WHY this formula? Think of probability as mass. Put mass pi at position xi on a number line. The balance point (center of mass) of that mass distribution is exactly ∑xipi. That's why E[X] is called the mean — it's the point where the distribution would balance on a knife edge.
HOW we use it (the most important tool): the Law of the Unconscious Statistician (LOTUS). To average a function of X, you do not need the distribution of g(X):
E[g(X)]=∑ig(xi)pior∫g(x)f(x)dx
Derivation of E[aX+b] from scratch (discrete):
E[aX+b]=∑i(axi+b)pi=a∑ixipi+b∑ipi=aE[X]+b⋅1.
Why each step? First we used LOTUS with g(x)=ax+b. Then split the sum (sums are linear). Then pulled constants out, and used ∑ipi=1 (probabilities sum to one). Done.
WHY squared, not just ∣X−μ∣? If we averaged X−μ directly we'd get 0 (positives cancel negatives — that's literally what the mean is). Squaring makes everything positive, punishes big deviations more, and gives clean algebra (it's differentiable, unlike ∣⋅∣).
Why each step? Expand the square (algebra). Apply linearity — note μ is a constant, so it pulls out. Then substitute E[X]=μ. The two middle terms collapse to −μ2.
Why? The +bcancels — shifting the whole distribution doesn't change its spread. The a factors out and gets squared because variance lives in squared units. Taking the root gives ∣a∣ (absolute value, because SD ≥ 0).
The first two terms are the variances; the last is 2Cov(X,Y). Why does independence kill it? Because for independent variables E[XY]=E[X]E[Y], making covariance 0.
State the computational formula for variance and derive it.
What is Var(aX+b) and why does b vanish?
When is Var(X+Y)=Var(X)+Var(Y)?
Why do we take a square root to get SD?
Is linearity of expectation true for dependent variables?
Recall Feynman: explain to a 12-year-old
Imagine throwing darts at a number line. The average spot you hit is the expected value — the middle of your cluster. The variance asks: how scattered are my darts around that middle? We measure each dart's distance to the middle, square it (so left and right misses both count as "bad"), and average those. The standard deviation is just that scatter measured back in normal distance units. If you slide the whole target sideways, your scatter doesn't change — but if you zoom the target picture to be twice as wide, your scatter doubles (and the squared scatter, variance, quadruples).
Dekho, koi bhi random variable X matlab ek aisa number jo chance se nikalta hai (jaise dice ka result). Iske baare mein hum do cheezein jaanna chahte hain: average kahaan baith raha hai (ye hai expected valueE[X], yaani center of mass — jahan distribution balance karega), aur values kitni bikhri hui hain center ke around (ye hai variance aur uska square root standard deviation).
Variance ka formula yaad rakhne ka aasaan tareeka: Var(X)=E[X2]−(E[X])2 — "mean of square minus square of mean". Hum deviation ko square isliye karte hain kyunki agar simple X−μ ka average lenge to wo hamesha zero aa jaayega (plus aur minus cancel ho jaate hain). Square karne se sab positive ho jaata hai aur bade misses zyada count hote hain.
Sabse important properties: agar tum X ko a se multiply karke b add karte ho, to E[aX+b]=aE[X]+b (linearity — ye hamesha sach hai, even dependent variables ke liye bhi). Lekin variance mein twist hai: Var(aX+b)=a2Var(X). Yahan +b (shift) variance ko change nahi karta — bas poori picture slide ho jaati hai — aur a (stretch) square ho ke aata hai. Yaad rakho: "Add shifts, Multiply stretches, Square the stretch."
Aur ek common galti: Var(X+Y)=Var(X)+Var(Y) sirf tab sach hai jab X aur Yindependent hon. Warna ek extra term 2Cov(X,Y) aata hai. Exam mein ye trap bahut aata hai, to dhyaan rakhna!
