Probability Distributions & Phenomena การแจกแจงความน่าจะเป็นและปรากฏการณ์

Summary of examples we discussed: binomial, Poisson, normal, and other / esoteric distributions. สรุปตัวอย่างที่เราพูดคุยกัน: การแจกแจงทวินาม, ปัวซง, ปกติ, และการแจกแจงอื่นๆ

Choose by Process, Not by Shape เลือกจากกระบวนการ ไม่ใช่จากรูปร่าง

A distribution is a claim about the process that generated the data. Fitting a curve checks that claim; it does not choose it. Start by asking what is happening. การแจกแจงคือข้อสันนิษฐานเกี่ยวกับกระบวนการที่สร้างข้อมูล การฟิตเส้นโค้งเป็นการตรวจสอบข้อสันนิษฐานนั้น ไม่ใช่การเลือก จึงควรเริ่มจากการถามว่าเกิดอะไรขึ้น

What is happening สิ่งที่เกิดขึ้น Reach for ควรใช้ Ask คำถาม
A fixed number of independent yes/no trials, each with the same chance จำนวนการทดลองแบบใช่/ไม่ใช่ที่กำหนดไว้ เป็นอิสระต่อกัน และมีโอกาสเท่ากันทุกครั้งBinomial ทวินามWhat is one trial? การทดลองหนึ่งครั้งคืออะไร?
Independent events arriving at a steady rate, counted in a window, with no fixed number of trials เหตุการณ์อิสระที่เกิดในอัตราคงที่ นับในหนึ่งช่วง โดยไม่มีจำนวนการทดลองที่กำหนดไว้Poisson ปัวซงWhat is the window? ช่วงที่นับคืออะไร?
Many small independent contributions added together ปัจจัยย่อยที่เป็นอิสระจำนวนมาก บวกกันNormal ปกติWhat is being added? อะไรถูกบวกกัน?
Many independent factors multiplied together, such as growth by percentages ปัจจัยอิสระจำนวนมาก คูณกัน เช่น การเติบโตเป็นเปอร์เซ็นต์Lognormal ล็อกนอร์มอลWhat is being multiplied? อะไรถูกคูณกัน?
Feedback: having more makes you likely to get more การป้อนกลับ: ยิ่งมีมาก ยิ่งมีโอกาสได้เพิ่มอีกPower law (Pareto, Zipf) กฎกำลัง (พาเรโต, ซิปฟ์)What feeds on itself? อะไรที่ส่งเสริมตัวเอง?
Waiting: how many trials until a success, or how long until an event การรอ: ต้องทดลองกี่ครั้งจึงสำเร็จ หรือต้องรอนานแค่ไหนจึงเกิดเหตุการณ์Geometric (trials), Exponential (time) เรขาคณิต (จำนวนครั้ง), เลขชี้กำลัง (เวลา)Waiting for what? รออะไร?
Drawing from a small finite group without putting anything back การหยิบจากกลุ่มขนาดเล็กที่มีจำกัดโดยไม่ใส่คืนHypergeometric ไฮเพอร์จีออเมตริกHow big is the group? กลุ่มใหญ่แค่ไหน?
The largest value in each of many blocks, such as the yearly highest river level ค่าสูงสุดในแต่ละกลุ่ม เช่น ระดับแม่น้ำสูงสุดในแต่ละปีGumbel กัมเบลMaximum of what? ค่าสูงสุดของอะไร?

How they connect ความเชื่อมโยงระหว่างการแจกแจง

Quick Parameter Cheat Sheet ตารางสรุปพารามิเตอร์ฉบับย่อ

Common distributions, their type, support, parameters, and basic mean/variance (where simple). การแจกแจงทั่วไป ประเภท ช่วงของค่า พารามิเตอร์ และค่าเฉลี่ย/ความแปรปรวนพื้นฐาน (ที่ง่าย)

Swipe the table sideways to see every column. ปัดตารางไปด้านข้างเพื่อดูทุกคอลัมน์

Distribution การแจกแจง Type ประเภท Support ช่วงของค่า Parameters พารามิเตอร์ Mean ค่าเฉลี่ย Variance ความแปรปรวน
Binomial ทวินาม Discrete ไม่ต่อเนื่อง \(k = 0,1,\dots,n\) \(n \in \mathbb{N}\), \(0 \le p \le 1\) \(np\) \(np(1-p)\)
Poisson ปัวซง Discrete ไม่ต่อเนื่อง \(k = 0,1,2,\dots\) \(\lambda > 0\) \(\lambda\) \(\lambda\)
Normal ปกติ Continuous ต่อเนื่อง \(x \in \mathbb{R}\) \(\mu \in \mathbb{R}\), \(\sigma > 0\) \(\mu\) \(\sigma^2\)
Geometric เรขาคณิต
(trials until 1st successการทดลองจนสำเร็จครั้งแรก)
Discrete ไม่ต่อเนื่อง \(k = 1,2,3,\dots\) \(0 < p \le 1\) \(\dfrac{1}{p}\) \(\dfrac{1-p}{p^2}\)
Negative binomial ทวินามนิเสธ
(trials until \(r\)th successการทดลองจนสำเร็จครั้งที่ \(r\))
Discrete ไม่ต่อเนื่อง \(k = r,r+1,\dots\) \(r \in \mathbb{N}\), \(0 < p \le 1\) \(\dfrac{r}{p}\) \(\dfrac{r(1-p)}{p^2}\)
Hypergeometric ไฮเพอร์จีออเมตริก
(sampling without replacementการสุ่มโดยไม่ใส่คืน)
Discrete ไม่ต่อเนื่อง see formula below ดูสูตรด้านล่าง \(N,K,n\) \(\dfrac{nK}{N}\) \(\dfrac{nK(N-K)(N-n)}{N^2(N-1)}\)
Exponential เลขชี้กำลัง Continuous ต่อเนื่อง \(x \ge 0\) \(\lambda > 0\) \(\dfrac{1}{\lambda}\) \(\dfrac{1}{\lambda^2}\)
Gamma แกมมา Continuous ต่อเนื่อง \(x \ge 0\) shape \(k > 0\), scale \(\theta > 0\) รูปร่าง \(k > 0\), สเกล \(\theta > 0\) \(k\theta\) \(k\theta^2\)
Uniform เอกรูป Continuous ต่อเนื่อง \(a \le x \le b\) \(a < b\) \(\dfrac{a+b}{2}\) \(\dfrac{(b-a)^2}{12}\)
Beta เบตา Continuous ต่อเนื่อง \(0 \le x \le 1\) \(\alpha,\beta > 0\) \(\dfrac{\alpha}{\alpha+\beta}\) \(\dfrac{\alpha\beta}{(\alpha+\beta)^2(\alpha+\beta+1)}\)
Pareto พาเรโต Continuous ต่อเนื่อง \(x \ge x_m > 0\) \(\alpha > 0, x_m > 0\) \(\dfrac{\alpha x_m}{\alpha - 1}\)
forเมื่อ \(\alpha > 1\)
\(\dfrac{\alpha x_m^2}{(\alpha-1)^2(\alpha-2)}\)
forเมื่อ \(\alpha > 2\)
Zipf ซิปฟ์ Discrete ไม่ต่อเนื่อง \(k = 1,2,3,\dots\) \(s > 1\) \(\dfrac{\zeta(s-1)}{\zeta(s)}\)
forเมื่อ \(s > 2\)
\(\dfrac{\zeta(s-2)}{\zeta(s)} - \dfrac{\zeta(s-1)^2}{\zeta(s)^2}\)
forเมื่อ \(s > 3\)
Lognormal ล็อกนอร์มอล Continuous ต่อเนื่อง \(x > 0\) \(\mu \in \mathbb{R}, \sigma > 0\)
\(\ln X \sim N(\mu,\sigma^2)\)
\(e^{\mu + \sigma^2/2}\) \((e^{\sigma^2}-1)e^{2\mu + \sigma^2}\)
Weibull ไวบูลล์ Continuous ต่อเนื่อง \(x \ge 0\) shape \(k > 0\), scale \(\lambda > 0\) รูปร่าง \(k > 0\), สเกล \(\lambda > 0\) \(\lambda \Gamma(1+1/k)\) \(\lambda^2\left[\Gamma\!\left(1+\frac{2}{k}\right) - \Gamma^2\!\left(1+\frac{1}{k}\right)\right]\)

Note: \(\Gamma(\cdot)\) is the gamma function, \(B(\cdot,\cdot)\) the beta function and \(\zeta(\cdot)\) the Riemann zeta function. Outside the conditions shown for Pareto and Zipf, the mean or variance is infinite, so sample averages may never settle down. หมายเหตุ: \(\Gamma(\cdot)\) คือฟังก์ชันแกมมา \(B(\cdot,\cdot)\) คือฟังก์ชันเบตา และ \(\zeta(\cdot)\) คือฟังก์ชันซีตาของรีมันน์ นอกเงื่อนไขที่ระบุสำหรับพาเรโตและซิปฟ์ ค่าเฉลี่ยหรือความแปรปรวนจะเป็นอนันต์ ค่าเฉลี่ยจากตัวอย่างจึงอาจไม่นิ่งเลย

Binomial Distribution การแจกแจงทวินาม

Discrete distribution for the number of successes in a fixed number of independent trials. การแจกแจงแบบไม่ต่อเนื่องสำหรับนับจำนวนความสำเร็จในการทดลองอิสระตามจำนวนครั้งที่กำหนด

Binomial with n 10 and p 0.5

Typical shape for \( \mathrm{Bin}(n,p) \) with \(p\) near 0.5. รูปร่างทั่วไปของ \( \mathrm{Bin}(n,p) \) เมื่อ \(p\) ใกล้ 0.5

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Bin}(n,p)\), then ถ้า \(X \sim \mathrm{Bin}(n,p)\) จะได้ว่า \[ P(X = k) = \binom{n}{k} p^k (1-p)^{n-k}, \quad k = 0,1,\dots,n. \]

When is binomial appropriate? เมื่อไหร่ถึงควรใช้ทวินาม?

Where it breaks เมื่อไหร่ที่ใช้ไม่ได้

  • Trials influence each other: flu spreading through a household, or faulty bulbs clustering in a bad batch. The spread comes out wider than the binomial predicts. การทดลองส่งผลต่อกัน: ไข้หวัดที่ติดกันในบ้าน หรือหลอดเสียที่กระจุกอยู่ในล็อตเดียว ทำให้การกระจายกว้างกว่าที่ทวินามทำนาย
  • \(p\) differs from trial to trial: students of different ability answering the same question. \(p\) ต่างกันในแต่ละครั้ง: นักเรียนที่มีความสามารถต่างกันตอบคำถามเดียวกัน
  • There is no fixed number of trials: typos on a page. Consider Poisson. ไม่มีจำนวนการทดลองที่กำหนดไว้: คำผิดในหนึ่งหน้า ควรพิจารณาปัวซง

1. Factory Quality Control (Defective Items) 1. การควบคุมคุณภาพในโรงงาน (สินค้ามีตำหนิ)

Phenomenon: ปรากฏการณ์: A factory produces light bulbs. Historically, 2% are defective. An inspector tests 100 bulbs. โรงงานผลิตหลอดไฟ โดยปกติมีหลอดเสีย 2% ผู้ตรวจสอบสุ่มตัวอย่าง 100 หลอด

Model: โมเดล: Each bulb is a trial; success = "defective". Number of trials \(n = 100\), probability of defect \(p = 0.02\). The number of defective bulbs \(X\) is \(X \sim \mathrm{Bin}(100, 0.02)\). หลอดไฟแต่ละหลอดคือการทดลอง; สำเร็จ = "มีตำหนิ". จำนวนครั้ง \(n = 100\), ความน่าจะเป็นที่จะเสีย \(p = 0.02\). จำนวนหลอดที่เสีย \(X\) คือ \(X \sim \mathrm{Bin}(100, 0.02)\)

Typical questions: คำถามทั่วไป: Probability of no defective bulbs? Probability of more than 5 defective bulbs? ความน่าจะเป็นที่จะไม่เจอหลอดเสียเลย? ความน่าจะเป็นที่จะเจอหลอดเสียมากกว่า 5 หลอด?

2. Genetics: Children With a Recessive Disease 2. พันธุศาสตร์: ลูกที่เป็นโรคทางพันธุกรรมแบบยีนด้อย

Phenomenon: ปรากฏการณ์: Two carrier parents have 4 children. Each child has a 25% chance of having a recessive disease. พ่อแม่ที่เป็นพาหะมีลูก 4 คน แต่ละคนมีโอกาส 25% ที่จะเป็นโรคทางพันธุกรรมแบบยีนด้อย

Model: โมเดล: Trial = one child, success = "child has disease", \(n = 4\), \(p = 0.25\). Then \(X \sim \mathrm{Bin}(4, 0.25)\). การทดลองหนึ่งครั้ง = ลูกหนึ่งคน, สำเร็จ = "ลูกเป็นโรค", \(n = 4\), \(p = 0.25\). ดังนั้น \(X \sim \mathrm{Bin}(4, 0.25)\)

Typical questions: คำถามทั่วไป: Probability none have the disease? Probability at least one has the disease? ความน่าจะเป็นที่ไม่มีใครเป็นโรค? ความน่าจะเป็นที่อย่างน้อยหนึ่งคนเป็นโรค?

3. Website A/B Testing (Clicks on a Button) 3. การทดสอบ A/B ของเว็บไซต์ (การคลิกปุ่ม)

Phenomenon: ปรากฏการณ์: A new button design has a 6% click-through rate. It is shown to 500 visitors. ปุ่มดีไซน์ใหม่มีอัตราการคลิก 6% และแสดงให้ผู้เข้าชม 500 คน

Model: โมเดล: Trial = one visitor, success = "clicked", \(n = 500\), \(p = 0.06\). Then \(X \sim \mathrm{Bin}(500, 0.06)\). การทดลองหนึ่งครั้ง = ผู้ชมหนึ่งคน, สำเร็จ = "คลิก", \(n = 500\), \(p = 0.06\). ดังนั้น \(X \sim \mathrm{Bin}(500, 0.06)\)

Typical questions: คำถามทั่วไป: Probability fewer than 20 clicks? Probability at least 40 clicks (8%)? ความน่าจะเป็นที่จะมีคนคลิกน้อยกว่า 20 ครั้ง? ความน่าจะเป็นที่จะมีคนคลิกอย่างน้อย 40 ครั้ง (8%)?

4. Clinical Trial: Treatment Success 4. การทดลองทางคลินิก: ความสำเร็จในการรักษา

Phenomenon: ปรากฏการณ์: A new drug is given to 80 patients; each has a 70% chance of improvement. ยาใหม่ถูกให้ผู้ป่วย 80 คน แต่ละคนมีโอกาส 70% ที่อาการจะดีขึ้น

Model: โมเดล: Trial = one patient, success = "improves", \(n = 80\), \(p = 0.70\). Then \(X \sim \mathrm{Bin}(80, 0.70)\). การทดลองหนึ่งครั้ง = ผู้ป่วยหนึ่งคน, สำเร็จ = "อาการดีขึ้น", \(n = 80\), \(p = 0.70\). ดังนั้น \(X \sim \mathrm{Bin}(80, 0.70)\)

Typical questions: คำถามทั่วไป: Probability that 60 or more patients improve? Is 45 improvements surprisingly low? ความน่าจะเป็นที่ผู้ป่วย 60 คนขึ้นไปอาการดีขึ้น? การที่ดีขึ้น 45 คนถือว่าน้อยจนน่าแปลกใจหรือไม่?

5. Satellite Solar Cell Failures During Launch 5. ความล้มเหลวของเซลล์แสงอาทิตย์บนดาวเทียมขณะส่ง

Phenomenon: ปรากฏการณ์: A satellite has 30 cells; each has a 1% chance of failing during launch. ดาวเทียมมีเซลล์ 30 เซลล์ แต่ละเซลล์มีโอกาส 1% ที่จะล้มเหลวขณะส่ง

Model: โมเดล: Trial = one cell, success = "fails", \(n = 30\), \(p = 0.01\). Then \(X \sim \mathrm{Bin}(30, 0.01)\). การทดลองหนึ่งครั้ง = เซลล์หนึ่งเซลล์, สำเร็จ = "ล้มเหลว", \(n = 30\), \(p = 0.01\). ดังนั้น \(X \sim \mathrm{Bin}(30, 0.01)\)

Typical questions: คำถามทั่วไป: Probability that no cells fail? Probability 2 or more cells fail? ความน่าจะเป็นที่ไม่มีเซลล์ใดล้มเหลว? ความน่าจะเป็นที่เซลล์ 2 เซลล์ขึ้นไปล้มเหลว?

Binomial pattern: รูปแบบทวินาม: fixed number of trials, two outcomes (success/failure), constant success probability \(p\), and independence between trials. จำนวนการทดลองคงที่, สองผลลัพธ์ (สำเร็จ/ล้มเหลว), ความน่าจะเป็นที่จะสำเร็จ \(p\) คงที่, และการทดลองเป็นอิสระต่อกัน

Poisson Distribution การแจกแจงปัวซง

Discrete distribution for counts of random events in a fixed time/space interval when events happen independently at a constant average rate. การแจกแจงแบบไม่ต่อเนื่องสำหรับนับเหตุการณ์สุ่มในขอบเขตเวลา/พื้นที่ที่กำหนด เมื่อเหตุการณ์เกิดขึ้นอย่างอิสระในอัตราเฉลี่ยคงที่

Poisson with mean 3

Typical shape for \( \mathrm{Poisson}(\lambda) \) with moderate \(\lambda\). รูปร่างทั่วไปของ \( \mathrm{Poisson}(\lambda) \) เมื่อ \(\lambda\) มีค่าปานกลาง

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Poisson}(\lambda)\), then ถ้า \(X \sim \mathrm{Poisson}(\lambda)\) จะได้ว่า \[ P(X = k) = e^{-\lambda} \frac{\lambda^k}{k!}, \quad k = 0,1,2,\dots. \]

When is Poisson appropriate? เมื่อไหร่ถึงควรใช้ปัวซง?

Where it breaks เมื่อไหร่ที่ใช้ไม่ได้

  • The rate changes during the window: calls at rush hour mixed with calls at night. อัตราเปลี่ยนไปในช่วงที่นับ: สายที่โทรเข้าตอนชั่วโมงเร่งด่วนปนกับตอนกลางคืน
  • Events come in bunches: one email thread triggers several replies at once. เหตุการณ์เกิดเป็นกลุ่ม: อีเมลหนึ่งฉบับทำให้มีการตอบกลับหลายฉบับพร้อมกัน
  • A quick check for both: a Poisson's variance equals its mean. If the variance is much larger, something other than independent events at a steady rate is going on. วิธีตรวจสอบอย่างรวดเร็ว: ความแปรปรวนของปัวซงเท่ากับค่าเฉลี่ย ถ้าความแปรปรวนมากกว่ามาก แสดงว่ามีสิ่งอื่นเกิดขึ้น ไม่ใช่แค่เหตุการณ์อิสระในอัตราคงที่

1. Number of Emails Per Hour 1. จำนวนอีเมลต่อชั่วโมง

Phenomenon: ปรากฏการณ์: The number of emails received in an hour, with an average rate of, say, 3 emails per hour. จำนวนอีเมลที่ได้รับในหนึ่งชั่วโมง โดยมีอัตราเฉลี่ย 3 ฉบับต่อชั่วโมง

Model: โมเดล: Let \(\lambda = 3\) emails/hour. The number of emails \(X\) in one hour is \(X \sim \mathrm{Poisson}(3)\). ให้ \(\lambda = 3\) อีเมล/ชั่วโมง จำนวนอีเมล \(X\) ในหนึ่งชั่วโมงคือ \(X \sim \mathrm{Poisson}(3)\)

Typical questions: คำถามทั่วไป: Probability of no emails in the next hour? Probability of 5 or more emails? ความน่าจะเป็นที่จะไม่มีอีเมลในชั่วโมงถัดไป? ความน่าจะเป็นที่จะมี 5 อีเมลขึ้นไป?

2. Radioactive Decay Counts 2. การนับการสลายตัวของกัมมันตรังสี

Phenomenon: ปรากฏการณ์: A Geiger counter records how many decays occur in a 10-second window, say with mean 12 decays/10 s. เครื่องไกเกอร์เคาน์เตอร์บันทึกการสลายตัวในหน้าต่าง 10 วินาที โดยมีค่าเฉลี่ย 12 ครั้งต่อ 10 วินาที

Model: โมเดล: Let \(\lambda = 12\). Then \(X \sim \mathrm{Poisson}(12)\) is the number of decays in 10 seconds. ให้ \(\lambda = 12\). ดังนั้น \(X \sim \mathrm{Poisson}(12)\) คือจำนวนการสลายตัวใน 10 วินาที

Typical questions: คำถามทั่วไป: Probability of exactly 10 decays? Probability of more than 20? ความน่าจะเป็นที่จะเกิดการสลายตัว 10 ครั้งพอดี? ความน่าจะเป็นที่จะเกิดมากกว่า 20 ครั้ง?

3. Cars Passing a Point on a Quiet Road 3. รถยนต์ที่ผ่านจุดหนึ่งบนถนนที่เงียบสงบ

Phenomenon: ปรากฏการณ์: Counting cars passing a fixed point in 5 minutes on a quiet road, with an average of 2 cars per 5 minutes. การนับรถที่ผ่านจุดคงที่ใน 5 นาทีบนถนนที่เงียบสงบ โดยมีค่าเฉลี่ย 2 คันต่อ 5 นาที

Model: โมเดล: Let \(\lambda = 2\). Then \(X \sim \mathrm{Poisson}(2)\) cars per 5 minutes. ให้ \(\lambda = 2\). ดังนั้น \(X \sim \mathrm{Poisson}(2)\) คันต่อ 5 นาที

Typical questions: คำถามทั่วไป: Probability no cars pass? Probability at least 3 cars pass in 5 minutes? ความน่าจะเป็นที่ไม่มีรถผ่านเลย? ความน่าจะเป็นที่มีรถอย่างน้อย 3 คันผ่านใน 5 นาที?

4. Typos on a Page 4. จำนวนคำผิดในหน้ากระดาษ

Phenomenon: ปรากฏการณ์: A printer averages 0.4 misprints per page, and misprints are rare and random. เครื่องพิมพ์มีค่าเฉลี่ยพิมพ์ผิด 0.4 ครั้งต่อหน้า และการพิมพ์ผิดนั้นหายากและสุ่ม

Model: โมเดล: Let \(\lambda = 0.4\). Then \(X \sim \mathrm{Poisson}(0.4)\) misprints per page. A page has no fixed number of "trials", which is why this is Poisson rather than binomial. ให้ \(\lambda = 0.4\). ดังนั้น \(X \sim \mathrm{Poisson}(0.4)\) คำผิดต่อหน้า หนึ่งหน้าไม่มีจำนวน "การทดลอง" ที่กำหนดไว้ จึงใช้ปัวซงแทนทวินาม

Typical questions: คำถามทั่วไป: Probability of zero misprints? Probability of two or more misprints on a page? ความน่าจะเป็นที่ไม่มีคำผิดเลย? ความน่าจะเป็นที่มีคำผิดสองคำขึ้นไปในหนึ่งหน้า?

5. Calls to a Call Center 5. จำนวนสายที่โทรเข้าคอลเซ็นเตอร์

Phenomenon: ปรากฏการณ์: A call center receives calls at an average of 5 per 10 minutes. คอลเซ็นเตอร์ได้รับสายเฉลี่ย 5 สายต่อ 10 นาที

Model: โมเดล: For a 10-minute interval, \(\lambda = 5\) and \(X \sim \mathrm{Poisson}(5)\). สำหรับช่วง 10 นาที, \(\lambda = 5\) และ \(X \sim \mathrm{Poisson}(5)\)

Typical questions: คำถามทั่วไป: Probability of exactly 3 calls in 10 minutes? Probability of no calls in the next 2 minutes (using scaled rate)? ความน่าจะเป็นที่จะมี 3 สายพอดีใน 10 นาที? ความน่าจะเป็นที่ไม่มีสายใน 2 นาทีถัดไป (ปรับอัตราตามความยาวช่วงเวลา)?

Poisson pattern: รูปแบบปัวซง: counting how many events occur in a fixed interval, with independent events, constant average rate, and events not happening in big synchronized bunches. การนับจำนวนเหตุการณ์ในขอบเขตที่กำหนด, เหตุการณ์เป็นอิสระ, อัตราเฉลี่ยคงที่, และเหตุการณ์ไม่เกิดพร้อมกันเป็นกลุ่มใหญ่

Normal Distribution การแจกแจงปกติ

Continuous, bell-shaped distribution often arising as the sum of many small independent effects (Central Limit Theorem). การแจกแจงต่อเนื่องรูปทรงระฆังที่มักเกิดจากผลรวมของปัจจัยย่อยๆ ที่เป็นอิสระต่อกัน (ทฤษฎีบทขีดจำกัดส่วนกลาง)

Normal

Typical symmetric bell-shaped \( N(\mu,\sigma^2) \) density. รูปทรงระฆังสมมาตรทั่วไปของความหนาแน่น \( N(\mu,\sigma^2) \)

Key formula: สูตรสำคัญ: If \(X \sim N(\mu,\sigma^2)\), then ถ้า \(X \sim N(\mu,\sigma^2)\) จะได้ว่า \[ f(x) = \frac{1}{\sqrt{2\pi}\,\sigma} \exp\bigg( -\frac{(x-\mu)^2}{2\sigma^2} \bigg), \quad x \in \mathbb{R}. \]

When is the normal distribution appropriate? เมื่อไหร่ถึงควรใช้การแจกแจงปกติ?

Where it breaks เมื่อไหร่ที่ใช้ไม่ได้

  • Effects multiply or feed on themselves: income, wealth, city sizes and word frequencies have long tails. ปัจจัยคูณกันหรือส่งเสริมตัวเอง: รายได้ ความมั่งคั่ง ขนาดเมือง และความถี่ของคำ มีหางยาว
  • The tail is what matters: a normal model gives extreme events almost no chance, so it can be badly wrong about the very cases that matter most. เมื่อส่วนหางคือสิ่งสำคัญ: โมเดลปกติให้โอกาสเหตุการณ์สุดขั้วน้อยมาก จึงอาจผิดอย่างมากในกรณีที่สำคัญที่สุด
  • The data mixes groups, such as adult heights with men and women pooled: two groups can blur a single bell. ข้อมูลปนกันหลายกลุ่ม เช่น ความสูงของผู้ชายและผู้หญิงรวมกัน ซึ่งสองกลุ่มอาจทำให้ไม่เป็นระฆังใบเดียว
  • The Central Limit Theorem is about sample means, under conditions such as finite variance. It does not say that data are normal. ทฤษฎีบทขีดจำกัดส่วนกลางพูดถึงค่าเฉลี่ยของตัวอย่าง ภายใต้เงื่อนไข เช่น ความแปรปรวนจำกัด ไม่ได้บอกว่าข้อมูลแจกแจงปกติ

1. Human Heights 1. ความสูงของมนุษย์

Phenomenon: ปรากฏการณ์: Heights of adult men, or of adult women, taken separately, cluster around a mean value with symmetry and few extremes. ความสูงของผู้ชาย หรือของผู้หญิง เมื่อแยกกันดู จะเกาะกลุ่มรอบค่าเฉลี่ยอย่างสมมาตรและมีค่าสุดขั้วน้อย

Model: โมเดล: If \(X\) is height, we might model \(X \sim N(\mu, \sigma^2)\), e.g. \(N(175, 7^2)\) cm. ถ้า \(X\) คือความสูง เราอาจใช้โมเดล \(X \sim N(\mu, \sigma^2)\), เช่น \(N(175, 7^2)\) ซม.

Typical questions: คำถามทั่วไป: Fraction taller than 185 cm? Fraction between 170 and 180 cm? สัดส่วนของคนที่สูงกว่า 185 ซม.? สัดส่วนของคนที่อยู่ระหว่าง 170 และ 180 ซม.?

2. Measurement Errors 2. ความคลาดเคลื่อนในการวัด

Phenomenon: ปรากฏการณ์: Repeated measurements of the same quantity (e.g. length of a rod) differ slightly due to many small random errors. การวัดค่าเดิมซ้ำๆ (เช่น ความยาวของแท่งไม้) จะแตกต่างกันเล็กน้อยเนื่องจากความคลาดเคลื่อนสุ่มเล็กๆ จำนวนมาก

Model: โมเดล: Error \(E \sim N(0, \sigma^2)\); measured value \(M = L + E\) is also normal around the true length \(L\). ความคลาดเคลื่อน \(E \sim N(0, \sigma^2)\); ค่าที่วัดได้ \(M = L + E\) ก็จะแจกแจงปกติรอบความยาวจริง \(L\)

Typical questions: คำถามทั่วไป: Probability a measurement is within a tolerance of the true length? Distribution of the sample mean of many measurements? ความน่าจะเป็นที่ค่าที่วัดได้จะอยู่ในช่วงความคลาดเคลื่อนที่ยอมรับได้? การแจกแจงของค่าเฉลี่ยตัวอย่างจากการวัดหลายครั้ง?

3. Exam Scores in a Large Class 3. คะแนนสอบในคลาสขนาดใหญ่

Phenomenon: ปรากฏการณ์: Exam scores from hundreds of students often cluster around an average. It is only an approximation: scores are capped at 0 and 100, and a very easy or very hard exam gives a skewed result. คะแนนสอบของนักเรียนหลายร้อยคนมักเกาะกลุ่มรอบค่าเฉลี่ย แต่เป็นเพียงการประมาณ เพราะคะแนนมีขอบเขตที่ 0 และ 100 และข้อสอบที่ง่ายหรือยากมากจะทำให้ผลเบ้

Model: โมเดล: \(X \sim N(\mu,\sigma^2)\), e.g. \(N(65, 10^2)\) for scores out of 100. \(X \sim N(\mu,\sigma^2)\), เช่น \(N(65, 10^2)\) สำหรับคะแนนเต็ม 100

Typical questions: คำถามทั่วไป: Proportion above 80? Score threshold for top 5%? สัดส่วนคนที่ได้คะแนนเกิน 80? คะแนนขั้นต่ำของกลุ่ม 5% สูงสุดคือเท่าไหร่?

4. Daily Returns of a Broad Market Index (A Warning) 4. ผลตอบแทนรายวันของดัชนีตลาด (ข้อควรระวัง)

Phenomenon: ปรากฏการณ์: Daily percentage returns of a large diversified index are roughly symmetric, with most days close to 0%. But very large moves happen far more often than a normal curve predicts. เปอร์เซ็นต์ผลตอบแทนรายวันของดัชนีขนาดใหญ่ที่มีการกระจายความเสี่ยงค่อนข้างสมมาตร โดยส่วนใหญ่อยู่ใกล้ 0% แต่การเคลื่อนไหวรุนแรงเกิดขึ้นบ่อยกว่าที่เส้นโค้งปกติทำนายมาก

Model: โมเดล: Return \(R \approx N(\mu,\sigma^2)\) is a rough first model for ordinary days. It is the wrong model for crashes, because the real tails are much heavier. ผลตอบแทน \(R \approx N(\mu,\sigma^2)\) เป็นโมเดลคร่าวๆ สำหรับวันปกติ แต่ไม่เหมาะกับช่วงตลาดล่ม เพราะหางของข้อมูลจริงหนักกว่ามาก

Typical questions: คำถามทั่วไป: Range that covers most ordinary days? And how far off is the normal model's chance of a very large one-day loss? ช่วงใดครอบคลุมวันปกติส่วนใหญ่? และโมเดลปกติประเมินโอกาสขาดทุนมากในวันเดียวคลาดเคลื่อนแค่ไหน?

5. Blood Pressure in a Population 5. ความดันโลหิตในประชากร

Phenomenon: ปรากฏการณ์: Systolic blood pressure in a reasonably homogeneous group is often unimodal and roughly symmetric. ความดันโลหิตซิสโตลิกในกลุ่มประชากรที่ค่อนข้างคล้ายคลึงกันมักมีโหมดเดียวและสมมาตร

Model: โมเดล: \(X \sim N(\mu,\sigma^2)\), e.g. \(N(120, 15^2)\) mmHg. \(X \sim N(\mu,\sigma^2)\), เช่น \(N(120, 15^2)\) มม.ปรอท

Typical questions: คำถามทั่วไป: Fraction with pressure above 140 mmHg? Interval covering about 68% of the population? สัดส่วนของคนที่มีความดันสูงกว่า 140 มม.ปรอท? ช่วงที่ครอบคลุมประมาณ 68% ของประชากร?

Normal pattern: รูปแบบปกติ: continuous variable, many small independent contributing factors, roughly symmetric, single peak (no strong skew or truncation). ตัวแปรต่อเนื่อง, ปัจจัยย่อยที่เป็นอิสระจำนวนมาก, ค่อนข้างสมมาตร, มีจุดสูงสุดเดียว (ไม่เบ้หรือถูกตัดขอบเขต)

Other Classical Distributions การแจกแจงคลาสสิกอื่นๆ

Beyond binomial/Poisson/normal: distributions for trials until success, waiting times, proportions, lifetimes, and heavy tails. นอกเหนือจากทวินาม/ปัวซง/ปกติ: การแจกแจงสำหรับการทดลองจนสำเร็จ, เวลารอคอย, สัดส่วน, อายุการใช้งาน, และหางหนัก

1. Geometric Distribution 1. การแจกแจงเรขาคณิต

Discrete ไม่ต่อเนื่อง Geometric with p 0.4

Phenomenon: ปรากฏการณ์: Number of attempts until first success, e.g. number of times a student retries connecting to Wi-Fi until it works. จำนวนครั้งที่พยายามจนกว่าจะสำเร็จครั้งแรก เช่น จำนวนครั้งที่นักเรียนพยายามเชื่อมต่อ Wi-Fi จนกว่าจะติด

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Geom}(p)\) counts trials until the first success, then ถ้า \(X \sim \mathrm{Geom}(p)\) นับจำนวนครั้งจนสำเร็จครั้งแรก จะได้ว่า \[ P(X = k) = (1-p)^{k-1} p, \quad k = 1,2,3,\dots. \]

Key idea: แนวคิดหลัก: Independent trials with success probability \(p\), count trials until the first success. การทดลองอิสระด้วยความน่าจะเป็นที่จะสำเร็จ \(p\), นับจำนวนครั้งจนกว่าจะสำเร็จครั้งแรก

2. Negative Binomial Distribution 2. การแจกแจงทวินามนิเสธ

Discrete ไม่ต่อเนื่อง

Phenomenon: ปรากฏการณ์: Number of shots needed to score a fixed number of goals, e.g. how many free throws to make 3 successes when each shot succeeds with probability \(p\). จำนวนครั้งที่ต้องยิงเพื่อทำประตูให้ได้ตามจำนวนที่กำหนด เช่น ต้องชู้ตบาสกี่ครั้งจึงจะลง 3 ลูก เมื่อความน่าจะเป็นที่จะลงคือ \(p\)

Key formula: สูตรสำคัญ: If \(X\) is the number of trials needed to achieve \(r\) successes, \(X \sim \mathrm{NegBin}(r,p)\), then ถ้า \(X\) คือจำนวนครั้งที่ต้องทดลองเพื่อให้สำเร็จ \(r\) ครั้ง, \(X \sim \mathrm{NegBin}(r,p)\) จะได้ว่า \[ P(X = k) = \binom{k-1}{r-1} p^r (1-p)^{k-r}, \quad k = r, r+1, \dots. \]

Key idea: แนวคิดหลัก: Generalizes geometric: number of trials needed for \(r\) successes. เป็นส่วนขยายของเรขาคณิต: จำนวนครั้งที่ต้องทดลองเพื่อให้สำเร็จ \(r\) ครั้ง

3. Hypergeometric Distribution 3. การแจกแจงไฮเพอร์จีออเมตริก

Discrete ไม่ต่อเนื่อง

Phenomenon: ปรากฏการณ์: Drawing balls without replacement from a finite population, e.g. number of red balls in 5 draws from an urn with 10 red and 20 blue balls. การสุ่มหยิบลูกบอลโดยไม่ใส่คืนจากประชากรจำกัด เช่น จำนวนลูกบอลสีแดงจากการหยิบ 5 ลูก จากโถที่มีสีแดง 10 ลูกและสีน้ำเงิน 20 ลูก

Key formula: สูตรสำคัญ: If population size is \(N\) with \(K\) "successes", and you sample \(n\) without replacement, then ถ้าประชากรขนาด \(N\) มี "สิ่งที่สนใจ" \(K\) ชิ้น, และสุ่ม \(n\) ชิ้นโดยไม่ใส่คืน จะได้ว่า \[ P(X = x) = \frac{\binom{K}{x}\binom{N-K}{n-x}}{\binom{N}{n}}, \quad \max(0,n-(N-K)) \le x \le \min(n,K). \]

Key idea: แนวคิดหลัก: Sampling without replacement changes probabilities after each draw. การสุ่มโดยไม่ใส่คืนทำให้ความน่าจะเป็นเปลี่ยนไปในแต่ละครั้งที่หยิบ

4. Exponential Distribution 4. การแจกแจงเลขชี้กำลัง

Continuous ต่อเนื่อง Exponential

Phenomenon: ปรากฏการณ์: Waiting time until the next random event, like the time until the next bus if buses arrive independently at a constant average rate. (Buses that keep to a timetable do not.) เวลารอคอยจนถึงเหตุการณ์สุ่มครั้งถัดไป เช่น เวลารอรถบัสคันถัดไป ถ้ารถบัสมาอย่างอิสระในอัตราเฉลี่ยคงที่ (รถบัสที่วิ่งตามตารางเวลาไม่เป็นแบบนี้)

Key formula: สูตรสำคัญ: If \(T \sim \mathrm{Exp}(\lambda)\), then ถ้า \(T \sim \mathrm{Exp}(\lambda)\) จะได้ว่า \[ f_T(t) = \lambda e^{-\lambda t}, \quad t \ge 0. \]

Key idea: แนวคิดหลัก: Models time between Poisson events; has the memoryless property. โมเดลเวลาระหว่างเหตุการณ์ปัวซง; มีคุณสมบัติไร้ความจำ (memoryless)

5. Gamma Distribution 5. การแจกแจงแกมมา

Continuous ต่อเนื่อง Gamma with shape 3

Phenomenon: ปรากฏการณ์: Time until the \(k\)-th event, such as time until the 3rd earthquake in a region when earthquakes occur randomly over time. เวลารอคอยจนถึงเหตุการณ์ที่ \(k\), เช่น เวลารอจนเกิดแผ่นดินไหวครั้งที่ 3 ในพื้นที่เมื่อแผ่นดินไหวเกิดขึ้นแบบสุ่ม

Key formula: สูตรสำคัญ: For shape \(k > 0\), scale \(\theta > 0\), if \(X \sim \mathrm{Gamma}(k,\theta)\), then สำหรับรูปร่าง \(k > 0\), สเกล \(\theta > 0\), ถ้า \(X \sim \mathrm{Gamma}(k,\theta)\) จะได้ว่า \[ f_X(x) = \frac{1}{\Gamma(k)\theta^k} x^{k-1} e^{-x/\theta}, \quad x \ge 0. \]

Key idea: แนวคิดหลัก: For whole-number \(k\), the sum of \(k\) independent exponential waiting times. เมื่อ \(k\) เป็นจำนวนเต็ม คือผลรวมของเวลารอคอยแบบ exponential ที่เป็นอิสระต่อกัน \(k\) ค่า

6. Uniform Distribution 6. การแจกแจงเอกรูป

Continuous ต่อเนื่อง Uniform

Phenomenon: ปรากฏการณ์: Any time in an interval is equally likely, like a train that is equally likely to arrive at any moment between 10:00 and 10:30. ทุกช่วงเวลาในขอบเขตมีความน่าจะเป็นเท่ากัน เช่น รถไฟที่มีโอกาสมาถึงเท่ากันทุกขณะระหว่าง 10:00 ถึง 10:30

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Unif}(a,b)\), then ถ้า \(X \sim \mathrm{Unif}(a,b)\) จะได้ว่า \[ f_X(x) = \frac{1}{b-a}, \quad a \le x \le b. \]

Key idea: แนวคิดหลัก: Constant probability density over an interval \([a,b]\). ความหนาแน่นของความน่าจะเป็นคงที่ตลอดช่วง \([a,b]\)

7. Beta Distribution 7. การแจกแจงเบตา

Continuous ต่อเนื่อง Beta with shape 2 and 5

Phenomenon: ปรากฏการณ์: Modeling an unknown probability or proportion, like the true support rate for a political candidate. การโมเดลความน่าจะเป็นหรือสัดส่วนที่ไม่ทราบค่า เช่น อัตราการสนับสนุนที่แท้จริงของผู้สมัครทางการเมือง

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Beta}(\alpha,\beta)\), then ถ้า \(X \sim \mathrm{Beta}(\alpha,\beta)\) จะได้ว่า \[ f_X(x) = \frac{1}{B(\alpha,\beta)} x^{\alpha-1}(1-x)^{\beta-1}, \quad 0 \le x \le 1. \]

Key idea: แนวคิดหลัก: Defined on \([0,1]\); commonly used as a prior for probabilities in Bayesian statistics. มีค่าในช่วง \([0,1]\); มักใช้เป็น prior สำหรับความน่าจะเป็นในสถิติแบบเบย์

8. Pareto Distribution 8. การแจกแจงพาเรโต

Continuous ต่อเนื่อง Pareto with shape 1.5

Phenomenon: ปรากฏการณ์: Heavy-tailed positive quantities such as high incomes or wealth, where a few individuals hold very large values. ปริมาณบวกที่มีหางหนัก เช่น รายได้หรือความมั่งคั่งสูงๆ ที่มีคนส่วนน้อยถือครองค่าจำนวนมาก

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Pareto}(\alpha,x_m)\) with shape \(\alpha\) and scale \(x_m\), then ถ้า \(X \sim \mathrm{Pareto}(\alpha,x_m)\) มีรูปร่าง \(\alpha\) และสเกล \(x_m\) จะได้ว่า \[ f_X(x) = \alpha x_m^\alpha x^{-(\alpha+1)}, \quad x \ge x_m. \]

Key idea: แนวคิดหลัก: Power-law tail; can produce "80/20" style concentration. The mean is infinite for \(\alpha \le 1\), and the variance for \(\alpha \le 2\). หางแบบ Power-law; ทำให้เกิดการกระจุกตัวแบบ "80/20" ได้ ค่าเฉลี่ยเป็นอนันต์เมื่อ \(\alpha \le 1\) และความแปรปรวนเป็นอนันต์เมื่อ \(\alpha \le 2\)

9. Lognormal Distribution 9. การแจกแจงล็อกนอร์มอล

Continuous ต่อเนื่อง Lognormal with sigma 0.7

Phenomenon: ปรากฏการณ์: Positive quantities formed by multiplicative effects, e.g. stock prices evolving as products of daily factors. ปริมาณบวกที่เกิดจากผลคูณ เช่น ราคาหุ้นที่เปลี่ยนแปลงเป็นผลคูณของปัจจัยรายวัน

Key formula: สูตรสำคัญ: If \(\ln X \sim N(\mu,\sigma^2)\), then ถ้า \(\ln X \sim N(\mu,\sigma^2)\) จะได้ว่า \[ f_X(x) = \frac{1}{x\sigma\sqrt{2\pi}} \exp\bigg( -\frac{(\ln x - \mu)^2}{2\sigma^2} \bigg), \quad x > 0. \]

Key idea: แนวคิดหลัก: If log of a variable is normal, the variable itself is lognormal; often right-skewed and strictly positive. ถ้าค่าลอการิทึมของตัวแปรแจกแจงปกติ ตัวแปรนั้นจะแจกแจงล็อกนอร์มอล; มักเบ้ขวาและมีค่าเป็นบวกเท่านั้น

10. Weibull Distribution 10. การแจกแจงไวบูลล์

Continuous ต่อเนื่อง Weibull with shape 2

Phenomenon: ปรากฏการณ์: Lifetimes of mechanical components (ball bearings, devices), capturing early failures or wear-out behavior. อายุการใช้งานของชิ้นส่วนเครื่องจักร (ตลับลูกปืน, อุปกรณ์), ใช้อธิบายการเสียในช่วงแรกหรือการสึกหรอ

Key formula: สูตรสำคัญ: If \(X \sim \mathrm{Weibull}(k,\lambda)\) with shape \(k\) and scale \(\lambda\), then ถ้า \(X \sim \mathrm{Weibull}(k,\lambda)\) มีรูปร่าง \(k\) และสเกล \(\lambda\) จะได้ว่า \[ f_X(x) = \frac{k}{\lambda}\left(\frac{x}{\lambda}\right)^{k-1} \exp\!\left[-\left(\frac{x}{\lambda}\right)^k\right], \quad x \ge 0. \]

Key idea: แนวคิดหลัก: Flexible failure rate (decreasing, constant, or increasing) depending on its shape parameter; widely used in reliability engineering. อัตราการล้มเหลวที่ยืดหยุ่น (ลดลง, คงที่, หรือเพิ่มขึ้น) ขึ้นอยู่กับพารามิเตอร์รูปร่าง; ใช้กว้างขวางในวิศวกรรมความน่าเชื่อถือ

Esoteric / Specialized Distributions การแจกแจงเฉพาะทาง / ซับซ้อน

Less common but important in advanced modeling: digits, ranks, angles, eigenvalues, extremes, and fractals. พบน้อยกว่าแต่สำคัญในการโมเดลขั้นสูง: ตัวเลข, อันดับ, มุม, ค่าไอเกน, ค่าสุดขั้ว, และแฟร็กทัล

1. Benford's Law (Benford Distribution) 1. กฎของเบนฟอร์ด (การแจกแจงเบนฟอร์ด)

Discrete ไม่ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: First digits of many real-world numerical data sets (financial data, populations, physical constants) are not uniform. Digit 1 appears more often than 9. เลขโดดตัวแรกของชุดข้อมูลในโลกจริง (ข้อมูลการเงิน, ประชากร, ค่าคงที่ทางฟิสิกส์) ไม่ได้มีการแจกแจงเอกรูป เลข 1 ปรากฏบ่อยกว่าเลข 9

Key formula: สูตรสำคัญ: For first digit \(d = 1,\dots,9\), สำหรับเลขโดดตัวแรก \(d = 1,\dots,9\), \[ P(D = d) = \log_{10}\!\left(1 + \frac{1}{d}\right). \]

Key idea: แนวคิดหลัก: Works best for data spanning several orders of magnitude. Used in fraud screening, where fabricated data may not match this pattern, but a mismatch is a reason to look closer, not proof of fraud. ใช้ได้ดีกับข้อมูลที่ครอบคลุมหลายอันดับขนาด ใช้คัดกรองการฉ้อโกงเพราะข้อมูลที่แต่งขึ้นอาจไม่ตรงกับรูปแบบนี้ แต่การไม่ตรงเป็นเหตุให้ตรวจสอบเพิ่ม ไม่ใช่หลักฐานว่าฉ้อโกง

2. Zipf's Law 2. กฎของซิปฟ์

Discrete ไม่ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Frequencies of words, city sizes, website hits often follow a rank-based power law: a few items are extremely common or large, many are rare or small. ความถี่ของคำ, ขนาดเมือง, ยอดเข้าชมเว็บไซต์ มักเป็นไปตามกฎ power law ตามอันดับ: มีบางส่วนที่พบบ่อยหรือมีขนาดใหญ่มาก และส่วนใหญ่ที่หายากหรือมีขนาดเล็ก

Key formula: สูตรสำคัญ: For rank \(k = 1,2,\dots\), สำหรับอันดับ \(k = 1,2,\dots\), \[ P(X = k) = \frac{k^{-s}}{\zeta(s)}, \] where \(s > 1\) and \(\zeta(s)\) is the Riemann zeta function. โดยที่ \(s > 1\) และ \(\zeta(s)\) คือฟังก์ชันซีตาของรีมันน์

Key idea: แนวคิดหลัก: Creates heavy tails and "80/20"-type behavior. The mean is infinite for \(s \le 2\), so a sample average never settles down. สร้างพฤติกรรมแบบหางหนักและแบบ "80/20" ค่าเฉลี่ยเป็นอนันต์เมื่อ \(s \le 2\) ค่าเฉลี่ยจากตัวอย่างจึงไม่นิ่ง

3. Wigner Semicircle Distribution 3. การแจกแจงครึ่งวงกลมของวิกเนอร์

Continuous ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Distribution of eigenvalues of large random symmetric matrices (random matrix theory), relevant in nuclear physics, number theory, and communications. การแจกแจงของค่าไอเกนของเมทริกซ์สมมาตรสุ่มขนาดใหญ่ (ทฤษฎีเมทริกซ์สุ่ม) เกี่ยวข้องกับฟิสิกส์นิวเคลียร์, ทฤษฎีจำนวน, และการสื่อสาร

Key formula: สูตรสำคัญ: With radius \(R\), the density is ด้วยรัศมี \(R\), ความหนาแน่นคือ \[ f(x) = \frac{2}{\pi R^2} \sqrt{R^2 - x^2}, \quad |x| \le R. \]

Key idea: แนวคิดหลัก: Describes how eigenvalues spread when matrix size goes to infinity. อธิบายว่าค่าไอเกนกระจายตัวอย่างไรเมื่อขนาดของเมทริกซ์เข้าใกล้อนันต์

4. Lévy Alpha-Stable Distributions 4. การแจกแจงเสถียรอัลฟาของเลวี

Continuous ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Heavy-tailed processes such as large financial jumps or anomalous diffusion where extremely large deviations occur more often than a normal distribution predicts. กระบวนการหางหนัก เช่น การกระโดดของราคาทางการเงิน หรือการแพร่ที่ผิดปกติ ที่ค่าเบี่ยงเบนสุดขั้วเกิดขึ้นบ่อยกว่าที่การแจกแจงปกติคาดการณ์

Formula remark: ข้อสังเกตเกี่ยวกับสูตร: In general there is no simple closed-form pdf, except for special cases (normal, Cauchy, Lévy). They are defined via their characteristic function. โดยทั่วไปไม่มี pdf ในรูปปิดที่ง่าย ยกเว้นกรณีพิเศษ (ปกติ, โคชี, เลวี) มันถูกนิยามผ่านฟังก์ชันลักษณะเฉพาะ (characteristic function)

Key idea: แนวคิดหลัก: Normal is the special case \(\alpha = 2\), and every other stable law has infinite variance. Sums of such variables stay in the same family (stability property). การแจกแจงปกติเป็นกรณีพิเศษที่ \(\alpha = 2\) ส่วนกฎเสถียรอื่นทั้งหมดมีความแปรปรวนอนันต์ ผลรวมของตัวแปรเหล่านี้ยังคงอยู่ตระกูลเดิม (คุณสมบัติความเสถียร)

5. Conway–Maxwell–Poisson (COM-Poisson) Distribution 5. การแจกแจงคอนเวย์–แมกซ์เวลล์–ปัวซง

Discrete ไม่ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Count data with over-dispersion or under-dispersion relative to Poisson, such as queue lengths or manufacturing counts that are more or less variable than Poisson assumes. ข้อมูลนับที่มีการกระจายตัวมากเกินไป (over-dispersion) หรือน้อยเกินไป (under-dispersion) เมื่อเทียบกับปัวซง เช่น ความยาวคิว หรือจำนวนนับในการผลิต ที่แปรปรวนมากกว่าหรือน้อยกว่าที่ปัวซงคาดไว้

Key formula: สูตรสำคัญ: For parameters \(\lambda > 0\) and \(\nu \ge 0\) (with \(\lambda < 1\) when \(\nu = 0\)), สำหรับพารามิเตอร์ \(\lambda > 0\) และ \(\nu \ge 0\) (โดย \(\lambda < 1\) เมื่อ \(\nu = 0\)) \[ P(X = k) = \frac{\lambda^k}{(k!)^\nu Z(\lambda,\nu)}, \quad k = 0,1,2,\dots, \] where \(Z(\lambda,\nu) = \sum_{j=0}^\infty \frac{\lambda^j}{(j!)^\nu}\) is a normalizing constant. โดยที่ \(Z(\lambda,\nu) = \sum_{j=0}^\infty \frac{\lambda^j}{(j!)^\nu}\) คือค่าคงที่ปรับฐาน (normalizing constant)

Key idea: แนวคิดหลัก: Two-parameter generalization of Poisson with an extra dispersion parameter. การขยายทั่วไปของปัวซงเป็นสองพารามิเตอร์ โดยเพิ่มพารามิเตอร์การกระจาย

6. Dirichlet Distribution 6. การแจกแจงดีริคเลต์

Continuous ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Probabilities themselves as random objects: topic proportions in documents, allele frequencies in genetics, or category probabilities in Bayesian models. ความน่าจะเป็นที่เป็นตัวแปรสุ่ม: สัดส่วนหัวข้อในเอกสาร, ความถี่อัลลีลในพันธุศาสตร์, หรือความน่าจะเป็นของหมวดหมู่ในโมเดลเบย์

Key formula: สูตรสำคัญ: For \(p_1,\dots,p_k \ge 0\) with \(\sum p_i = 1\) and parameters \(\alpha_i > 0\), สำหรับ \(p_1,\dots,p_k \ge 0\) โดยที่ \(\sum p_i = 1\) และพารามิเตอร์ \(\alpha_i > 0\), \[ f(p_1,\dots,p_k) = \frac{1}{B(\boldsymbol{\alpha})} \prod_{i=1}^k p_i^{\alpha_i-1}, \] where \(B(\boldsymbol{\alpha})\) is a multivariate beta function. โดยที่ \(B(\boldsymbol{\alpha})\) คือฟังก์ชันเบตาหลายตัวแปร

Key idea: แนวคิดหลัก: Distribution over vectors \((p_1,\dots,p_k)\) that sum to 1. การแจกแจงบนเวกเตอร์ \((p_1,\dots,p_k)\) ที่มีผลรวมเป็น 1

7. Von Mises Distribution 7. การแจกแจงฟอนไมซีส

Continuous (circular) ต่อเนื่อง (แบบวงกลม) Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Directional data on a circle, such as wind direction or orientation of animals' movement. ข้อมูลทิศทางบนวงกลม เช่น ทิศทางลม หรือทิศทางการเคลื่อนที่ของสัตว์

Key formula: สูตรสำคัญ: For angle \(\theta\), mean direction \(\mu\), and concentration \(\kappa\), สำหรับมุม \(\theta\), ทิศทางเฉลี่ย \(\mu\), และความเข้มข้น \(\kappa\), \[ f(\theta) = \frac{1}{2\pi I_0(\kappa)} \exp\big(\kappa \cos(\theta - \mu)\big), \] where \(I_0\) is the modified Bessel function of the first kind. โดยที่ \(I_0\) คือฟังก์ชันเบสเซลชนิดปรับปรุงประเภทที่หนึ่ง

Key idea: แนวคิดหลัก: A "circular normal" distribution. การแจกแจง "ปกติแบบวงกลม"

8. Gumbel Distribution (Extreme Value Type I) 8. การแจกแจงกัมเบล (ค่าสุดขั้วประเภทที่ 1)

Continuous ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Maxima of samples over time, such as yearly maximum rainfall, river levels, or extreme wind speeds. ค่าสูงสุดของตัวอย่างในช่วงเวลา เช่น ปริมาณน้ำฝนสูงสุดรายปี, ระดับแม่น้ำ, หรือความเร็วลมสุดขั้ว

Key formula: สูตรสำคัญ: For location \(\mu\) and scale \(\beta > 0\), สำหรับตำแหน่ง \(\mu\) และสเกล \(\beta > 0\), \[ f(x) = \frac{1}{\beta} \exp\!\left(-\frac{x-\mu}{\beta}\right)\exp\!\left(-e^{-(x-\mu)/\beta}\right), \quad x \in \mathbb{R}. \]

Key idea: แนวคิดหลัก: Comes from extreme value theory; models the distribution of maxima for many underlying distributions under certain conditions. มาจากทฤษฎีค่าสุดขั้ว; โมเดลการแจกแจงของค่าสูงสุดสำหรับการแจกแจงพื้นฐานหลายชนิดภายใต้เงื่อนไขบางอย่าง

9. Skellam Distribution 9. การแจกแจงสเกลลัม

Discrete ไม่ต่อเนื่อง Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Difference of two independent Poisson counts, e.g. goal difference in a match when each team’s goals are Poisson. ผลต่างของการนับแบบปัวซงสองค่าที่เป็นอิสระต่อกัน เช่น ผลต่างประตูในการแข่งขันเมื่อประตูของแต่ละทีมเป็นแบบปัวซง

Key formula: สูตรสำคัญ: If \(N_1 \sim \mathrm{Poisson}(\lambda_1)\), \(N_2 \sim \mathrm{Poisson}(\lambda_2)\) independent, and \(X = N_1 - N_2\), then ถ้า \(N_1 \sim \mathrm{Poisson}(\lambda_1)\), \(N_2 \sim \mathrm{Poisson}(\lambda_2)\) เป็นอิสระต่อกัน, และ \(X = N_1 - N_2\) จะได้ว่า \[ P(X = k) = e^{-(\lambda_1+\lambda_2)}\left(\frac{\lambda_1}{\lambda_2}\right)^{k/2} I_{|k|}\!\big(2\sqrt{\lambda_1\lambda_2}\big), \] where \(I_{|k|}\) is a modified Bessel function. โดยที่ \(I_{|k|}\) คือฟังก์ชันเบสเซลชนิดปรับปรุง

Key idea: แนวคิดหลัก: Takes values on all integers (negative, zero, positive); natural once you model goals for each team with Poisson distributions. มีค่าเป็นจำนวนเต็มทั้งหมด (ลบ, ศูนย์, บวก); เป็นธรรมชาติเมื่อคุณโมเดลประตูของแต่ละทีมด้วยการแจกแจงปัวซง

10. Cantor Distribution 10. การแจกแจงแคนเทอร์

Continuous (singular) ต่อเนื่อง (เอกฐาน) Esoteric เฉพาะทาง

Phenomenon: ปรากฏการณ์: Mostly theoretical: used as a classic example of a distribution supported on a fractal set (the Cantor set) with very unusual properties. ส่วนใหญ่เป็นทฤษฎี: ใช้เป็นตัวอย่างคลาสสิกของการแจกแจงที่อยู่บนเซตแฟร็กทัล (เซตแคนเทอร์) ที่มีคุณสมบัติแปลกประหลาด

Formula remark: ข้อสังเกตเกี่ยวกับสูตร: Defined via an iterative construction of its cumulative distribution function; it has no pdf in the usual sense. นิยามผ่านการสร้างฟังก์ชันการแจกแจงสะสม (cdf) แบบวนซ้ำ; ไม่มี pdf ในความหมายทั่วไป

Key idea: แนวคิดหลัก: Continuous cdf but no density; supported on a set of measure zero. Important as a pathological example in probability and analysis. cdf ต่อเนื่องแต่ไม่มีฟังก์ชันความหนาแน่น; อยู่บนเซตที่มีเมเชอร์เป็นศูนย์ มีความสำคัญในฐานะตัวอย่างที่ผิดปกติ (pathological) ในทฤษฎีความน่าจะเป็นและการวิเคราะห์