Frequenist Methods

Frequentist school of statistics

Introduction

  • ๋ฒ ์ด์ง€์•ˆ ํ†ต๊ณ„ ๋‹ค์Œ์œผ๋กœ ๋“ฑ์žฅํ•œ ํ†ต๊ณ„ํ•™ํŒŒ

  • 20์„ธ๊ธฐ์— ์ง€๋ฐฐ์ ์ธ ํ•™ํŒŒ์˜€์Œ

  • confidence intervals, p-values, t-test, x^2-test๊ฐ€ ์ด์— ํ•ด๋‹น๋จ

  • ์ปดํ“จํ„ฐ์˜ ๋“ฑ์žฅ ์ดํ›„๋กœ๋Š” bayesian method๊ฐ€ ๊ฐ€์žฅ ๋ณดํŽธ์ ์ธ ๋ฐฉ๋ฒ•์ด ๋จ

The fork in the road

  • Bayesian inference: ์ง€๊ธˆ๊นŒ์ง€ ๊ณต๋ถ€ํ–ˆ๋˜ ๋ฐฉ๋ฒ•

    • H๊ฐ€ hypothesis, D๊ฐ€ data

    • prior๋ฅผ ํ™•์‹คํžˆ ์•ˆ๋‹ค๋ฉด ์ •ํ™•ํ•˜๊ฒŒ ์ž‘๋™ํ•จ: ๊ทธ๋Ÿฌ๋‚˜ ์‹ค์ œ๋กœ ์™„๋ฒฝํ•œ prior๋Š” ์—†์Œ

      -> Bayesian์€ prior, Frequentist๋Š” likelyhood func.์„ ์‚ฌ์šฉํ•จ

What is probability?

  • Frequentist

    • ๊ณ ์ • ๊ฐ’์„ ๊ฐ€์ง„ ํŒŒ๋ผ๋ฏธํ„ฐ์˜ ํ™•๋ฅ  ๋ถ„ํฌ๋ฅผ ๋ฌด์˜๋ฏธํ•˜๋‹ค๊ณ  ์ƒ๊ฐํ•จ, ๊ฐ€์„ค์— ๋Œ€ํ•œ ์‹ ๋ขฐ๋„๋ฅผ ์ •๋Ÿ‰ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ํ™•๋ฅ  ์‚ฌ์šฉ์„ ๊ฑฐ๋ถ€ํ•จ

      • ex) ๋™์ „์„ ๋˜์กŒ์„ ๋•Œ ์•ž๋ฉด์ด ๋‚˜์˜ฌ ํ™•๋ฅ  1/2 -> ๋™์ „์„ ๋˜์ง€๋Š” ํšŸ์ˆ˜๊ฐ€ ๋ฌดํ•œ๋Œ€๋กœ ๊ฐˆ์ˆ˜๋ก 1/2์— ๊ฐ€๊นŒ์›Œ์ง

    • ๊ฐ€์„ค์ด ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ์— ํ™•๋ฅ  ๋ถ„ํฌ(๋žœ๋ค, ๋ฐ˜๋ณต ๊ฐ€๋Šฅ, ์‹คํ—˜์ )๋ฅผ ์ ์šฉํ•จ

  • Bayesian

    • ๊ณ ์ •๋œ ํŒŒ๋ผ๋ฏธํ„ฐ์— ๋Œ€ํ•ด ๋ถˆ์™„์ „ํ•œ ์ง€์‹์„ ์„ค๋ช…ํ•˜๊ธฐ ์œ„ํ•ด ํ™•๋ฅ ์„ ์‚ฌ์šฉํ•จ

    • ๋ชจ๋“  ๊ฒƒ(๊ฐ€์„ค, ๋ฐ์ดํ„ฐ)์— ํ™•๋ฅ  ๋ถ„ํฌ๋ฅผ ์ ์šฉํ•จ


Null Hypothesis Significance Testing 1

Introduction

  • Neyman-Pearson ํŒจ๋Ÿฌ๋‹ค์ž„์„ ์ฃผ๋กœ ์‚ฌ์šฉํ•จ

    • ex) ๋™์ „์„ 10๋ฒˆ ๋˜์กŒ์„ ๋•Œ ์•ž๋ฉด์ด ๋‚˜์˜ค๋Š” ํšŸ์ˆ˜

    Null_hypothesis

Types of error

Composite hypothesis

ex) ๋™์ „์„ ๋˜์กŒ์„ ๋•Œ ๋‚˜์˜ฌ ์ˆ˜ ์žˆ๋Š” ์•ž๋ฉด์˜ ๊ฐœ์ˆ˜์˜ ํ‰๊ท 

Null distribution:

High and low power test

  • both standard normal

  • High power: ๋†’์€ ๊ฒ€์ •๋ ฅ

  • Low power: ๋‚ฎ์€ ๊ฒ€์ •๋ ฅ

Designing a hypothesis test

  • Pick the null hypothesis H0.

  • Decide if HA is one-sided or two-sided

  • Pick the test statistic.

    • ex) z-test, t-test, x^2-test

  • Pick a significance level and determine the rejection region

    • ex) 0.1, 0.05, 0.001

  • Determine the power

Critical values

ex) critical value = 0.05

p-values

  • reject region์˜ ๋ฉด์ 

ex) IQ์˜ normal distribution = N(100, 15^2)

9๋ช…์˜ ํ•™์ƒ๋“ค์˜ ํ‰๊ท  IQ๊ฐ€ 112์ผ ๋•Œ ์‹ ๋ขฐ์ˆ˜์ค€ 0.05๋กœ H0์„ rejectํ•  ์ˆ˜ ์žˆ๋Š”๊ฐ€?

reject H0

+์˜ˆ์‹œ

Last updated

Was this helpful?