Foundations — Confidence intervals

1.3.21 · D1 · AI-ML › Probability & Statistics › Confidence intervals

Parent note ki derivation padhne se pehle, tumhe har letter aur squiggle ka concrete matlab pata hona chahiye jo woh use karta hai. Neeche hum har ek ko zero se banate hain, us order mein jisme woh ek doosre par depend karte hain. Koi bhi cheez use nahi hoti jab tak woh draw na ho jaaye.

1. Population vs. Sample — do duniyaein

Is topic ki har cheez do cheezoon ke beech ek tension mein rehti hai: poori duniya jise hum poori tarah dekh nahi sakte, aur choti si scoop jo humne actually maapi.

Figure dekho: baayein taraf ka bada cloud population hai (hazaaron dots). Daayein taraf ka chota circle tumhara sample hai — sirf kuch dots jo tumne pakde. Confidence intervals ka poora khel yeh hai: chhote circle ka use karke bade cloud ke baare mein koi fact guess karo.

Yeh topic isliye zaroorat padti hai: neeche har symbol ko ya toh population fact (chhupa hua, fixed, usually Greek letters) ya sample fact (measured, har scoop mein badalta hai, usually Latin letters) tag kiya gaya hai. In do duniyaon ko alag rakhna woh ek sabse common cheez hai jo beginners mix up karte hain.

2. $\mu$ — sach-much mean (joh fish hum dhoondh rahe hain)

Picture: bade cloud par, $\mu$ bilkul balance ka center hai — woh ek jagah jahan cloud ek pin par perfectly balance ho jaaye.

Yeh topic isliye zaroorat padti hai: $\mu$ woh fish hai jise hum pakadne ki koshish kar rahe hain. Poora confidence interval is ek chhupe hue number ko bracket karne ke liye exist karta hai. Kyunki yeh fixed hai, hum kabhi nahi kehte "the probability $\mu$ is here" — $\mu$ hilta nahi.

3. $\sigma$ aur ${\sigma }^2$ — duniya kitni spread out hai

Picture mein do bell-shaped clouds dikhte hain jin ka same center $\mu$ hai lekin alag-alag widths hain. Narrow wala small $\sigma$ hai (sab average ke paas); wide wala large $\sigma$ hai (values door tak bikhri hain). $\sigma$ literally beech se typical distance hai.

Yeh topic isliye zaroorat padti hai: ek confidence interval ki width $\sigma$ ke saath badhti hai. Agar duniya bahut spread out hai, toh tumhara jaal trustworthy hone ke liye wider hona chahiye. Hum square karte phir square-root isliye ki $\mu$ ke upar aur neeche ki distances cancel na ho jaayein — squaring minus signs khatam kar deta hai.

4. $\bar{x}$ aur $\bar{X}$ — woh average jo tumne actually compute kiya

Chalte hain is formula ko symbol by symbol decode karte hain, kyunki parent note ise baar baar use karta hai:

$\bar{x}$ ki picture: tumhare chote sample circle par, $\bar{x}$ uska balance point hai — sample ka apna center. Yeh chhupe hue $\mu$ ke liye humara best single guess hai, lekin yeh wobble karta hai.

5. $n$ — sample size (tumhari scoop kitni badi hai)

Yeh topic isliye zaroorat padti hai: $n$ woh knob hai jo tum actually control kar sakte ho. Yeh interval width ke denominator mein $\sqrt{n}$ ke roop mein aata hai, isliye zyada measure karna tumhara jaal chhota kar deta hai. Yahi poori wajah hai "zyada data = zyada confidence."

6. Random variable & distribution — ek average ka shape kyun hota hai

Figure: socho "ek fresh sample lo, $\bar{x}$ compute karo" hazaaron baar repeat karna. Har $\bar{x}$ ek dot hai. Unhe pile karo aur ek bell shape milti hai jo $\mu$ par centred hai. Yeh pile mean ki sampling distribution hai. Khaas baat yeh hai ki yeh raw data cloud se narrower hai — averages individual measurements se kam jitter karte hain.

Yeh topic isliye zaroorat padti hai: poori derivation poochh rahi hai "$\bar{x}$ $\mu$ se kitna door ja sakta hai?" Woh sawaal is bell ki shape se answer hota hai. Distribution nahi toh interval nahi.

Yeh bell ka normal hona (bade $n$ ke liye) exactly wahi hai jo Central Limit Theorem guarantee karta hai — parent usse rely karta hai, aur yeh apni alag study deserve karta hai.

7. Normal distribution $N(\text{mean},\text{variance})$

Width ${\sigma }^2/n$ note karo: sample-mean bell ka population spread ${\sigma }^2$ hai jo $n$ se divide karke shrunk hai. Yahi maths hai "averages kam jitter karte hain" ke peeche — bada $n$, tight bell.

8. $Z$ — standardized ruler

Yeh topic isliye zaroorat padti hai: hum ek standard bell par exact cut-off points jaante hain. $Z$ mein convert karke, hum kisi bhi problem ke liye woh universal cut-offs lete hain. Yahi parent mein "standardize to control probability" step hai.

9. $1-\alpha$, $\alpha$, aur $z_{\alpha /2}$ — confidence aur uska critical value

Picture: do chote shaded tails wala standard bell, har ek mein area $\alpha /2$ hai. Vertical lines jahaan shading shuru hoti hai woh $\pm z_{\alpha /2}$ par hain. Unshaded middle mein $1-\alpha$ area hai — yahi region hum trust karte hain.

10. Standard error — guess ki apni spread

$\sqrt{n}$ kyun aur $n$ kyun nahi? Variance $n$ se divide hota hai; standard deviation uska square root hai, isliye spread $\sqrt{n}$ se divide hota hai. Yahi wajah hai ki apna data chaar guna karne se sirf error aadha hota hai — poori kahani ke liye Standard Error dekho.

11. $s$, $n-1$, aur t-distribution — jab $\sigma$ unknown ho tab kya karein

Prerequisite map

Equipment checklist

Cover the right side; you're ready when each reveal matches your own answer.

Aage: parent note par wapas jao aur dekho kaise yeh har symbol derivation mein fit hota hai. Deeper roads: Central Limit Theorem, Standard Error, Hypothesis Testing, Bootstrap Methods.