You are on page 1of 19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

Introduction
Bayesian Statistics continues to remain incomprehensible in the ignited minds of many analysts.
Being amazed by the incredible power of machine learning, a lot of us have become unfaithful to
statistics.Ourfocushasnarroweddowntoexploringmachinelearning.Isntittrue?
Wefailtounderstandthatmachinelearningisonlyonewaytosolverealworldproblems.Inseveral
situations,itdoesnothelpussolvebusinessproblems,eventhoughthereisdatainvolvedinthese
problems. To say the least, knowledge of statistics will allow you to work on complex analytical
problems,irrespectiveofthesizeofdata.
In1770s,ThomasBayesintroducedBayesTheorem.Evenaftercenturieslater,theimportanceof
Bayesian Statistics hasnt faded away. In fact, today this topic is being taught in great depths in
someoftheworldsleadinguniversities.
With this idea, Ive created this beginners guide on Bayesian Statistics. Ive tried to explain the
concepts in a simplistic manner with examples. Prior knowledge of basic probability & statistics is
desirable. By the end of this article, you will have a concrete understanding of Bayesian Statistics
anditsassociatedconcepts.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

1/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

TableofContents
1.FrequentistStatistics
2.TheInherentFlawsinFrequentistStatistics
3.BayesianStatistics
ConditionalProbability
BayesTheorem
4.BayesianInference
Bernoullilikelihoodfunction
PriorBeliefDistribution
PosteriorbeliefDistribution
5.TestforSignificanceFrequentistvsBayesian
pvalue
ConfidenceIntervals
BayesFactor
HighDensityInterval(HDI)

Before we actually delve in Bayesian Statistics, let us spend a few minutes


understandingFrequentistStatistics,themorepopularversionofstatisticsmostofuscomeacross
andtheinherentproblemsinthat.

1.FrequentistStatistics
Thedebatebetweenfrequentistandbayesianhavehauntedbeginnersforcenturies.Therefore,itis
important to understand the difference between the two and how does there exists a thin line of
demarcation!
Itisthemostwidelyusedinferentialtechniqueinthestatisticalworld.Infact,generallyitisthefirst
schoolofthoughtthatapersonenteringintothestatisticsworldcomesacross.
FrequentistStatisticstestswhetheranevent(hypothesis)occursornot.Itcalculatestheprobability

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

2/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

of an event in the long run of the experiment (i.e the experiment is repeated under the same
conditionstoobtaintheoutcome).
Here, the sampling distributions of fixed size are taken. Then, the experiment is theoretically
repeated infinite number of times but practically done with a stopping intention. For example, I
perform an experiment with a stopping intention in mind that I will stop the experiment when it is
repeated1000timesorIseeminimum300headsinacointoss.
Letsgodeepernow.
Now,wellunderstandfrequentiststatisticsusinganexampleofcointoss.Theobjectiveisto
estimatethefairnessofthecoin.Belowisatablerepresentingthefrequencyofheads:

Weknowthatprobabilityofgettingaheadontossingafaircoinis0.5. No.ofheads representsthe


actualnumberofheadsobtained. Difference isthedifference
between 0.5*(No.oftosses)no.ofheads .
An important thing is to note that, though the difference between the actual number of heads and
expected number of heads( 50% of number of tosses) increases as the number of tosses are
increased, the proportion of number of heads to total number of tosses approaches 0.5 (for a fair
coin).
Thisexperimentpresentsuswithaverycommonflawfoundinfrequentistapproachi.e.Dependence
oftheresultofanexperimentonthenumberoftimestheexperimentisrepeated.
Toknowmoreaboutfrequentiststatisticalmethods,youcanheadtothisexcellentcourseon

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

3/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

inferentialstatistics.

2.TheInherentFlawsinFrequentistStatistics
Tillhere,weveseenjustoneflawinfrequentiststatistics.Well,itsjustthebeginning.
20thcenturysawamassiveupsurgeinthefrequentiststatisticsbeingappliedtonumericalmodelsto
checkwhetheronesampleisdifferentfromtheother,aparameterisimportantenoughtobekeptin
themodelandvariousothermanifestationsofhypothesistesting.Butfrequentiststatistics suffered
some great flaws in its design and interpretation which posed a serious concern in all real life
problems.Forexample:
1. pvalues measuredagainstasample(fixedsize)statisticwithsomestoppingintentionchanges
with change in intention and sample size. i.e If two persons work on the same data and have
different stopping intention, they may get two different p values for the same data, which is
undesirable.
Forexample:PersonAmaychoosetostoptossingacoinwhenthetotalcountreaches100whileB
stopsat1000.Fordifferentsamplesizes,wegetdifferenttscoresanddifferentpvalues.Similarly,
intentiontostopmaychangefromfixednumberofflipstototaldurationofflipping.Inthiscasetoo,
weareboundtogetdifferentpvalues.
2 Confidence Interval (C.I) like pvalue depends heavily on the sample size. This makes
thestoppingpotentialabsolutelyabsurdsincenomatterhowmanypersonsperformthetestsonthe
samedata,theresultsshouldbeconsistent.
3ConfidenceIntervals(C.I)arenotprobabilitydistributionsthereforetheydonotprovidethemost
probablevalueforaparameterandthemostprobablevalues.
Thesethreereasonsareenoughtogetyougoingintothinkingaboutthedrawbacksof
thefrequentistapproachandwhyisthereaneedforbayesianapproach.Letsfinditout.
Fromhere,wellfirstunderstandthebasicsofBayesianStatistics.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

4/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

3.BayesianStatistics
Bayesian statistics is a mathematical procedure that applies probabilities to statistical problems. It
providespeoplethetoolstoupdatetheirbeliefsintheevidenceofnewdata.
Yougotthat?Letmeexplainitwithanexample:
Suppose,outofallthe4championshipraces(F1)between NikiLaudaand Jameshunt, Niki won 3
timeswhileJamesmanagedonly1.
So,ifyouweretobetonthewinnerofnextrace,whowouldhebe?
IbetyouwouldsayNikiLauda.
Heresthetwist.WhatifyouaretoldthatitrainedoncewhenJameswonandoncewhenNikiwon
anditisdefinitethatitwillrainonthenextdate.So,whowouldyoubetyourmoneyonnow?
Byintuition,itiseasytoseethatchancesofwinningforJameshaveincreaseddrastically.Butthe
questionis:howmuch?
Tounderstandtheproblemathand,weneedtobecomefamiliarwithsomeconcepts,firstofwhichis
conditionalprobability(explainedbelow).
Inaddition,therearecertainprerequisites:
PreRequisites:
1.LinearAlgebra:Torefreshyourbasics,youcancheckoutKhansAcademyAlgebra.
2.ProbabilityandBasicStatistics:Torefreshyourbasics,youcancheckoutanothercoursebyKhan
Academy.

3.1ConditionalProbability

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

5/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

It is defined as the: Probability of an eventA given B equals the probability of B andA happening
togetherdividedbytheprobabilityofB.
Forexample:AssumetwopartiallyintersectingsetsAandBasshownbelow.
Set A represents one set of events and Set B represents another. We wish to calculate the
probabilityofAgivenBhasalreadyhappened.LetsrepresentthehappeningofeventBbyshadingit
withred.

NowsinceBhashappened,thepartwhichnowmattersforAisthepartshadedinbluewhichis
interestingly

.So,theprobabilityofAgivenBturnsouttobe:

Therefore,wecanwritetheformulaforeventBgivenAhasalreadyoccurredby:

or

Now,thesecondequationcanberewrittenas:

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

6/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

ThisisknownasConditionalProbability.
Letstrytoanswerabettingproblemwiththistechnique.
Suppose,BbetheeventofwinningofJamesHunt.Abetheeventofraining.Therefore,
1.P(A)=1/2,sinceitrainedtwiceoutoffourdays.
2.P(B)is1/4,sinceJameswononlyoneraceoutoffour.
3.P(A|B)=1,sinceitrainedeverytimewhenJameswon.

Substituting the values in the conditional probability formula, we get the probability to be around
50%,whichisalmostthedoubleof25%whenrainwasnottakenintoaccount(Solveitatyourend).
ThisfurtherstrengthenedourbeliefofJameswinninginthelightofnewevidencei.erain.Youmust
be wondering that this formula bears close resemblance to something you might have heard a lot
about.Think!
Probably,youguesseditright.ItlookslikeBayesTheorem.
BayestheoremisbuiltontopofconditionalprobabilityandliesintheheartofBayesianInference.
Letsunderstanditindetailnow.

3.2BayesTheorem
BayesTheoremcomesintoeffectwhenmultipleevents

formanexhaustivesetwithanotherevent

B.Thiscouldbeunderstoodwiththehelpofthebelowdiagram.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

7/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

Now,Bcanbewrittenas

So,probabilityofBcanbewrittenas,

But

So,replacingP(B)intheequationofconditionalprobabilityweget

ThisistheequationofBayesTheorem.

4.BayesianInference
Thereisnopointindivingintothetheoreticalaspectofit.So,welllearnhowitworks!Letstakean
exampleofcointossingtounderstandtheideabehindbayesianinference.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

8/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

Animportantpartofbayesianinferenceistheestablishmentofparametersandmodels.
Modelsarethemathematicalformulationoftheobservedevents.Parametersarethefactorsinthe
modelsaffectingtheobserveddata.Forexample,intossingacoin,fairnessofcoinmaybedefined
astheparameterofcoindenotedby.TheoutcomeoftheeventsmaybedenotedbyD.
Answerthisnow.Whatistheprobabilityof4headsoutof9tosses(D)giventhefairnessofcoin().
i.e P(D|)
Wait,didIasktherightquestion?No.
We should be more interested in knowing : Given an outcome (D) what is the probbaility of coin
beingfair(=0.5)
LetsrepresentitusingBayesTheorem:
P(|D)=(P(D|)XP())/P(D)

Here, P() isthepriori.ethestrengthofourbeliefinthefairnessofcoinbeforethetoss.Itis


perfectlyokaytobelievethatcoincanhaveanydegreeoffairnessbetween0and1.
P(D|) isthelikelihoodofobservingourresultgivenourdistributionfor.Ifweknewthatcoinwas

fair,thisgivestheprobabilityofobservingthenumberofheadsinaparticularnumberofflips.
P(D) is the evidence. This is the probability of data as determined by summing (or integrating)

acrossallpossiblevaluesof,weightedbyhowstronglywebelieveinthoseparticularvaluesof.
Ifwehadmultipleviewsofwhatthefairnessofthecoinis(butdidntknowforsure),thenthistellsus
the probability of seeing a certain sequence of flips for all possibilities of our belief in the coins
fairness.
P(|D) istheposteriorbeliefofourparametersafterobservingtheevidencei.ethenumberof

heads.
Fromhere,welldivedeeperintomathematicalimplicationsofthisconcept.Dontworry.Onceyou

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=F...

9/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

understandthem,gettingtoitsmathematicsisprettyeasy.
To define our model correctly , we need two mathematical models before hand. One to represent
thelikelihoodfunctionP(D|)andtheotherforrepresentingthedistributionofpriorbeliefs.The
productofthesetwogivestheposteriorbeliefP(|D)distribution.
Sincepriorandposteriorarebothbeliefsaboutthedistributionoffairnessofcoin,intuitiontellsus
thatbothshouldhavethesamemathematicalform.Keepthisinmind.Wewillcomebacktoitagain.
So, there are several functions which support the existence of bayes theorem. Knowing them is
important,henceIhaveexplainedthemindetail.

4.1.Bernoullilikelihoodfunction
Letsrecapwhatwelearnedaboutthelikelihoodfunction.So,welearnedthat:
It is the probability of observing a particular number of heads in a particular number of flips for a
givenfairnessofcoin.Thismeansourprobabilityofobservingheads/tailsdependsuponthefairness
ofcoin().
P(y=1|)=

P(y=0|)=

[Ifcoinisfair=0.5,probabilityofobservingheads(y=1)is0.5]
[Ifcoinisfair=0.5,probabilityofobservingtails(y=0)is0.5]

It is worth noticing that representing 1 as heads and 0 as tails is just a mathematical notation to
formulate a model. We can combine the above mathematical definitions into a single definition to
representtheprobabilityofboththeoutcomes.
P(y|)=
ThisiscalledtheBernoulliLikelihoodFunctionandthetaskofcoinflippingiscalledBernoullis
trials.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

10/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

y={0,1},=(0,1)

And,whenwewanttoseeaseriesofheadsorflips,itsprobabilityisgivenby:

Furthermore,ifweareinterestedintheprobabilityofnumberofheadszturningupinNnumberof
flipsthentheprobabilityisgivenby:

4.2.PriorBeliefDistribution
This distribution is used to represent our strengths on beliefs about the parameters based on the
previousexperience.
But,whatifonehasnopreviousexperience?
Dont worry. Mathematicians have devised methods to mitigate this problem too. It is known
as uninformative priors . I would like to inform you beforehand that it is just a misnomer. Every
uninformativeprioralwaysprovidessomeinformationeventtheconstantdistributionprior.
Well,

the

mathematical

function

used

to

represent

the

prior

beliefs

is

known

as betadistribution .Ithassomeverynicemathematicalpropertieswhichenableustomodelour
beliefsaboutabinomialdistribution.
Probabilitydensityfunctionofbetadistributionisoftheform:

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

11/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

where, our focus stays on numerator. The denominator is there just to ensure that the total
probabilitydensityfunctionuponintegrationevaluatesto1.
and arecalledtheshapedecidingparametersofthedensityfunction.Here isanalogousto

numberofheadsinthetrialsand correspondstothenumberoftails.Thediagramsbelowwillhelp
youvisualizethebetadistributionsfordifferentvaluesof and

YoutoocandrawthebetadistributionforyourselfusingthefollowingcodeinR:

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

12/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

>library(stats)
>par(mfrow=c(3,2))
>x=seq(0,1,by=o.1)
>alpha=c(0,2,10,20,50,500)
>beta=c(0,2,8,11,27,232)
>for(iin1:length(alpha)){
y<dbeta(x,shape1=alpha[i],shape2=beta[i])
plot(x,y,type="l")
}

Note: and are intuitive to understand since they can be calculated by knowing the mean ()
andstandarddeviation()ofthedistribution.Infact,theyarerelatedas:

Ifmeanandstandarddeviationofadistributionareknown,thenthereshapeparameterscanbe
easilycalculated.
Inferencedrawnfromgraphsabove:
1.Whentherewasnotosswebelievedthateveryfairnessofcoinispossibleasdepictedbytheflatline.
2.When there were more number of heads than the tails, the graph showed a peak shifted towards the
rightside,indicatinghigherprobabilityofheadsandthatcoinisnotfair.
3.Asmoretossesaredone,andheadscontinuetocomeinlargerproportionthepeaknarrowsincreasing
ourconfidenceinthefairnessofthecoinvalue.

4.3.PosteriorBeliefDistribution
The reason that we chose prior belief is to obtain a beta distribution. This is because when we
multiplyitwithalikelihoodfunction,posteriordistributionyieldsaformsimilartothepriordistribution

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

13/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

whichismucheasiertorelatetoandunderstand.Ifthismuchinformationwhetsyourappetite,Im
sureyouarereadytowalkanextramile.
Letscalculateposteriorbeliefusingbayestheorem.
CalculatingposteriorbeliefusingBayesTheorem

Now,ourposteriorbeliefbecomes,

This is interesting. Just knowing the mean and standard distribution of our belief about the
parameter andbyobservingthenumberofheadsinNflips,wecanupdateourbeliefaboutthe
modelparameter( ).
Letsunderstandthiswiththehelpofasimpleexample:
Suppose,youthinkthatacoinisbiased.Ithasamean()biasofaround0.6withstandarddeviation
of0.1.
Then,
=13.8 , =9.2

i.eourdistributionwillbebiasedontherightside.Suppose,youobserved80heads( z=80 )in100


flips( N=100 ).Letsseehowourpriorandposteriorbeliefsaregoingtolook:
prior=P(|,)=P(|13.8,9.2)

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

14/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

Posterior=P(|z+,Nz+)=P(|93.8,29.2)

Letsvisualizeboththebeliefsonagraph:

TheRcodefortheabovegraphisas:
>library(stats)
>x=seq(0,1,by=0.1)
>alpha=c(13.8,93.8)
>beta=c(9.2,29.2)
> for(iin1:length(alpha)){
y<dbeta(x,shape1=alpha[i],shape2=beta[i])
plot(x,y,type="l",xlab="theta",ylab="density")
}

Asmoreandmoreflipsaremadeandnewdataisobserved,ourbeliefsgetupdated.Thisisthereal
powerofBayesianInference.

5.TestforSignificanceFrequentistvsBayesian
Without going into the rigorous mathematical structures, this section will provide you a
quickoverviewofdifferentapproachesoffrequentistandbayesianmethodstotestforsignificance

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

15/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

anddifferencebetweengroupsandwhichmethodismostreliable.

5.1.pvalue
In this, the tscore for a particular sample from a sampling distribution offixed size is calculated.
Then,pvaluesarepredicted.Wecaninterpretpvaluesas(takinganexampleofpvalueas0.02for
adistributionofmean100):Thereis2%probabilitythatthesamplewillhavemeanequalto100.
Thisinterpretationsuffersfromtheflawthatforsamplingdistributionsofdifferentsizes,oneisbound
togetdifferenttscoreandhencedifferentpvalue.Itiscompletelyabsurd.Apvaluelessthan5%
does not guarantee that null hypothesis is wrong nor a pvalue greater than 5% ensures that null
hypothesisisright.

5.2.ConfidenceIntervals
Confidence Intervals also suffer from the same defect. Moreover since C.I is not a probability
distribution,thereisnowaytoknowwhichvaluesaremostprobable.

5.3.BayesFactor
Bayes factor is the equivalent of pvalue in the bayesian framework. Lets understand it in an
comprehensivemanner.
The null hypothesis in bayesian framework assumes probability distribution only at a particular
valueofaparameter(say=0.5)andazeroprobabilityelsewhere.(M1)
The alternative hypothesis is that all values of are possible, hence a flat curve representing the
distribution.(M2)
Now,posteriordistributionofthenewdatalookslikebelow.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

16/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

Bayesianstatisticsadjustedcredibility(probability)ofvariousvaluesof.Itcanbeeasilyseenthat
theprobabilitydistributionhasshiftedtowardsM2withavaluehigherthanM1i.eM2ismorelikelyto
happen.
Bayesfactordoesnotdependupontheactualdistributionvaluesofbutthemagnitudeofshiftin
valuesofM1andM2.
InpanelA(shownabove):leftbar(M1)isthepriorprobabilityofthenullhypothesis.
InpanelB(shown),theleftbaristheposteriorprobabilityofthenullhypothesis.
Bayesfactorisdefinedastheratiooftheposterioroddstothepriorodds,

Torejectanullhypothesis,aBF<1/10ispreferred.
We can see the immediate benefits of using Bayes Factor instead of pvalues since they are

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

17/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

independentofintentionsandsamplesize.

5.4.HighDensityInterval(HDI)
HDIisformedfromtheposteriordistributionafterobservingthenewdata.SinceHDIisaprobability,
the95%HDIgivesthe95%mostcrediblevalues.Itisalsoguaranteedthat95%valueswillliein
thisintervalunlikeC.I.
Notice, how the 95% HDI in prior distribution is wider than the 95% posterior distribution. This is
becauseourbeliefinHDIincreasesuponobservationofnewdata.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

18/19

20/6/2016

BayesianStatisticsExplainedinSimpleEnglishForBeginners

EndNotes
Theaimofthisarticlewastogetyouthinkingaboutthedifferenttypeofstatisticalphilosophiesout
thereandhowanysingleofthemcannotbeusedineverysituation.
Its a high time that both the philosophies are merged to mitigate the real world problems by
addressing the flaws of the other. Part II of this series will focus on the Dimensionality Reduction
techniquesusingMCMC(MarkovChainMonteCarlo)algorithms.PartIIIwillbebasedoncreatinga
BayesianregressionmodelfromscratchandinterpretingitsresultsinR.So,beforeIstartwithPart
II,Iwouldliketohaveyoursuggestions/feedbackonthisarticle.
Didyoulikereadingthisarticle?Asabeginner,wereyouabletounderstandtheconcepts?Letme
knowincomments.

http://www.analyticsvidhya.com/blog/2016/06/bayesianstatisticsbeginnerssimpleenglish/?utm_source=feedburner&utm_medium=email&utm_campaign=...

19/19

You might also like