hypothesis testing (๊ฐ€์„ค๊ฒ€์ •)

null hypothesis(h0, ๊ท€๋ฌด๊ฐ€์„ค)

The hypothesis that we want to varify during the research.
์—ฐ๊ตฌ์—์„œ ์ฃผ์žฅํ•˜๊ณ  ์‹ถ์€(์ฑ„ํƒํ•˜๊ณ  ์‹ถ์€) ๋‚ด์šฉ๊ณผ ๋ฐ˜๋Œ€๋˜๋Š” ๊ฐ€์„ค.


alternative hypothesis(h1, ๋Œ€๋ฆฝ๊ฐ€์„ค)

The opposite of the null hypothesis.
๊ท€๋ฌด๊ฐ€์„ค์ด ๊ธฐ๊ฐ๋˜์—ˆ์„ ๋•Œ, ๋Œ€์‹  ์ฑ„ํƒ๋˜๋Š” ๊ฐ€์„ค. ์—ฐ๊ตฌ์—์„œ ์ฃผ์žฅํ•˜๊ณ  ์‹ถ์€ ๋‚ด์šฉ.


hypothesis testing(๊ฐ€์„ค๊ฒ€์ •)

๊ด€์ธก๋œ ๋ฐ์ดํ„ฐ๋กœ๋ถ€ํ„ฐ ๋ชจ์ง‘๋‹จ์— ๋Œ€ํ•œ ๊ฐ€์„ค์„ ์„ค์ •ํ•˜๊ณ  ํ‘œ๋ณธ์„ ํ†ตํ•ด ์–ป๋Š” ์ •๋ณด์— ๋”ฐ๋ผ ํƒ€๋‹น์„ฑ์„ ๊ฒ€์ฆ.

  1. ๊ฐ€์„ค ์„ค์ • (๊ธฐ๊ฐํ•˜๊ณ  ์‹ถ์€ ๊ท€๋ฌด๊ฐ€์„ค / ์ฑ„ํƒํ•˜๊ณ  ์‹ถ์€ ๋Œ€๋ฆฝ๊ฐ€์„ค)
  2. ํ™•๋ฅ ๋ถ„ํฌ์™€ ํŒ์ • ๊ธฐ์ค€(์œ ์˜์ˆ˜์ค€) ๊ฒฐ์ •
  3. ๊ฒ€์ •ํ†ต๊ณ„๋Ÿ‰ ๊ณ„์‚ฐ
  4. ํ™•๋ฅ ๊ณ„์‚ฐ
  5. ๊ฐ€์„ค ํŒ์ •
    • ํ™•๋ฅ ์ด ๊ธฐ์ค€๋ณด๋‹ค ์ž‘๋‹ค = reject h0 –> ๊ท€๋ฌด๊ฐ€์„ค์€ ์˜ณ์ง€ ์•Š๋‹ค –> ๋Œ€๋ฆฝ๊ฐ€์„ค์ด ์˜ณ๋‹ค
    • ํ™•๋ฅ ์ด ๊ธฐ์ค€๋ณด๋‹ค ํฌ๋‹ค = accept h0 –> ๊ท€๋ฌด๊ฐ€์„ค์ด ์˜ณ์„์ง€๋„ ๋ชจ๋ฆ„(ํ”ํžˆ ์žˆ๋Š” ์ผ์ด ์ผ์–ด๋‚จ)


Significance level(์œ ์˜์ˆ˜์ค€) ฮฑ

๊ท€๋ฌด๊ฐ€์„ค์„ ๊ธฐ๊ฐํ•  ํŒ๋‹จ์˜ ๊ธฐ์ค€์ด ๋˜๋Š” ์ •ํ™•๋„.
๋ณดํ†ต ๋ถ„ํฌ์˜ ์–‘์ธก์—์„œ 5%(ฮฑ = 0.05)


p-value

๊ท€๋ฌด๊ฐ€์„ค ๋ถ„ํฌ์—์„œ ๊ฒ€์ •ํ†ต๊ณ„๋Ÿ‰๋ณด๋‹ค ๊ทน๋‹จ์ ์ธ ๊ฐ’์ด ๊ด€์ธก๋  ํ™•๋ฅ  = ๊ท€๋ฌด๊ฐ€์„ค์„ ๊ธฐ๊ฐํ•  ์ˆ˜ ์žˆ๋Š” ๊ฐ€์žฅ ๋‚ฎ์€ ์œ ์˜์ˆ˜์ค€. ์ž‘์„์ˆ˜๋ก ๋ฐ”๋žŒ์ง.


Type 1 error (1์ข… ์˜ค๋ฅ˜)

Reject h0 when h0 is actually true.
e.g
h0: I don’t have covid virus.
Test result shows that I have covid virus but I don’t have covid virus in reality.


Type 2 error (2์ข… ์˜ค๋ฅ˜)

Accept h0 when h0 is actually false.
e.g
h0: I don’t have covid virus.
Test result says that I don’t have covid virus but I actually have.


Type 1 vs. Type 2 error
When null hypothesis is True False
Rejected Type 1 error
False positive
(p = ฮฑ)
Correct decision
True positive
(p = 1 - ฮฒ)
Not rejected Correct decision
True negative
(p = 1 - ฮฑ)
Type 2 error
False negative
(p = ฮฒ)


โ˜… ๋” ๋งŽ์€ ๋‚ด์šฉ ๋ณด๋Ÿฌ๊ฐ€๊ธฐ click click! โ˜