Measures of dispersion — variance, standard deviation: Visual walkthrough

2.7.3 · D2 · Maths › Statistics & Probability — Intermediate › Measures of dispersion — variance, standard deviation

Hum poore waqt ek hi running dataset use karte hain: marks $3, 5, 5, 7, 10$ (10 mein se). Yeh paanch dots poori kahani mein chalte dekho.

Step 1 — Data ko line par rakho

KYA. Hum paanch numbers lete hain aur unhe ek horizontal line (ek axis) par dots ki tarah place karte hain. Axis bas ek ruler hai: position = value.

KYUN. "Spread" ki baat karne se pehle, humein raw data dekhna hoga. Spread ek fact hai ki yeh dots kaise arrange hain — tum ise apne dimag mein list se measure nahi kar sakte, lekin line par dekh sakte ho.

PICTURE. Paanch dots. Kuch paas-paas hain (do 5s ek doosre ke upar baithe hain), ek straggler 10 par bahut door baith gaya hai.

Figure — Measures of dispersion — variance, standard deviation

Step 2 — Centre dhundho (mean)

KYA. Hum ek special number, mean, compute karte hain aur use usi line par mark karte hain.

$\overset{x}{ˉ} = \frac{1}{n} \sum_{i = 1}^{n} x_{i} = \frac{3 + 5 + 5 + 7 + 10}{5} = \frac{30}{5} = 6$

Term by term padhein: $i = 1 \sum n x_{i}$ ka matlab hai "saare dots add karo" $= 30$ ; $\frac{1}{n} = \frac{1}{5}$ se divide karo ki kitne hain; $\overset{x}{ˉ}$ (padho "x-bar") result hai, $6$ .

KYUN. Spread ka matlab hai "kisi cheez ke aas-paas spread." Mean woh cheez hai — yeh balance point hai, woh jagah jahan line balance hogi agar har dot equal weight ka ek coin ho. (Dekho Mean, Median, Mode — measures of central tendency ki yeh kyun balance karta hai.)

PICTURE. Ek neeche triangle (see-saw pivot) $6$ par baitha hai. $10$ par door wala dot balance ko daayein kheenchta hai exactly utna jitna $3$ aur $5$ wale do neeche dots baayein kheenchte hain.

Step 3 — Har dot ki centre se distance measure karo

KYA. Har dot ke liye hum centre $\overset{x}{ˉ}$ se us dot tak arrow kheenchte hain. Uski signed length hai deviation:

$d_{i} = x_{i} - \overset{x}{ˉ}$

Yahan $\overset{x}{ˉ} = 6$ hai, toh deviations hain $3 - 6 = - 3, 5 - 6 = - 1, 5 - 6 = - 1, 7 - 6 = + 1, 10 - 6 = + 4.$

Har symbol: $x_{i}$ woh jagah hai jahan dot hai, $\overset{x}{ˉ}$ centre hai, aur unka difference $d_{i}$ hai "kitna door, aur kis direction mein." Minus matlab dot centre ke baayein baitha hai; plus matlab daayein.

KYUN. Distance-from-centre hi spread ka raw material hai. Centre se door wala dot ( $+ 4$ ) zyada contribute karta hai; centre ke paas wala dot ( $\pm 1$ s) kam contribute karta hai.

PICTURE. Orange arrows neeche wale dots se baayein (negative) ki taraf point karte hain, teal arrows upar wale dots se daayein (positive) ki taraf. Length = kitna door.

Step 4 — Kyun hum arrows ka average nahi le sakte

KYA. Naive idea: "deviations ka average nikalo." Karte hain try: $\sum d_{i} = (- 3) + (- 1) + (- 1) + (+ 1) + (+ 4) = 0.$

YEH KYUN FAIL HOTA HAI. Baayein-pointing arrows aur daayein-pointing arrows ki total length equal hoti hai — exactly iska hi matlab hai Step 2 se "balance point." Toh yeh zero par cancel ho jaate hain har baar, chahe data kitna bhi scattered ho. Ek spread measure jo hamesha $0$ ho woh hume kuch nahi bata ta.

PICTURE. Orange (baayein) arrows end-to-end stack karo toh teal (daayein) arrows ki total length utni hi hogi — ek visual tug-of-war jo $0$ par perfect tie se khatam hota hai.

Step 5 — Signs khatam karne ke liye har deviation ko square karo

KYA. Har arrow ki jagah us par bana ek square rakho. Square ko positive side length chahiye, toh hum side ko absolute deviation $∣ d_{i} ∣$ lete hain (arrow ki length, sign hatakar). Uska area phir hoga $side = ∣ d_{i} ∣ \Rightarrow area = ∣ d_{i} ∣^{2} = d_{i}^{2},$ aur numerically $(- 3)^{2} = 9, (- 1)^{2} = 1, (- 1)^{2} = 1, (+ 1)^{2} = 1, (+ 4)^{2} = 16.$

$∣ d_{i} ∣^{2}$ aur $d_{i}^{2}$ same kyun hain? Kyunki $∣ d_{i} ∣^{2} = (\pm d_{i})^{2} = d_{i}^{2}$ — kisi length ko square karna already sign erase kar deta hai, toh side length $∣ d_{i} ∣$ aur raw $d_{i}$ dono same area dete hain. Chhota $^{2}$ matlab "number ko khud se multiply karo," aur $(- 3)^{2}$ aur $(+ 3)^{2}$ dono $9$ hain: squaring sign khatam kar deta hai lekin size rakhta hai.

KYUN squaring aur absolute value nahi? Do tarike signs khatam karte hain: $∣ d_{i} ∣$ aur $d_{i}^{2}$ . Hum squaring choose karte hain kyunki (a) yeh smooth hai — square ki area smoothly change hoti hai, toh hum baad mein isme calculus kar sakte hain (isi liye Least Squares Regression squares use karta hai); (b) yeh bade misses ko zyada punish karta hai — $+ 4$ ek bhaari $16$ ban jaata hai, $\pm 1$ s wale just $1$ ke maqable; (c) squared distance woh geometry hai jis par Normal Distribution bana hai.

PICTURE. $∣ d_{i} ∣$ length ka har arrow usi side ka ek literal square ban jaata hai. $+ 4$ deviation ka square ( $16$ ) tiny unit squares par clearly dominate karta hai.

Step 6 — Squares ka average karo → variance

KYA. Saare square-areas add karo aur kitne hain usse divide karo: $σ^{2} = \frac{1}{n} \sum_{i = 1}^{n} (x_{i} - \overset{x}{ˉ})^{2} = \frac{9 + 1 + 1 + 1 + 16}{5} = \frac{28}{5} = 5.6.$

Term by term: $(x_{i} - \overset{x}{ˉ})^{2}$ ek square ki area hai; $\sum$ paanchon areas add karta hai $= 28$ ; $\frac{1}{n} = \frac{1}{5}$ total area ko average area mein badalta hai; $σ^{2}$ (padho "sigma-squared") woh average hai, $5.6$ .

KYUN average, sirf sum nahi? Plain sum badhta hai agar tum zyada data points add karo, chahe data equally spread ho — yeh fair nahi hai. $n$ se divide karne par typical square-area per point milti hai, ek number jo spread ko describe karta hai, sample size ko nahi.

PICTURE. Paanchon squares melt karke ek bade square mein dobara daale jaate hain jiska area average hai, $5.6$ — "typical square" jo poore dataset ki jagah khadaa hai.

Step 7 — Real units mein wapas aane ke liye square-root lo (standard deviation)

KYA. Variance $5.6$ marks² hai — ek squared unit, jiske baare mein baat karna awkward hai. Square root lo: $σ = σ^{2} = 5.6 \approx 2.37 marks .$

Step 5 ki squaring ko undo karta hai: yeh ek area ko wapas length mein badal deta hai.

KYUN. Hum jawaab dena chahte hain "typically ek mark average se kitna door hai, marks mein?" Sirf ek length hi yeh jawaab de sakti hai. $σ \approx 2.37$ matlab: average par marks $6$ se lagbhag $2.37$ door baithte hain. Yeh ek sentence hai jo ek insaan use kar sakta hai.

PICTURE. Area $5.6$ wale bade "average square" ki side length wapas original number line par draw ki jaati hai: centre $6$ ke dono taraf $σ \approx 2.37$ width ka ek band, jo typical dots ko pakadta hai.

Step 8 — Degenerate case: bilkul bhi spread nahi

KYA. Maano har mark same hai: $6, 6, 6, 6, 6$ . Phir $\overset{x}{ˉ} = 6$ , har deviation $d_{i} = 6 - 6 = 0$ , har square $0$ hai, toh $σ^{2} = \frac{0 + 0 + 0 + 0 + 0}{5} = 0, σ = 0.$

KYUN IMPORTANT HAI. Yahi ek tarika hai variance zero ho sakta hai, aur yeh sanity floor hai: variance kabhi negative nahi hoti (yeh squares ka average hai, aur squares $\geq 0$ hote hain), aur yeh exactly $0$ hoti hai jab saare values identical hon. No spread ⇒ no arrows ⇒ no squares ⇒ zero.

PICTURE. Paanchon dots ek hi point $6$ par pile ho jaate hain. Har arrow ki length $0$ hai; draw karne ke liye koi squares nahi hain.

Recall Edge case khud check karo

Data $4, 4, 4$ : mean hai 4, saare deviations 0 hain, toh variance $0$ ke barabar hai, aur SD bhi $= 0$ .

Ek-picture summary

Ek hi frame mein har step: dots → centre → arrows → squares → average square → square-root band. Numbers follow karo $3, 5, 5, 7, 10 \to \overset{x}{ˉ} = 6 \to d = - 3, - 1, - 1, 1, 4 \to d^{2} = 9, 1, 1, 1, 16 \to σ^{2} = 5.6 \to σ \approx 2.37$ .

Recall Feynman: poori walk simple words mein batao

Apne paanch test scores ko ek ruler par line up karo. Balance point dhundho — woh average hai, 6. Balance point se har score tak ek arrow kheencho: kuch baayein point karte hain, kuch daayein. Agar tum un arrows ko simply add karo toh woh cancel hokar zero ho jaate (iska hi matlab hai "balance point"), toh instead tum har arrow ko ek square tile mein badal do jiska side arrow ki length hai — ab left aur right cancel nahi ho sakte, aur door wala score ek huge tile banata hai. Saari tiles ka average ek "typical tile" mein karo; woh average area hai variance (5.6). Uski side length — square root lo — hai standard deviation (lagbhag 2.37), jo finally plain marks mein jawaab deta hai: "ek typical score 6 se lagbhag 2.37 door baithta hai." Aur agar tumhare saare scores identical hote, koi arrows nahi hote, koi tiles nahi hoti, aur spread exactly zero hoti.

Connections

Measures of Dispersion — variance, standard deviation — parent note jise yeh walkthrough build karta hai.
Mean, Median, Mode — measures of central tendency — kyun mean Step 2 ka balance point hai.
Normal Distribution — squares-of-distances ka idea 68-95-99.7 rule ko power karta hai.
Coefficient of Variation — $σ$ ko $\overset{x}{ˉ}$ se divide karta hai unitless spread ke liye.
Covariance and Correlation — same arrows-and-products idea, lekin ek saath do variables ke liye.
Least Squares Regression — "sum of squared arrows minimise karo," exactly Step 5 ki tiles.

Visual walkthrough — Measures of dispersion — variance, standard deviation