When we have only two samples we can use the t-test to compare the means oI the samples

but it might become unreliable in case oI more than two samples. II we only compare two
means, then the t-test (independent samples) will give the same results as the ANOVA.
It is used to compare the means oI more than two samples. This can be understood better with
the help oI an example.
ONE WAY ANOVA
EXAMPLE: Suppose we want to test the eIIect oI Iive diIIerent exercises. For this, we recruit
20 men and assign one type oI exercise to 4 men (5 groups). Their weights are recorded aIter
a Iew weeks.
We may Iind out whether the eIIect oI these exercises on them is signiIicantly diIIerent or not
and this may be done by comparing the weights oI the 5 groups oI 4 men each.
The example above is a case oI one-way balanced ANOVA.
It has been termed as one-way as there is only one category whose eIIect has been studied
and balanced as the same number oI men has been assigned on each exercise. Thus the basic
idea is to test whether the samples are all alike or not.
WHY NOT MULTIPLE T-TESTS?
As mentioned above, the t-test can only be used to test diIIerences between two means. When
there are more than two means, it is possible to compare each mean with each other mean
using many t-tests.
But conducting such multiple t-tests can lead to severe complications and in such
circumstances we use ANOVA. Thus, this technique is used whenever an alternative
procedure is needed Ior testing hypotheses concerning means when there are several
populations.
ONE WAY AND TWO WAY ANOVA
Now some questions may arise as to what are the means we are talking about and why
variances are analyzed in order to derive conclusions about means. The whole procedure can
be made clear with the help oI an experiment.
Let us study the eIIect oI Iertilizers on yield oI wheat. We apply Iive Iertilizers, each oI
diIIerent quality, on Iour plots oI land each oI wheat. The yield Irom each plot oI land is
recorded and the diIIerence in yield among the plots is observed. Here, Iertilizer is a Iactor
and the diIIerent qualities oI Iertilizers are called levels.
This is a case oI one-way or one-Iactor ANOVA since there is only one Iactor, Iertilizer. We
may also be interested to study the eIIect oI Iertility oI the plots oI land. In such a case we
would have two Iactors, Iertilizer and Iertility. This would be a case oI two-way or two-Iactor
ANOVA. Similarly, a third Iactor may be incorporated to have a case oI three-way or three-
Iactor ANOVA.
HANE AUSE AND ASSIGNABLE AUSE
In the above experiment the yields obtained Irom the plots may be diIIerent and we may be
tempted to conclude that the diIIerences exist due to the diIIerences in quality oI the
Iertilizers.
But this diIIerence may also be the result oI certain other Iactors which are attributed to
chance and which are beyond human control. This Iactor is termed as 'error. Thus, the
diIIerences or variations that exist within a plot oI land may be attributed to error.
Thus, estimates oI the amount oI variation due to assignable causes (or variance between the
samples) as well as due to chance causes (or variance within the samples) are obtained
separately and compared using an F-test and conclusions are drawn using the value oI F.
ASSUMPTIONS
There are Iour basic assumptions used in ANOVA.
O the expected values oI the errors are zero
O the variances oI all errors are equal to each other
O the errors are independent
O they are normally distributed

Read more: http://www.experiment-resources.com/anova-test.html#ixzz1ciFVUyul






Wby not |ust use tbe `test?
1he tLesL Lells us lf Lhe varlaLlon beLween Lwo groups ls slgnlflcanL Why noL [usL do tLesLs for all
Lhe palrs of locaLlons Lhus flndlng for example LhaL leaves from medlan sLrlps are slgnlflcanLly
smaller Lhan leaves from Lhe pralrle whereas shade/pralrle and shade/medlan sLrlps are noL
slgnlflcanLly dlfferenL MulLlple tLesLs are noL Lhe answer because as Lhe number of groups grows
Lhe number of needed palr comparlsons grows qulckly lor 7 groups Lhere are 21 palrs lf we LesL 21
palrs we should noL be surprlsed Lo observe Lhlngs LhaL happen only 3 of Lhe Llme 1hus ln 21
palrlngs a l03 for one palr cannoL be consldered slgnlflcanL AnCvA puLs all Lhe daLa lnLo one
number (l) and glves us ooe l for Lhe null hypoLhesls


O aslc ldeas
4 1he arLlLlonlng of Sums of Squares
4 MulLllacLor AnCvA
4 lnLeracLlon LffecLs
O omplex ueslgns
4 eLweenCroups and 8epeaLed Measures
4 lncompleLe (nesLed) ueslgns
O Analysls of ovarlance (AnCvA)
4 llxed ovarlaLes
4 hanglng ovarlaLes
O MulLlvarlaLe ueslgns MAnCvA/MAnCvA
4 eLweenCroups ueslgns
4 8epeaLed Measures ueslgns
4 Sum Scores versus MAnCvA
O onLrasL Analysls and osL hoc 1esLs
4 Why ompare lndlvldual SeLs of Means?
4 onLrasL Analysls
4 osL hoc omparlsons
O AssumpLlons and LffecLs of vlolaLlng AssumpLlons
4 uevlaLlon from normal ulsLrlbuLlon
4 omogenelLy of varlances
4 omogenelLy of varlances and ovarlances
4 SpherlclLy and ompound SymmeLry
O MeLhods for Analysls of varlance
A general introduction to ANOVA and a discussion oI the general topics in the analysis oI
variance techniques, including repeated measures designs, ANCOVA, MANOVA,
unbalanced and incomplete designs, contrast eIIects, post-hoc comparisons, assumptions, etc.
For related inIormation, see also Variance Components (topics related to estimation oI
variance components in mixed model designs), Experimental Design/DOE (topics related to
specialized applications oI ANOVA in industrial settings), and Repeatability and
Reproducibility Analysis (topics related to specialized designs Ior evaluating the reliability
and precision oI measurement systems).
See also, General Linear Models and General Regression Models; to analyze nonlinear
models, see Generalized Linear Models.
Basic Ideas
be Purpose of nalysis of ariance
In general, the purpose oI analysis oI variance (ANOVA) is to test Ior signiIicant diIIerences
between means. Elementary Concepts provides a brieI introduction to the basics oI statistical
signiIicance testing. II we are only comparing two means, ANOVA will produce the same
results as the t test Ior independent samples (iI we are comparing two diIIerent groups oI
cases or observations) or the t test Ior dependent samples (iI we are comparing two variables
in one set oI cases or observations). II you are not Iamiliar with these tests, you may want to
read Basic Statistics and Tables.
Why the name analysis of variance? It may seem odd that a procedure that compares means
is called analysis oI variance. However, this name is derived Irom the Iact that in order to test
Ior statistical signiIicance between means, we are actually comparing (i.e., analyzing)
variances.
O 1he arLlLlonlng of Sums of Squares
O MulLllacLor AnCvA
O lnLeracLlon LffecLs
More introductory topics:
O omplex ueslgns
O Analysls of ovarlance (AnCvA)
O MulLlvarlaLe ueslgns MAnCvA/MAnCvA
O onLrasL Analysls and osL hoc 1esLs
O AssumpLlons and LffecLs of vlolaLlng AssumpLlons
See also, Methods Ior Analysis oI Variance, Variance Components and Mixed Model
ANOVA/ANCOVA, and Experimental Design (DOE).
be Partitioning of Sums of Squares
At the heart oI ANOVA is the Iact that variances can be divided, that is, partitioned.
Remember that the variance is computed as the sum oI squared deviations Irom the overall
mean, divided by n-1 (sample size minus one). Thus, given a certain n, the variance is a
Iunction oI the sums oI (deviation) squares, or SS Ior short. Partitioning oI variance works as
Iollows. Consider this data set:
Group 1 Group 2
Cbservat|on 1
Cbservat|on 2
Cbservat|on 3
2
3
1
6
7
3
Mean
Sums of Squares (SS)
2
2
6
2
Cvera|| Mean
1ota| Sums of Squares
4
28

1he means for Lhe Lwo groups are qulLe dlfferenL (2 and 6 respecLlvely) 1he sums of squares wltblo
each group are equal Lo 2 Addlng Lhem LogeLher we geL 4 lf we now repeaL Lhese compuLaLlons
lgnorlng group membershlp LhaL ls lf we compuLe Lhe LoLal 55 based on Lhe overall mean we geL
Lhe number 28 ln oLher words compuLlng Lhe varlance (sums of squares) based on Lhe wlLhlngroup
varlablllLy ylelds a much smaller esLlmaLe of varlance Lhan compuLlng lL based on Lhe LoLal varlablllLy
(Lhe overall mean) 1he reason for Lhls ln Lhe above example ls of course LhaL Lhere ls a large
dlfference beLween means and lL ls Lhls dlfference LhaL accounLs for Lhe dlfference ln Lhe 55 ln facL
lf we were Lo perform an AnCvA on Lhe above daLa we would geL Lhe followlng resulL

MAIN LIILC1
SS df MS I p
Lffect
Lrror
240
40
1
4
240
10
240

008


As can be seen ln Lhe above Lable Lhe LoLal 55 (28) was parLlLloned lnLo Lhe 55 due Lo wltblogroup
varlablllLy (2+24) and varlablllLy due Lo dlfferences beLween means (28(2-2)24)
SS Error and SS Effect. The within-group variability (SS) is usually reIerred to as Error
variance. This term denotes the Iact that we cannot readily explain or account Ior it in the
current design. However, the SS Effect we can explain. Namely, it is due to the diIIerences in
means between the groups. Put another way, group membership explains this variability
because we know that it is due to the diIIerences in means.
Significance testing. The basic idea oI statistical signiIicance testing is discussed in
Elementary Concepts, which also explains why very many statistical tests represent ratios oI
explained to unexplained variability. ANOVA is a good example oI this. Here, we base this
test on a comparison oI the variance due to the between-groups variability (called Mean
Square Effect, or MS
effect
) with the within-group variability (called Mean Square Error, or
Ms
error
; this term was Iirst used by Edgeworth, 1885). Under the null hypothesis (that there
are no mean diIIerences between groups in the population), we would still expect some minor
random Iluctuation in the means Ior the two groups when taking small samples (as in our
example). ThereIore, under the null hypothesis, the variance estimated based on within-group
variability should be about the same as the variance due to between-groups variability. We
can compare those two estimates oI variance via the F test (see also F Distribution), which
tests whether the ratio oI the two variance estimates is signiIicantly greater than 1. In our
example above, that test is highly signiIicant, and we would in Iact conclude that the means
Ior the two groups are signiIicantly diIIerent Irom each other.
Summary of the basic logic of ANOVA. To summarize the discussion up to this point, the
purpose oI analysis oI variance is to test diIIerences in means (Ior groups or variables) Ior
statistical signiIicance. This is accomplished by analyzing the variance, that is, by
partitioning the total variance into the component that is due to true random error (i.e.,
within-group SS) and the components that are due to diIIerences between means. These latter
variance components are then tested Ior statistical signiIicance, and, iI signiIicant, we reject
the null hypothesis oI no diIIerences between means and accept the alternative hypothesis
that the means (in the population) are diIIerent Irom each other.
Dependent and independent variables. The variables that are measured (e.g., a test score)
are called dependent variables. The variables that are manipulated or controlled (e.g., a
teaching method or some other criterion used to divide observations into groups that are
compared) are called factors or independent variables. For more inIormation on this
important distinction, reIer to Elementary Concepts.
ultiactor
In the simple example above, it may have occurred to you that we could have simply
computed a t test Ior independent samples to arrive at the same conclusion. And, indeed, we
would get the identical result iI we were to compare the two groups using this test. However,
ANOVA is a much more Ilexible and powerIul technique that can be applied to much more
complex research issues.
Multiple factors. The world is complex and multivariate in nature, and instances when a
single variable completely explains a phenomenon are rare. For example, when trying to
explore how to grow a bigger tomato, we would need to consider Iactors that have to do with
the plants' genetic makeup, soil conditions, lighting, temperature, etc. Thus, in a typical
experiment, many Iactors are taken into account. One important reason Ior using ANOVA
methods rather than multiple two-group studies analyzed via t tests is that the Iormer method
is more efficient, and with Iewer observations we can gain more inIormation. Let's expand on
this statement.
ontrolling for factors. Suppose that in the above two-group example we introduce another
grouping Iactor, Ior example, ender. Imagine that in each group we have 3 males and 3
Iemales. We could summarize this design in a 2 by 2 table:

Lxper|menta|
Group 1
Lxper|menta|
Group 2
Ma|es


2
3
1
6
7
3
Mean 2 6
Iema|es


4
3
3
8
9
7
Mean 4 8

efore performlng any compuLaLlons lL appears LhaL we can parLlLlon Lhe LoLal varlance lnLo aL leasL
3 sources (1) error (wlLhlngroup) varlablllLy (2) varlablllLy due Lo experlmenLal group membershlp
and (3) varlablllLy due Lo gender (noLe LhaL Lhere ls an addlLlonal source lotetoctloo LhaL we wlll
dlscuss shorLly) WhaL would have happened had we noL lncluded qeoJet as a facLor ln Lhe sLudy buL
raLher compuLed a slmple t LesL? lf we compuLe Lhe 55 lgnorlng Lhe qeoJet facLor (use Lhe wlLhln
group means lqootloq or collopsloq octoss qeoJet Lhe resulL ls 5510+1020) we wlll see LhaL Lhe
resulLlng wlLhlngroup SS ls larger Lhan lL ls when we lnclude gender (use Lhe wlLhln group wlLhln
gender means Lo compuLe Lhose 55 Lhey wlll be equal Lo 2 ln each group Lhus Lhe comblned SS
wlLhln ls equal Lo 2+2+2+28) 1hls dlfference ls due Lo Lhe facL LhaL Lhe means for moles are
sysLemaLlcally lower Lhan Lhose for femoles and Lhls dlfference ln means adds varlablllLy lf we
lgnore Lhls facLor onLrolllng for error varlance lncreases Lhe senslLlvlLy (power) of a LesL 1hls
example demonsLraLes anoLher prlnclpal of AnCvA LhaL makes lL preferable over slmple Lwogroup L
LesL sLudles ln AnCvA we can LesL each facLor whlle conLrolllng for all oLhers Lhls ls acLually Lhe
reason why AnCvA ls more sLaLlsLlcally powerful (le we need fewer observaLlons Lo flnd a
slgnlflcanL effecL) Lhan Lhe slmple t LesL
nteraction Effects
There is another advantage oI ANOVA over simple t-tests: with ANOVA, we can detect
interaction eIIects between variables, and, thereIore, to test more complex hypotheses about
reality. Let's consider another example to illustrate this point. (The term interaction was Iirst
used by Fisher, 1926.)
Main effects, two-way interaction. Imagine that we have a sample oI highly achievement-
oriented students and another oI achievement "avoiders." We now create two random halves
in each sample, and give one halI oI each sample a challenging test, the other an easy test.
We measure how hard the students work on the test. The means oI this (Iictitious) study are
as Iollows:

Ach|evement
or|ented
Ach|evement
avo|ders
Cha||eng|ng 1est
Lasy 1est
10
3
3
10

ow can we summarlze Lhese resulLs? ls lL approprlaLe Lo conclude LhaL (1) challenglng LesLs make
sLudenLs work harder (2) achlevemenLorlenLed sLudenLs work harder Lhan achlevemenL avolders?
nelLher of Lhese sLaLemenLs capLures Lhe essence of Lhls clearly sysLemaLlc paLLern of means 1he
approprlaLe way Lo summarlze Lhe resulL would be Lo say LhaL challenglng LesLs make only
achlevemenLorlenLed sLudenLs work harder whlle easy LesLs make only achlevemenL avolders
work harder ln oLher words Lhe Lype of achlevemenL orlenLaLlon and LesL dlfflculLy lotetoct ln Lhelr
effecL on efforL speclflcally Lhls ls an example of a twowoy lotetoctloo beLween achlevemenL
orlenLaLlon and LesL dlfflculLy noLe LhaL sLaLemenLs 1 and 2 above descrlbe socalled molo effects
Higher order interactions. While the previous two-way interaction can be put into words
relatively easily, higher order interactions are increasingly diIIicult to verbalize. Imagine that
we had included Iactor ender in the achievement study above, and we had obtained the
Iollowing pattern oI means:
Iema|es

Ach|evement
or|ented
Ach|evement
avo|ders
Cha||eng|ng 1est
Lasy 1est
10
3
3
10
Ma|es

Ach|evement
or|ented
Ach|evement
avo|ders
Cha||eng|ng 1est
Lasy 1est
1
6
6
1

ow can we now summarlze Lhe resulLs of our sLudy? Craphs of means for all effecLs greaLly
faclllLaLe Lhe lnLerpreLaLlon of complex effecLs 1he paLLern shown ln Lhe Lable above (and ln Lhe
graph below) represenLs a tbteewoy lnLeracLlon beLween facLors

Thus, we may summarize this pattern by saying that Ior Iemales there is a two-way
interaction between achievement-orientation type and test diIIiculty: Achievement-oriented
Iemales work harder on challenging tests than on easy tests, achievement-avoiding Iemales
work harder on easy tests than on diIIicult tests. For males, this interaction is reversed. As
you can see, the description oI the interaction has become much more involved.
A general way to express interactions. A general way to express all interactions is to say
that an eIIect is modiIied (qualiIied) by another eIIect. Let's try this with the two-way
interaction above. The main eIIect Ior test diIIiculty is modiIied by achievement orientation.
For the three-way interaction in the previous paragraph, we can summarize that the two-way
interaction between test diIIiculty and achievement orientation is modiIied (qualiIied) by
ender. II we have a Iour-way interaction, we can say that the three-way interaction is
modiIied by the Iourth variable, that is, that there are diIIerent types oI interactions in the
diIIerent levels oI the Iourth variable. As it turns out, in many areas oI research Iive- or
higher- way interactions are not that uncommon.