. : randomisation statistics.
I
• exact binomial test : .excenter
a statistical test referring to the ~ the center of an excircle.
binomial distribution in its ex I • excircle
act algebraic form, rather than
through continuous approxima ~ an escribed circle of a triangle.
tions which are used especially I
where sample sizes are substan :
tial.
• exact number
a numerical result that has not I
been rounded or estimated.
• exact test
 the characteristic of a re
randomisation test based upon ~ • exclusive OR
exhaustive rerandomisation, ~ one or the other, but not both
that the value of alpha will be
fixed irrespective of any random I • exhaustive re
sampling of randomisations or ~ randomisation
uponanydistributionalassump : a series of samples from a
tions. Notable examples are the ~ randomisation set which is
exact binomial test, fisher test, ; known to generate every
the Pitman permutation tests ( 1 randomisation. In particular,
and 2), and various nonpara I sampling which generates every
metric tests based upon ranked I randomisation exactly once.
data. ; • exhaustive
• exactstats : a collection of events {AI, A2,
this is the name of the academic ~ A3, ... } is exhaustive if at least
initiative which produced this ; one of them must occur; that
present glossary. exactstats is : is, if S = Al U A2 U A3 U ...
a closed email based discussion ~ where S is the outcome space.
II existentUdstll:tement I e:cpectedi =nxpi 57
*~~==~~~
• existential statement ~ pected value of a constant a
a conditional that uses the word times a random variable X is the
;
'same' ~ constant times the expected
: value of X (E(axX ) = aX
• expansion I E(X».
a size change where k is greater
than I t • expected i = nxpi
MsthlJllllilties============ II
58 txperiment I ~tment /I
II =======MJJthemlJnes
I
II Fisher'sl'&ll&ttest Iflexibly 63
II = = = = = = = = = = M a t h _ t i & s
II form space Ifranu. mmpli"9franu *========6=5
orientation are said to have been ~ (structural complexity at
"removed." ; smaller scales is mathematically
_ form space : indistiguishable from that at
~ largerscales) over all scales con
th~ space of figures with differ
; sidered.
ences due to location and ori
entation removed. It is of 2p3 I BON Size Offset
Width 540
dimensions for twodimen 2 2 2 Height 110
sional coordinate data and 3p6 3 .. 3
dimensions for threedimen I . 13 .. ~
5 22 5
sional coordinate data. C!!!D
6 33 6
_ formula 44 7 I Dd.ulll
8 55 8
a concise statement expressing 9 66 9
O!!D
the symbolic relationship be I 10 77 .10
tween two or more qu;mtities. 88 ~
"12 111 "
12
00
 Fourier analysis I
in morphometries, the decom :  fraction
position of an outline into a ~ an expression of the form alb . .
weighted sum of sine and co
sine functions. The chapter by ~  frame, sampling frame
Rohlf in the Blue Book pro I a sampling frame is a collec
vides an overview of this and ~ tion of units from which a
other methods of analysing out : sample .will be drawn. Ideally,
line data. I the frame is identical to the
MRthmuJti&s======= II
66 frelJumcy I funetion rule II
~~~~~~~=*
might desire to estimate the ~ (fraction or percentage) of ob
current annual average in ; servations in different ranges,
come of 1998 graduates of : called class intervals.
the University of Delhi. I pro I
pose to use the sample mean
income of a sample of gradu
ates drawn at random . To fa I
cilitate taking the sample and I
contacting the graduates to
obtain income information I 1
from them, I might draw I
names at random from the list
J
of 1998 graduates for whom
the alumni association has an I  .
accurate current address. The I
population is the collection of :
1998 graduates; the frame is I
those graduates who have cur  frustum


rent addresses on file with the for a given solid figure, a re
alumni association. If there is I lated figure formed by two par
a tendency for graduates with I allel planes meeting the given
higher incomes to have upto solid. In particular, for a cone
date addresses on file with the I or pyramid, a frustum is deter
alumni association, that would I mined by the plane of the base
introduce a positive bias into and a plane parallel to the base.
the annual average income es I NOTE: this word is frequently
timated from the sample by I incorrectly misspelled as
the sample mean. frustrum.
_ frequency _ function rule
the number of times a value I the set of operations that de
occurs in some time interval. I scribes the process that takes
the independent variable and
_ frequency table transforms it into the dependent
a table listing the frequency I variable in a consistent way.
(number) or relative frequency
67
.==~=============
• fundamental region I variance and correlation of the
a region used in a tesselation I variables in measuring dis
tances between points, i. e., dif
• fundamental rule of
I ferences in directions in which
counting
I there is less variation within
if a sequence of experiments or
groups are given greater weight
trials TI, T2, T3, ... , Tk could I
than are differences in directions
result, respectively, in nl, n2 n3,
I in which there is more varia
· .. , nk possible outcomes, and
the numbers nl, n2 n3, ... ,nk I tion.
do not depend on which out
comes actually occurred, the
entire sequence of k experi I
.:~
p
ments has nl x n2 x n3 x dj(~) = mill
ceo}
IIx  eil Vd;(%)=~
x nk possible outcomes.
• game theory
a field of study that bridges I
mathematics, statistics, eco
nomics, and psychology. It is • generalised superimposi
used to study economic I
tion
behaviour, and to model conflict I the superimposition .ofa set of
between nations, for example, configurations. onto their con
"nuclear stalemate" during the I
sensus configuration. The fit
Cold War. I ring may involve leastsquares,
• gaussian curve resistantfit, or other algo
a normal curve. rithms and may be strictiy or
I thogonal or allow affine trans
• generalised distance I formations.
d. A ,synonym for Mahalanobis
distance. Defined by the equa ; • geoboard
tion for two row vectors x. and a flat board into which nails
•
Xj for two individuals, and p I have been driven in a regular
variables as: , where S is the pxp I rectangular pattern. These nails
variancecovariance matrix. It represent the · lattice points in
takes into consideration the the plane.
=======~* =D=eodeS'ie='=I=Beomem&=='morp==~=='= "
=68='
• geodesic ~ The expected value of the geo
the arc on a surface of shortest . metric distribution is lip, and
length joining two given points. its SE is (lp)V2/p.
• geodesic distance • geometric mean
the length of the shortest path I the nonnegative number whose
between two points in a suitable , square is the product of two
geometric space (one for which given nonnegative numbers;
curving paths have lengths). On I the side of a square having the
a sphere, it is the distance be I same area as a rectangle whose
tween two points as measured length and width are given.
along a great circle.
• geodesy
a oranch of mathematics deal
ing with the shape, size, and
curvature of the Earth.
.,~_ O____
:!
~~ 0 OJ
,. ________________. __,
____
I
i
I
• geometric distribution
,.~
II =~=======.MsJthmultiu
IIBeometrie probllhility 1 probllhility IB~~1 symmetry . 69
flection.
 gnomon magic square
a 3 X 3 array in which the ele
ments in each 2 X 2 corner I
MIJ. . .n u = = = = = = = =___ 11
76 hyperplllne I hypothesis tutino II
~~~~~~~~~.
• iff : real is y.
I
if and only if. : • implies, logical implication
• iff, if and only if ~ logical implication is an opera
if p and q are two logical propo I tion on two logical proposi
sitions, then(p iff q) is a propo ~ tions. If p and q are two logical
sition that is true when both p : propositions,(p IMPLIES q) is
and q are true, and when both ~ a logical proposition that is true
p and q are false. If is logically ; if P is false, or if both P and q
equivalent to the proposition ( : are true. The proposition (p
(p IMPLIES q) AND (q IM ~ IMPLIES q) is logicallyequiva
PLIES p) ) and to the proposi ; lent to the proposition ((NOT
: p) ORq).
I
MAtbmuJncs======= II
,,;,,78==========im~7 subset I irulepnulmt, irukperulm&e II
• improper subset I • independent and identi
a subset that includes the entire I cally distributed
parent set. a collection of two or more ran
I dom variables {Xl, X2, . . . , }
• incenter I is independent and identically
the point of concurrency of a
distributed if the variables have
triangle's three angle bisectors.
I the same probability distribu
• incircle I tion, and are independent .
the circle inscribed in a given
I •independent variable
figure.
in regression, the independent
• included angle ~ variable is the one that is sup
an angle formed between two I posed to explain the other; the
given sides of a triangle. term is a synonym for "explana
• inclusive OR tory variable." Usually, one re
one or the other, or both; and/ I gresses the "dependent vari
or able" on the "independent vari
able." There is not alwavs a
• incoming angle 
I clear choice of the independent
the angle formed between the variable. The independent vari
path of an approaching object able is usually plotted on the
(a billiard ball, a light ray) and I horizontal axis. Independent in
the surface it rebounds against I this context does not mea.., the
(a cushion, a mirror) . same thing as statistically inde
pendent .
• independent, indepen
dence
two events A and B are (statis
ti~ally) independent if the
I chance that they both happen
simultaneously is the product of
tl:ie chances that each occurs in
I dividually; i.e., if P(AB) =
P(A)P(B). This is essentially
equivalent to saying that learn
II = = = = = = = M J J t__ri&s
79
II iruJ,icaturrandmn variable I indirta;s"=rem&n==t=========
mined by Z is independent of ~
MlJthenuJrics======= II
80 indirect proof I inst:ribetl1l1llJk "
~~~~=~~~=.
ally calculated using a formula I • infinitesimal
or a known relationship. I a variable that approaches 0 as
a limit.
• indirect proof
a proof that begins by assum • inflection
ing the conclusion is not true ~ a point of inflection of a plane
and leads to a contradiction of ; curve is a point where the
either the assumption or a pre curve has a stationary tangent,
viously proved theorem. at which the tangent is chang
I ing from rotating in one di
• indirect technique
the method used to determine I rection to rotating in the
oppostie direction.
an indirect measurement.
• inductive reasoning • initial side
I the side that the measurement
the process of observing data,
of an angle starts from ..
recognizing patterns, and maleI
ing conjectures about • injection
generalisations. a onetoone mapping.
• inequality • inscribed
the statement that one quantity I (in a polygon or polyhedron)
is less than (or greater than) Intersecting each side or face
another. of a figure exactly once. Usu
I ally referring to circles in
• infinite
becoming large beyond bound. I scribed in polygons or spheres
inscribed in polyhedrons. The
I figure outside is circum
scribed around the inscribed
figure.
• inscribed angle
an angle formed by two chords
I of the circle with a common
endpoint (the vertex of the
angle).
II instance ofa sentence I intersecting p:es===========""8,,,1
y is called extrapolation.
• interior angles
angles between two lines cut by I
• interquartile range
a transversal. the interquartile range of a list
I of nwnbers is the upper quartile
• interior of a circle minus the lower quartile.
the set of points whose dis I
tance from the center of the I • intersecting planes
planes that share a line
MR.thematics=================== II
"",8"",2========'"",'n"",tersec==~oftwosetSAllndB I i~verseoperation II
I fully be added or subtracted and
I that the mean is a representa
tive measure of central ten
I dency. Such data are common
I in the domain of physical sci
ences or engineering e.g.
lengths or weights .
• intersection of two sets A • invar.iant
andB I an invariant, generally speaking,
the set of elements which are is a quantity that is unchanged
in both A and B. (even though its formula may
I have changed) when one
A.B or A.,,B
The above operation is cOIIum.alive. associative, changes some inessential aspect
anddistribttive of a measurement. For in
AB=BA (AB)C=A.(BC) A(B+C)=AB+AC I stance, Euclidean distance is an
We note that if A. c B, thenAB =A.. Hence invariant under translation or
AA =.'1. {$} A = {$} AF= A rotation of one's coordinate sys
• intersection I tern, and ratio of distances in
the intersection of two or more I the same direction is an invari
sets is the set of elements that ant under affine transforma
all the sets have in common; the I tions. In the morphometrics of
elements contained in every one I triangles, the invariants of a
of the sets. The intersection of : particular transformation are
I
the events A and B is written the shape variables that do not
'1\ and B" and '1\B." C.f. union. I change under that transforma
tion.
• interval scale
a characteristic of data such that • inverse
the difference between two val a form of conditional; if not p,
ues measured on the scale has I then not q.
the san1e substantive meaning! I • inverse operation
significance irrespective of the examples of inverse operations
common level of the two val I are addition and subtraction,
ues being compared. This im I multiplication and division, ex
plies that scores may meaning
II ===================MJJthemllti&s
II in17eNe sine, cosine, or tangent I isome; 83
~.
•• sional object in one view. The
isometric drawing of a cube
I shows all the edges equal, but
I • isometry
: an isometry is a transformation
of a geometric space that leaves
Mllthemntics==================== II
84 isosceles tetrahedron I isotropic II
=======*
distances between points un
changed. If the space is the Eu I
clidean space of a picture or an
organism, and the distances are
distances between landmarks,
II =======MR.th_ti.cs
II jointprobability distribution 'I Kenda;'S""S""hR""ifJ""esp=ac""e=~=====8~5
tion in all directions implying ~  Jordan matrix
equal variance and zero corre ; a matrix whose diagonal ele
lation between the original : ments are all equal (and non
variables (e.g., axis coordi ~ zero) and whose elements
nates). ; above the principal diagonal are
_ joint probability distri : equal to 1, but all other ele
I
. ments are O.
bution
if Xl, X2, ... , Xk are ran 0 0 0 1 0 0 0
dom variables, their joint I 0 0 0 0 0 0
probability distribution gives 0 0 2 0 0 0 0
the probability of events de ..
termined by the collection of I A:= 0 0 2 0 0 0
random variables: for any col 0 0 1 0 2 0 0
lection of sets of numbers 3 2 0 1 2
{AI, ... ,Ale}, the joint prob I 0 1 0 0 0 0
ability distribution deter I •
mines P( (Xl is in AI) and:  Joule
(X2 is in A2) and ... and (Xk ~ a unit of energy or ~ork.
is in Ak) ). ~ _ jump discontinuity
_ joint probability function ; a discontinuity in a function
a function that gives the prob : where the left and righhand
ability that each of two or ~ limits exist but are not equal to
more random variables takes ; each other.
at a particular value.
;  justify
 Jordan curve : give a logical explanation or in
a simple closed curve. ~ formal proof of a mathemati
; cal situation, computation or
: property.
I
:  Kendall's shape space
~ the fundamental geometric con
I strUction, due to David Kendall,
: underlying geometric
~ morphometrics. Kendall's
MRthmuJri&s======= II
86 kilonIeml I II
~~~~~~~~=*
shape space provides a com I sides are called the vertex
plete geometric setting for I angles. The angles between the
analyses of Procrustes distances pairs of noncongruent sides are
among arbitrary sets of land I called the nonvertex angles.
methods of geometric
!'t~
I
morphometrics are
linearisations of statistical I
II ======~MII""'tiu
87
*================~
reasons . The equation of the ~ • lateral faces
thinplate spline has coeffi ; the faces of the lateral surface
cients Vlh, where h is. a vec : of a prism, or a face of a pyra
tor of the xor ycoordinates of ~ mid that is not a base
the landmarks in a target
~ • lateral surface
fo;m, followed by three D's
; the surface not included in the
(for two dimensional data, : base(s)
four D's for threedimensional I
M..themiJtUs======= II
88 ~awofaverages I law ofla1lJenumhers II
II =======MRthmMtiu
II law ofsines I likelihood ratio test 89
*=================
of successes, n X p, tends to ~ • leastsquares estimates
grow as n grows. The follow ; parameter estimates that
ing tool illustrates the law of : minimise the sum of squared
large numbers; the button ~ differences between observed
toggles between displaying the ; and predicted sample values.
difference between the number ; • leg of a right triangle
of successes and the expected : a side of a right triangle that
number of successes, and the ~ include the 90 degree angle
difference between the percent I
age of successes and the ex : • legs
pected percentage of successes. I (of an isosceles triangle), the
MAthemtltics======= II
90 limit lline~t II
=========*
• limit
the actual area of a region
• line
an undefined term in most de E
L I\. ~ E )
r v
• line of symmetry
the line of reflection of a figure
y I having reflectional symmetry.
i I • line perpendicular to a
plane
t ~1++7~ ·· ~ I
a line perpendicular to every line
in the plane that it intersects (or
anyone of them)
v A
/
IE
• line of best fit I
given a collection of points, a : /
/ .I
/1
line that passes closest to all of ~ .'
them, as measured by some . Li __________.___ j I
given criterion.
I • line segment
• line of reflection two points (the endpoints of the
the line over which every point I line segment) and all the points
of a figure is moved by a reflec I between them that are on the
tion. line containing them. The line
segment connects the points.
I The measure of a line segment
I is its length.
II = = = = = = = M J J d l m u l r i u
of
lilinesyrnmetry Ilineartenn anequa;===========9=1
• line symmetry ~ • linear operation
a figure has line symmetry if ; suppose f is a function or op
there is a least one line that di
: eration that acts on things we
vides the figure into two parts ~ shall denote generically by the
that are mirror images of each ; lowercase Roman letters x and
other. : y, Suppose it makes sense to
~, multiply x and y by numbers
• linear association
two variables are linearly asso I (which we denote by a), and
ciated if a change in one is as I that it makes sense to add
sociated with a proportional things like x and y together, We
change in the other, with the I say that f is linear if for every
. f ' al I number a and every value of x
same constant 0 propornon  d fi hi h C() d C )
· h h h f : an y or w c l' x an 1(y are
lty t roug out t e range · 0 I d efimed , (') C(
1 l' a x x
).IS d efimed
measurement, The corre lanon: d al £:() d ") £:(
coefficient measures the degree I an equ s aX x, an (u x
of linear association on a scale : + y) is defined and equals f(x)
of 1 to 1. ~ + f(y) .
~ • linear pair
• linear combination
a sum of values each multiplied ; 2 supplementary adjacent
: angles whose noncommon
by some coefficient. A linear
combination can be expressed ~ sides form a line.
as the inner product of two vec ~ • linear pair of angles
tors, one representing the data ; two adjacent angles whose dis
and the other a vector of coelli tinct sides lie on the same line.
cients.
• linear equation
ax + By + C = 0 <Angles 1 and 2 form a )
Linear Pair
• linear function
a function that, when applied to
I • linear term of an equation
consecutive whole numbers,
; the term w~th a variabl~, but no
generates a sequence with a
: exponent m an equauon; ex
constant difference between
I ample: By in a linear equation
consecutive terms .
.MiJt"'nu======= II
",,92===========* lineartransfOrmatUm IlocatUm II
• linear transformation I ments, usually bits of geometry
II =======MlJthenuuies
111OCRtion, measureofllongitudinaI7==========~9=3
• location, measure of I sion. An argument is valid if the
a measure of location is a way I conclusion has been arrived at
of summarising what a "typi through deductive reasoning.
cal" element of a list isit is a I
• logistic regression
onenumber summary of a dis I this relates to an experimental
tribution.
I design for predicting a binary
• locus categorical (yes/no) olltome on
the path of a moving point; the the basis of predictor variables
set of all points in a plane satis I measured on interval scales. For
fying some given condition or each of a set of values of the
property. predictor variables, the out
• logarithm I comes are regarded as repre
a logarithm of a number is the I
senting a binomial process,
with the binomial parameter 'p'
exponent to which a given base
I depending upon the value of the
must be raised to produce the
given number. predictor variable. The model
ling accounts for the logarithm
• logarithmic growth I of the odds ratio as a linear
a set of values which are ap I function of the predictor vari
proximated by an equation of able. Fitting is via a weighted
the form y = log b x. I 1eastsquares .
regreSSIOn
• logarithmic notation I method. randomisation tests
use of the symbols "log" or "in" for this purpose have been de
in context. veloped by Mehta & Patel.
• logic • longitudinal study
the study of the formal laws of I a study in which individuals are
reasonmg. followed over time, and corn
pared with themselves at differ
• logical argument I ent times, to determine, for ex
a set of premises followed by ample, the effect of aging on
statements, each of which relies I some measured variable. Lon
on the premises or on previous I gitudinal studies provide much
statements, and ending with a more persuasive evidence about
final statement called a conclu
Mathemlltics==================== II
94 loJl1er bountl I mRBie tour II
=========*
the effect of aging than do cross I numbers. Lo =2, L} =1, Ln =Ln_
sectional studies. I } + Ln_r
• lower bound I • lune
any number below which a the portion of a sphere between
function value may approach I two great semicircles having
but not pass. I common endpoints (including
the semicircles).
• lowest common denorni
nator • magic square
the smallest number that is ex a square array of n numbers
actly divisib.1e by each denomi I such that sum of the n numbers
nator of a set of fractions. in any row, column, or main di
.loxodrome agonal is a constant (known as
I the magic sum).
on a sphere, a curve that cuts
all parallels under the same I
angle. 4 3 8
9 5 1
2 7 6
I • magic tour
if a chess piece visits each square
I of a chessboard in succession,
• ltetromino I this is called a tour of the chess
a tetromino in the shape of the board. If the successive squares
letter L. I of a tour on an n X n chess
I board are numbered from 1 to
• lucas number n A 2, in order, the tour is called
a member of the sequence 2, 1, I a magic tour if the resulting
3, 4, 7, ... where each number is I square is a magic square.
the sum of the previous two
II =======MAtlmnRtics
II magnitude IMannWhitney test 95
*================
• magnitude ~ • manipulatives
the value of a number; its dis ; objects that can be arranged,
tance from the origin : built, and moved around by
~ hand,
• magnitude of a rotation
the amount of rotation in de ~ • MannWhitney test
grees ; this is a test of difference in
: location for an experimen
• magnitude of a translation
the distance between any point ~ tal design involving two
I samples with data mea
and its image
: sured on an ordinal scale or
• major arc ~ better. The test statistic is
an arc whose endpoints form ~ a measure of .<?:rdinal prece
an angle over 180 degrees with ; dence. For each possible
the center of the circle; written : pairing of an observation in
the extra letter is used to dis ~ one group with an observa
tinguish it from a minor arc. ; tion in the alternate group,
: the pair is classified in one
• major axis
the major axis of an ellipse is ~ of three ways according to
it's longest chord. ; whether the difference is
: positive, zero or negative;
~ the numbers in these three
; categories are tallied over
: the randomisation set. The
~ randomisation set is the
I same as that for the Pitman
~ permutation test. This test
Reference srs1ltml
: is generally recommended
~ for comparisons involving
; ordinalscale data but is not
• malfatti circles
three equal circles that are mu : confined to this scaletype.
~ An equivalent formulation
tually tangent and each tangent
to two sides of a given triangle. ; of the test, based upon
: ranking the data and sum
~ ming ranks within groups,
; is the Wilcoxon test.
MsthmuJties======= II
!!!!!!96~~~~~~~~==.MANi~.OVA I mmeimumliltelihootlestimtlte II
.MANOVA ~ world situation is abstracted to
see multivariate analysis of vari I a model, the related math
ance. ematical problem is posed and
I solved, and the mathematical
.mapping
I solution is interpreted back into
making a transformation
the realworld situation as a
• margin of error solution to the realworld prob
a measure of the uncertainty in I lem.
an estimate of a parameter;
I • mathematical notation
unfortunately, not everyone
agrees what it should mean. correct use of labels, symbols,
The margin of error of an esti ~ and abbreviations in a math
I ematics contett.
mate is typically one or two
times the estimated standard I • matrix
error of the estimate. arrangement of pixels
• Markov's inequality • maximum
for lists: If a list contains no the largest of a set of values.
negative numbers, the fraction I
• maximum likelihood
of numbers in the list at least estimate
as large as any given constant the maximum likelihood esti
a>O is no larger than the arith I
mate of a parameter from data
metic mean of the list, divided I
is the possible value of the pa
by a. For random variables: if a rameter for which the chance of
random variable X must be I
observing the data largest. That
nonnegative, the chance that X I
is, suppose that the parameter
exceeds any given constant a>O I is p, and that we observe data
is no larger. than the expected x. Then the maximum likeli
value of X, divided by a. hood estimate of p is estimate
• mathematical model I p by the value q that makes
a mathematical q,bject (such as P(observing x when the value
a geometric fig~re, graph, of p is q) as large as possible.
table, or equation) representing I For example, suppose we are
a realworld situation. In math trying to estimate the chance
ematical modeling the real that a (possibly biased) coin
II = = = = = = = M A . t h e m s r i e s
II mean I meamremmttype 97
*==~~====~~==
lands heads when it is tossed. ~ symbols, if X is an estimator of
Our data will be the number of ; the parameter t, then
times x the coin lands heads in ; • mean, arithmetic mean
n independent tosses of the coin.
: the sum of a list of numbers,
The distribution of the number
~ divided by the number of num
of times the coin lands heads is
i bers. See also average.
binomial with parameters n
(known) and p (unknown). The I.meaning
chance of observing x heads in : a version of a conditional that
I
n trials if the chance of heads in : defmes a term, where the term
a given trial is q is nex qx( lq)n
x. The maximum likelihood es
timate of p would be the value
of q that makes that chance
I.
I is in the antecedent.
~
measure
the amount of openness in an
: angle
largest. We can find that value I
of q explicitly using calculus; it : • measure of an angle
turns out to be q = xjn, the frac I the smallest amount of rotation
tion of times the coin is ob ~ necessary to rotate from one
served to land heads in the n : ray of the angle to the other,
tosses. Thus the maximum ~ usually measured in degrees.
likelihood estimate of the I • measure of an arc
chance of heads from the num ~ the measure of minor arc or
ber of heads in n independent : major arc is the measure of its
tosses of the coin is the observed ~ central angle.
fraction of tosses in which the
~ • measurement type
coin lands heads.
; this is a distinction regarding
• mean : the relationship between a phe
average ~ nomenon being measured and
• mean squared error i the data as recorded. The main
the mean squared error of an : distinctions are concerned with
estimator of a parameter is the ~ the meaningfulness of numeri
expected value of the square of ~ cal comparisons of data (nomi
the difference between the es . nal scale versus ordinal scale
I . al
timator and the parameter. In : versus mterv scale versus ra
Mathematics================= II
98 ;w mangle rmethod ofcomptlrison II
tio scale this is known as I in the list after sorting the list
Stevens' typology), whether the I into increasing order. IT the list
scale of the measurements has an even number of entries,
(other than nominal scale mea I the median is the smaller of the
surements) should be regarded I two middle numbers after sort
as essentially continuous or dis ing. The median can be esti
crete, and whether the scale is mated from a histogram by
bounded or unbounded. I fmding the smallest number
such that the area under the his
• medial triangle
togram to the left of that num
the triangle whose vertices are
I ber is 50%.
the midpoints of the sides of a
2. (of a triangle), a line segment
given triangle.
connecting a vertex to the mid
A I point of the opposite side.
I • member of a set
something is a member (or el
ement) of a set if it is one of
I the things in the set.
8
I • mersenne number
c I a number of the form 2pl
where p is a prime.
• ~edian size • mersenne prime
a SIZe measure based on the re I a Mersenne number that is
peated . median of ~ prime.
interlandmark distances. Used·
in resistantfit methods. ; • method of comparison
: the most basic and important
• median ~ method of determining
1. "Middle value" of a list. The ; whether a treatment has an ef
smallest number suc~ that .at : fect: compare what happens to
least half the numbe.rs m the ~t ~ individuals who are treated (the
are no greater than It. IT the ~t I treatment group) with what
has an odd number of entrIes, : happens to individuals who are
the median is the middle entry ~ not treated (the control group).
I
1\ metric space I minor lire 99
MathBm4tics======= II
100 minor IlI4ds I monic polynomi,lll II
~ a representation of a math
i ematical relationship or situ
: ation.
I
: • modulo
I the integers a and b are said to
I be congruent modulo m if ab
is divisible by m .
• modus ponens
the type of valid reasoning that
• minor axis uses "if P then Q' and the state
the minor axis of an ellipse is ment P to conclude that Q must
its smallest chord. be true.
.mira • modus tollens
a plastic device which is used I the type of valid reasoning that
to determine and complete uses "if P then Q" and state
symmetries by reflecting im ment "not Q' to conclude that
ages and allowing the user to I "not P" must be trUe.
also see through the reflect I • moment
ing surface. the kth moment of a list is the
• mixed variation average value of the elements
variation that contains both di I raised to the kth power; that is,
rect and inverse variation. if the list consists of the N ele
ments xl, x2, ... ,xN, the kth
• mode
I moment of the list is ( xlk' +
for lists, the mode is a most
x2k + xNk )fN. The kth mo
common (frequent) value. A I
ment of a random variable X is
list can have more than one I
the expected value of Xk,
mode. For histograms, a mode
I B(Xk).
is a relative maXImum
("bump"). I ., monic polynomial
a polynomial in which the coef
• model ficient of the term of highest
to create, using concrete ma
I degree is 1.
terials, drawings or symbols;
II =======MJJthem4tics
• monochromatic triangle . ~ randomisation set, sampled
a triangle whose vertices are all ; without replacement, and using
colored the same. : the values of the test statistic to
~ generate an estimate of the
• monohedral tiling ; form of the full randomisation
a tessellation in which all tiles
are c;ongruent. : distribution. This procedure is
~ in contrast to the bootstrap pro
• monomial I cedure in that the sampling of
an algebraic expression consist ~ the randomisation set is with
ing of just one term. : out replacement. An advantage
I of the MonteCarlo test over the
• monotone
a sequence is monotone if its ~ bootstrap is that with succes
terms are increasing or decreas : sive resamplings it converges to
mg. ~ the gold standard form of the
. . ; exact test. An effective necessity
• mon~to~c funCtlO~ . : for the MonteCarloprocedure
a funct10n IS monotone if It only I is a source of random codes or
~ncreases or only de~reases:. f ~ an effective pseudorandom
Increases monotOnIcally (IS : generator.
monotonic increasing) if x > y, ~ .
implies thatf(x) > = f(y). A ; • morphometrlcs
function f decreases monotoni . from the Greek: "morph,"
cally (is monotonic decreasing) I meaning "shape," and
if x > y, implies thatf(x) < = I : "metron '
" meaning "measure
f(y),. A function f is strictly : ment." Schools of
monotonically increasing if x > I morphometrics are
y, implies thatf(x) > f(y), and ~ characterised by what aspects of
strictly monotonically decreas : biological "form" they are con
ing if if x > y, implies thatf(x) ~ cerned with, what they choose
< f(y). ; to measure, and what kinds of
: biostatistical questions they ask
• MonteCarlo test . ~ of the measurements once they
named after the famous SIte of . are made. The methods of this
gambling. casinos. A m~nte ~ glossary emphasise configura
carlo test mvolves generaong a ~ tions of landmarks from whole
random subset of the;
Msth_ties======= II
102 mse(x) =;( (#&')2) I multinomial tlistributiun II
organs or organisms analysed ~ of the bias and SE of the esti
by appropriately invariant bio ; mator: MSE(X) = (bias (X) )2
metric methods (covariances of : + (SE(X))2.
taxon, size, cause or effect with I • multimodal distribution
position in Kendall's shape I a distribution with more than
space) in order to answer bio lone mode. The histogram of a
logical questions. Another sort multimodal distribution has
of morphometries studies tis I more than one "bump."
sue sections, measures the den
sities of points and curves, and .4
uses these patterns to answer I ••
questions about the random ••
processes that may be control
ling the placement of cellular I 4
II =======MathimuJries
II multiple I multi"'ariatemultipleregr~ 103
MIIthernatics======= II
*".
!!::!lO!!::!4=====!!::!m!!::!ul!!::!tm!!::!'!!::!'I1II'!!::!Ut'!!::!te!!::!regremon: I negstiPe binomiRldistrihution II
or more independent vari ~ • nearly normal distribution
abies. ; a population of numbers (a list
• . : of numbers) is said to have a
• multivanate regression I
.. nearly normal distribution if the
the prediction of two or more
I histogram of its values in stan
dependent variables using one
independent variable.. : dard units nearly follows a nor
. ~ mal curve. More precisely, sup
• nadir I pose that the mean of the list is
the point on the celestial : JL and the standard deviation of
~ the list is SD. Then the list is
spehere in the direction down
wards of the plumbline. I nearly normally distributed if,
I • negation
/' I (of a statement) A statement
: that is false if the original state
~ ment is true, and true if the
; original statement is false. The
: negation can usually be made by
~ appropriately adding the word
I not to the statement, or by pre
: ceding the statement with the
phrase "It is not the case that. .
• natural number I .. " A double negation of a state
anyone of the numbers 1,2,3, ment is a negation of the nega
4,5, .... tion of that statement.
• navigational system • negative binomial distribu
compass directions or bearings I tion
in a variety of formats. consider a sequence of inde
pendent trials with the same
II negllti:"enumbQ' I ninepointcircle .. 105
M/lthemtI,tics=======:;;;;;;;;;:' II
~1~06~~~~~~~~~no~M~~~wi~mdm8ni~~I~mJ II
• no causation without ~ alpha criterion level, other
manipulation ; wise not ('nonsignificant').
a slogan attributed to Paul : The commonest conventional
Holland. If the conditions I values for the nominal alpha
were not deliberately manipu I criterion level are 0.05 and
lated (for example, if the situ 0.01.
ation is an observational study ~ • nominal scale
rather than an experiment), it I
this is a type of measurement
is unwise to conclude that :
scale with a limited number of
there is any causal relation I
possible outcomes which cannot
ship between the outcome and I
be placed in any order repre
the conditions. See post hoc I senring the intrinsic properties
ergo propter hoc. of the measurements. Ex
• node amples : Female versus Male;
a description of a point in a net I the collection of languages in
work where it is possible for which an international treaty is
two different segments to share published.
the same endpoints • nomograph
• nominal alpha criterion I a graphical device used for com
level I putation which uses a straight
a publicly agreed value for edge and several scales of num
typel error, such that the I bers.
outcome of a statistical test is I
classified in terms of whether • nonagon
I a ninesided polygon.
the obtained value of alpha is :
extreme as compared with I 0cId_
this criterion level. The fine :
detail of the comparison in
vol ves the tail definition I
policy. The outcome is classi
fied as showing statistical sig
nificance ('significant') if the I
outcome has low alpha as
compared with the nominal
II nf11'Ull9onal number 11I01IIPfJ//rametric:==========~1~O~7
• nonagonal number ~ • nonlinear association
a number of the form n(7n5)/ ; the relationship between two
2. : variables is nonlinear if a
~ change in one is associated with
.nonary
; a change in the other that is
associated with 9
: depends on the value of the
• nonconstant rate of ~ first; that is, if the change in the
change ; second is not simply propor
set of data or table of values : tional to the change in the: first,
in which the amount of the ~ independent of the value of the
dependent variable does not I first variable.
change by a constant value as
the value of the independent I • nonoverlapping regions
variable changes by a constant ~ regions that don't share interior
value. : points
I •
: • nonparametric test
• non~on~ex ~et hi h all I a number of statistical tests
a set 0 POInts In ~ c ~ot 0 ~ were devised, mostly over the
segments connectIng POInts
. I . th of. . perlO. d 19301960 , Wit . h the
the set lie entlre y In e set; .I spec!°fi!C 0 b'Jectlve . 0 f bypass
synonym: concave.
; ing assumptions about sam
• noneuclidean geometries : pIing from populations with
hyperbolic geometry ~ data supposedly conforming
a geometry in which, through ; to theoretically modeled sta
a point not on a line, there are : tistical distributions such as
infinitely many lines parallel ~ the normal distribution. Sev
to the given line. ; eral of these tests were
• noneuclidean geometry : explictly concerned with ordi
solid geometry ~ nalscale data for which mod
I eling based upon continuous
• nonincluded side ~ functions is clearly inappro
the side of a triangle that is not : priate. These tests are implic
included by 2 given angles ~ itly rerandomisation tests.
MIIthenumes======~ II
108 nonpersp:edrtllwing I1UJrmtJl4ppyoximation II
• nonperspective drawing I ciable, the survey suffer from
a threedimensional drawing I large nonresponse bias.
that doesn't use perspective I • nonresponse
• nonregular shape in surveys, it is rare that every
a shape that does not have all lone who is "invited" to partici
sides congruent and all angles I pate (everyone whose phone
. congruent. number is called, everyone who
is mailed a questionnaire, ev
• nonresponse bias
I eryone an interviewer tries to
in a survey, those who re
spond may differ from those I stop on the street . . . ) in fact
who do not, in ways that are responds. The difference be
related to the effect one is try I tween the "invited" sample
ing to measure. For example, I sought, and that obtained, is the
a telephone survey of how nonresponse.
many hours people work is • nonrigid transformation
likely to miss people who are I a transformation that does not
working late, and are there preserve the size and shape of
fore not at home to answer the the original figure.
phone. When that happens, I
• nonvertex angles
the survey may suffer from I (of a kite) The two angles be
nonresponse bias. I tween consecutive noncongru
Nonresponse bias makes the ent sides of a kite.
result of a survey differ sys I
tematically from the truth. • normal
I perpendicular
• nonresponse rate
the fraction of nonresponders ~ • normal approximation
in a survey: the number of I the normal approximation to
nonresponders divided by the : data is to approximate areas
number of people invited to ~ under the histogram of data,
participate (the number sent I transformed into standard
questionnaires, the number of ~ units, by the corresponding
interview attempts, etc.) If the : areas under the normal curve.
nonresponse rate is appre ~ Many probability distribu
tions can be approximated by
II =======MRthemtlBes
II nomudcurve I nomuddistributUm *=========1~O~9
a normal distribution, in the ~ The tool on this page illus
sense that the area under the ; trates the normal approxima
probability histogram is close : tion to the binomial probabil
to the area under a corre ~ ity histogram. Note that the
sponding part of the normal ; approximation gets worse
curve. To fmd the correspond : when p gets close to 0 or 1,
ing part of the normal curve, ~ and that the approximation
the range must be converted ~ improves as n increases.
to standard units, by subtract
I • normal curve
ing the expected value and di : the normal curve is the famil
viding by the standard error.
~ iar "bell curve:," illustrated on
For example, the area under
I this page. The mathematical
the binomial probability his
~ expression for the normal
togram for n = 50 and p =
: curve is y = (2xpi)V2Ex2/2,
30% between 9.5 and 17.5 is
~ where pi is the ratio of the cir
74.~%. To use the normal ap ; cumference of a circle to its
proximation, we transform : d·lameter. (3 14159265 . .. ) ,
the endpoints to standar d ~ and E is the base of the natu
units, by subtracting the ex ; rallogarithm (2.71828 ... ).
pected value (for the Bino Th al .
.I d . bl : e norm curve IS symmet
mla ran om vana e, n X p = .I nc ..
aroun t e pomt x  0 ,and
d h .
15 fior these vaIues 0 f nan· d .. c. al f
d di .di th ul b I pOSItiVe lor every v ue 0 x.
p ) an Vi ng e res t .y : The area under the normal
the ~tandard error (for a BI ~ curve is unity, and the SD of
nOffilal, (n x p x (lp)) 1/2 = ; the normal curve suitably de
3.24 for these values of n an~ : fined, is also unit~ Many (but
p). The area normal approXI ~ not most) histograms con
mation is the area under the ; verte d Into . stan d ar d'units,
.
normaI curve b etween (9 .5: . I c. 11 th
15)/3.24 = 1.697 and (17.5 I approXimate y 10 ow e nor
. ; mal curve.
15)/3 .24 =.77;
0 2 t h at area IS I
73.5%, slightly smaller than : • normal distribution
the corresponding area under I the normal distribution is a
the binomial histogram. See ; theoretical distribution appli
also the continuity correction. : cable for continuous interval
Msth_tUs========== II
lIO
II = = = = = = = = =   M l J t h t m U l M
II null set I oblique triangle 111
*================
distributed by independent cir I
cular normal noise of the same I
variance in the original
digitising plane or space and I
drawn from a single, homoge I
neous population. It is exactly
analogous to the usual assump
tion of "independent identically I
distributed error terms" in con I
ventionallinear models (regres :  oblique
sion, ANOVA) . ~ at an angle that is not a mul
 numerator
in the fraction x/y, x is called the
numerator and y is called the I
denominator. 5
• numerical analysis R
the study of methods for ap
proximation of solutions of ~  oblique line
various classes of mathematical ; a line that has a definite slope
problems including error analy : not equal to zero
I
SIS.
:  oblique prism or cylinder
 oblate spheroid ~ a nonright prism or cylinder
an ellipsoid produced by rotat ~  oblique triangle
ing an ellipse through 360 0 ; a triangle that is not a right tri
about its minor axis. angle.
MR.thematics==================== II
112
=========*
• observational study I • odd number
c.f. controlled experiment. I an integer that is not diVisible
by 2.
• obtuse angle
an angle whose measure is • odds
greater than 90 but less than I the odds in favour of an event
180 degrees. I is the ratio of the probability
I of unfavourable outcomes.
~. ..
Note that odds are not syn
• octant onymous with probability, but
anyone of the 8 portions of ; the two can be converted back
space dtermined by the 3 coor and forth. If the odds in
dinate planes. favour of an event are q, then
• odd function I the probability of the event is
II = = = = = = = M l l t h m u J n e s
II oddsmtio I orderofopemtions 113
*================
 odds ratio ~  open interval
an alternative characterisation ; an interval that does not include
of the parameter 'p' for a bi : its two endpoints.
nomial process is the ratio of ~ "t C
. . f .  °PPOSI e laces
t h e InCIdences 0 the two al I f: th l i ' all I I
ternatives : p/( Ip) ; this : aces at e m par e panes
I "
quantity is termed the odds .  opposIte rays
ratio; the value may range ; two rays with a common end
from zero to infinity. This re : point that form a line
I
lates to a possible view of a : _ opposite side
binomial process as the com  ~ (of an angle of a triangle) The
bined activity of two Poisson I side that is not a side of the
MIIthem4.tics======= II
114 orderedpair lordinatiml II
=========*
cation and division, then addi I • ordinal variable
tion and subtraction. a variable whose possible val
ues have a natural order, such
• ordered pair
I as {short, medium, long},
the two numbers that (called
{cold, warm, hot}, or {O, 1,
coordinates) are used to iden I
2, 3, ... }. In contrast, a vari
tify a point in a plane; written I
able whose possible valufs are
(x, y)
I {straight, curly} or {Arizona,
• ordered pair rule California, Montana, New
a rule that uses ordered pairs York} would not naturally be
to describe a transformation. I ordinal. Arithmetic with the
For example, the ordered pair possible values of an ordinal
rule (x, y) ??(x + h, y + k) de variable does not necessarily
scribes a translation horiwn I make sense, but it does make
tally by h units and vertically by sense to say that one possible
k units. value is larger than another.
• ordered triple • ordinate
the three numbers (called coor I the ycoordinate of a point in
dinates) that are used to iden I the plane.
tify a point in space; written (x,
• ordination
y, z)
a representation of objects with
• ordinal scale I respect to one or more coordi
a measurement type for which I nate axes. There are many kinds
the relative values of data are of ordinations depending upon
defined solely in terms of be I the goals of the ordination and
ing lesser, equalto or greater I criteria used. For example, plot
as compared with other data ting objects according to their
on the ordinal scale. These scores on the first two princi
characteristics may arise from I pal component axes provides
categorical rating scales, or the twodimensional ordination
from converting interval scale best summarising the total vari
data to become ranked data. I ability of the objects in the
original sample space. Biplots
combine an ordination of speci
II = = = = = = = M s J t b m u J t i c s
II orientation I orthographic drawing .. ===========1~1~5
MMhemlJtics===================== II
116 ..orthommnal I operlapping triangles II
• orthonormal I fied, and the outlier is deter
a set of vectors is orthonormal I mined to be spurious. Other
if each has length unity and all wise, discarding outliers can
pairs are orthogonal with re I cause one to underestimate
spect to some relevant matrix, I the true variability of the mea
P, such as the identity matrix. A surement process.
matrix is orthogonal if its rows • overlapping angles prop
(columns) are orthonormal as I
erty
a set of vectors.
I the property that, if two
II =======MRrh_tics
II palindrome I parameter 117
*================
• palindrome ~ • parallel
a positive integer whose digits ; (lines, rays, or line segments),
read the same forward and : lying in the same plane and not
I . .
backwards. : mtersectmg.
• palindromic ~ • parallel lines
a positive integer is said to be ; two or more coplanar lines
palindromic with respect to a : that have no points in com
base b if its representation in ~ mon or are identical (eg, the
base b reads the same from left I same line)
to right as from right to left.
I •parallel planes
• pandiagonal magic square ~ planes that have no points in
a magic square in which all the : common
broken diagonals as well as the
~ • parallelepiped
main diagonals add up to the
I a prism whose bases are
magic constant.
paraIleograms.
• pandigital
a decimal integer is called I
pandigital if it contains each of
the digits·from 0 to 9.
• paraboloid
a paraboloid of revolution is a I
surface of revolution produced by ; • parallelogram
rotating a parabola about its axis. : a quadrilateral in which both
~ pairs of opposite sides are par
I aIle!.
I • parameter
: in general, a parameter is a
~ number (an integer, a deci
I mal) indexing a function. For
MRt'nu======= II
118 parameter I partial warp srores !I
========*
eters, both integers: the I • partial least squares
counts of the degrees of free I partial Least Squares is a mul
dom for the two variances tivariate statistical method for
whose ratio is being tested. In I assessing relationships among
morphometrics, there are I two or more sets of variables
four main kinds of param measured on the same enti
eters: nuisance parameters, ties. Partial Least Squares
which must be estimated to I analyses the covariances be
account for differences not of ~ tween the sets of variables
particular scientific interest; rather than optimizing linear
the geometric parameters, I combinations of variables in
such as shape coordinates, in the various sets. Their compu
which landmark shape is ex tations usually do not involve
pressed; statistical param I the inversion of .
eters, such as mean differ I • partial warp scores
ences or correlations, by
partial warp scores are the
which biological interpreta I
quantities that characterise
tion is confronted with that I the location of each specimen
data; and another set of geo I in the space of the partial
metric parameters, such as
warps. They are a rotation of
partial warp scores or I the Procrustes residuals
Procrustes residuals, in which I around the Procrustes mean
the findings of the statistical configuration. For the nonuni
analysis are expressed.
form partial .warps, the coef
• parameter I ficients for the rotation are the
II ====================MlJthmuJrics
II partial warps I pentomino 119
*=========
• partial warps ~ • patterns in fractals recur
partial warps are an auxiliary ; sive rule
structure for the interpretation : a rule used to fmd terms in num
of shape changes and shape ~ ber sequences using recursion.
variation in sets of landmarks. I ff '
. • payo matrIX
Geometrically, partial warps are
; a way of representing what each
an orthonormal basis for a
player in a game wins or loses,
space tangent to Kendall's shape I
as a function of his and his
space. Algebraically, the partial
~ opponent's strategies.
warps are eigenvectors of the
bending energy matrix that de I • pedal triangle
scribes the net local informa . the pedal triangle of a point P
tion in a deformation along with respect to a triangle ABC
each coordinate axis. Except for I is the triangle whose vertices
MAthem4rics======= II
120 percentile Ipennutation II
=================*
• • perfect square
~ ....
,...... .
I an integer is a perfect square if
it is of the form m 2 where m is
an integer.
• • perimeter of a polygon
• the sum of the lengths of the
sides of the polygon
• percentile • periodic curve
the pth percentile of a list is the • a curve that repeats in a regu
smallest number such that at • lar pattern.
least p% of the numbers in the
Closed (Periodic) Cubic S5pline
list are no larger than it. The • A
pth percentile of a random vari  •
able is the smallest number
B
such that the chance that the •
random variable is no larger •
than it is at least p%. C.f.
quantile.
c:
• perfect cube
an integer is a perfect cube if it • • permutation
is of the form m 3 where m is an a permutation of a set is an ar
integer. rangement of the elements of
• the set in some order. If the
• perfect number set has n things in 'it, there are
a positive integer that is equal I
nl different orderings of its
to the sum of its proper divi
• elements. For the first ele
sors. For example, 28 is per
ment in an ordering, there are
fect because 28 = 1 + 2 + •
n possible choices, for the sec
4+7+14.
• ond, there remain n1 pos
• perfect power • sible choices, for the third,
an integer is a perfect power there are n2, etc., and for the
if it is of the form mn where • nth element of the ordering,
m and n are integers and n> 1. • there is a single choice re
mammg. By the fundamental
II = = = = = = = = = = M l l t b m u J t i & s
II perpendicular I Pitman permutation;==========~1~2~1
rule of counting, the total ~ • perspective
number of sequences is thus ; feeling of depth
nx(n1)x(n2)x .. . xl. ; • perspective drawing
Similarly, the number of : a technique of representing
orderings of length k one can
~ threedimensional relation
form from n> = k things is ; ships realistically in a draw
nx(n1)x(n2)x . . . x(n
: ing, by drawing objects
k+ 1) = n!j(nk)1. This is de
~ smaller as they recede into
noted np k' the number of per
mutations of n things taken k
at a time. C.f. combinations.
• perpendicular
I.
I the distance.
pi
: written 7t the ratio C[D where
~ C is the circumference and D is
intersecting at right angles.
I the diameter of a circle;
P B
I valscale data.
N
• C
MIIthematics================== II
122 pixel I point ofaverages !I
=================*
• pixel I thickness and is therefore con
small dot of color that makes I sidered twodimensional.
up computer'and TV screens
• plane figure
• placebo effect a set bf pO'ints that are on a
the belief or knowledge that one I plane
is being treated can itself have I • plane geometry
an effect that confounds with
I the study of twodimensional
the real effect of the treatment.
figures in a plane
Subjects given a placebo as a I
painkiller report statistically • plane section
significant reductions in pain in I the intersection of a figure with
randomised experiments that I a plane
compare them with subjects I • point
II =======Mnthmultics
II pointofconcurrency I polyhedron 123
*=========
• point of concurrency I improve any prediction of the
the point at which more than I position (time) of the next event
two concurrent lines, line seg by reference to the detail of any
ments, or rays intersect. I number of preceding observa
I tions. The corresponding distri
• point of tangency
bution of intervals between
(of a circle) The single point I
events is an exponential distri
where a tangent line touches a
I bution. The conventional ex
circle.
ample of a Poisson processes is
• Poisson distribution concerned with occurence of
the Poisson distribution is a dis I radioactive emissions in a sub
crete probability distribution stantial sample of radioactive
that depends on one paranleter, with a halflife very much
m. If X is a random variable I longer than the total observa
with the Poisson distribution tion period.
with parameter m, then the
• polarity of a variable
probability that X = k is Em I
x mkjk!, k = 0, 1,2, ... , where the positivity or negativity of a
I variable; its direction
E is the base of the naruralloga
ritlun and ! is the factorial func I • polygon
tion. For all other values of k, I a closed planar geometric fig
the probability is zero. The ex ure consisting of line segments
pected value the Poisson distri I (tlle sides), each of which inter
bution with parameter m is 111, I sects exactly two others at end
and the standard error of the points forming the polygon'S
Poisson distribution with pa angles. Each point of intersec
rameter m is m 1/2 . I tion is a vertex of the polygon.
• Poisson process I • polygonal region
a process whereby t;vents occur the union of a polygon and its
independently in some COI1 lntenor
tinuwn (in many applications, I • polyhedron
time), such that the overall den I a solid whose surface consists
sity (rate) is statistically con of polygons and tlleir interiors,
stant but that it is impossible to I each of which is a face. A line
Mathematics===================== I
124 P01,;,i1W I population stand4rd deviation II
segment where two faces inter I of a box of numb.ered tickets
sect is an edge. A point of in I is the mean of the list com
tersection of three or more prised of all the numbers on
edges is a vertex. I all the tickets. The population
I mean is a parameter. C.f.
 polyomino :
sample mean.
a planar figure consisting 0 f I
congruent squares joined edge :  population percentage
toedge. the percentage of units in a
I population that possess a speci
registered as Republicans. If
each unit that possesses the
I property is labelled with "1,"
II = = = = = = = M J J t__#U
II populatum I preshape space 125
*=================
• population I provement of performance of
a collection of units being stud I a statistical test.
ied. Units can be people, places,
I • practical number
objects, epochs, drugs, proce
a practical number is a positive
dures, or many other things.
I integer m such that every natu
Much of statistics is concerned
I ral munber n not exceeding m
with estimating numerical
is a sum of distinct divisors of
properties (parameters) of an I
entire population from a ran m.
dom sample of units from the • precision
population. I the closeness of repeated mea
surements to the same value.
• post hoc ergo propter hoc I
• power I • premises
this is the probability that a (in a deductive sy~tem) State
statistical test will detect a I ments (including undefined
defined pattern in data and I terms, definitions, properties of
declare the extent of the pat I algebra and equality, postulates,
tern as showing statistical sig and theorems) used to prove
nificance. power is related to I further conclusions.
type2 error by the simple for I • preshape space
mula: power = (Ibeta) ; the the space corresponding to fig
motive for this redefinition is I
ures that have been centered
so that an increase in value for I and scaled but not rotated to
power shall represent im alignment. It is of k(Pl)1 di
mensions.
Mathematics===================== II
""12""6==========,,,,p"=·nu;aCie I p1incipal components atlalysis I
• prima facie I its effect on a circle or sphere.
latin for "at tirst glance." "On I An affine transformation takes
the face of it." Prima facie evi circles into ellipses. The princi
dence for something is infor I pal axes of the shape change are
mation that at first glance sup I the directions of the diameters
ports the conclusion. On closer of the circle that are mapped
examination, that might not be into the major and minor axes
true; there could be another I of the ellipse. The principal
explanation for the evidence. strains of the change are the
ratios of the lengths of the axes
• prime I of the ellipse to the diameter of
a prime number is an integer
the circle. In the case of the tet
larger than 1 whose only posi I
rahedron, there are three prin
tive divisors are 1 and itself.
I cipal axes, the axes of the ellip
• prime factorisation soid into which a sphen: is de
the unique set of factors of a formed. One has the greatest
number, all of which are prime I principal strain (ratio of axis
numbers. I length to diameter of sphere),
• primitive pythagorean one the least, and there is a third
triangle perpendicular to both, having
a right triangle whose sides are I an intermediate principal strain.
relatively prime integers. I • principal components
• primitive root of unity analysis
the complex number z is a the eigenanalysis of the
primitive nth root of unity if ~ sample covariance matrix.
Zll = 1 but Zk is not equal to 1 for ; Principal components (PC's)
any positive integer k less than can be defined as the set of
I vectors that are orthogonal
n.
both with respect to the iden
• principal axes and strains tity matrix and the sample
a change of one triangle into I covariance matrix. They can
another, or of one tetrahedron also be defined sequentially:
into another, can be modelled the first is the linear combi
as an affine transformation I nation with the largest vari
which can be parameterised by
II ====================MiJthemllhcs
II principal wa1'"Ps I prism.. 12 7
MRthemilties=================== II
""12,,,,8========== .probability I probabilitydistrilmtilm "
II = = ; ; ; ; ; ; = = = = = = = M a t "  t i c s
II probability wmts I probability} th~0""if==========1""2,,,,9
8 5/36 ~ • probability sarr.ple
9 4/36 I a sample drawn from a popula
10 3/36 : tion using a random mecha
11 2/36 ~ nism so that every element of
12 1/36 ; the population has a known
The probability distribution of : chance of ending up in the
a continuous random variable ~ sample.
can be characterised by its prob
~ • probability, theories of
ability density ftillction.
I a theory of probability is a way
• probability events : of assigning meaning to prob
the set of all possible outcomes ~ ability statements such as "the
of an experiment is the sample I chance that a thumbtack lands
0.25
; cumstances. According to the
: subjective theory of probabil~
Mslt'tics======= "
=13=0========'1'== ;ocrusteidistance Iprocrustes methods 11
0% to 100%, with 0% indicat I most subsequent morphomet
ing that we are completely I ric analyses .
sure it won't occur, and 100%
I • procrustes methods
indicating that we are com
a term for leastsquares meth
pletely sure that it will occur. I
ods for estimating nuisance
According to the theory of :
parameters of the Euclidean
equally likely outcomes, if an ~ similarity transformations.
experiment has n possible out I
The adjective "Procrustes"
comes, ana (for example, by
I refers to the Greek giant who
symmetry) there is no reason
would stretch or shorten vic
that any of the n possible out I
tims to fit a bed and was first
comes should occur preferen
I used in the context of super
tially to any of the others, then
imposition methods by Hudey
the chance of each outcome is I
and Cattell, 1962, The
100%jn. Each of these theo
I Procrustes program: produc
ries has its limitations, its pro
ing a direct rotation to test an
ponents, and its detractors. hypothesised factor structure,
• procrustes distance I Behav. Sci. 7:258262. Mod
11 =    = = = = = = M i J t h m u l r i e s
II procrustes residttals I proportional 131
*=================
to the construction of mean I • procrustes superimposi
configurations from the super tion
imposition of multiple speci the construction of a twoform
mens using either the stan I superimpOSitIOn by least
dard registrations of Boas' I squares using orthogonal or af
method (Phelps, 1932). The fine transformations.
latter being essentially a Gen
• pronic number
eralized Procrustes Analysis. I
a number of the form n(n+ 1) .
• procrustes residuals
the set of vectors connecting the • proof
I a sequence of justified conclu
landmarks of a specimen to cor
sions used to prove the validity
responding landmarks in the I
of an ifthen statement
consensus configuration after a
Procrustes fit. The sum of • proper divisor
squared lengths of these vectors I the integer d is a proper divisor
is approximately the squared of the integer n if O<d<n and
Procrustes distance between the d is a divisor of n.
specimen and the consensus in I • proper subset
MRthem4tics=================== II
132 proportWnality Ippalue II
=======~.
• proportionality I process. Successive pseudoran
a relationship described by a I dom data are produced by a
constant ratio. fLxed calculation process acting
upon preceding data from the
• proposition, logical
I pseudorandom sequence. To
proposition
start the sequence it is necessary
a logical proposition is a state I
ment that can be either true or to decide arbitrarily upon a first
I datum, which is termed the
false. For example, "the sun is
seed value ..
shining in Berkeley right now" I
is a proposition. • pvalue
suppose we have a family of
• protractor I hypothesis tests of a null hy
a tool used to measure the size
pothesis that let us test the hy
of an angle if'. degrees.
pothesis at any significance level
I p between 0 and 100% we
choose. The P value of the null
hypothesis given the data is the
I smallest significance level p for
I which any of the tests would
have rejected the null hypoth
esis. For example, let X be a test
I statistic, and for p between 0
and 100%, let xp be the small
est number such that, under the
• prove I null hypothesis, P( X < = x )
use logical arguments, defini > = p. Then for any p between
tions, theorems, and properties o and 100%, the rule reject the
to show that a relationship is I null hypothesis if X < xp tests
true for all numbers or specific the null hypothesis at signifi
set of figures. cance level p. If we observed X
I = x, the Pvalue of the null hy
• pseudorandom pothesis given the data would
a source of data which is effec I
be the smallest p such that x <
tively unpredictable although I
xp.
generated by a determinate
II ==========MRth_ties
II pyramid I qef 133
.==============~
• pyramid I squares of the lengths of the
MAthmuJti&s======= II
134 .. qw:ulrangle I quantitative l1ariable 1/
II = = = = = = = M 1 I t__ries
IllJUllrtic polynomial I radical axis 135
*================
quantitative. For example, add ~ est integer that is at least as
ing and subtracting social secu ; big as the number of entries
rity numbers does not make : in the list divided by two. Call
sense. Quantitative variables ~ that integer 1. The lth element
typically have units of measure ; of the sorted list is the me
ment, such as inches, people, or : dian. Find the smallest integer
pounds. ~ that is at least as large as the
1 number of entries in the list
 quartic polynomial
~ times 3/4. Call that integer m.
a polynomial of degree 4.
: The mth element of the sorted
 quartiles 1 list is the upper quartile.
there are three quartiles. The ~ Quota Sampling.
first or lower quartile (LQ) of : .. I 'al
. IS
a hst . a number (not neces 1
 qUlnnc po .
ynoml
sarily a number in the list) ~ a polynoffilal of degree 5.
such that at least 1/4 of the :  quotient
numbers in the list are no 1 the result of a division.
larger than it, and at least 3/4 ;_R
of the numbers in the list are · rotation
I
no smaller than it. The second
quartile is the median. The ; r
third or upper quartile (UQ) : radius
is a number such that at least ~ _ radian
3/4 of the entries in the list are 1 a unit of angular measurement
no larger than it, and at least such that there are 2 pi radians
1/4 of the numbers in the list in a complete circle. One radian
are no smaller than it. To find ~ = ISOjpi degrees. One radian
the quartiles, first sort the list ; is approximately 57.30.
into increasing order. Find the
smallest integer that is at least ;  radical axis
as big as the number of entries : the locus of points of equal
1 • h
in the list divided by four. Call : power Wit respect to two
that integer k. The kth ele 1 circle.
.MR.thmulties=================== II
136 radical center I random sample II
========*
 A '"',\.
 I
I
each time a measurement is
made, and behaves like a num
ber drawn with replacement
.
J
\
'\ random experiment
I •
J \
,
, \ an experiment or trial whose
\. outcome is not perfectly pre
I dictable, but for which the long
• radical center run relative frequency of out
the radical center of three circles I comes of different types in re
is the common point of; peated trials is predictable. Note
interesection of the radical axes that "random" is different from
of each pair of circles. "haphazard," which does not
I necessarily imply longterm
• radii
plural form of radius regularity.
II ===================MJJt__ncs
111 llndomVilriable I mndomislltiunset . .
o
137
I'
0 0 0
MIIth_tics================== /I
",1""3,,,,8===========;ndmnisation test I ratio ofsimilitude II
• randomisation test I signed to treatment, and the
the rationale of a randomisation I rest to control. Randomising
test involves exploring re the assignment tends to de
randomisations of the actual I crease confoimding of the treat
data to form the randomisation I ment effect with other factors,
distribution of values of the test by making the treatment and
statistic. the outcome value control groups roughly compa
value of the test statistic is I rable in all respects but the
judged in terms of its relative treatment.
position within the re
randomisation distribution. if I
• range
the range of a set of num bers is
the outcome value is near to one I the largest value in the set mi
extreme of the re nus the smallest value in the set.
randomisation distribution then I
Note that as a statistical term,
it may be judged that it is in the I the range is a single number,
extreme tail of the distribution, not a range of numbers.
with reference to a nominal al I
II = = = = = = = M R . t h e m 4 t i c s
II ratio scale I recursiveform 139
*================
 ratio scale ~  real part
this is a type of measurement ; the real number x is called ther
scale for which it is meaning : eal part of the complex number
ful to reason in terms of dif ~ x+iy where x and y are real and
ferences in scores and also in ; i=sqrt( 1).
terms of ratios of scores. Such
;  real variable
a scale will have a zero point : a variable whose value ranges
which is meaningful in the ~ over the real numbers.
sense that it indicates complete
I • I
absence of the property which :  reclproca
the scale measures. The ratio I the reciprocal of the number x
II ===================MRther_#&s
II rejlectWnsymmetricM'tre I regres~oward the mmn, regression effict 141
MRth_tics======= II
142 regres;, linearngnssWn I regularpyramid II
group . Suppose the father's I zontal axis, the regression line
height is 3SD above the aver I passes through the point of av
age of all the fathers' heights. erages, and has slope equal to
(The SD is the standard devia I the correlation coefficient times
tion of the fathers' heights.) I the SD ofY divided by the SD
Then the son's height is also of X.
~kely to be above the average
• regular hexagon
of the sons' heights, but by I
a sixsided figure whose sides
fewer than 3SD (here the SD is
I are of equal length and whose
the standard deviation of the
angles are of equal measure.
sons' heights). In an hypothesis I
test using a test statistic, the I • regular polygon
rejection region is the set of : a convex polygon whose angles
values of the test statistic for I and sides are all congruent
which we reject the null hypoth·· I • regular polyhedron
eSlS. a polyhedron whose faces are
• regression, linear regres enclosed by congruent, regular
sion I polygons that meet at all verti
II =========i;;;;;;;;;;;;=========MR.thcmancs
II reguJartessellatWn I repeatedmeaJ1t~ ==========14~3
ter of the polygon perpendicu ~ differences. Relative warps can
lar to its plane ; be computed from Procrustes
_ regular tessellation : residuals or from partial warps.
I
an edgetoedge tessellation in :  remote interior angles
which tiles are congruent regu ~ (of the exterior angle of a tri
lar polygons. ; angle), the two interior angles
: that do not share a vertex with
 relative power ~ the e:A'terior angle.
a comparison of two or more
statistical tests, for the same ~  rep digit
experimental design, sample I an integer all of whose digits
size, and nominal alpha crite : are the same.
I
rion value, in terms of the re : _ repeated median
spective values of power. ~ a median of medians. Repeated
_ relative warps I medians are used to estimate
relative warps are principal com ~ some superimposition param
ponents of a distribution of : eters in the resistantfit meth
shapes in a space tangent to ~ ods. For example, the resistant
Kendall's shape space. Theyare ; fit rotation estimate is the me
the axes of the "ellipsoid" oc : dian of the estimates obtained
cupied by the sample of shapes ~ for each landmark, which is, in
in a geometry in which spheres ; tmn, the median of angular dif
are defined by Procrustes dis : ferences between the reference
rance. Each relative warp, as a ~ configuration and the configu
direction of shape change about ; ration being fit of the line seg
the mean form, can be inter : ments defined using that land.:
preted as specifying multiples of ~ mark and the other nl land
one single transformation, a I marks. Repeated medians are
transformation that can often : insensitive to larger subsets of
be usefully drawn out as a thin ~ extremely deviant values than
plate spline. In a relative warps I simple medians.
analysis, the parameter can be .I _ repeatedmeasures
used to weight shape variation ~ this is a feature of an experi
By the geometric scale of shape : mental design whereby sev
I
MJJthematU:s======= II
144 replications I resitlfuU II
~~~~~~~==*
eral observations measured ~ • repunit
on a common scale refer to ; an integer consisting onlyofl's.
the same sampling unit. Iden
tification of the relation of the I • rerandomisation
individual observations to the I the process of generating al
. ternative arrangements of
expenmental design is crucial I given data which would be
to this definition. Examples :
the measurement of water I consistent "\vith the experi
mental design.
Ievel at a particular site on
several systematicallydefined : • rerandomisation statistics
occasions; measurement of I also known as permutation or
reactiontime of an individual : randomisation statistics. These
using right hand and left hand are the specific area of concern
separatel. I of this present glossary.
II =======Msthem4ries
II residualplot I resultant'flC&tor' 145
*================
• residual plot ~ mean is not resistant; the me
a residual plot for a regression ; dian is.
is a plot of the residuals from i • resistantfit superimposi
the regression against the ex tion
planatory variable. ~ superimposition methods that
• residual ; use medianand repeatedme
the difference between a datum : dianbased estimates of fitting
and the value predicted for it by ~ parameters rather than least
a model. In linear regression of I squares estimates. Resistantfit
a variable plotted on the verti : procedures are less sensitive to
cal axis onto a variable plotted ~ subsets of extreme values than
on the horizontal axis, a re I those of comparable least
sidual is the "vertical" distance ~ squares methods. As such, their
from a datum to the line. Re : results may provide a simple
siduals can be positive (if the ~ description of differences in
datum is above the line) or ; shape that are due to changes
negative (if the datum is below : in the positions of just a few
the line). Plots of residuals can ~ landmarks. However, resistant
reveal computational errors in i fit methods lack the welldevel
linear regression, as well as con : oped distributional theory asso
ditions under which linear re ~ ciated with the leastsquares fit
gression is inappropriate, such I ting methods.
as nonlinearity and; • resolution
heteroscedasticity. If linear re : the smallest scale distinguish
gression is performed properly, ~ able by a digitising, imaging, or
the sum of the residuals from
~ display device.
the regression line must be
zero; otherwise, there is a com I • resultant vector
putational error somewhere. ~ the result of combining two vec
: tors. To fmd the resultant vec
• resistant I tor, slide the original vectors so
a statistic is said to be resistant
~ that their tails intersect. The
if corrupting a datum cannot
: resultant vector's tail is the com
change the statistic much. The
~ mon tail. Its head is the image
146
=================*
of the head of one of the vec ~ • right cylinder
tors after you translate it along; a cylinder whose direction of
the other vector. Also known as sliding is perpendicular to the
a vector sum. I plane of the base
I • right prism
I a prism whose direction of slid
ing is perpendicular to the plane
of the base
Light Ray
:~I"""""
• review mandala
a circular design arranged in I
concentric arcs.
• rhombus
a parallelogram with four equi
lateral sides • right triangle
I a triangle with a right angle.
• ridge curve
ridge curves are curves on a sur The side opposite the right
:
face along which the curvature ~ angle is the hypotenuse. The
perpendicular to the curve is a I other two sides are the legs.
local maximum. For instance on I • rigid motion
a skull, the line of the jaw or a motion that preserves shape
the rim of an orbit. and size.
• right angle • rigid rotation
an angle whose measure is 90 I an orthogonal transformation
degrees of a real vector space with re
spect to the Euclidean distance
• right cone
a cone whose axis is perpen I metric. Such transformations
dicular to the plane containing I
leave distances between points
its base and angles between vectors un
I changed. A principal compo
II rmserrorofregression I rutatWn 147
*================~
nents analysis represents a rigid ~ of the elements in the list. It is
rotation to new orthogonal ; a measure of the average "size"
axes. A canonical variates : of the elements of the list. To
analysis does not. ~ compute the rms of a list, you
; square all the entries, average
Y2
: the numbers you get, and take
I
: the squareroot of that average.
I
:  RootMeanSquare Error
I (RMSE)
: the RMSE of an an estimator
 rms error of regression ~ of a parameter is the square
the rms error of regression is ~ root of the mean squared error
the rms of the vertical residu ; (MSE) of the estimator. In sym
als from the regression line. : boIs, if X is an estimator of the
For regressing Y on X, the ~ parameter t, then RMSE(X) =
nns error of regression is ; (E( (Xt)2 ) ) !f2. The RMSE of
equal to (I r2)!f2xSDY, where : an estimator is a measure of the
r is the correlation coefficient ~ expected error of the estimator.
between X and Y and SDY is ; The units of RMSE are the
the standard deviation of the : same as the units of the estima
values of Y. ~ tor.
 RNG ~  rotation
acronym for Random Number ; in effect, a rotation is a turning
Generator. This is a process : of the plane about a point (the
which uses a arithmetic algo ~ center of rotation) by an angle
rithm to generate seyuences of I (the angle of rotation). For
pseudorandom numbers. : mally, a rotation is an isometry
~ that is the composition of re
 root of unity I flections through two lines that
a solution of the equation xn = I,
~ intersect at the center of the
where n is a positive integer.
: rotation. The angle of rotation
 rootmeansquare (rms) ~ has twice the measure of the
the rms of a list is the square ; smaller angle formed by the
root of the mean of the squares : lines.
Mnth_tics======== II
=l48==~==~====;tUnudsymmetry I samplepercentage II
• rotational symmetry ~ • sample mean
a figure has rotational symme the arithmetic mean of a ran
try if it can be rotated (turned) dom sample from a popula
less than 360 degrees about a I tion. It is a statistic commonly
point so that it appears the same used to estimate the popula
as the original figure. tion mean. Suppose there are
n data, {xl, x2, ... ,xn}. The
• ruled surface sample mean is (xl + x2 + ..
a surface formed by mqving a
straight line (called the genera I
. + xn)jn. The expected value
tor). of the sample mean is the
population mean. For sam
• rusty compass pling with replacement, the
a pair of compasses that are SE of the sample mean is the
fixed open in a given position. population standard devia
tion, divided by the square
• sa root of the sample size. For
surface area
sampling without replace
• sample ment, the SE of the sample
a set of individual units, drawn mean is the finitepopulation
from some definable popula correction ((Nn)j(Nl»V2
tion of units, and generally a times the SE of the sample
small proportion of the popu mean for sampling with re
lation, to be used for a statis I
placement, with N the size of
tical examination of which the the population and !1 the size
findings are intended to be of the sample.
applied to the population. it is I
essential for such inference I • sample percentage
that the sample should be rep the percentage of a random
resentative. in re I sample with a certain prop
randomisation statistics the I erry, such as the percentage of
process of applying inferences voters registered as Demo
based upon the sample to the I crats in a simple random
population is essentially infor I sample of voters. The sample
rna.I I
mean is a statistic commonly
used to estimate the popula
II ===================Msthematies
II samplesize I sample sum .. 149
MnthfR'llBtW======== II
150 .. samplesuwey I samplingerror II
sum is the sample size times I the sampling distribution of an
the population mean. For sam I estimator is the probability dis
pIing with replacement, the tribution of the estimator when
SE of the sample sum is the ~ it is applied to random samples.
population standard devia ; The tool on this page allows you
tion, times the squareroot of : to explore empirically the sam
the sample size. For sampling ~ pling distribution of the sample
without replacement, the SE I mean and the sample percent
of the sample sum is the fi age of random draws with or
nitepopulation correction ~ without replacement draws
«Nn)j(Nl))V2 times the SE I from a box of numbered tick
of the sample sum for sam ~ ets.
pIing with replacement, with : _ lin
N h . f th I· I samp g error
t e Slze .0 e popu atl~n : in estimating from a random
and n the Slze of the samp e. I sample, the difference be
 sample survey tween the estimator and the
a survey based on the responses parameter can be written as
of a sample of individuals, I the sum of two components:
rather than the entire popula I bias and sampling error. The
tion. bias is the average error of the
I estimator over all pos,sible'
 sample variance
I samples. The bias is not ran
the sample variance is the
dom. Sampling error is the
square of the sample standard I
component of error that var
deviation S. It is an unbiased
I ies from sample to sample.
estimator of the square of the
: The sampling error is ran
population standard deviation,
which is also called the variance
~ dom: it comes from "the luck
I of the draw" in which units
of the population.
happen to be in the sample. It
 sample is the chance variation of the
a sample is a collection of units I estimator. The average of the
from a population. sampling error over all pos
 sampling distribution sible samples (the expected
I value of the sampling error)
II sampling unit I s c o r e . . 151
MRthematics======= II
152 sd line I selection bias II
=========*
• sd line I • secular trend
for a scatterplot, a line that I a linear association (trend)
goes through the point of av with time.
erages, with slope equal to the I
ratio of the standard devia • segment
I aka line segment; the set of
tions of the two plotted vari I points consisting of two dis
ables. If the variable plotted tinct points and all inbetween
on the horizontal axis is called I
them.
X and the variable plotted on
the vertical axis is called Y, the • segment of a circle
slope of the SD line is the SD I the region between a chord
of Y, divided by the SD of X. and the included arc.
• se(sample mean)
= n V2xSD(box,where
SD(box) is the standard devia I
II = = = = = = = M 4 t h e m 4 t k s
II selfseIeaitm I set 153
*================
units for the sample; it tends ~  semicircle
to be eliminated by probabil i an arc of a circle whose end
ity sampling schemes in which : points are the endpoints of a
the interviewer is told exactly ~ diamete~ 
whom to contact (with no
~  semimagic square
room for individual choice).
i a square array of n numbers
 selfselection : such that sum of the n numbers
selfselection occurs when in ~ in any row or column is a con
dividuals decide for them I stant (known as the magic
selves whether they are in the ~ sum).
control group or the treat . ".. _ semtre 'gu1ar t esse11a0'on
ment" group" " " " a esse 11 a t"lon conS1S
SelfselectIon IS I t " t'lllg 0 f
qUlte common III studles of i l l all f h
human behaviour. For ex : reg~ ar PI? ygons hOw" ose
" f h ffi f I vertIces le on ot er vertIces,
amp1e, stud les 0 tee ect "0 : an d"tn W h"lC h every vertex IS "
ki h h alth
smo ng on ~an "e " . tn I surrounded by the same ar
r
instance, what happens when : (of an angle) The two rays, hav
you slide the top of a square ~ ing a common endpoint, that
sideways without altering its ~ form an angle.
vertical position or the length
of the horizontal edges. The I • Sierpinski triangle
MJr.themtlties======~ II
!!!!:1!!!!!56~~~~~~~~~~:laritytransjimJmtiun lsimulatiun II
corresponding sides are propor I • simple random sample
tional. I a simple random sample of n
units from a population is a ran
6
I dom sample drawn by a proce
L
I dure that is equally likely to
Mathematics=================== II
158 slope I sptUe II
• slope I • slopeintercept form
the ratio of the increase in the I the form of a linear equation y
yvalues to the increase in the = fiX + b where m represents
xvalues between any two or I the slope and b represents the
dered pairs. I yintercept.
• slope of a line I • small circle
in a coordinate plane, the the circle formed by the inter
amount of vertical change : sectiCJn of a sphere and a plane
(change in y) for each unit of I that doesn't contain the center
horizontal change (change inx). ; • solid
The slope of a vertical line is the union of the surface and the
undefined. You can calculate the I
region of space enclosed by a 3
slope m of a line (or line seg I D figure; examples: conic solid,
ment) through points with co cylindric solid, rectangular solid
ordinates (xl,yl) and (x2,y2) I
using the formula m = • solid geometry
(Y2  Yl) the study of figures in threedi
I mensional space
II ====================Mathematics
II sphere IsttuuJ,srd units 159
*================
generalisation of these intuitive ~ of deviations between each ele
strUctur~s. Examples are: Eu ; ment of the set and the mean
clidean spaces, sample spaces, : of the set.
I
shape spaces, linear vector : • standard error
spaces, etc. ~ the Standard Error of a random
• sphere ; variable is a measure of how far
the locus of pointsin threespace : it is likely to be from its ex
that are a fIXed distance froma ~ pected value; that is, its scatter
given point (called the center). I in repeated experiments. The
:I SE of a random variable X is
.sphericru oigonometty
the branch of mathematics deal : defmed to be SE(X) = [E( (X
I E(X))2 )] V2. That is, the stan
ing with measurements on the
~ dard error is the squareroot of
sphere.
: the expected squared difference
• square I between the random variable
a quadrilateral with 4 equal ; and its expected value. The SE
sides and 4 right angles. : of a random variable is analo
~ gous to the SD of a list.
• square free
an integer is said to be square ~ • standard form
free if it is not divisible by a ; the form of a number expressed
perfect square, n2, for n> 1. : as a sum of products involving
• square number ~ powers of ten.
a number of the form n2 • ~ • standard units
• squareroot law ; a variable (a set of data) is said
the SquareRoot Law says that: to be in standard units if its
the standard error (SE) of the ~ me~n .is z.ero and its standard
sample sum of n random draws ~ dCV1anon IS ~ne. You transfor.m
with replacement from a box of ; a set of dat~ mto standard uruts
tickets with numbers on them : by subtracnng the mean from
IS I each element of the list, and di
~ viding the results by the stan
• standard deviation : dard deviation. A random vari
the standard deviation of a set ~ able is said to be in standard
of numbers is the rms of the set
Mathcmatics======= II
160 . standa1"1lise I stmtifted II
units if its expected value is zero I scale versus interval scale ver
and its standard error is one. I sus ratio scale.
You transform a random vari
I • straight angle
able to standard units by sub an angle whose measure is 180
tracting its expected value then I
degrees, forming a line with its
dividing by its standard error.
I sides
• standardise
to transform into standard I • straightedge
a tool used to construct straight
units.
lines.
• statistic : • straightedge, unmarked
a number that can be computed I just how it sounds, an un
from data, involving no un : marked tool used to draw
known parameters. As a func I straight lines
tion of a random sample, a sta
tistic is a random variable. Sta • stratified
tistics are used to estimate pa I this is a feature of an experi
rameters, and to test hypoth mental design whereby a
eses. scheme of observations is re
I peated entirely using further
• stem and leaf plots sets (strata) of experimental
a method of displaying data I units, with each such further
where the leading digit (s) are I set distinguished by a level of
the stem and the ending single a categorical variable which is
dio1ts are arranged in ascend I
0 distinct from any categorical
ing order to the side represent I variables used to define the
ing the leaves.
I experimnad design within a
Msth_ti&s======= \I
162 ~ ~tslmbjea I superimposifitm II
~~~~~~~~. .
be approximated accurately ~ ment of the subset must be
by the normal curve. How ; long to the original set, but
ever, for small and intermedi not every element of the origi
ate values of n, Student's t 1 nal set need' be in a subset
curve with n I degrees of free I (otherwise, a subset would
dom gives a better approxima always be identical to the set
tion. That is, P(a < T < b) is it came from).
approximately the area under I . ••
Student's T curve with n I :  successlve apprOXimatiOn
1 a sequence of approximations,
degrees of freedom, from a to· h cl th
b. Student's t curve can be ; eac one oser to e desired
: value,
used to test hypotheses about
the population mean and con ~  sufficient condition
struct confidence intervals for 1 a version of a conditional that
the population mean, when tells you when you can use the
the population distribution is term defined, where the term
known to be nearly normally 1 is in the consequent; a condition
distributed. This page con 1 that implies a preset conclusion.
tains a tool that shows _ superimposition
Student's t curve and lets you ~ the transformation of one or
find the area under parts of the 1 more figures to achieve some
curve. I geometric relationship to an
_ subject, experimental other figure. The transforma
subject I tions are usually affine trans
MnthemRncs============ II
164 systematic random sample I tltm II
=================*
everything measured with it I they were random, if the or
will appear to be longer than I der in which the units appears
it really is (ignoring random in the list is haphazard. Sys
error) . If your watch runs I tematic samples are a special
fast, every time interval you I case of cluster samples.
measure with it will appear to I • t test
be longer than it really is an hypothesis test based on ap
(again, ignoring random er I
proximating the probability his
ror). Systematic errors do not
I togram of the test statistic by
tend to average out. Student's t curve. t tests usually
• systematic random sample are used to test hypothes.es
a systematic sample starting I about the mean of a population
at a random point in the list when the sample size is inter
ing of units in the of frame, mediate and the distribution of
instead of starting at the first I the population is known to be
unit. Svstematic random sam nearly normal.
pIing is better than systematic • t2 statistic
sampling, but typically not as I a multivariate generalisation of
good as simple random sam I the univariate t 2 statistic. It is
piing. I t he square 0 f t h e ratIo
. 0 f th e
• systematic sample group mean difference to the
a systematic sample from a I standard error of that differ
frame of units is one drawn by I ence. Used in the 'Ptest.
listing the units and selecting I • t 2 test
every kth element of the list. a test due to Hotelling for com
For example, if there are N I
paring an observed mean vec
units ifi the frame, and we I tor to a parametric mean; or
want a sample of size N/I0, comparing the difference be
we would take every tenth I tween two mean vectors to a
unit: the first unit, the eJev I parametric difference (usually
enth unit, the 21st unit, eLc. the zero vector). If the obser
Systematic samples are not I
vations are independently mul
random samples, but they of I tivariate normal, then the 'P
ten behave essentially as if :
II = = = = = = = M R . t h e m R . t i c s
II :able ojJlalues I tangent space \
I 165
*=================
test may be used to test null I actual outcome as part of the
hypotheses using the Fdistribu I tail.
tion. T2 is also closely related I • tangent
to Mahalanobis D2.
(of an acute angle) The ratio of
• table of values I the length of the opposite side
a table of two colLUnns, the first I to the length of the adjacent side
representing values of the inde in any right triangle containing
pendent variable, the second the angle.
representing the values of the I • 1
: . tangent clr~ es
dependent variable. I circles that are tangent to the
MRth_ties================== II
166 tangmtitU veIoeity I tensor II
========*
space to S at P is a linear space I it left the circle along a tangent
T having points with the same I line.
"names" as the points in S and
I • tautology
in which the metric on S "in the a sentence that is true because
vicinity of P" is very nearly the I of its logical structure.
ordinary Euclidean metric on T.
One can visualise T as the pro I • tensor
jection of S onto a "tangent an example of a tensor in
plane" "touching" at P just like morphometrics is the represen
a map is a projection of the sur tation of a uniform component
face of the earth onto flat pa I of shape change as a transfor
per. In geometric I mation matrix. The transforma
morphometrics, the most rel tion matrix assigns to each vec
evant tangent space is a linear I tor in a starting (or average)
vector space that is tangent to form a vector in a second form.
Kendall's shape space at a point A rigorous, general definition
corresponding to the shape of ~ of a tensor would be beyond
a reference configuration (usu . the scope of this glossary, but a
ally taken as the mean of a reasonably intUltIve
sample of shapes). If variation I characterisation comes from
in shape is small then Euclid I Misner, Thorne, and Wheeler,
ean distances in the tangent Gravitation (Freeman, 1973): a
space can be used to approxi I tensor is a "geometric machine"
mate Procrustes distances in I that is fed one or more vectors
Kendall's shape space. Since the in an arbitrary Cartesian coor
tangent space is linear, it is pos dinate system and that produces
sible to apply conventional sta I scalar values (ordinary decimal
tistical methods to study varia numbers) that are independent
tion in shape. of that coordinate system. In
I morphometrics, these "num"
• tangential velocity bers" will be ordinary geomet
(0f an object moving in a circle) I
ric entities like lengths, areas,
The speed of the moving object I
in the direction it would take if : or angles: anything that doesn't
I change when the coordinate sys
tem changes. For the represen
II=======MII~
II terminal side I tetrahedron 167
*=================
tation of a uniform component ~.  tessellation
as a transformation matrix, the I an arrangement of shapes
"scalars" of the Misner.:rhorne : (called tiles) that completely
Wheeler metaphor arc the ~ LOvers a plane without overlaps
lengths of the resulting vectors I or gaps.
and the angles among them. A
different tensor representing
the same uniform transforma I
tion is the relative metric tensor,
which you probably know as the
ellipse of principal axes and I
principal strains. This tensor
produces the necessary numeri
cal invariants (distances in the ~
second form as a function of ; . .
.  test statistic
coordmates on the first form) : . . d h th
directly. Other tensors include ~ a sta~ch use hto . test ypo ~
the metric tensor of a curving ; eses. dYPOb t des1s'dt~st can
, di . constructe y eCl 109 to re
sun.ace Which expresses stance·. th ull h h' h th
on the surface as a function of ~ Ject e n ypot eSls.w. e~ . e
the parameters 10 . wh'1Ch sunc.ace I. value of the test statistic. IS 10
. d d h some range or collection of
pomts are expresse an t e T . h
cUrPature tensor of the same sur I ran~es. J.~ ~et a test WIt a
c.
lace Which expresses teh way·10 I speCified slgmficance level, the .
· h the surlace
w h1C C.
lauS awa y"
"C._II chance
. when the null hvpothes1s
. .
fi' I I IS true that the test statistic falls
r~m itS tangent p ane at any ~ in the range where the hvpoth
pomt. . wo uld be rqecte
: eS1S . dmust ' be
 terminal side ~ at most the specified signifi
the side that the measurement ; cance level. The Z statistic is a
of an angle ends at : common test statistic.
I
 tesselate :  tetrahedron
the ability of a regIOn to ~ a polyhedron with four faces.
tessalate ; The regular tetrahedron is one
: of the Platonic solids.
I
MR.,hmuJhu======= II
168
II =======MRthmulriu
hometrics
II tied rilniu I trlUlitiolUJl mM1 • =========1=6=9
MRtm#Cs==================== II
=17=O=========~talnumber I transformation II
in that even though the dis I formations are used to put
tances or measurements are I variables in standard units . In
defined to record biologically that case, you subtract the
meaningful aspects of the or I mean and divide the results by
ganism, but the geometrical I the SD. This is equivalent to
relationships between these multiplying by the reciprocal
measurements are not taken of the SD and adding the
into account. Traditional I negative of the mean, divided
morphometrics makes no ref by the SD, so it is an affine
erence to Procrustes distance transformation. Affine trans
or any other aspect of ~ formations with positive mul
Kendall's shape space. . tiplicative constants have a
• transcendental number simple effect on the mean,
I median, mode, quartiles, and
a number that is not algebraic.
other percentiles: the new
• transformation value of any of these is the old
transformations turn lists into lone, transformed using ex
other lists, or variables into I acdy the same formula. When
other variables. For example, the multiplicative constant is
to transform a list of tem I negative, the mean, median,
peratures in degrees Celsius I mode, are still transformed by
into the corresponding list of : the same rule, but quartiles
temperatures in degrees Fahr ~ and percentiles are reversed:
enheit, you multiply each ele I the qth quantile of the trans
ment by 9/5, and add 32 to formed distribution is the
each product. This is an ex transformed value of the 1
ample of an affine transforma I qth quantile of the original
tion: multiply by something distribution (ignoring the ef
and add something (y = ax + fect of data spacing). The ef
b is the general affine trans I feet of an affine transforma
formation of X; it's the familtion on the SD, range, and
iar equation of a straight line). IQR, is to make the new value
In a linear transformation, I the old value times the abso
you only multiply by some I lute value of the number you
thing (y = ax). Affine trans multiplied the first list by:
II tmnsformationnotation I treatment~==========17=1
~ of the translation), and have the
; same length (the distance of the
: translation) .
I
:  translation vector
~ see translation
~  transversal
; a line that intersects 2 others
;  transversible
2 a network in which all arcs can
~
what you added does not affect be traced without going over
them. lone more than once
II =======MlJtbem.ries
II tromino I twosided hypothesis test 173
*================
 tromino ~ _ twopoint shape coordi
a threesquare polyomino. nates
a conveniel1t system of shape
'. truncated pyramid
coordinates, originally Francis
part of a pyramid remaining
I Galton's, rediscovered by
after truncating the vertex
Bookstein, consisting (for two
with a plane parallel to the
~ dimensional data) of the coor
base.
I dinates of landmarks 3, 4, ...
 twin primes : after forms are rescaled and
I ..
two prime numbers that differ : repositloned so that landmark
by 2. For example, 11 and 13 I 1 is fixed at (0,0) and landmark
are twin primes. ~ 2 is fixed at (1, 0) in a Cartesian
_ twocolumn proof : coordinate system. Also re
~ ferred to as Bookstein coordi
a form of proof in which each
statement in the argument is ; nates or Bookstein's shape co
written in the left column, and : ordinates.
I
the reason for each statement :  twosided hypothesis test
is written directly across from ~ c.f. onesided test. An hypoth
it in the right column. ; esis test of the null hypothesis
_ twocolumn proof : that the value of a parameter,
a form of proof in which each ~ /L, is equal to a null value, /LO,
I designed to have power against
statement in the argument is
written in the left column, and : the alternative hypothesis that
the reason for each statement ~ either /L < /LO or /L > /LO (the
I alternative hypothesis contains
is written directly across from
it in the right column. ~ values on both sides of the null
: value). For example, a signifI
 twodimensional I cance level 5%, twosided z test
having both width and length, ~ of the null hypothesis that the
but no thickness : mean of a population equals
 twopoint perspective ~ zero against the alternative that
a method of perspective draw ; it is greater than zero would
ing that uses two vanishing : reject the null hypothesis for
points. ~ values of
1\
174
=================*
~ true. A 1YPe 2 error occurs if
(sample mean) I ; the null hypothesis is not re
Izl= : jected when it is in fact false.
I
> 1.96.SE(sample mean)
• type 2 landmark
I a mathematical point whose
• twoway table I claimed homology from case to
a representation of suitable data case is supported only by geo
in a table organised as rows and metric, not histological, evi
columns, such that the rows rep I dence: for instance, the sharp
resent one scheme of alterna est curvature of a tooth.
tives covering the whole of the
• type 3 landmark
the data represented, the col I
a landmark having at least one
umns represent a further
I deficient coordinate, for in
scheme of alternatives covering
stance, either end of a longest
the whole of the data repre I
diameter, or the bottom of a
sented, and the entries in the I
concavity. Type 3 landmarks
twoway table are the counts of :
characterise more than one re
numbers of observations con I
gion of the form. The multi
forming to the respective cells I variate machinery of geometric
of the twoway classification.
I morphometries permits them
MIIthmuJtics========== II
""1,,,,76===========* unilateralsuiftue I unitarydi'Pisor \I
_ unimodular
 unit square
a square matrix is unimodular I a unit square is a square of side
if its determinant is 1. length 1.
_ union of two sets a and b _ unitary divisor
the set of elements in A, B, or I a divisor d of c is called unitary
both; written AUB I if gcd(d,c/d) = 1.
II ===================MJJth_tics
177
*==~~~~~~~
 unity ~ _ variable
one i a numerical value or a charac
: teristic that can differ from in
 univariate
~ dividual to individual. Variance,
having or having to do with a
; population variance The vari
single variable. Some univariate
: ance of a list is the square of the
techniques and statistics include
~ standard deviation of the list,
the histogram, IQR, mean,
; that is, the average of the
median, percentiles, quantiles,
: squares of the deviations of the
and SD. C.f. bivariate.
~ numbers in the list from their
_ universal statement I mean. The variance of a random
a conditional that uses the words ~ variable X, Var(X), is the ex
'all' or 'everything' : pected value of the squared dif
_ universe I ference between the variable
I • vertical angles
nonadjacent angles formed by
the intersection of two lines .
• vertical line
I a line that goes straight up
and down, and whose slope is
defined as infinite or unde
• vertex I fined .
(of a polygon) The point of in I • vertices
tersection of three or more I plural form of vertex; the point
edges. of intersection of the rays of an
• vertex angles I angle, "corner" point of any
(of a kite) The angles between I
geometric figure bounded by
the pairs of congruent sides. lines, planes, or lines and
planes.
II =======M#them4ties
11l'iew I WU~ test 179
Mtldlmufriu======;;;;;;;;;;; II
180 wintling number I xpentoltlitllJ II
of the differences. The ranks are I proceeds to fmd additional fac
I • zintercept
~ the point at which a line crosses
: the zaxis.
I
: • zone
~ the portion of a sphere between
• z statistic ; two parallel planes.
a Z statistic is a test statistic whose ; • zscore
distribution under the null hy : the observed value of the Z sta
pothesis has expected value zero I risric.
and can be approximated well by
the normal curve. Usually, Z sta I • ztest
tistics are constructed by ; an hypothesis test based on ap
standardising some other statis : proximating the probability his
tic. The Z statistic is related to ~ togram of the Z statistic under
the original statistic by Z = ; the null hypothesis by the nor
(original expected value of origi : mal curve.
I