Probability Cheatsheet v1.1.1
CCompiied ty Willams Chen (np: //asetun con) with eontibutiong
1eep: //sthub con/enchen/probabi sey chastahest
Counting
Maliplication Rule bars roy we have = compound caperiner
Naive Definition of Probability =f the lisihood of each
Probability and Thinking Conditionally
Independence
Pace)
PAB) = PA)
at giver © uk PIA“ BIC) = PLAI
net dace not imply Independence, abd independence doce
Simpson’s Paradox
PUA IB.) < PAL BSC) and PA | B.C) < PA
yet ell P(A B) > PLA |B)
Bayes’ Rule and Law of Total Probability
P(A) = PAB.) PB.) + PLAIB.) PCB.) + PLAID} PB
P{AIC) = P(AIRs, C)PURIC) + «PAIR, CPUC)
P(AIC) = P(A By ©) + PLAN BIC) + P(A BalC)
al Probabity with B and B® (apecal case of « parttioning
nd with extra conditioning (just aC)
P(A) = PAB) PUB) + PLAIB) PO)
P(A) = PAN) + PANS”
PAC) = PAIS, C)PUBIC) + PI
Plc) = PAP BIC) + PAN B"C)
caipy ~ ADB) _ POBIA)PIA}
(Al) = Fon IR)
PIANBIC) _ POBIA.cyP(AIC)
rane) = me = eae
FURST) ~ POA) PLAS)
POAIB.G) _ POBIA,C) PLAC)
PIASIB, G) ~ PURIAS G) PASC)
Random Variables and their Distributions
PMF, CDF, and Independence
Probability Mase Function (PMF) (Discrete Only) gives the
Pele) = PC
Cumulative Distribution Function (CDF) sive the probabitty
Fxlzo) = POX S20)
Expected Value,
Indicator Rando
probability of the event
Variance
Expectation and
Continuous RY
Continuous Ran:
the PDP. ee below)
ala i the itera [aHow do I nd the expected value of a CRV? Whore in discrete
Soa Jouoe
Law of the Unconscious Statistician (LotUS)
Be
Bx)
teeP(X =2)
Bix) = [7 *f(eree
otUS states that you can find the expected value of a funetion of @
‘Pondove warble gX) this mays
292) PX
[sono
‘amie, if Xs the sumer of
‘you seein an hour then o(X) = 2X could be the number f bike
thee you seein ass hour. Both are random vartablee
What's the point? You don't ned to know the PDE/PMF of st)
Universality of Uniform
When you plug any random vacable into Ss own CDF, you get «
Unie se Varah” When ou p's Union in a
fbetion then we get a unify steer roator vaeiable
Similar, since FLX) ~ U then X ~ BAG). The ey point i that
Jorany continuous random earible X, we can traferm H nfo ®
‘Snorm random torible and beck by sing te ODE.
Moment Generating Functions (MGFs)
The dratve ofthe Nomen gnciatng funtion catated Wi
mene Hh = BUX) = MYO)
eo ene -8( eX) mare
M(t) = Bie
ager
Bo) = BPN) «BY = a,
My (0) = BION) = BEEN) =o Ala)
Uniqueness ofthe MCP. if ev the MCP wrguly dfs
Yi ehey are dtsbuted the same (het ODEs/PDFs ave equal tant
Sou eve fo radon vate that ave the same MG
Yteintepentens then, eV DY Mutiniving MGPe. 16% aod
Mey 79) = Ble) = BEI) BLE)
M(t) My)
GPs of those two random variables mene
Joint PDFs and CDFs
Toint Distributions
eviews Joint Probability of events A and: PLAN B)
Bath the Joint POE apd Jint PDE mut be nonnegative nd
Siimintegrate to LOS, 35, PX BY We)
(Jy facta) = 1), ik Ue univasiate cause, you sum integrate
Mx(t)- My(e)
Moca
Conditional Distributions
eviews By Bayes Rule, P(A|B) = SUMAYELAY sian conditions
poly to condiNional dstibutions of random viable
or dccatereniom variable
Pee
7a
PY =¥1K =) a0) eo
Independence of
Puna) Para)
Smile conditions opel
independent two vend
dlasebution function ie
“geitioa, ov that th
tS
Multivariate Lot
Review: Hla(%)) = 5
BUG) = f, ateife
For dlacaterasdom va
aX.)
For continuous random
BiviX.¥))
Covariance an
‘Covariance and ¢
Cow, ¥) = BI
ov X.%)
eoneeX,¥)
Independent, then they
vatCovariance and Invariance - Corelation, Covatlanes, and Valance
vari $9 =
comtX HEY Fe)
ComiX $8. 4)
Gore(2X,3¥) = ComiX)
Continuous Transformations
Why do we need the Jacobian? We need the Jacobian to teteale
One Variable Transformations Lats say that we have a random
‘We cal hie function Y~ 9(X). Note that Ysa
eae elf go ferential and one-tasone every
|
To fod Jy(p) a «function of y, lag in
Fete)
re(2) Fx(2)
moo e|
mior= moon)
‘|
ays
se dotw
Convolutions
Definition Ifyou want to nd the PDF ofa um of two independent
Fairdom variable, sou te the eouvoblion of thes india
‘daeibutione
evwt = [7 senna
Lat X,¥ ~ id N(O,1), Tat Casa constant Integrate a
teovto= fo
Order Statistics
Metis oust a east acs bt
‘ect haea istebution. For any vaie of Xo Xuyay XO
peep? (2) and PDF f(z), the CDF and PDF of Xjy are ae
Prrgy(2) = PX) se-D (rete = Pay"
Jao) ey a= oN Fe)
Lnlversaty af the Uniorn We can so eas he diteation
FOX) ~ Yop
Beta Dinrbution a Order Sateen of Union - The mate
{tive Ulome h deate.bpy sa) Se ate
Sana ssw Spr walt toe ae
gy = Beta 1), The ditebtion ofthe the 5 onder sti
HED Waltons
w Batagin 540)
Tae
Conditional Expectation and Variance
Conditional Expectation
Conditioning on an Event.» Wr can fd the expected val of Y
tata of Bia and B(YIX = 2) Note that conditioning in an event
Expectation aod Song the sonttional expecation. Tae expected
Sahota ice roll given thats pomele Jat do-+ 45294. The
‘Sie cna fine that you a ut eth cn
{CSraming thatthe wating time i = Bxpots)) given that you have
Srcndy wate minutes 10'morc minute bythe memesylest
BE veo)
worn) EePty = ax =)
Conditional Vari
vast)
MVN, LLN, €
Law of Large Nu
Let ue have Xi Mas
i peed
nm, Ba BOO
Central Limit Th
Approximation us
fanetn Lid. odo
BI) ey and Vari)
When we ut contra i
Yom ehyte +x
aha
Asymptotic Distri
We use 4 to donate
re the sate reute
an’
Markov Chain
Definition
trbich cement of the
Chain ithe not of ran
seit ina‘Transition Matrix
Element uy i aguare transition matt Q is the probability that Ube
hain oot from stat toate J, or tore formal
Peas =I =H)
Reps take the (5) element of
a = Pain
Xs i dite according to roweector PME 5 (og
bs = Pie =) then the PMP ot Nye FO"
Chain Properties
iEendueble chain vad have al of ie eater secures, Achaia ie
ride any of estates are perdi, and aperiodie i ome of
Ae'tates aro periodic. Iman lvedcible chan, al sates have the same
period
igrbore the Marky ehnin would be sationeey with rrpoct to
cae
Reversitility Condition Implies Stationarity - If you have a PMP
Pina Marlov chain wish transition mates thon toy oyu fn
Stationary Distribution
Let us aay that the vector = (ps.Pa, Pat) S¢ a pombe and val
PPM of whee the Rfarow Chale iat ta cota te We wil eal
this ctor the stationary ett Phen
Unigie, ant»; the longron probability of chan being at atte
‘smberofstepe to ream back to starting fom is
ive forthe ratlonsry istibation, you ean solve for
T)GA} =o. The weationary dtibtion i wxform if the colamne
Sg num ob
Random Walk on Undirected Network
you have a certain amber of nods wih edges between hem, and a
‘chun can pick any edge randomly and move to anther node, then tht
2'Datdomm walk on an undirected network The stationary
Normal
Let ue aay that X la datibuted A/G, 02). We kom the following
intribation, we change to another normal dneibation. Uf we add ¢
torn normaly distrosted reodom variable, thn fe mean inconeee.
tahny hy - i we rip t normaly dieribated reo warble
very niall distributed random variate 3 = Ng e*), we ean
frtorm it the standard 0,1) by the flowing tanstornction
=x)
fertral mit theorem, the sampling average fom a population i also
on)
CDF = 10 too fies to write thie one out, 40 we enprote ita the
fisetion #2)
Exponential yution
et teeny that X fa datibuted Kxpo(A). We now the flowing
Story You've slitting on an open meadow right befor the brea of
‘Your wasting time ie metmoryesenesn, which meane Vat the time util
the ext shooting star comes dove not depend oe how long you've
‘ited ales
Example The wating tiene until the next shooting war ls deed
Bipot rhe ae 5 the ate putamen how may |
¥ = Bapotd) 9X
AY ~ Bxpost)
Mamorylatsness The Exponential Distribution i the sole
‘ostimuoue tetoryles dacibuton: ‘This meane that f alwaye “ae
Gamma Distribu
Let urea that X i
Beta Distributio
Conjugate Prior oft
posterior dation
Beciietbated: Tee
Relationship with Gs
Reasoning OY Represen
x? Distribution
et we tay that Xie di
Example The sun of
Bod)